Predictability of radon airborne measurements based on surrogate measures

Transcription

1 University of Iowa Iowa Research Online Theses and Dissertations 2012 Predictability of radon airborne measurements based on surrogate measures Nirmalla Barros University of Iowa Copyright 2012 Nirmalla G. Barros This dissertation is available at Iowa Research Online: Recommended Citation Barros, Nirmalla. "Predictability of radon airborne measurements based on surrogate measures." PhD (Doctor of Philosophy) thesis, University of Iowa, Follow this and additional works at: Part of the Occupational Health and Industrial Hygiene Commons

2 PREDICTABILITY OF RADON AIRBORNE MEASUREMENTS BASED ON SURROGATE MEASURES by Nirmalla Barros An Abstract Of a thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Occupational and Environmental Health in the Graduate College of The University of Iowa May 2012 Thesis Supervisor: Professor R. William Field

3 1 ABSTRACT This research focuses on the evaluation of temporal and spatial variability associated with radon airborne measurements and the predictive utility of surrogate measures to estimate radon concentrations within the same environment as well as in other environments. This dissertation consists of three components. In Chapter II, Evaluation of agreement of time-integrated basement residential radon measurements and correctness of further radon testing indicators, we investigated the temporal variability between short-term and annual residential radon measurements collected on the lowest livable level and identified housing/occupant factors that influenced each measurement as well as their differences. The false negative rate of how often the short-term test incorrectly indicated that further radon testing was unnecessary was 12 percent at the action level of 148 Bq m -3, but dropped to two percent at a 74 Bq m -3 reference level. The foundation wall material of the basement was the only significant factor to have an impact on the absolute difference between both measurements. This study has the potential to significantly influence public health policy concerning radon testing protocols, specifically the need to re-assess the EPA s current radon mitigation guidance level of 148 Bq m -3. In Chapter III, Temporal and spatial variation associated with residential airborne radon measurements, we investigated the temporal and spatial variability between basement winter short-term and annual radon measurements performed in upper floors of the home and identified housing/occupant factors that influenced each measurement as well as their differences. This study found that individuals would be falsely overestimating their potential exposure to radon half the time at the EPA s action level of 148 Bq m -3 based on basement short-term tests and much more frequently (80 percent of the time) at a lower reference level of 74 Bq m -3. The

4 2 presence of a sump was the only factor that was significantly associated with the absolute difference between both these measurements. This study has the potential to influence public health policy in regard to exposure surrogate measures, specifically to encourage testing of radon in living areas of the home and not relying solely on a screening measurement to estimate the concentration of radon in the entire home. In Chapter IV, Comparative survey of outdoor, residential, and workplace radon concentrations, we investigated occupational radon concentrations in above ground workplaces in Missouri and compared them to above ground radon concentrations in nearby homes and outdoor locations and evaluated the utility of above ground annual home and outdoor concentrations to predict above ground radon concentrations at a nearby workplace. Employees at county agencies, schools, and businesses were recruited to participate in the study. Annual above ground workplace radon concentrations were found to be similar to annual radon concentrations in the upper floor of homes. Annual non-basement first floor home and outdoor radon concentrations were poor predictors of annual above ground radon concentrations at a nearby workplace. This study provides insights into the potential for above-ground radon exposures in the workplace and the potential agreement between workplace and residential radon concentrations. Abstract Approved: Thesis Supervisor Title and Department Date

5 PREDICTABILITY OF RADON AIRBORNE MEASUREMENTS BASED ON SURROGATE MEASURES by Nirmalla Barros A thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Occupational and Environmental Health in the Graduate College of The University of Iowa May 2012 Thesis Supervisor: Professor R. William Field

6 Copyright by NIRMALLA BARROS 2012 All Rights Reserved

7 Graduate College The University of Iowa Iowa City, Iowa CERTIFICATE OF APPROVAL PH.D. THESIS This is to certify that the Ph.D. thesis of Nirmalla Barros has been approved by the Examining Committee for the thesis requirement for the Doctor of Philosophy degree in Occupational and Environmental Health at the May 2012 graduation. Thesis Committee: R. William Field, Thesis Supervisor Renee Anthony David Osterberg Mary Kathryn Cowles Lucie Laurian Daniel Steck

8 ACKNOWLEDGMENTS I am grateful to many people who supported me to complete this project. Thank you to my committee members for their collaboration and guidance. I am also appreciative of assistance from Dan Olson of the Iowa Cancer Registry, Dan Field, Phil Jalbert, Paul Kotrappa, Mark Salasky, and Michael Kitto. I am grateful notably to Paul Skinner for his endless assistance with geocoding and numerous other aspects of this project. I am especially thankful to my advisor, Dr. Bill Field, for his expert advice, inspiration, and exceptional support. Finally, I am grateful to my family for their constant encouragement and endless reassurance. ii

9 ABSTRACT This research focuses on the evaluation of temporal and spatial variability associated with radon airborne measurements and the predictive utility of surrogate measures to estimate radon concentrations within the same environment as well as in other environments. This dissertation consists of three components. In Chapter II, Evaluation of agreement of time-integrated basement residential radon measurements and correctness of further radon testing indicators, we investigated the temporal variability between short-term and annual residential radon measurements collected on the lowest livable level and identified housing/occupant factors that influenced each measurement as well as their differences. The false negative rate of how often the short-term test incorrectly indicated that further radon testing was unnecessary was 12 percent at the action level of 148 Bq m -3, but dropped to two percent at a 74 Bq m -3 reference level. The foundation wall material of the basement was the only significant factor to have an impact on the absolute difference between both measurements. This study has the potential to significantly influence public health policy concerning radon testing protocols, specifically the need to re-assess the EPA s current radon mitigation guidance level of 148 Bq m -3. In Chapter III, Temporal and spatial variation associated with residential airborne radon measurements, we investigated the temporal and spatial variability between basement winter short-term and annual radon measurements performed in upper floors of the home and identified housing/occupant factors that influenced each measurement as well as their differences. This study found that individuals would be falsely overestimating their potential exposure to radon half the time at the EPA s action level of 148 Bq m -3 based on basement short-term tests and much more frequently (80 percent of the time) at a lower reference level of 74 Bq m -3. The iii

10 presence of a sump was the only factor that was significantly associated with the absolute difference between both these measurements. This study has the potential to influence public health policy in regard to exposure surrogate measures, specifically to encourage testing of radon in living areas of the home and not relying solely on a screening measurement to estimate the concentration of radon in the entire home. In Chapter IV, Comparative survey of outdoor, residential, and workplace radon concentrations, we investigated occupational radon concentrations in above ground workplaces in Missouri and compared them to above ground radon concentrations in nearby homes and outdoor locations and evaluated the utility of above ground annual home and outdoor concentrations to predict above ground radon concentrations at a nearby workplace. Employees at county agencies, schools, and businesses were recruited to participate in the study. Annual above ground workplace radon concentrations were found to be similar to annual radon concentrations in the upper floor of homes. Annual non-basement first floor home and outdoor radon concentrations were poor predictors of annual above ground radon concentrations at a nearby workplace. This study provides insights into the potential for above-ground radon exposures in the workplace and the potential agreement between workplace and residential radon concentrations. iv

11 TABLE OF CONTENTS LIST OF TABLES... vi LIST OF FIGURES... viii CHAPTER I. INTRODUCTION AND LITERATURE REVIEW...1 History...2 Radioactive decay...3 Radioactivity units...5 Health effects of radon...5 Exposure guidelines...7 Sources of radon exposure...8 Measurement of radon...11 Short- versus long-term radon measurements...11 Radon detectors...12 Testing conditions...15 Relationship between short- and long-term radon measurements collected at the same location...16 Ability of short-term radon measurements to predict annual radon concentrations at the same location and factors that affect their variability...18 Relationship between short- and long-term radon measurements collected on different floor levels...20 Ability of short-term radon measurements to predict annual average radon concentrations on different floor levels and factors that affect their variability...21 Exposure to radon in other settings...23 Comparison of home and workplace radon concentrations...25 Relationship between indoor and outdoor radon concentrations...27 Aims of the dissertation...27 II. EVALUATION OF AGREEMENT OF TIME-INTEGRATED BASEMENT RESIDENTIAL RADON MEASUREMENTS AND CORRECTNESS OF FURTHER RADON TESTING INDICATORS...30 Abstract...30 Introduction...31 Methods...33 Sample selection...33 Survey questionnaire...34 Radon measurements...34 Geocoding of home addresses...36 Data analysis...36 Quality assurance...43 Results...43 Sample characteristics...44 Agreement between measurements...47 Simple linear regression...51 v

12 Backward stepwise regression...52 Quality assurance...61 Discussion...62 Limitations...66 Strengths...67 Conclusions...67 III. TEMPORAL AND SPATIAL VARIATION ASSOCIATED WITH RESIDENTIAL AIRBORNE RADON MEASUREMENTS...69 Abstract...69 Introduction...70 Methods...71 Sample selection...71 Radon measurements...72 Data analysis...72 Results...79 Sample characteristics...80 Agreement between measurements...82 Simple linear regression...86 Backward stepwise regression...88 Discussion...87 Limitations Strengths Conclusions IV. COMPARATIVE SURVEY OF OUTDOOR, RESIDENTIAL, AND WORKPLACE RADON CONCENTRATIONS Abstract Introduction Methods Sample selection Radon measurements Survey questionnaire Sampling handling Geocoding of home and workplace addresses Data analysis Quality assurance Results Sample characteritistics Agreement between measurements Simple linear regression Quality assurance Discussion Limitations Strengths Conclusions V. CONCLUSIONS vi

13 APPENDIX A. HUMAN SUBJECTS OFFICE/INSTITUTIONAL REVIEW BOARD APPROVAL FOR DISSERTATION STUDIES APPENDIX B. IOWA RADON LUNG CANCER STUDY (IRLCS) QUESTIONNAIRE ABOUT THE HOME (CHAPTERS II AND III) APPENDIX C. IRLCS QUESTIONNAIRE ABOUT THE PARTICIPANT (CHAPTERS II AND III) APPENDIX D. IRLCS PLACEMENT OF ALPHA TRACK RADON DETECTORS (CHAPTERS II AND III) APPENDIX E. IRLCS MOBILITY FLOW CHART (CHAPTERS II AND III) APPENDIX F. IRLCS SHORT-TERM E-PERM RADON MEASUREMENT FORM (CHAPTERS II AND III) APPENDIX G. ADDITIONAL METHODS AND RESULTS FOR CHAPTER II APPENDIX H. SOURCES OF ERROR FOR RADON DETECTORS APPENDIX I. ADDITIONAL RESULTS FOR CHAPTER III APPENDIX J. MISSOURI INDOOR RADON DETECTOR PLACEMENT (CHAPTER IV) APPENDIX K. MISSOURI RADON SURVEY QUESTIONNAIRE (CHAPTER IV) APPENDIX L. MISSOURI OUTDOOR RADON DETECTOR PLACEMENT (CHAPTER IV) APPENDIX M. ADDITIONAL METHODS AND RESULTS FOR CHAPTER IV REFERENCES vii

14 LIST OF TABLES Table 1. Radon gas measurement devices and their characteristics Interpretation of short-term and annual radon measurements that were equal to or above or less than the respective reference level Characteristics of basement winter short-term and basement annual radon measurements (Bq m -3 ) as well as their differences from 158 residences in Iowa Number of basement short-term and basement annual radon measurements that were equal to or above or less than the respective reference level Summary of regression predictor parameters from 158 homes in Iowa Backward regression parameters (final model) for predicting ln(basement annual radon concentration) Backward regression parameters (final model) for predicting ln(basement annual radon concentration given ln(basement short-term radon concentration) and housing/occupant factors Backward regression parameters (final model) for predicting ln(basement short-term radon concentration) Backward regression parameters (final model) for predicting the absolute difference, ln( basement short-term Rn-radon conc. basement annual radon conc. ) Backward regression parameters (final model) for predicting the signed difference, ln(basement short-term radon conc.) ln(basement annual radon conc.) Characteristics of basement short-term and upper floor radon measurements (Bq m -3 ) as well as their paired differences from 158 residences in Iowa Number of basement short-term and first floor annual radon measurements that were equal to or above or less than the respective reference level Summary of additional regression predictor parameters from 158 homes in Iowa Backward regression parameters (final model) for predicting ln(annual first floor radon concentration) Backward regression parameters (final model) for predicting ln(first floor annual radon conc.) given ln(basement short-term radon concentration) and housing/occupant factors...93 viii

15 16. Backward regression parameters (final model) for predicting ln(basement winter short-term Rn-222 conc.) Backward regression parameters (final model) for predicting the absolute difference, ln( basement short-term Rn-222 conc. annual first floor Rn-222 conc. ) Backward regression parameters (final model) for predicting the signed difference, ln(basement short-term radon conc.) ln(first floor annual radon conc.) Summary of annual radon detector placement counts by type of environment and date Characteristics of radon concentrations (Bq m -3 ) by sampling environment ix

16 LIST OF FIGURES Figure 1. Radon decay chain Percent of total effective dose for average individual in U.S. population from various radiation sources (percent values rounded to nearest 1%, except for those <1%) Photo of alpha track detectors and alpha tracks etched on a plastic chip as viewed through a light microscope Open and closed E-PERM chambers Map of basement winter short-term and basement annual radon concentrations with respect to EPA s action level of 148 Bq m -3 in Iowa Box plot of radon concentration by type of measurement Plots of agreement of short-term and annual radon concentrations versus their ratio Percentages of diagnostic indicators comparing basement short-term radon tests to basement annual radon tests by reference level Scatter plot of ln(basement winter short-term Rn-222 conc.) versus ln(basement annual Rn-222 conc.) and its residuals Box plot of Rn-222 concentration by type of measurement Percentages of diagnostic indicators comparing basement short-term radon tests to first floor annual radon tests by reference level Scatter plots of working level months versus basement and upper level radon measurements Scatter plots of working level months versus basement short-term, basement annual, first floor annual, and annual average bed/living radon concentrations after removing a basement annual radon concentration outlier Scatter plot of ln(basement winter short-term Rn-222 conc.) versus ln(first floor annual Rn-222 conc.) and its residuals Annual radon concentrations (Bq m -3 ) by type of test environment across counties in Missouri Box plots of radon concentration by type of sampling environment Box plots of radon concentration by type of workplace x

17 18. Range of differences between annual home, outdoor, and work radon concentrations (Bq m -3 ) by increasing work concentrations Scatter plot of annual ln(home Rn-222 conc.) versus annual ln(work Rn-222 conc.) and its residuals Scatter plot of annual ln(outdoor Rn-222 conc.) versus annual ln(work Rn- 222 conc.) and its residuals G-1. Histograms of radon concentration by type of measurement and their paired difference G-2. Histogram and quantile-quantile plot of the paired difference after natural log (ln) transformation G-3. Histograms of residuals for model predicting basement annual radon concentrations based on basement short-term radon concentrations G-4. Quantile-Quantile plots of residuals for model predicting basement annual radon concentrations based on basement short-term radon concentrations G-5. Diagnostics for regression of ln(basement winter short-term radon conc.) on ln(basement annual radon conc.) I-1. Histograms of radon concentration by type of measurement and their paired differences I-2. Histograms and quantile-quantile plots of paired differences after a logarithmic (ln) data transformation I-3. Histograms of residuals for models predicting first floor and annual average of bed/living room annual radon concentrations based on basement short-term radon concentrations I-4. Quantile-Quantile plots of residuals for models predicting first floor and annual average of bed/living room radon concentrations based on basement short-term radon concentrations M-1. Histograms of radon concentration by sampling environment and their paired differences M-2. Histograms and quantile-quantile plots of the paired differences after lntransformations M-3. Scatter plots of home versus workplace radon concentrations M-4. Scatter plots of outdoor versus workplace radon concentrations M-5. Histograms of residuals for model predicting workplace radon concentrations based on nearby home radon concentrations M-6. Histograms of residuals for prediction of work radon concentrations based on nearby outdoor radon concentrations xi

18 M-7. Quantile-Quantile plots of residuals for prediction of work radon concentration based on a nearby home radon concentration M-8. Quantile-Quantile plots of residuals for prediction of work radon concentration based on nearby outdoor radon concentration M-9. Diagnostics for regression of ln(home) on ln(work) M-10 Diagnostics for regression of ln(outdoor) on ln(work) xii

19 1 CHAPTER I INTRODUCTION AND LITERATURE REVIEW Radon (Rn-222) is a radioactive gas that is formed from the breakdown of Uranium-238 and subsequently Radium-226. From this decay, radon gas is released into the atmosphere and can move through the earth where it easily dilutes as it mixes with outdoor air. Radon is found naturally in soil, rock, and groundwater supplies. Radon is ubiquitous; it is present everywhere outdoors and can reach high concentrations in enclosed areas such as homes and workplaces. It is the predominant source of natural radiation, accounting for about 37 percent of the overall radiation exposure in the United States (1). Radon is the leading environmental cause of cancer mortality in the United States (2,3). Based on the analysis by the National Academy of Sciences on the Biological Effects of Ionizing Radiation VI Report, the U.S. Environmental Protection Agency (EPA) estimates that out of a total of 157,400 lung cancer deaths nationally in 1995, 21,100 (13.4%) were radon related making radon the second leading cause of lung cancer after only tobacco smoke and the primary cause of lung cancer among individuals who have never smoked (4). Smokers exposed to radon are at a greater risk of developing lung cancer than are those who never smoked but with the same level of exposure. This greater risk is due to the synergistic effect (i.e., sub-multiplicative effect) between tobacco and radon (4).

20 2 History Radon is a noble gas, and, like other noble gases radon is colorless, odorless, and tasteless. In 1898, Marie Curie named a radioactive element that she isolated radium. A year later, Ernest Rutherford identified the types of radiation that are emitted from uranium through radioactive decay. The findings from Curie s and Rutherford s experiments were preceded by Henri Becquerel s work with uranium (5). Radium emanation (radon) was discovered by the German physicist Friedrich Dorn in 1900 (6). In the 1500s, the high mortality of pulmonary disease among underground metal miners was reported by Agricola at Schneeberg in the Erz Mountains of Central Europe. Härting and Hesse made the association of high mortality rates of pulmonary disease in miners with lung cancer several hundred years later in 1879 (7). Although Dorn is credited with the discovery of radium emanations in 1900, which would later be called radon in 1923, it was not until H.E. Müeller, in 1913, and Margaret Uhlig, in 1921, who had linked lung cancer occurrence with radium emanations. By the 1950s, the decay products of radon, rather than radon gas itself, were identified to be the primary cause of lung cancer. By the 1970s, risk estimates were being calculated for lung cancer in radonexposed miners. The American press brought attention about the potential of elevated residential radon exposure in December 1984 after a construction engineer named Stanley Watras set off portal radiation monitors at the newly constructed Limerick Nuclear Power Plant in Pennsylvania (8). After the source of his radiation exposure was not traced to the plant, but rather his home, a team of scientists tested his home for radiation and discovered radon concentrations as high as 99,900 becquerels per cubic meter of air (2,700

21 3 picocuries per liter of air). The source of the radiation that was causing the alarms to go off at the plant was traced to the radon decay products that attached to his clothing before he left home. Because of investigations and studies involving the potential of elevated residential radon concentrations, the EPA implemented a National Radon Program in Radioactive decay With over 30 known isotopes of radon, only three are found naturally (9). The three isotopes are Rn-219 (actinon), Rn-220 (thoron), and Rn-222 (radon). Radon has the longest half-life of 3.8 days compared to the other two, with half-lives of less than a minute. Radon-222 is formed from the radioactive decay of uranium s most plentiful isotope, Uranium-238, which undergoes a series of radioactive decays generating radium (Ra-226), which in turn decays to radon (Rn-222). The subsequent decay of radon bring about the decay of polonium (Po-218, Po-214, and Po-210), lead (Pb-214 and Pb-210), and bismuth (Bi-214 and Bi-210) isotopes, ending with a stable non-radioactive lead atom (Pb-206) (Figure 1). Radon decays to both short- and long-lived radioactive decay products having half-lives, for instance, as short as about 200 microseconds (Po-214) to as long as 22 years (Pb-210).

22 4 Figure 1. Radon decay chain (Adapted from Barros N, Field RW (2011) Radon, The Praeger Handbook of Environmental Health, Editor Robert H. Friis, ABC-CLIO, Inc., Santa Barbara, CA (bookchapter)) From the decay of radon and its decay products, ionizing radiation is emitted. This type of radiation releases energy in the form of alpha, beta, and gamma radiation. Alpha radiation emits energy as particles, containing two protons and two neutrons, which is equivalent to a helium nucleus. Compared to other types of radiation, alpha radiation has the heaviest mass and delivers the greatest radiation dose to the lungs from radon s decay products. Beta radiation also releases energy in the form of particles. But, unlike alpha radiation, beta particles have very little mass allowing them to move faster

23 5 and penetrate deeper into human tissue. Gamma radiation, in contrast, has no mass, emits energy as a ray, and moves at the speed of light. Alpha particles are considered a type of high linear energy transfer (LET) radiation. High LET radiation imparts a large amount of energy to biological tissues, often resulting in dense ionization, over its relatively short path. Beta and gamma radiation are considered low LET radiation since they transfer less energy per unit path length as compared to high LET radiation (4). Radioactivity units The radioactivity of a material is measured by the rate in which it decays. The International System of Units (SI) adopted the becquerel (Bq) in 1960 as a standard unit to measure radioactivity, which replaced the Curie (Ci), an older non-si unit of measurement (10). One becquerel is equivalent to one disintegration per second. A Curie equals 3.7 x becquerels or 3.7 x disintegrations per second, i.e., the disintegration rate of one gram of radium (Ra-226) (11). Airborne measurements of radon are often expressed in picocuries per liter of air (pci L -1 ) in the United States, but in becquerels per cubic meter of air (Bq m -3 ) elsewhere. One pci L -1 is equivalent to 37 Bq m -3. Health effects of radon From the decay of radon gas, a series of solid decay products are produced. Some of the decay products attach to aerosols in the atmosphere, some remain attached, and some of the attached particles deposit on room surfaces. The attachment rate of the decay products to aerosols is based on the size and quantity of aerosols in the air as well as air circulation in the room (4). The deposit of these attached particles in the lungs depends on a number of factors including aerosol size, breathing frequency, nasal versus oral breathing, lung tidal volume, and lung architecture.

24 6 Inhalation of two of the short-lived decay products, polonium-218 and polonium- 214, deliver the majority of the radiation dose by alpha decay to the respiratory epithelium (Figure 1). Alpha particles cannot penetrate very far into the lung tissue, but impart their energy on tissue where they deposit, therefore, providing a more localized dose than other types of radiation. This irradiation occurring in the tissue is capable of causing disruptions to DNA including gene mutations, chromosomal damage, induction of micronuclei, and sister chromatid exchanges with the subsequent possibility of a gene mutation leading to lung cancer. The primary adverse effects of radon, therefore, originate from its decay products, rather than from radon gas itself. Given that a single alpha particle can cause significant biological damage to a cell and because cancer is considered monoclonal in origin (i.e., cancer can derive from a single cell that has undergone malignant transformation), a safe threshold for radon-induced lung cancer is unlikely to exist. The inhalation of the short-lived particles, especially the unattached ultrafine particles (around one nanometer), contributes the majority of the dose to the lung because they deposit efficiently (4). Once deposited in the lung, alpha particles cannot penetrate very far into the lung tissue, but impart their energy on tissue where they deposit, therefore, providing a more localized dose than other types of radiation. This irradiation occurring in the tissue is capable of causing disruptions to DNA including gene mutations, chromosomal damage, induction of micronuclei, and sister chromatid exchanges. Lung cancer is the most established adverse health outcome documented in the literature as being positively associated with protracted exposure to radon and its decay products. The World Health Organization s International Agency for Research on Cancer (IARC) has classified radon and its decay products a Group 1 carcinogen since 1988 (12). Group 1 is the highest ranking risk potential for a carcinogen (e.g., tobacco is also classified as a Group 1 carcinogen). This classification is met when there is sufficient

25 7 evidence of carcinogenicity in humans as documented from studies in experimental animals and humans exposed to the agent. Lung cancer, compared to other cancers, is the cause of the most deaths in the U.S. in both males and females with tobacco smoking as the main risk factor (13). Radon is also the seventh cause of cancer mortality exceeding the number of deaths for many other types of cancer (e.g., liver, ovarian, kidney, melanoma) from all causes. Pooled studies of radon-exposed underground uranium and hard rock miners (4) as well as pooled North American and European residential radon studies (14,15) provides strong evidence of a linear dose-response relationship between protracted radon exposure and lung cancer No evidence of a threshold was detected in these pooled studies. The Biological Effects of Ionizing Radiation VI (BEIR VI) report, released by the National Academy of Sciences, explored other adverse health effects from radon exposure, besides lung cancer, using the miner data, but no significant dose-response related increases in mortality were found (4). This finding may be due to adverse health outcomes not being observable in mortality studies as cancer and other chronic conditions are often not reported on death certificates. A case-cohort cancer incidence study by Řeřicha et al. (16) detected a positive association between protracted radon exposure and leukemia, including chronic lymphocytic leukemia in their case-cohort cancer incidence study. In addition, a small statistically non-significant increased risk for residential radon exposure and both chronic lymphocytic leukemia and chronic myelocytic leukemia were observed in a Bayesian study using a hierarchical risk model and data from the Iowa Cancer Registry (17). Exposure guidelines Currently, the EPA advises building occupants to mitigate radon concentrations in the home at 150 Bq m -3 (~4 pci L -1 ) or larger (18). This radon action level was established in 1986 (19) and is not a health-based guideline. In fact, the EPA notes that exposure below this level still poses a risk. The World Health

26 8 Organization, on the other hand, set a reference level of 100 Bq m -3 (2.7 pci L -1 ) in 2009 to help minimize the health hazard due to radon exposure in the home (20). In the workplace, the U.S. Occupational Safety and Health Administration (OSHA) enforces a radon exposure limit of 100 pci L -1 (3,700 Bq m -3 ) over 40 hours in a consecutive sevenday workweek (21). The OSHA guidelines for exposure to ionizing radiation in general industry (29 CFR ) refers to the Nuclear Regulatory Commission s regulations for exposure limits to radon (Table 1 of Appendix B, 10 CFR part 20) (22). Sources of radon exposure Due to the extremely long half-life (4.5 billion years) of Uranium-238, radon has long been present in the earth in soils, rocks, and water. From the decay of Uranium-238, and its long-lived decay products (i.e., Uranium-234, Thorium-230, Radium-226), radon gas is released and can move through the earth into the atmosphere where it easily dilutes with outdoor air. However, because most homes are not built radon resistant, the gas moves up through the soil and can enter a building where radon can accumulate under the usual ventilation conditions maintained by the occupant. From this buildup, radon can reach elevated concentrations and inhaling its decay products over an extended period can present a significant health risk. In most regions of the United States, the greatest source of indoor radon concentration is from the soil and rocks below the foundation of a building. Radon contributes 37 percent of the average person s overall radiation dose in the United States (Figure 2). While the contribution of radon to one s overall radiation exposure has decreased over the past 30 years, it is not because individuals are being exposed less to

27 9 radon, but because an increasing contribution of the total radiation exposure is attributed to a larger exposure from computed topography (CT) scans and nuclear medicine procedures. Figure 2. Percent of total effective dose for average individual in U.S. population from various radiation sources (percent values rounded to nearest 1%, except for those <1%) (Reprinted with permission of the National Council on Radiation Protection and Measurements, 2009, NCRP Report No. 160 Section 1) Both indoor and outdoor radon concentrations vary across geographical regions in the U.S. based on the radium content in the soil and underlying geology. Studies conducted by the EPA have found that Iowa has both the highest mean residential radon

28 10 concentrations (geometric mean of 241 Bq m -3 compared to the next highest geometric mean of 202 Bq m -3 for North Dakota) and the greatest percentage (71% versus the next highest percentage of 61% for North Dakota) of screening radon measurements above the EPA s action level of 150 Bq m -3 (~4 pci L -1 ) compared to any other state surveyed in the U.S. (23). According to the U.S. Geological Survey, the elevated radon potential in the upper Midwest compared to other regions can be explained by continental glacial deposits (24). The glacial deposits in the upper Midwest are formed from the breakdown of granitic materials (i.e., large granitic rocks) into smaller and smaller pieces, which increases their surface area and, in turn, their radon emanating potential. In Iowa, glaciers produced this high level of radon potential by effectively transporting glacial materials from the Canadian Shield to Iowa. The soils formed from glacial deposits are distinct by the type of bedrock radon sources common in most other areas of the U.S. Studies for this dissertation take place in Iowa and an adjacent state, Missouri. According to a geologic map developed by the U.S. Geological Survey and the EPA to assess indoor radon potential in counties within the U.S., Iowa is located in a zone with the highest radon potential where the predicted average screening radon level is larger than 148 Bq m -3, whereas Missouri is situated in a zone where the predicted average screening radon level is between 74 (2 pci L -1 ) and 148 Bq m -3 (25). High radon concentrations, however, can be observed in any region of the country, even in parts found to have low radon potential. The only way to know if a building has an elevated radon concentration is to test it. Radon concentrations within the home can exhibit significant temporal and spatial variability (26,27). Meteorological conditions, influencing wind speed and direction,

29 11 atmospheric pressure, and temperature can affect the driving forces of radon transport. Indoor radon concentrations are also affected by the characteristics of the home (e.g., foundation wall material, presence of a crawl space, presence of a sump pump) (28) and daily activities of the occupant (e.g., air exchange rates) (27). Radon can pass more easily through a block concrete foundation wall via joints between the blocks or holes/cracks in the block compared to a poured concrete wall. Another housing factor, a crawl space, is a site for the entry of radon into a home from the soil beneath it. A sump pump can also be a source of radon entry in homes because it breaks the barrier between the soil below the foundation of a home and the basement. In addition, air movement related to the presence of a forced air heating system or the use of a central air conditioner can move radon away from its entry area in the lowest level to upper floors within a residence. Measurement of radon Short- versus long-term radon measurements The measurement of radon is frequently obtained using short-term measurements in the home. A short-term radon test is considered a screening test as they are thought to indicate the potential for a home to have elevated radon concentrations. The measurement period ranges from two to 90 days depending on detector type. Most commonly, the screening testing period lasts for two to five days in the United States because of the need for rapid testing during real estate transactions. Long-term testing, however, lasts more than 90 days with an optimal measurement period of one year. Long-term tests are advised to determine the yearly radon concentration as compared to short-term monitoring because the former covers several months and accounts for changes in meteorological conditions and occupant behavior (e.g., window opening, air conditioner usage, etc.). Long-term measurements

30 12 allow radon concentrations to be averaged over months or years rather than days or weeks. Lung cancer is the most established adverse health outcome associated with protracted exposure to radon and its decay products, so a long-term radon measurement would be more suitable for a radon exposure assessment, as it is an indicator of long-term exposure versus a short-term test. However, the EPA recommends that if a short-term radon test is above the action level, this test can be followed by a second short-term test to make decisions on further radon testing (18). Given the wide use of short-term tests to assess exposure to radon and their influence on motivating individuals to take further action as needed, it is important to determine how well these measurements compare to the long-term radon measurements, evaluate their predictive value, and determine which factors affect their variability. Radon detectors Radon measurement devices can be classified in several ways: 1) by measurement of radon or radon decay products, 2) as passive (i.e., no pump or electricity required) or active, 3) by duration of measurement, or 4) by type of detector. Devices measuring radon decay products are primarily used by radon professionals and researchers. Overall, radon gas concentrations are considered a good indicator of radon decay product concentrations. Some of the more frequently used radon measurement devices can be found in Table 1. Table 1. Radon gas measurement devices and their characteristics TYPE OF DEVICE COST TYPICAL MEASUREMENT PERIOD PASSIVE / ACTIVE

31 13 Activated Charcoal Detector Low 2 days 1 week Passive Alpha Track Detector Low 1-12 months Passive Electret Ion Chamber Medium 2 days - 1 year Passive Electronic Integrating Device Medium 2 days year(s) Active Continuous Radon Monitor High 2 hours year(s) Active (Adapted from Barros N, Field RW (2011). Radon, The Praeger Handbook of Environmental Health, Editor Robert H. Friis, ABC-CLIO, Inc., Santa Barbara (bookchapter)) Alpha track detectors (Figure 3) are commonly deployed for long-term radon measurements. These devices contain a piece of CR-39 plastic that detects the alpha decay of radon. After the device is opened and exposed to air, radon gas penetrates the detector through the filter found in either the top or side of the detector depending on vendor. Once the radon enters the detector, the gas radioactively decays where some of the emitted alpha particles strike the plastic inside the detector forming small pits (Figure 3). These pits can be counted at 100x using a light microscope to determine the number of tracks per unit of area once the chemical etching of the plastic is applied. The laboratory that provides the detectors calibrates them so the radon concentration can be computed based on the duration of exposure and number of tracks. A major advantage of this detector is that it can be placed in the home for as long as a year to obtain an average integrated year-long radon concentration. Another advantage of this detector is that humidity, temperature, and background gamma radiation do not influence does not influence its performance (20).

32 14 Figure 3. Photo of alpha track detectors and alpha tracks etched on a plastic chip as viewed through a light microscope (Original photos) Other types of radon measurement devices, used by professional radon testers and researchers, include the electret ion chamber (sometimes called an E-PERM), continuous radon monitor, and radon progeny detector. The E-PERM (Figure 4) is the only passive device that can be used for both short- and long-term measurements. It contains a positively charged electret and an ionization chamber made of electrically conductive plastic. The electret serves as a source of an electric field and a sensor in the ion chamber. Radon penetrates the filter, which excludes the entry of radon s decay products, by passive diffusion to enter the ion chamber. Radiation produced from the decay of radon and its decay products ionizes the air within the chamber volume. The positively charged electret collects negative ions, which has the effect of reducing the charge, thus causing the electret s voltage to change. The discharge of the electret over a known time interval

33 15 is a measure of time-integrated ionization during this interval and can be measured using an electret reader before and after the exposure sampling period. Figure 4. Open and closed E-PERM chambers (Reprinted with permission of Wolters Kluwer Health from Kotrappa P, Dempsey JC, Ramsey RW, et al. A practical E-PERM (electret passive environmental radon monitor) system for indoor 222Rn measurement. Health Physics 1990;58: ) Testing conditions The EPA recommends that either short- or long-term detectors be placed in an occupied area of the lowest lived-in level of the home. The detector should be placed at least 4 inches from other objects, 3 feet away from windows or doors, and 20 inches above the floor. The EPA also advises that short-term measurements be taken in the winter under closed house conditions prior to and during obtaining a shortterm test (18). For closed house conditions, the windows and outside doors should be kept shut as much as possible. Closed house conditions are suggested so that the measured radon concentrations will in most cases represent a worst case scenario (i.e., often obtain larger short-term radon measurements compared to long-term

34 16 measurements). Short-term measurements, obtained under the testing conditions above, provide an indicator of the potential for a home to have elevated radon concentrations. Long-term measurements, on the other hand, are obtained under the usual ventilation conditions maintained by the occupant(s) of the home (19). Relationship between short- and long-term radon measurements collected at the same location Studies comparing short- and long-term measurements of radon have shown contrasting results. Vaupotič and Kobal (29) measured radon in 890 schools in Slovenia using alpha track detectors exposed during two winter months and a continuous radon monitor for 7 to 10 days during this two-month measurement period with both devices placed in the same room. Even though one continuous monitor was deployed, various types of monitors were used (Alpha-Guard (Genitron, Germany); System-30 (Scintrex, Canada); EQF-3010, EQF-3020, and EQF (Sarad, Germany). The rooms were closed nightly during the testing. Comparisons between both types of measurements resulted in only about 33 percent of the measurements to be in agreement within 20 percent. A study conducted in the Spanish region of Galicia by Ruano-Ravina and colleagues (30) tested for radon in 391 residences. Charcoal detectors were used for the short-term measurement period lasting a total of three to four days that were placed throughout the year with windows kept closed the day before sampling until the end of the test. Alpha track detectors were deployed for a five-month test period beginning at the same time the charcoal detectors were placed in the homes. Measurement devices were positioned one meter from the other detector in rooms of high occupancy (e.g., bedroom). The correlation between the two types of measurements was 0.61 (p< 0.001). The strongest correlations were observed in several categories including: dwellings over 149 years old (p=0.829, p=0.021) and 50 to 99 years old (p=0.808, p<0.001), when windows

35 17 were not closed (p=0.799, p<0.001), and in coastal areas (p=0.699, p<0.001). The median long-term readings were larger than the median short-term measurements regardless of month of placement. The greatest differences between both types of measurements occurred in the summer months. Wysocka et al. (31) performed a similar survey in residences in Poland, but with alpha track detectors exposed for one year. An opposing trend, however, was observed. The short-term monitors were deployed in the months of January and February for four days in more than 300 residences and found to overestimate the annual average measurements, an opposite trend to the previous study. The short-term average radon concentrations were larger in bedrooms (70 Bq m -3 ) of ground floor dwellings and in basements (118 Bq m -3 ) compared to their respective long-term measurements of 46 Bq m -3 and 84 Bq m -3. The highest radon concentrations and greatest difference between short- and long-term measurements were found in buildings with an underlying ground structure of outcrops of Triassic rocks containing fissures and cracks which allow gases to be more mobile thereby increasing the mobility of radon. A radon measurement comparison survey in New Jersey, performed as part of an epidemiologic study examining residential radon exposure and lung cancer, deployed charcoal canister radon devices in the basement over four days from October through April (when windows were closed during the heating season). Annual alpha track measurements were also placed in the basement and a non-basement living area, mostly in bedrooms, in 709 houses. The survey indicated that short-term radon measurements overestimated annual radon concentrations, but both measurements were noted to have low radon concentrations (GM of basement canister measurements = 56 Bq m -3 versus GM of basement alpha track measurements= 48 Bq m -3 ) (32). In a joint survey evaluating charcoal canister measurements, Steck (33) noted that screening and annual radon measurements were well correlated (correlation coefficient not provided) (p < 0.001). The charcoal detectors were exposed for a period of two days,

36 18 while the alpha track measurement period ranged from 8 to 12 months. Detectors were placed on the lowest two levels (unable to locate a more specifically site where they were deployed in the paper) in 76 Minnesotan houses. The ratio of the geometric average of annual measurements to the geometric average of screening measurements was But, this ratio increases to 1.2 when the lowest livable level was sampled for the annual measurements indicating that there was a larger disparity between the screening and annual measurements when the lowest livable level was sampled compared to the lowest level for the annual tests. Ability of short-term radon measurements to predict annual radon concentrations at the same location and factors that affect their variability The weak correlation between short- and long-term radon concentrations obtained in the same location can often be attributed to temporal radon variation. These variations include changes in meteorological conditions and behaviors by the occupants in the home (e.g., window opening). The impact of seasonal influences on time-integrated radon measurements was demonstrated in a survey of 110 residences in Colorado by Borak et al. (34). Radon was measured by alpha track detectors over an annual period as well as three months in the summer and another three months in the winter with both detectors (six months of the year was unaccounted for) placed adjacent to each other on the ground floor. The investigators found that radon concentrations in the winter were about 1.7 times larger than the measured annual mean concentrations, whereas annual measurements were about 2.5 times larger than the summer measurements. A joint analysis of short- and long-term radon monitoring from the U.S. National Residential Radon Survey and the State Residential Radon Surveys demonstrated that data obtained from winter screening measurements made in the lowest level of ground contact can be used to characterize annual average living area measurements (35). The prediction of annual average living area radon concentrations from screening tests can be performed by multiplying a conversion factor relevant to the type of house (i.e., no

37 19 basement, living area basement, non-living area basement) in different geographical regions to each screening measurement. This analysis was comprised of over 50,000 screening measurements from 41 states and annual tests covering 125 counties. Klotz et al. (32) examined the ability of canister measurements to predict an annual test on the same floor with respect to the EPA s action level of 150 Bq m -3 (~4 pci L -1 ) for making decisions regarding radon mitigation. In a sample of 709 residences (noted to have very low radon concentrations), there were no instances where the living area annual measurements exceeded the EPA s action level when the living area canister test measured less than 150 Bq m -3 (0% false negative rate). It should be mentioned, however, that there were a very low number (N=8) of living area alpha track measurements larger than the action level. Among screening tests measuring larger than 150 Bq m -3, 76 percent (false positive rate) exhibited annual measurements less than 150 Bq m -3. The false negative rate slightly increased to 2 percent when a level of 74 Bq m -3 (2 pci L -1 ) was considered. Studies have been performed examining factors that influence time-integrated residential radon measurements. Field et al. (28) described factors affecting screening measurements in a study of 582 rural Iowan households. Houses with larger radon concentrations were positively and significantly associated with larger energy efficiency (p=0.0246), presence of a basement (p<0.0001), access to the basement from inside the house (p=0.0397), and presence of a non-poured concrete wall in unfinished basements (p= ). The importance of a particular housing characteristic on these measurements depended on the location of the radon detector. The presence of a crawl space, for instance, was not significantly different from radon concentrations in residences without one (p=0.1713). Thus, when the placement of the radon monitor is considered, homes with crawl spaces were significantly associated with larger radon concentrations in unfinished basements and main floor rooms (p= and p=0.0003, respectively).

38 20 Factors affecting annual average radon concentrations, on the other hand, were examined as part of follow-up testing for the Iowa Radon Lung Cancer Study (IRLCS) (26). One to two additional years of annual radon measurements were obtained from study participants homes to quantify the degree of temporal radon variation from one year to the next. In total, homeowners at 196 homes participated in an additional year of testing. An additional 64 homeowners agreed to two additional years of follow-up measurements. The significant predictors for these measurements, when accounting for type of house and floor where the radon measurements took place, included: unfinished basement; lack of an insulated ceiling; presence of crawl space; year of home construction; location of lowest home level relative to the ground; and presence of a toilet, bathtub, shower, washing machine, or hot tub in the basement. Another survey conducted in Minnesota was comprised of annual average radon measurements in 196 sites within 98 houses in the two lowest levels from 1983 through This survey found average hours of snowfall, exposure to wind, heating-ventilation system changes, basement expansions, and other major structural additions, that significantly influenced year-to-year radon variation (36). Relationship between short- and long-term radon measurements collected on different floor levels There are few studies comparing short- and long-term measurements of radon in different floors within a residence. One of these is a New Jersey survey (noted to have very low radon concentrations) of 709 houses by Klotz and colleagues (32) where the variation between four-day (during the heating season) charcoal canister radon measurements in the basement and annual alpha track measurements from a nonbasement living area, usually a bedroom, was evaluated. The geometric mean ratio of the basement canister measurement to the living area alpha track device measurement was 3.1(geometric standard deviation (GSD) = 2.3) for all residences but increased to 5.6

39 21 (GSD=2.0) for basement canister measurements larger than 150 Bq m -3 (~4 pci L -1 ) and decreased to 2.7 (GSD=2.2) for basement concentrations less than 150 Bq m -3. For homes with forced air heat, the geometric mean ratio was 2.4 (GSD=2.0) and rose to 4.1 (GSD=1.7) for basement measurements larger than 150 Bq m -3 but dropped to 2.2 (GSD=2.0) when basement measurements were less than 150 Bq m -3. Among residences with other heating systems, the geometric mean ratio was 3.5 (GSD=2.3) and increased to 6.8 (GSD=2.1) among basement measurements larger than 150 Bq m -3 while it decreased to 3.1 (GSD=2.2) for basement canister measurements less than 150 Bq m -3. Another study performed a comparison of winter basement canister measurements and first floor annual alpha track radon measurements as part of a five county survey in New York in over 700 houses (37). The geometric means of winter basement radon concentrations were about twice the geometric means of first floor annual measurements for all foundations and heating system types. The geometric mean ratio of the basement winter radon measurement to the first floor annual radon measurement was greatest in houses with a brick foundation (3.6), although only six houses had this foundation type, and a hot water heating system (4.1). The lowest ratios were observed in residences with a fieldstone foundation (2.2) and a wood/coal heating system (1.9). Ability of short-term radon measurements to predict annual average radon concentrations on different floor levels and factors that affect their variability The accuracy of forecasting an annual average concentration in a non-basement living area based on a four day charcoal canister basement screening test was evaluated for the purpose of making mitigation decisions in a study by Klotz and colleagues (32). There were no cases when the basement canister measurement was less than the EPA s action level of 150 Bq m -3 (~4 pci L -1 ) but the living area alpha track device measured larger than 150 Bq m -3 (i.e., false negative rate). It should be mentioned, however, that there were a very low number (N=8) of living area alpha track measurements larger than the action level. Among the 111 basement canister measurements larger than 150 Bq m -3,

40 22 there were 103 living area alpha track measurements less than 150 Bq m -3 (93% false positive rate). The false positive rates dropped to 77 percent when a lower action level of 74 Bq m -3 (2 pci L -1 ) was taken into account. Another measurement that has been compared to a screening test as a surrogate marker of radon exposure in a living area is the average of annual measurements from each level of a residence. White (38) compared this annual mean measurement to a single two-day winter screening test obtained in the lowest livable level as well as the mean of two 2-day winter screening tests with reference to the EPA s action level of 148 Bq m -3 (4 pci L -1 ) for measurements available from 1,449 residences participating in the U.S. State Residential Radon Survey. The proportion of cases in which the screening measurement was less than 148 Bq m -3, but the annual average living area test was larger than 148 Bq m -3 (referred to as a false negative) was 0.3 percent for screening measurements less than 74 Bq m -3 (2 pci L -1 ). This rate increases as the screening measurement rises. The false negative rate reaches 13 percent for screening measurements between 111 and 129 Bq m -3 and increases further to 18 percent for those screening tests between 130 and 147 Bq m -3. This rate rises slightly to 19 when the mean of two screening tests is less than 148 Bq m -3. In contrast, the false positive rate decreases as the mean of the two 2-day screening measurements increases. An error rate of 56 percent occurs when the mean of the two screenings tests is between 148 and 184 Bq m -3 but the annual measurement is less than 148 Bq m -3 and drops to 36 percent when the mean is between 222 and 258 Bq m -3. A predictive analysis by White and colleagues (39) was performed to forecast the annual living area average (ALAA) based on a seasonal two day charcoal canister screening measurement as part of the EPA s state residential radon surveys. More than 300 screening measurements were available from each season. Seasonal equations for predicting the ALAA measurement demonstrated that a single screening measurement obtained in the summer was within a factor of 2.8 of the ALAA versus a factor of 2.3 for

41 23 a screening test taken in the spring at a 95 percent confidence limit. The winter and fall screening measurements were both within a factor of 2.5 of the ALAA test value. Factors that affect spatial variability among long-term radon measurements may in turn affect short-term agreement with long-term measurements. A study by Fisher et al. (27) investigated spatial variation of annual mean radon concentrations in 918 Iowan homes as part of the Iowa Radon Lung Cancer Study where multiple year-long measurements were conducted on each house level. The ratio of first floor to basement average radon concentrations differed for one and two story homes, 0.61(SD=0.19) and 0.53 (SD=0.39), respectively. There was less variation in average radon concentrations between the second and the first floor (1.02, SD=0.24) versus between the second floor and the basement (0.51, SD=0.21) in two story residences. There were significantly larger ratios for residences with forced air heating systems compared to homes with a nonforced air system for all types of residences with the exception of the ratio of the second floor to the first floor in two story homes. The prediction of a first floor annual radon concentration based on a basement annual radon measurement was evaluated for one and two story residences. Larger first floor radon measurements and larger first floor to basement ratios were significantly associated with higher basement radon concentrations, a forced air heating system, and a larger basement volume. This variation between floors demonstrates that multiple measurements within the home need to be performed rather than using a lowest livable short-term radon concentration to predict the radon concentration on another floor in to order to obtain more precise estimates for radon exposure assessments. Exposure to radon in other settings Many studies have described the distribution of radon concentrations that occur in the home. However, because of the potential for high radon concentrations occurring

42 24 outside the home, the distribution of radon concentrations and duration of exposure in other settings deserves consideration. Since individuals spend part of their day in the workplace and outdoors, exposures in these areas should be also evaluated to determine how it contributes to one s total radon exposure. Numerous scientific studies have linked occupational radon exposure to increased rates of lung cancer in cohorts of underground miners (4), but data regarding radon exposure in above ground workplaces (e.g., offices) especially in the United States are sparse. In addition, there are fewer studies providing comparative analyses of radon concentrations in distinct environments within close proximity to each other. The few workplace studies that have been performed have exhibited contrasting findings where differences in measuring radon concentrations were found. A national survey of radon levels in U.S. public schools was conducted in the 1990 through 1991 school year by the EPA (40). Short-term radon measurements were obtained in a randomly selected sample of 927 schools using diffusion barrier charcoal canisters placed for seven days in the months of February and March. Long-term radon measurements lasting five months were made from December through May using alpha track detectors (ATDs) in a random sample of 100 schools. Radon detectors were deployed in frequently occupied rooms with ground contact (i.e., lowest level). About two percent (73,000 rooms) of the total number of school rooms had short-term radon measurements that were equal to or larger than 150 Bq m -3 (~4 pci L -1 ). There was a smaller percentage (1.5%) of the long-term radon measurements equal to or above this action level. A smaller survey of short- and long-term radon detectors positioned side-by-side in the same rooms of these schools was also performed. The decision to mitigate radon with respect to the action level was correct 99 percent of the time based on the short-term measurement when compared to the long-term measurement. In total, close to 19 percent (15,000 schools) of U.S. schools were estimated to have at least one frequently occupied room in contact with the ground when a short-term radon measurement was equal to or above 148

43 25 Bq m -3 (4 pci L -1 ). These concentrations were averaged, however, during hours when the buildings were not occupied. In another survey of about 3,100 places of work in the United Kingdom where ATDs were deployed for one month, which also did not account for occupancy differences, found that greater than 80 percent of the work areas measured less than the national action level of 200 Bq m -3 (5.4 pci L -1 ) (41). Variations in radon concentrations have been noted in studies that take into consideration occupied versus unoccupied building hours. In a Finnish survey of workplaces with 4,500 one-month radon measurements in mostly ground floor areas, lower mean radon concentrations were observed during hours when the workplaces were occupied as compared to monthly mean radon concentrations which averaged radon concentrations hours when the workplaces were not occupied (42). These differences were attributed to mechanical ventilation, which was only in operation during work hours. Continuous radon monitoring in a school and day care center in Norway exhibited similar findings. Radon concentrations over a six-day measurement period were below the action level of 200 Bq m -3 (5.4 pci L -1 ) for both the school and day care center when the buildings were occupied during the week day hours, whereas radon concentrations during week nights were several times higher and well above the action level (43). Comparison of home and workplace radon concentrations The relationship between radon concentrations in the home and workplace has been investigated using mostly short-term measurements and produced inconsistent findings. In an Italian survey of the Tuscany region, geometric mean radon concentrations in dwellings and workplaces were strongly correlated (r=0.88) in the same geographical area with measurements from 1,541 dwellings and 1,159 workplaces (44). Radon concentrations in living rooms and bedrooms were measured in each dwelling and two measurements on each floor in each workplace were taken over two consecutive six month periods using passive nuclear track detectors.

44 26 In contrast, Poffijn et al. found a greater percentage of dwellings (24%) that exceeded 200 Bq m -3 (5.4 pci L -1 ) compared to schools (12%) or public buildings (10%) in a Belgian study covering 79 houses (45), 421 schools, and 36 other buildings (46). Charcoal measurements were mostly taken in closed conditions over a weekend. Similar findings were exhibited in a survey in Finland consisting of two month home and workplace alpha track measurements during the spring (47). A poor correlation was observed (r=0.18, p=0.04) between 333 workplace and 939 home (447 living room and 492 bedroom) measurements. The geometric mean radon concentration in the dwelling (68 Bq m -3 ) was more than three times larger than the concentration in the workplace (20 Bq m -3 ). A survey in New Mexico (United States) also found the mean radon concentration in the dwelling to be three times larger than the levels in the office (48). Sixty-five offices and 47 homes were measured for radon for an exposure period of three months in the same county. However, another Italian study (49) found schools to have larger annual average radon concentrations than dwellings in the same geographical region in almost all of the six regions measured with 2,173 schools (about 76% of rooms measured were on the ground level) and 4,866 dwellings (generally in the bedroom) (50) tested with alpha track measurements. Radon concentrations from schools were seasonally corrected ( annualised ) to make comparisons with the national survey in dwellings. There was more agreement between both types of measurements when adjusting for the location of the detector. Likewise, Iyogi et al. (51) found evidence that workplace and home radon concentrations were statistically different from workplaces exhibiting a larger annual geometric mean radon concentration (20 Bq m -3 ) as compared to dwellings (13 Bq m -3 ). The study was performed in a region of Japan and covered a ten-month measurement period with 107 workplaces and 109 dwellings tested. However, the researchers noted that average workplace radon concentrations were lower when measured during working hours when air conditioning is activated compared to non-working hours.

45 27 Relationship between indoor and outdoor radon concentrations The few studies that have been performed that compare indoor and outdoor radon concentrations exhibit inconsistent findings. A study by Wenbin et al. (52) conducted in a province of China observed indoor geometric mean radon concentrations more than 2.5 times larger than outdoor geometric mean levels. These findings were based on levels tested, using a scintillation and two filter method, mainly during autumn and spring in 397 rooms in urban and rural dwellings and 227 outdoor measurements. Based on air samples taken over several days, maximum radon concentrations were observed during the early morning for the indoor measurements and at midnight for those outdoors. The minimum radon concentrations were measured in the afternoon both inside and outdoors. As part of the Iowa Radon Lung Cancer Study, annual atmospheric average outdoor concentrations were measured in 111 and 64 locations in Iowa and Minnesota, respectively (53). Annual average indoor concentrations were measured in dwellings near the outdoor detectors (unable to locate distance between indoor and outdoor detectors in paper) with a total of 1,039 measurements in Iowa and 128 in Minnesota. Geometric mean annual concentrations in the bedrooms were slightly larger than three times than outdoor levels in Iowa and fivefold larger in Minnesota. In contrast, a smaller survey in Belgium reported larger mean outdoor levels than those measured in dwellings for measurements covering five years of observation in seven homes and three years of outdoor observation in 21 outdoor locations covering a three month exposure period (54). Aims of the dissertation Previous studies have provided mixed findings concerning the temporal and spatial agreement between short- and long-term radon measurements. Because of the wide use of short-term radon tests, obtained in the lowest livable level, and their reliance on them on whether to conduct further radon testing in homes as well as to generate

46 28 estimates for radon exposure assessments, further evaluation of these measurements is warranted to determine how well they compare to long-term radon measurements. Many studies have documented the distribution of radon exposures that occur in the home. However, because of the potential for high radon exposure occurring outside the home as well as the health effects of protracted radon exposure, the magnitude of radon concentrations in other settings deserves further assessment. Since individuals spend part of their day in the workplace and outdoors, exposures in these areas should be evaluated to determine how they contribute to one s total radon exposure. Specific Aims 1 and 2 (discussed in Chapter II) involve assessing the temporal agreement, whereas Specific Aims 3 and 4 (Chapter III) evaluate the temporal and spatial variability between short- and long-term radon measurements in Iowa residences where a substantial proportion of basement radon concentrations exceed 150 Bq m -3 (~4 pci L -1 ). Specific Aim 5 (Chapter IV) characterizes radon concentrations in above ground workplaces in Missouri and compares those concentrations to above ground radon concentrations in both nearby homes and nearby outdoor locations. Specific Aim 6 (Chapter IV) evaluates whether annual outdoor or residential radon concentrations have any utility in predicting the annual occurrence of radon in nearby workplaces. The Specific Aims of this dissertation are as follows: Specific Aim 1: To evaluate the degree of agreement between short-term (electret ion chamber detector-based test) and annual (alpha track detector-based test) radon test results for measurements collected at the same location in the lowest livable level within a residence. Specific Aim 2: To evaluate the ability of a short-term radon test to predict an annual radon concentration placed at the same location and possible factors (e.g., presence of forced heating, house volume, presence of a crawl space) that may influence this agreement.

47 29 Specific Aim 3: To evaluate the degree of agreement between short-term (electret ion chamber detector-based test) and annual (alpha track detector-based test) radon test results for measurements placed on different levels within the same home. Specific Aim 4: To evaluate the ability of a short-term radon test to predict an annual mean radon concentration for detectors placed at different levels within the home and possible factors (e.g., presence of forced heating, house volume, presence of a crawl space) that may influence this agreement. Specific Aim 5: To evaluate the agreement between annual radon measurements that were obtained in above ground workplaces, non-basement first floor residences near the workplaces, and outdoor locations near the workplace. Specific Aim 6: To evaluate the utility of the outdoor or residential annual radon concentrations to predict the annual radon concentration in nearby workplaces. The University of Iowa s Human Subjects Office/Institutional Review Board issued approval for these studies on May 31, 2011 (Appendix A). The studies were supported by grant number R01 ES05653 from the National Institute of Environmental Health Sciences and grant number R01 CA85942 (partly funding Missouri studies) from the National Cancer Institute.

48 30 CHAPTER II EVALUATION OF AGREEMENT OF TIME-INTEGRATED BASEMENT RESIDENTIAL RADON MEASUREMENTS AND CORRECTNESS OF FURTHER RADON TESTING INDICATORS Abstract This study investigated the temporal variability between short-term and annual residential radon measurements collected on the lowest livable level. Another objective was to evaluate the utility of predicting the annual radon concentration in the lowest livable level based on a short-term radon measurement at the same location and the types of housing factors and occupant practices that affected predictor performance. Electret ion chamber (short-term) and alpha track (annual) radon measurements were obtained in residences between 1992 and 1997 as part of the Iowa Radon Lung Cancer Study. The geometric mean of the basement winter short-term radon concentrations (199 Bq m - 3 ) slightly overestimated the geometric mean of the basement annual concentrations (181 Bq m -3 ). About 60 percent of both types of measurements were equal to or above the EPA s radon action level of 148 Bq m -3. The basement winter short-term tests predicted correctly, when additional measurements of radon were recommended 88 percent of the time based on the EPA s radon action level of 148 Bq m -3. Its performance improved greatly at a lower reference radon level of 74 Bq m -3 (98%). The false negative rate of how often the short-term test incorrectly indicated that further radon testing was unnecessary based on an annual measurement was 12 percent at the action level of 148 Bq m -3, but dropped to two percent at a 74 Bq m -3 reference radon level. The significant common housing factors that were suggestive of influencing the basement winter shortterm and the basement annual radon concentrations were the presence of central conditioning, the presence of a clothes dryer in the basement, and the presence of a sump in the basement. The foundation wall material of the basement was the only significant

49 31 factor to have an impact on the absolute difference between both measurements and a significant factor for influencing the basement annual radon concentrations. Introduction The measurement of radon (Rn-222) is frequently obtained using short-term measurements in the lowest livable level of the home under closed house conditions when windows and outside doors are kept closed as much as possible (18). The shortterm measurement period typically ranges from two to 90 days depending on detector type. In the United States, the testing period usually lasts for two to five days because of the need for rapid testing during real estate transactions. Long-term radon measurements, on the other hand, are collected under the usual ventilation conditions generally used by the occupant of the home (19). Long-term testing lasts more than 90 days with an optimal measurement period of one year. Long-term measurements are preferable for assessing an individual s yearly radon exposure, as compared to short-term radon monitoring, because the long-term measurement period covers at least three months and accounts for changes in meteorological conditions and occupant behavior (e.g., window opening, air conditioner usage). Long-term testing allows radon concentrations to be averaged over months or years rather than days or weeks generating more accurate estimates for radon exposure assessments. Short-term measurements, obtained under closed house conditions, are considered screening tests as they provide an indicator of the potential for a home to have elevated radon concentrations. Closed house testing conditions are recommended by the EPA so that the measured radon concentrations will in most cases represent a worstcase scenario (i.e., maximizing the conditions for higher radon concentrations in a home) and as a result, would often times be larger than the long-term measurements (19). Lung cancer is the most established adverse health outcome associated with protracted exposure to radon and its decay products. A long-term radon measurement,

50 32 therefore, would be more reflective of long-term exposure versus a short-term test. However, the EPA recommends that if a short-term radon test is above the radon action level of 148 Bq m -3, this test can be followed by a second short-term test to make radon mitigation decisions (18). Because of both the wide use of short-term tests to assess exposure to radon and the test measurements influence on guiding individuals to take further action as needed, it is important to determine how well these measurements approximate long-term radon measurements as well as to evaluate their ability to predict long-term measurements for decisions concerning the need for further radon testing. Indoor radon concentrations can exhibit significant temporal variability (29-34) and are affected by housing factors (e.g., forced air heating system) and occupant practices (e.g., air conditioner usage) (28). A specific aim of this study was to determine the temporal agreement between short-term and annual radon airborne measurements that were both collected at the same location in the lowest livable level of homes in the state of Iowa. Studies conducted by the EPA have found that Iowa has both the highest mean residential radon concentrations (241 Bq m -3 ) and the greatest percentage (71%) of screening radon measurements above the EPA s action level of 148 Bq m -3 compared to any other state surveyed in the U.S. (23). According to the U.S. Geological Survey, the high radon emanating potential of the glacial deposits (24) that were deposited in Iowa are responsible for the state s elevated radon potential. Another study aim was to evaluate the ability of the short-term radon measurement to predict the annual average radon measurement as well as to assess the effect of certain housing factors and occupant practices on the short-term and annual measurements as well as their differences. Few studies have examined the impact of these factors on short-term and annual measurements and their agreement. The influence of these factors on these measurements is important to consider if homes with particular housing characteristics and occupants engaging in these practices contribute to larger radon concentrations compared to homes without these characteristics. The agreement

51 33 between short- and long-term annual radon measurements and the effect of housing and occupant factors on these measurements as well as their differences is an essential consideration in developing policies and communication materials on radon measurement testing and mitigation decision making. Methods The study used existing data from the Iowa Radon Lung Cancer Study (IRLCS) (Principal Investigators: Charles Lynch, Department of Epidemiology, R. William Field, Department of Occupational and Environmental Health, Department of Epidemiology, University of Iowa). The IRLCS, conducted from 1992 through 1997, was a populationbased, case-control study that evaluated the association between protracted exposure to radon and its decay products and lung cancer among women in Iowa (55). Sample selection The criteria for inclusion in the IRLCS limited enrollment to female residents of Iowa between the ages of 40 and 84 years. Further requirements included residence in the current home for at least 20 consecutive years and no reports of modifications to one s home based on any previous radon testing. Lung cancer cases were included in the study if they were newly diagnosed between May 1, 1993 and April 30, 1996 with a microscopically confirmed primary invasive lung cancer without any prior lung cancer as determined by the State Health Registry of Iowa (SHRI). Controls met the following additional criteria: 1) no prior malignant (invasive) lung cancer as determined by the SHRI, 2) no malignant lung cancer within the last two years as reported by the control at the time of initial contact, and 3) alive at initial contact. There were 413 cases and 614 controls included in the IRLCS. Once case and control ascertainment was completed, a contact letter describing the study was mailed to the potential participants. Shortly thereafter, the potential study participants were contacted via phone to explain the study further, answer any questions,

52 34 determine eligibility, and obtain informed consent. After receiving informed consent, the home survey visit was scheduled and questionnaires were mailed to be completed by the study participant before the visit took place. Survey questionnaire The questionnaire about the home (Appendix B) collected information about rural versus urban residence, drinking water source(s), levels of the home, heating and cooling systems, ventilation, circulation, and weatherization. The questionnaire about the resident (Appendix C) collected information about demographics, occupational history and exposures, active and passive tobacco exposure, personal and family health history, vitamin usage, and diet and cooking practices. Field assistants reviewed the questionnaire for completeness during the home survey, and if critical information was incomplete, the assistant facilitated completion of the questionnaire. Radon measurements As part of the on-site residential radon assessment survey, yearlong radon measurements were collected using alpha track detectors (ATD) (Figure 3). The number and location of ATD placements in each residence was based on the floor plan of the house and weighted by participant mobility (i.e., where most time was spent in the home) (Appendices D and E). At least one Radtrak ATD (Landauer, Inc., Glenwood, Illinois) was placed on each level of the home, in current and historic bedrooms, and in work area(s) within the home (e.g., office). This device was chosen to provide an integrated mean radon concentration of temporally varying residential radon gas. The Radtrak ATD was selected based on its past record of good accuracy and precision over a multi-year period when compared to other commercially available ATDs (56) as well as its low cost, mailability, and ease of use. With the exception of duplicates (i.e., collocated detectors), up to seven ATDs were placed in each residence for an exposure period of a year. Field assistants retrieved the detectors after the one-year placement period and inspected them for damage as well as any changes in placement from their initial deployment. All initial detector placements were identified on a floor plan that was drawn at the time of placement. The field assistants checked to assure the

53 35 detector was in the same location as indicated on the floor plan at the time they were retrieval. After retrieval, the ATDs were stored in a low radon environment (less than 7.4 Bq m -3 ) prior to their return to Landauer Inc. for processing. For the short-term measurements, an E-PERM (Electret-Passive Environmental Radon Monitor) (Rad Elec, Inc., Frederick, Maryland) (Figure 4) was placed in the lowest livable area in a subset of the homes during the winter months. Cases and controls who had their initial site visit during November, January, or February when closed house conditions are maintained were eligible to have an E-PERM placed in their residence. The possibility of failure to assure closed house conditions during the other months, primarily because of the potential for warmer weather outdoors and the potential for window opening, precluded placement of short-term measurements between March and October. The lowest level of the house was considered livable if the participant spent at least 15 minutes per day in that level. Participants who completed short-term and annual measurements of radon were eligible for this study. E-PERMs were selected as the detector of choice to obtain the short-term measurements based on excellent accuracy and precision in a comparison of commercially available short-term radon detectors (57). The study researchers noted that the mean of the absolute values of the relative error (MARE), an indicator of accuracy, was the lowest for the E-PERM (4%) compared to various commercial brands for the charcoal canister (ranging from 9 to 30%) (57). For precision, the E-PERM had low coefficient of variation values of 4.3 and 7.3 percent for exposures lasting three and four days, respectively, compared to 16.8 percent for one company s diffusion barrier charcoal adsorption liquid scintillation short-term detector (57). The E-PERM was positioned five to seven inches away from the ATD. The E- PERMs were left in the home for a 7 to10 day measurement period. Beginning and ending voltages of the internal electret were recorded. The radon concentration was determined using the manufacturer s equation employing the difference between the

54 36 beginning and ending voltage, duration of exposure period, level of background gamma radiation, and a detector calibration factor (58). Background gamma measurements were collected at the E-PERM placement sites and other areas in the home during deployment using a Ludlum Measurements, Inc. (Sweetwater, Texas) Model 19 micro-r survey meter. This information was recorded on a short-term radon measurement form (Appendix F). A reference electret with known voltage was used to assess proper calibration of the electret reader prior to measuring the voltage on the E-PERMs (i.e., E- PERMS should have the same voltage prior to measuring the detectors on their way out the door and before measuring detectors when they are returned). This survey resulted in 189 lowest livable winter short-term and 930 lowest livable annual radon measurements. Geocoding of home addresses Zip codes of the homes were geocoded using ArcMap 8.3 (ESRI, Inc., Redlands, CA) and a 2009 TIGER/line file of Iowa as the reference address locator. Data analysis The calculations for the basement short-term E-PERM radon concentrations can be found in Appendix G. Descriptive statistics (e.g., geometric mean, geometric standard deviation) were computed for each type of radon measurement as well as their differences. The absolute differences between measurements was evaluated to demonstrate the magnitude of the tests differences. The distribution of each measurement and their differences was evaluated using the Shapiro-Wilks test to assess normality. If the test results supported normality, the distributions were described with an arithmetic mean and standard deviation. In the cases when the data did not follow a normal distribution, the data were transformed using a natural ln (log base e) and the lntransformed values were tested to verify they were normally distributed. If the lntransformed values followed a normal distribution, they were described by a geometric mean and geometric standard deviation. The following analyses were conducted using the normalized data.

55 37 A Student s paired t-test was performed to evaluate whether significant differences existed between the short-term and annual radon concentrations. Paired differences between the radon measurements were examined to assess normality using Shapiro-Wilks test (an assumption of the Student s paired t-test). Short-term and annual measurements were paired for the t-test based on the site identification number where they were placed. The data were ln-transformed in cases when they did not follow a standard normal distribution. A Shapiro-Wilks test was performed on the ln-transformed data to verify they were normally distributed. The ability of the short-term measurement to classify the annual measurement (considered the gold standard for this comparison), with respect to being below or equal to or above the EPA s radon action level of 148 Bq m -3 (Table 2), was used as an indication to make an informed decision for the need of additional radon testing. This classification procedure was repeated using a lower reference radon level of 74 Bq m -3 in order to compare the results to similar work performed by other investigators (32). Findings at a reference level of 74 Bq m -3 are of interest since the EPA states that radon levels below 148 Bq m -3 also carries some risk for cancer and in most cases, radon concentrations can be reduced to 74 Bq m -3 or below (19). A lower reference level of 111 Bq m -3 was also examined for the purpose to evaluate how the short-term tests categorically classified its annual measurement close to the proposed World Health Organization s radon reference value of 100 Bq m -3 (20). The classification rates at these reference levels were defined as follows: I. P (correct classification) = Number of occurrences when the short-term and annual measurements were both < reference level or both reference level Total number of sites with short-term and annual measurements II. P (misclassification) =

56 38 Number of occurrences when the short-term measurement was < reference level, but the annual measurement was reference level or the short-term measurement was reference level, but the annual measurement was < reference level Total number of sites with short-term and annual measurements III. Sensitivity (true positive) = P (short-term test reference level annual test reference level) = Number of occurrences when both the short-term and annual measurements were reference level Total number of sites with annual measurements reference level IV. Specificity (true negative) = P (short-term test < reference level annual test < reference level) = Number of occurrences when both the short-term and annual measurements were < reference level Total number of sites with annual measurements < reference level V. Positive predictive value = P (annual test reference level short-term test reference level) = Number of occurrences when both the short-term and annual measurements were reference level Total number of sites with short-term measurements reference level VI. Negative predictive value = P (annual test <reference level short-term test < reference level) = Number of occurrences when both the short-term and annual measurements were < reference level Total number of sites with short-term measurements < reference level VII. False positive (1 specificity) = P (short-term test reference level annual test < reference level) = Number of occurrences when the short-term measurement was reference level, but the annual measurement was < reference level

57 39 Total number of sites with annual measurements < reference level VIII. False negative (1 sensitivity) = P (short-term test < reference level annual test reference level) = Number of occurrences when the short-term measurement was < reference level, but the annual measurement was reference level Total number of sites with annual measurements reference level In making these comparisons, the false negative rates are especially important as they are a measure of how often the short-term test incorrectly indicates that further radon testing is unnecessary, based on an annual measurement. Table 2. Interpretation of short-term and annual radon measurements that were equal to or above or less than the respective reference level Basement annual alpha track measurement (Bq m -3 ) Basement winter short-term electret ion chamber measurement (Bq m -3 ) < < True Negative False Negative < ST test < 74 True Negative False Negative False Positive True Positive < ST test < True Negative True Negative False Positive False Negative False Negative True Positive

58 40 True Negative - short-term test correctly indicates that further radon testing is not necessary False Negative - short-term test incorrectly indicates that further radon testing is not necessary True Positive - short-term test correctly indicates that further radon testing is necessary False Positive - short-term test incorrectly indicates that further radon testing is necessary In addition to the agreement of the short-term radon test to classify the annual radon test by a reference level cutoff, it would be helpful to provide information to an individual obtaining a short-term radon test on what can be concluded about the radon potential in a home (i.e., what one would obtain if one eventually did a long-term radon test). This type of interpretation is critical for providing evidence for the confidence of short-term radon measurements that can be incorporated in public information materials on radon measurement testing and mitigation decision making. The assessment of the short-term test was evaluated that would provide the individual taking the test with a 95 percent level of confidence or more with respect to the action level of 148 Bq m -3 to make decisions about additional radon testing. The conclusive negative short-term measurement value (i.e., lower limit of uncertainty range) was determined when five percent of the short-term tests were less than a short-term measurement value with annual tests equal to or greater than 148 Bq m -3 : i. P (annual test 148 short-term test < conclusive negative value) = 0.05 To illustrate this further, if the short-term test result was 80 Bq m -3 and the conclusive negative value was 100 Bq m -3, for instance, the individual would be more than 95 percent confident that the annual test was less than 148 Bq m -3. The conclusive positive short-term measurement value (i.e., upper limit of uncertainty range) was determined when five percent of the short-term tests were greater than a short-term measurement value with annual tests less than the action level: ii. P(annual test < 148 short-term test > conclusive positive value) = 0.05

59 41 If the short-term test result was 250 Bq m -3 and the conclusive positive value was 200 Bq m -3, for instance, the individual would be more than 95 percent confident that the annual test was equal to or larger than 148 Bq m -3. If the short-term test result was between the conclusive negative and conclusive positive values, there would be less than 95 percent confidence that the annual test was less than 148 Bq m -3. This type of interpretation is especially important if an individual obtains a short-term test value close to the action level. This clarification would inform the individual how certain one can be that the annual test will also be below the action level. The ability of the short-term radon measurement to predict the annual radon measurement (i.e., predictor performance) using simple linear regression was also evaluated. The key assumptions of simple linear regression are: 1) a linear relationship between the response and predictor variable; 2) independent errors; 3) normally distributed errors; and 4) constant variance of errors. Pearson s correlation coefficients were computed to determine the strength of the linear relationship between the short-term and annual radon measurements. Dependent errors occur when repeated measurements are obtained on the same site. To examine whether the errors (i.e., residuals) from this model were normally distributed, the Shapiro-Wilks test was performed. If the errors did not follow a standard normal distribution, the measurements were ln-transformed. Normality of the ln-transformed data was further verified using the Shapiro-Wilks test. All regression analyses were conducted using the normalized data. To check for constant variance of the model s errors (i.e., homoscedasticity), the residuals (i.e., differences between sample values and fitted values) versus the model s predicted responses were plotted and their spread was evaluated. Additional methods for diagnostics of the model are presented in Appendix G. Multiple regression models were also examined. These included: 1) the prediction of the annual radon measurements given certain housing factors and occupant practices; 2) the prediction of the short-term radon measurements given certain housing factors and

60 42 occupant practices; 3) the prediction of the annual radon measurements given certain housing factors and occupant practices as well as the short-term radon measurements; 4) the prediction of the absolute difference between the short-term and annual radon measurements given certain housing factors and occupant practices; and 5) the prediction of the signed difference (comprised of positive and negative values) between the shortterm and annual radon measurements given certain housing factors and occupant practices. The absolute difference model was used to examine the magnitude of the difference between the short-term and annual measurements using a priori selected housing factors and occupant practices. Regression with absolute differences was selected to model the overall predictive performance of the difference (expected bias and variance) as a function of these factors since we were primarily interested in modeling mean absolute error as a function of housing and occupant factors rather than solely modeling average bias. However, regression with signed differences was also used to explore the effect of certain housing factors and occupant practices on the magnitude and direction of the difference between the short-term and annual measurements. In other words, it modeled the expected bias or degree (magnitude of difference) in which the basement short-term test over or underestimates (direction of difference) the basement annual test. Based on previous studies, important predictors of temporal radon variability have included type of heating system, presence of crawl space, fireplace usage, foundation wall construction material, presence of toilet, shower, washing machine or hot tub in the basement, presence of a sump pump, and proportion of first level below ground (26). The case and control status of the study participants was also evaluated to assess whether radon measurements varied between cases and controls. Occupant behavior that affected radon concentrations (e.g., window opening, time spent in home) in the case population may have been modified once a diagnosis of cancer was made. A backward

61 43 variable selection process was used to generate final regression models with predictor variables being retained given p-values less than 0.20 (26). Analyses were performed using SAS version 9.2 (SAS Institute, Inc., Cary, North Carolina). Quality assurance This study followed the EPA s guidelines for quality assurance to measure radon including: 1) quality assurance objectives for measurement data; 2) calibration procedures and frequency; 3) analytical procedures; 4) data validation and reporting; 5) internal quality control; 6) quality assurance audits and reports; 7) preventative maintenance of monitoring equipment; and 8) procedures to assess measurement precision, accuracy, and completeness (59-63). Duplicate detectors (a priori plan to designate 10% of the detectors), which were treated in an identical fashion to the primary device, were positioned ten centimeters from the primary detector. Detectors were also exposed to known radon concentrations (spikes) (a priori plan to designate 5% of the detectors) in the EPA s former National Air and Radiation Environmental Laboratory in Montgomery, Alabama to test for detector accuracy and precision. Field and laboratory control detectors (blanks) (a priori plan to designate 5% of the detectors)) were deployed to evaluate possible radon exposure encountered in the field and in transit to the laboratory. Duplicate, spiked, and blank detectors were labeled in the same manner as primary devices to ensure identical processing by the laboratory. All data were double entered into a database and verified to eliminate entry errors. Results All lowest livable level measurements in this study were obtained in the basement. Seven hundred seventy two annual basement and 18 winter short-term radon measurements were excluded due to a lack of a corresponding winter short-term basement and annual basement measurement, respectively. This resulted in the inclusion

62 44 of 158 basement short-term tests and 158 respective basement annual measurements for this study. Sample characteristics The mean duration of the basement winter short-term radon measurements was 8.4 days (SD=2.7) and ranged from three to 27 days with a median of 7.5 days. The greatest percentage (37%) of the tests was collected in Six (3.8%) were conducted in 1993, 52 (33%) in 1994, and 42 (27%) in Thirty-six (23%) measurements were obtained in the month of January, 30 (19%) in February, 34 (21%) in March, 4 (2.5%) in April, 25 (16%) in November, and 29 (18%) in December. The average background gamma exposure rate of the area where a detector was deployed was 9.7 microroentgens (μr) per hour and ranged from 3 to 17 μr per hour. Figure 5 displays the approximate location of the study participants and their relative basement winter short-term and corresponding basement annual radon measurements. Graduated symbols distinguish areas with regard to the EPA s radon action level of 148 Bq m -3. Paired radon measurements were collected in homes located in 65 of the 99 counties in Iowa. Among the residences with both radon measurements equal to or over the action level, these homes tended to be more concentrated (57%) in the eastern part of Iowa compared to the western part (43%), whereas those residences with both measurements under the action level were located mostly in the central region of Iowa. Differences between short-term and annual measurements relative to the action level were found mostly at sites in the central eastern regions of Iowa.

63 45 Figure 5. Map of basement winter short-term and basement annual radon concentrations with respect to EPA s action level of 148 Bq m -3 in Iowa After both measurements were ln-transformed, the Shapiro-Wilks test supported normality of the distributions for the basement winter short-term (p=0.14) and basement annual (p=0.27) measurements. The basement winter short-term radon concentrations ranged from 42 to 1,331 Bq m -3, with a geometric mean of 199 Bq m -3 (GSD=2.0) (Table 3, Figure 6). The basement annual concentrations had a lower geometric mean (181 Bq m -3 ), but both measurements had about the same median radon concentrations. Seventy five percent of the basement short-term tests were less than or equal to 297 Bq m -3 and 266 Bq m -3 for the annual measurements. About 63 percent of the short-term tests and a slightly lower percent (60%) of the annual tests were equal to or above the EPA s radon action level of 148 Bq m -3. The absolute difference between the basement short-term and

64 46 annual radon concentrations ranged from 0.39 to 738 Bq m -3. Half of the differences between the measurements were below 38 Bq m -3. There was a larger percentage (61%) of the short-term radon measurements greater than their matching annual measurements compared to annual measurements greater than their respective short-term measurements (39%) and none of the measurements had the same value (not presented in Table 3). Table 3. Characteristics of basement winter short-term and basement annual radon measurements (Bq m -3 ) as well as their differences from 158 residences in Iowa Basement Basement Winter Short-Term Annual Difference Characteristic Electret Ion Chamber Test Alpha Track Test Short-Term - Annual Arithmetic Mean (SD) Geometric Mean (GSD) Range 25th Percentile Median 75th Percentile 257 (217) 233 (206) 67 (99) 199 (2.0) 181 (2.0) 34 (3.6) < 148 Bq m 37% 40% Bq m 63% 60%

65 47 Figure 6. Box plot of radon concentration by type of measurement Agreement between measurements The ln-transformed basement short-term and ln-transformed basement annual radon concentrations were compared and found to be significantly different (p=0.001). The degree of agreement between both types of measurements is displayed in Figure 7. There is a similar distribution when the ratio of the basement short-term radon concentration and the basement annual concentration was compared with either the short-term or annual measurement. The mean of their ratio was 1.2 (SD=0.57). As seen in Figure 7, a slightly larger percentage of the ratio values were primarily between zero and one versus between one and two. There were only a few measurements having ratios exceeding two standard deviations from the ratio s mean (represented by dotted lines).

66 48 Figure 7. Plots of agreement of short-term and annual radon concentrations versus their ratio The ability of the short-term radon measurement to classify the categorical annual radon concentration by the specified reference level is presented in Table 4 and Figure 8. The correct classification rate of the number of occurrences for which the short-term radon test matched its corresponding annual radon test category, as both larger than or equal to or less than the action level of 148 Bq m -3 was 83 percent. This rate increased to 92 percent when the reference level was lowered to 74 Bq m -3. Among the short-term measurements between 111 and 148 Bq m -3, about 40 percent had annual measurements equal to or above 148 Bq m -3. Fifty percent had annual measurements equal to or above 148 Bq m -3 when the short-term measurements were between 55.5 and 74 Bq m -3 (N = 4) at a 74 Bq m -3 reference level. Given the annual radon concentration was equal to or larger than the action level of 148 Bq m -3, the probability of the short-term radon concentration also being equal to or larger than this level (i.e., sensitivity) was 88 percent and increased to 98 percent when

67 49 the reference level was reduced to 74 Bq m -3. The probability of an annual test being equal to or larger than the action level of 148 Bq m -3 given the short-term test was also equal to or larger than this level (i.e., positive predictive value) was 84 percent. The positive predictive value remained at 84 percent for a lower reference level of 111 Bq m - 3, but increased to 94 percent when it was further lowered to a 74 Bq m -3 reference level (Figure 8). The probability of the short-term test being less than 148 Bq m -3 when the annual measurement was equal to or larger than 148 Bq m -3 (i.e., false negative rate) was about 12 percent. At lower reference levels of 111 and 74 Bq m -3, the false negative rates dropped considerably to one and two percent, respectively. Table 4. Number of basement short-term and basement annual radon measurements that were equal to or above or less than the respective reference level Basement winter short-term electret ion chamber measurement (Bq m -3 ) Basement annual alpha track measurement (Bq m -3 ) < < < ST test < < ST test <

68 50 Figure 8. Percentages of diagnostic indicators comparing basement short-term radon tests to basement annual radon tests by reference level % Positive Predictive Value Negative Predictive Value False Positive (1 - Specificity) False Negative (1 - Sensitivity) 74 Bq m-3 (2 pci L-1) 111 Bq m-3 (3 pci L-1) 148 Bq m-3 (4 pci L-1) Rn-222 conc. reference level The assessment of a basement winter short-term test was evaluated that would provide an individual taking the test about information for the need of further radon testing with respect to the action level of 148 Bq m -3. The conclusive negative shortterm measurement value (i.e., lower limit of uncertainty range) was 122 Bq m -3 and the conclusive positive (i.e., upper limit of uncertainty range) was 214 Bq m -3. If the shortterm test value was less than 122 Bq m -3, the individual would be more than 95 percent confident that the annual test was less than 148 Bq m -3. If a short-term test value was larger than 214 Bq m -3, the individual would be more than 95 percent confident that the

69 51 annual test was equal to or above 148 Bq m -3. If the short-term test was between 122 Bq m -3 and 214 Bq m -3, the individual would be less than 95 percent confident that the annual test was less than 148 Bq m -3. Simple linear regression The best fitted line for predicting basement annual radon concentrations based on basement winter short-term radon concentrations after both measurements were ln-transformed was generated and plotted in Figure 9 (information concerning model assumptions are presented in Appendix G). A strong linear relationship was noted (Pearson s correlation coefficient (r) = 0.87, p <0.0001). The r-squared value (represented by lower case r-squared to indicate a model comprised of only one predictor variable) indicated that 75 percent of the variability in the basement annual radon measurements can be explained by the basement winter short-term radon measurements. As indicated by the best fitting line equation, a one log unit increase in the basement winter short-term radon concentration would yield a 0.86 log unit increase in the basement annual radon concentration (i.e., a 10 log unit increase in the basement winter short-term radon concentration would yield a 8.6 log unit increase in the basement annual radon concentration). The prediction equation to relate the non-transformed short-term and annual measurements given the prediction equations based on ln-transformed measurements with a constant of e (approximate value of 2.72) is: annual radon concentration = e 0.64 (short-term radon concentration) 0.86 To check for non-constant variance (another assumption of linear regression), the residuals (i.e., differences between sample values and fitted values) versus the predicted responses were plotted. The constant variance assumption appears reasonable in the lntransformed data, as the spread is similar throughout the distribution.

70 52 Figure 9. Scatter plot of ln(basement winter short-term Rn-222 conc.) versus ln(basement annual Rn-222 conc.) and its residuals Backward stepwise regression The housing and occupant factors that were selected for the regression models can be found in Table 5. Overall, slightly more than 50 percent of the residences surveyed had a concrete block foundation wall. Among the

71 53 participants, about seventy percent (N=110) were controls (i.e., individuals without lung cancer). Homes had, on average, two (SD=0.9) major plumbing penetrations in the basement and 82 percent (SD=18) of the basement below ground. About 70 percent of the homes had central air conditioning, 60 percent had a clothes dryer in the basement, 46 percent had a crawl space in the basement, 82 percent had a forced air heating system, and 21 percent had a sump in the basement. The mean volume of the basement was 5,779 cubic feet (SD=2,826) with a median of 5,618. Homes with neither a concrete block nor a poured concrete foundation wall in the basement were removed from the backward elimination procedures.

72 54 Table 5. Summary of regression predictor parameters from 158 homes in Iowa Characteristic N (%) Mean (SD), Median Basement foundation wall material - Concrete block 82 (51.9) Poured concrete 45 (28.5) Neither concrete block nor poured concrete 31 (19.6) Case/control status of participants - Case 48 (30.4) Control 110 (69.6) Number of major plumbing penetrations in basement 123 (77.8) 2.0 (0.9), 2.0 Missing 35 (22.2) Percentage of basement underground 150 (94.9) 82 (18), 90 Missing 8 (5.1) Presence of central air conditioning - Yes 112 (70.9) No 46 (29.1) Presence of clothes dryer in basement - Yes 96 (60.8) No 62 (39.2) Presence of crawl space in basement - Yes 72 (45.6) No 86 (54.4) Presence of forced air heating system - Yes 130 (82.3) No 28 (17.7) Presence of sump in basement - Yes 33 (20.9) No 122 (77.2 ) Missing 3 ( 1.9) Volume of basement 158 (100) 5779 (2826), 5618 Variables that remained in the backward elimination multiple regression model for annual radon concentration were the foundation wall material in the basement (p=0.06), presence of central air conditioning (p=0.004), presence of a clothes dryer in

73 55 the basement (p=0.17), and presence of a sump in the basement (p=0.02) (Table 6). The factors that were positively associated with the annual radon concentrations were the presence of central air conditioning and presence of a sump in the basement and negatively associated were having a poured concrete basement foundation wall and a clothes dryer in a basement. The final model s R-squared value (represented by upper case R-squared to indicate a model comprised of more than one predictor variable) showed that 16 percent of the variability in the basement annual radon concentrations can be explained by all these factors. The changes in the annual concentrations in homes with a particular characteristic compared to homes without this factor were computed as a percent change. This is illustrated further for the percent change of the annual concentration in homes with a sump pump, for instance, compared to homes without a sump pump given all the other variables in the model were fixed (i.e., all other variables were set to reference groups except the variable of interest): Percent change = (annual conc) sump=1 (annual conc) sump=0 (annual conc) sump=0 (annual conc) sump=0 = = (e ) 100 There was no need to include the intercept value because it canceled out when examining the difference between homes with a particular characteristic versus homes without the factor given all the other variables were fixed in the model. This is illustrated further, for instance, when comparing homes with a poured concrete foundation wall to homes with a concrete block foundation wall when all other variables were fixed in the model: I. ln(annual test) = β 0 + β 1 (poured concrete = 1) + β 2 (central ac = 0) + β 3 (clothes dryer = 0) + β 4 (sump = 0) II. ln(annual test) = β 0 + β 1 (concrete block = 0) + β 2 (central ac = 0) + β 3 (clothes dryer = 0) + β 4 (sump = 0)

74 56 III. Difference of ln(annual test) between homes with poured concrete and homes with concrete block wall = β 1 (poured concrete = 1) = 0.28 With all the other variables fixed in the model, the basement annual radon concentration was: 24 percent smaller, i.e., (e ) 100, in homes with a poured concrete versus a concrete block foundation wall; 65 percent larger, i.e., (e ) 100, in homes with central air conditioning versus homes without central air conditioning; 20 percent smaller (e ) in homes with a clothes dryer in the basement compared to those without one; and 45 percent larger (e ) in homes with a sump in the basement versus homes without a sump. Table 6. Backward regression parameters (final model) for predicting ln(basement annual radon concentration) Covariate VIF a β Coeff. b (SE c ) Student's t-test (p) Percent change d Basement foundation wall material 1.04 Concrete block Ref. e - Poured concrete (0.15) -1.9 (0.06) 24 Presence of central air conditioning 1.03 Yes 0.50 (0.17) 3.0 (0.004) 65 No Ref. - Presence of clothes dryer in basement 1.04 Yes (0.16) -1.4 (0.17) 20 No Ref. - Presence of sump in basement 1.01 Yes 0.37 (0.16) 2.3 (0.02) 45 No Ref. - R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Given all other variables were fixed in model, e Reference category

75 57 Variables that remained in the multiple regression model for predicting the basement annual radon concentration given the basement short-term radon concentration and housing/occupant characteristics were the basement short-term radon concentration (p<0.0001), foundation wall material in the basement (p=0.07), presence of clothes dryer in the basement (p=0.12), and presence of a crawl space in the basement (p=0.20) (Table 7). The factors that were positively associated with basement annual radon concentrations were basement short-term radon concentrations, presence of a clothes dryer in the basement, and presence of a crawl space in the basement. The only factor that was significantly associated with smaller annual radon concentrations when adjusting for short-term radon concentrations was having a poured concrete foundation wall in the basement compared to homes with a concrete block foundation wall. The final model s R-squared value showed that 79 percent of the variability in the basement annual radon concentrations can be explained by all these factors. In comparison to the previous model when not adjusting for the short-term measurement (Table 6), housing and occupant factors alone explained very little of the variability in the annual measurements (R 2 =0.16). With all the other variables fixed in the model and adjusting for basement shortterm measurements, the basement annual radon concentration was 12 percent smaller in homes with a poured concrete foundation wall versus a concrete block foundation wall in the basement, 14 percent larger in homes with a clothes dryer in the basement versus homes without a dryer, and 11 percent larger in homes with a crawl space in the basement compared to those without one.

76 58 Table 7. Backward regression parameters (final model) for predicting ln(basement annual radon concentration) given ln(basement short-term radon concentration) and housing/occupant factors Covariate VIF a β Coeff. b (SE c ) Student's t-test (p) Ln(basement short-term radon conc.) (0.05) 19 (<0.0001) Basement foundation wall material 1.03 Concrete block Ref. d - Poured concrete (0.07) -1.8 (0.07) Presence of clothes dryer in basement 1.14 Yes 0.13 (0.08) -1.6 (0.12) No Ref. - Presence of crawl space in basement 1.09 Yes 0.10 (0.07) 1.3 (0.20) No Ref. - R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Reference category Variables that remained in the multiple regression model for predicting basement winter short-term radon concentrations were the presence of central air conditioning (p=0.01), presence of a clothes dryer in the basement (p=0.01), and presence of a sump in the basement (p=0.003) (Table 8). The factors that were positively associated with basement winter short-term radon concentrations were the presence of central air conditioning and presence of a sump in the basement. The factor that was negatively associated with basement winter short-term radon concentrations was the presence of a clothes dryer in the basement. The final model s R-squared value showed that 17 percent of the variability in the basement winter short-term radon concentrations can be explained by all these factors.

77 59 With all the other variables adjusted for in the model, the basement short-term concentration was 62 percent larger in homes with central conditioning versus homes without central air conditioning, 32 percent smaller in homes with a clothes dryer in the basement compared to those without one, and 63 percent larger in homes with a sump in the basement versus homes without a sump. Table 8. Backward regression parameters (final model) for predicting ln(basement winter short-term Rn-222 conc.) Covariate VIF a β Coeff. b (SE c ) Student's t-test (p) Percent change d Presence of central air conditioning 1.02 Yes 0.48 (0.17) 2.9 (0.01) 62 No Ref. e - Presence of clothes dryer in basement 1.14 Yes (0.16) -2.5 (0.01) 32 No Ref. - Presence of sump in basement 1.01 Yes 0.49 (0.16) 3.0 (0.003) 63 No Ref. - R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Given all other variables were fixed in model, e Reference category The only variable that remained in the multiple regression model for predicting the absolute paired difference was the foundation wall material in the basement (p=0.08) (Table 9). The final model s R-squared value showed that only three percent of the

78 60 variability in the absolute paired differences can be explained by the foundation wall material. The difference between the basement winter short-term and the basement annual radon concentration decreased by 38 percent in homes with a poured concrete versus a concrete block foundation wall. Table 9. Backward regression parameters (final model) for predicting the absolute difference, ln( basement short-term radon conc. - basement annual radon conc ) Covariate VIF a β Coeff. b (SE c ) Student's t-test (p) Percent change d Basement foundation wall material 1.00 Concrete block Ref. e - Poured concrete (0.27) -1.8 (0.08) 38 R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Given all other variables were fixed in model, e Reference category Variables that remained in the multiple regression model for predicting the nonabsolute difference between the basement short-term and basement annual radon measurements was the foundation wall material in the basement (p=0.15), presence of a clothes dryer in the basement (p=0.11), and volume of the basement (p=0.15) (Table 10). The final model s R-squared value showed that seven percent of the variability in the signed differences can be explained by all these factors. The basement short-term measurement underestimated the basement annual measurement (i.e., negative difference) in homes with a concrete block foundation wall in the basement, a clothes

79 61 dryer in the basement, and a basement volume that was equal to or less than 4,228 cubic feet. The variance inflation factors (VIF) for each variable in all the final regression models were close to one, indicating that the predictor variables were not strongly correlated (i.e., multicollinearity). Table 10. Backward regression parameters (final model) for predicting the signed difference, ln(basement short-term radon conc.)- ln(basement annual radon conc.) Covariate VIF a β Coeff. b (SE c ) Student's t-test (p) Basement foundation wall material 1.03 Concrete block Ref. d - Poured concrete 0.11 (0.08) 1.5 (0.15) Presence of clothes dryer in basement 1.03 Yes (0.08) -1.6 (0.11) No Ref. - Volume of basement E-5 (1.4E-5) 1.5 (0.15) R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Reference category Quality assurance Twelve percent of the ATDs had a collocated detector placement (duplicate). The mean coefficient of variation (COV = standard deviation / mean of duplicate measurements) for the duplicate ATDs was about seven percent. There were five percent of the ATDs that were annually exposed to known radon concentrations (spikes) in the EPA s National Air and Radiation Environmental Laboratory. The spiked detectors had a mean COV of 8.3 percent for an exposure at 74 Bq m -3, a COV of 7.5

80 62 percent at 148 Bq m -3, and 5.7 percent at 222 Bq m -3. The mean absolute relative error for the spiked ATDs (( MV i TV i TV i ) 100 where MV i is the measured value for the ATD i and TV i is the target radon chamber value for ATD i) was 12.8 percent at 74 Bq m -3, 11.4 percent at 148 Bq m -3, and 8.5 percent at 222 Bq m -3. At least five percent of the ATDs were field controls (i.e., blanks). The control detectors indicated no extraneous radon exposure above Landauer s lowest level of detection of 4 Bq m -3. E-PERMS underwent annual periodic testing for accuracy in the EPA s radon test chamber. The relative error of the measurements was within 15 percent for one week equivalent radon concentrations of 74 Bq m -3, 148 Bq m -3, and 222 Bq m -3. The COVs were all less than 10 percent for all three exposure levels. Ten percent of placements had collocated detectors to assess detector precision. The COVs for the duplicate placements were all less than 10 percent. Ten percent of the E-PERMs were field controls (i.e., blanks) and no extraneous radon exposure was detected as indicated by no decrease in voltage. Additional findings are presented in Appendix G. Sources of error associated with the radon detectors can be found elsewhere (Appendix H). Discussion Among the 158 residences tested for radon in this study, the geometric mean of the basement winter short-term radon concentrations was 199 Bq m -3 (GSD=2.0). In comparison, a larger geometric mean (241 Bq m -3, GSD=2.5) was found in the EPA state radon survey for Iowa (23) for two-day screening radon concentrations obtained during the winter, examining 1,208 residences with basements. More than half (63%) of the basement winter short-term radon concentrations were at or above the EPA s action level of 148 Bq m -3. In contrast, the EPA state radon survey for Iowa found a larger percentage (71%) of screening radon measurements above this guidance level. The basement weeklong winter short-term radon concentrations were significantly different from their respective basement annual radon concentrations. The geometric

81 63 mean of the basement short-term concentrations overestimated the geometric mean of the basement annual concentrations (181 Bq m -3, GSD=2.0) by 18 Bq m -3. A similar trend was observed in a survey of more than 300 residences in Poland by Wysocka et al. (31) and in a study of 709 New Jersey homes by Klotz and colleagues (32) where the average of four-day winter basement radon measurements also overestimated their corresponding annual average basement radon concentrations by 34 and 8 Bq m -3, respectively. More than half (61%) of the short-term measurements were larger than their matching annual measurements. It was surprising, however, to find a considerable number of the annual measurements (39%) larger than their respective short-term measurements given the annual measurements were obtained during conditions when windows and doors were either open or closed. The basement annual radon concentrations had a slightly lower percentage (60%) at or above the EPA s action level compared to the basement winter short-term radon concentrations. An opposing trend was observed in a study conducted in the Spanish region of Galicia (30). The authors found the geometric mean of the 3 to 4 day short-term charcoal measurements (38.6 Bq m -3, GSD=2.9) to overestimate the geometric mean of the 5- month alpha track measurements (69.4 Bq m -3, GSD=2.8) that were collected throughout the year in 391 residences. Windows were kept closed the day before sampling until the end of the short-term test. Measurement devices were positioned one meter from the other detector in rooms of high occupancy (e.g., bedroom), but the floor levels of where these measurements were obtained was not available. The median long-term readings were larger than the median short-term measurements regardless of month of placement. The percentage of homes with long-term measurements exceeding 148 Bq m -3 was much smaller (22%) than the percentage of homes with annual measurements equal to or above 148 Bq m -3 in this study. The larger geometric mean of the long-term radon concentrations compared to the geometric mean of the short-term radon concentrations may be attributed to poor ventilation conditions during the 5-month sampling period.

82 64 These measurement differences may be also due to the poor performance of the charcoal detectors in relative humidity conditions exceeding 80 percent, which have been observed in this region. A single weeklong basement short-term measurement to classify correctly its corresponding basement annual measurement as below or equal to or above the EPA s action level of 148 Bq m -3 was 83 percent. This rate improved considerably to 92 percent at a reference level of 74 Bq m -3 that was previously used by Klotz (32) and White (38). The effectiveness of the short-term test to correctly indicate when additional radon testing was warranted (i.e., sensitivity) was very strong (88%) and became much stronger at a lower reference value of 74 Bq m -3 (98%). In making these comparisons, the false negative rates are especially important as they are a measure of how often the short-term test incorrectly indicates that further radon testing is unnecessary, based on an annual measurement. At the action level of 148 Bq m - 3, the false negative was high (12%), but dropped considerably (2%) at a 74 Bq m -3 reference level. This is critical for national policy making insofar as to provide evidence for the need to reduce the current action level from 148 Bq m -3 to 74 Bq m -3. Based on estimates made by the National Research Council s BEIR VI Committee (4), about one-third of cancers attributable to radon could be prevented by lowering radon concentrations below the action level of 148 Bq m -3 nationwide. To reduce radon-related lung cancer deaths in the U.S. by half, the Committee further estimated that radon concentrations in all homes in the U.S. could not exceed 74 Bq m -3. Decisions made based on these short-term test results at the current action level would incorrectly indicate to individuals taking these tests that further radon testing is not needed 12 percent of the time and therefore, contribute to poor decision making and no action to continue to evaluate the radon potential in one s home. The findings from this study provide evidence of a very low likelihood of obtaining a false negative result from a single short-term test when the reference level is lowered to 74 Bq m -3. The findings

83 65 from this study are not representative of the usual screening measurement(s) which is often a single, very short-term (i.e., two-day), placement in any season of side-by-side charcoal canister detectors and consequently, may result in larger false negative rates for these typical short-term measurements. There may be less certainty of obtaining an annual measurement below 148 Bq m - 3 when the short-term measurement is near this level. In this study, the likelihood of an annual measurement being equal to or above the action level when the short-term measurement was between 111 and 148 Bq m -3 was 40 percent. This certainty increased to 50 percent near the 74 Bq m -3 reference level (i.e., between 55.5 and 74 Bq m -3 ), although there were only four measurements within this range. It was also determined that if a short-term test value is below 122 Bq m -3, the individual taking the test would be more than 95 percent confident that the annual test was less than 148 Bq m -3 and if the decision to pursue additional action was based on a single short-term test, further radon testing would not be needed. If the short-term test value was between 122 and 214 Bq m - 3, however, the individual would be less than 95 percent confident that the annual test was less than 148 Bq m -3 and additional radon testing would be advised. This conclusive short-term test value would vary for radon measurements collected in other seasons, in other locations within a residence, and for longer-term tests. The common significant housing factors that were suggestive of influencing the basement winter short-term and the basement annual radon concentrations were the presence of central conditioning, the presence of a clothes dryer in the basement, and the presence of a sump in the basement. The foundation wall material of the basement was the only significant factor to have an impact on the absolute difference between both measurements and a significant factor for influencing the basement annual radon concentrations. The paired differences between the short-term and annual measurements were smaller in homes with a poured concrete foundation wall in the basement compared to homes with a concrete block foundation wall.

84 66 An earlier study in Iowa by Field et al. (28) examined the effect of housing factors on screening measurements in 582 households. Field and colleagues found houses with higher radon screening concentrations to be positively and significantly associated with a crawl space in the basement and presence of a non-poured concrete wall in unfinished basements, which were not significantly associated with short-term measurements in this study. The presence of a crawl space was not significantly different from radon concentrations in residences without one (p=0.17). Having a crawl space, however, was significantly associated with larger radon concentrations when the placement of the radon detector was considered. Homes with crawl spaces were significantly associated with larger radon concentrations in unfinished basements compared to homes without a crawl space and in main floor rooms compared to homes without a crawl space (p=0.004 and p=0.0003, respectively). The significant factors affecting annual average radon concentrations as part of follow-up testing for the Iowa Radon Lung Cancer Study (26) when accounting for type of house and floor where the radon measurements took place, included an unfinished basement; lack of an insulated ceiling; presence of a crawl space; year of home construction; location of lowest home level relative to the ground; and presence of a toilet, bathtub, shower, washing machine, or hot tub in the basement. This study, however, did not find the number of major plumbing penetrations in the basement (e.g., toilet, washing machine) to be significantly associated with annual radon concentrations. Another survey conducted in the Midwestern area of Minnesota (33), found other distinct factors not considered in this study to significantly influence annual average radon measurements. These included average hours of snowfall, exposure to wind, heatingventilation system changes, basement expansions, and other major structural additions. Limitations A limitation of the study is the lack of climate-related data to assess the influence of climate conditions on local geological conditions (e.g., soil porosity) and therefore, its effect on the variability in the agreement between winter short-term and

85 67 annual radon concentrations. It should be noted, however, that the findings from this study are most generalizable to residences in the Midwestern region of the United States with high average screening radon concentrations where similar climate patterns and construction practices are found. Strengths This study explores the relationship between radon short-term and yearlong measurements in the lowest livable level in a robust data set of measurements that included a stringent quality assurance/quality control plan. It was carried out in a manner to minimize the potential of measurement errors by a number of approaches. The first approach was to test residences for radon in the winter months in the Midwestern region of the U.S. when closed house conditions are generally maintained. The possibility of failure to assure closed house conditions during the other months, primarily because of the potential for warmer weather outdoors and the potential for window opening, precluded placement of short-term measurements during non-heating seasonal months. Previous studies that compared short-term radon measurements to long-term measurements on the same floor did not incorporate methods that maximized the reliability (e.g., accuracy, precision) of the short-term radon test (30-32). For example, the E-PERMs used in this study are not affected by humidity (58) and provide a true integrated mean radon measurement as opposed to charcoal canisters that weights the measured radon concentration toward the radon concentrations in the home at the end of the measurement period (20). In addition, the 7 to 10 day duration of the screening measurement period minimized the effect of fluctuations in radon concentrations due to weather-related events. Conclusions The data generated from this study provide insight into how often a single shortterm radon test appropriately classified the advised long-term annual test with regard to being below or equal to or above the EPA s radon action level of 148 Bq m -3 as as well at

86 68 lower reference levels. The basement winter short-term tests predicted correctly, when additional measurements of radon were recommended 88 percent of the time based on the action level of 148 Bq m -3. Its performance improved greatly at a lower reference level of 74 Bq m -3 (98%). The false negative rate of how often the short-term test incorrectly indicated that further radon testing was unnecessary based on an annual measurement was 12 percent at the action level of 148 Bq m -3, but dropped considerably to 2 percent at a 74 Bq m -3 reference level. This study was performed in a state with the highest mean radon concentrations and the greatest percentage of screening radon measurements above the EPA s action level of 148 Bq m -3 compared to any other state surveyed in the U.S. This study has the potential to significantly influence public health policy concerning radon testing protocols, specifically the need to re-assess the EPA s current radon mitigation guidance level of 148 Bq m -3 by providing evidence of how basement winter short-term radon measurements approximate basement annual radon measurements at a lower reference level of 74 Bq m -3. Additional studies are needed to evaluate whether the significant housing factors influencing the temporal radon variation in our study affect temporal radon variation in other regions outside of the Midwestern area of the United States.

87 69 CHAPTER III TEMPORAL AND SPATIAL VARIATION ASSOCIATED WITH RESIDENTIAL AIRBORNE RADON MEASUREMENTS Abstract This study investigated the temporal and spatial variability between basement winter short-term and annual radon measurements obtained on upper floors (i.e., nonbasement) of the home. The ability of the basement winter short-term measurements to predict upper floor annual radon concentrations was evaluated as well as housing factors and occupant practices that may affect this predictive value. Electret ion chamber (shortterm) and alpha track (annual) upper floor radon measurements were collected in residences between 1992 and 1997 as part of the Iowa Radon Lung Cancer Study. The geometric mean of the basement winter short-term radon concentrations was double the geometric mean of the upper floor annual radon concentrations. About 60 percent of the basement winter short-term and 30 percent of the upper floor annual radon measurements were equal to or above the EPA s radon action level of 148 Bq m -3. This study found that individuals would be falsely overestimating their potential exposure to radon half the time at the current radon action level of 148 Bq m -3 based on false positive diagnostic indicators comparing the basement short-term tests to upper floor annual radon tests. If the reference radon level is reduced to 74 Bq m -3, individuals would incorrectly overestimate their potential radon exposure much more frequently (80 percent of the time). The significant common factors that were suggestive of influencing the upper floor annual and basement short-term radon concentrations were the presence of central air conditioning and the presence of a sump in the basement. The presence of a sump was the only factor that was significantly associated with a larger absolute difference between the basement short-term and the first floor annual radon concentrations.

88 70 Introduction Indoor residential measurements of radon (Rn-222) gas are often obtained using short-term radon measurements on the lowest livable level. The short-term measurement period typically ranges from two to 90 days. In the United States, the testing period usually lasts for two to five days based on the need for rapid results to complete real estate transactions. Short-term radon tests, collected under closed house conditions, are recommended initially to become familiar with a home s potential to have elevated radon concentrations. A single measurement of this type, frequently obtained in the basement where residential occupancy is often limited (64), is commonly used to approximate radon concentrations in the upper floors of the home so as to estimate an individual s yearly average radon exposure. Long-term testing, however, lasts more than 90 days (18) with an optimal measurement period of one year to account for seasonal radon variations. As opposed to a short-term radon test, long-term radon detectors are placed in living areas of the home, usually in a non-basement first floor, that usually have higher occupancy than the areas selected for short-term tests (18). Several investigators have reported overall radon exposure was approximated more closely by a first floor radon measurement rather than a measurement of the basement radon concentration (65,66). Estimates of radon exposure can vary significantly by placement location between floor levels within a residence (27,32). The variation of these concentrations between floors can limit the ability of both a short-term and long-term radon measurement collected in the basement to predict the long-term radon concentration on an upper floor of the home. Examining factors that can influence the level of agreement between shortand long-term radon measurements is helpful in understanding the predictive ability (i.e., predictor performance) of a short-term test to characterize the distribution of longer-term measurements on another level within a residence.

89 71 We previously found that the geometric mean of basement winter short-term radon concentrations (199 Bq m -3 ) slightly overestimated the geometric mean of basement annual concentrations (181 Bq m -3 ) in 158 residences as part of the Iowa Radon Lung Cancer Study (Chapter II). About 60 percent of both the basement short-term and basement annual measurements were equal to or above the EPA s radon action level of 148 Bq m -3. It was also previously found that the significant common factors suggestive of influencing the basement short-term and basement annual concentrations were the presence of central conditioning, presence of a clothes dryer in the basement, and presence of a sump in the basement. The foundation wall material of the basement was the only significant factor that was indicative of influencing the difference between the short-term and annual measurements as well as a significant factor for influencing the basement annual radon concentrations. The primary aim of this chapter was to examine the temporal and spatial variability between the lowest livable short-term and upper floor (i.e., non-basement) annual radon measurements in the home. Secondary aims of the study were to evaluate the ability of short-term radon measurements, obtained on the lowest livable area of a home, to predict the upper floor annual radon concentrations, and to assess the effect of a priori selected housing factors and occupant practices on the lowest livable short-term and upper floor radon measurements as well as their differences. Methods Sample selection Radon measurement data for this study were obtained from the Iowa Radon Lung Cancer Study (IRLCS) (51). The study criteria included being a resident of Iowa for at least 20 years, female, aged 40 to 84 years, residence in one s current home for at least twenty consecutive years or more, and no reports of making modifications to one s home based on previous radon testing. There were 413 cases and

90 controls included in the IRLCS. Participants from the IRLCS who met all the inclusion criteria and completed lowest livable winter short-term, lowest livable annual, and upper floor annual measurements of radon were eligible for this study. A questionnaire was mailed to each study participant to complete as time permitted before a home visit was conducted for the IRLCS. It collected information about heating, ventilation, and air conditions (HVAC) systems, smoking history, weatherization, characteristics of the home, ventilation patterns, and occupant practices that could influence concentrations of radon (Appendices B and C). Radon measurements An on-site residential radon assessment survey was also conducted. Field study workers drew a home floor plan for each study home and recorded location of detector placement within the home and room and date of placement. The onsite measurement of radon was collected by placing at least one Radtrak Alpha Track Detector (ATD) (Landauer, Inc., Glenwood, Illinois) on each level (i.e., location and number of room placements was weighted by where participants spent the most time), in current and historical bedrooms, and in work area(s) within the home (e.g., office) (Appendix D). Field study workers retrieved the detectors after one year of exposure and compared the original placement location based on the floor plan to the pick-up location within the home to assess whether the detector was in the original location where it was placed. At least one E-PERM (Electret-Passive Environmental Radon Monitors) (Rad Elec, Inc., Frederick, Maryland) was placed in a subset of the homes during the winter months alongside the lowest livable level alpha track detector. The lowest livable level E- PERM detectors were left in the home for 4 to 7 days. All E-PERM placement locations were adjusted for background gamma radiation using a calibrated Ludlum Measurements, Inc. (Sweetwater, Texas) Model 19 micro roentgen (μr) Meter. Data analysis Since the placement protocol of alpha track detectors resulted in a range of detectors being placed in each residence, additional summary parameters, distinct from Chapter II, were evaluated: 1) the annual average radon concentration of all

91 73 alpha track measurements on the non-basement first floor and 2) the average of the bedroom annual and living room annual alpha track radon measurements. Annual radon measurements in these locations were selected based on an earlier study by Field et al. of higher rates of occupancy for participants in the IRLCS on the first floor and in common living areas such as the bedroom (64). The bedroom and living room annual radon measurements were averaged to compare the results to the pooled analyses of seven North American case-control studies, including the IRLCS, that deployed radon detectors in living room and bedroom areas where participants spent the most time (14). These case-control studies examined the risk of lung cancer from protracted residential radon exposure. Descriptive statistics (e.g., geometric mean, geometric standard deviation) were computed for radon measurements by location as well as by differences between the short-term and the upper floor measurements. The absolute differences between measurements were evaluated to assess the magnitude of the tests differences. The absolute difference was examined so as to avoid including negative differences in the case a lowest livable short-term measurement was larger than its respective upper floor annual radon measurement. The distribution of each measurement and their differences was evaluated using the Shapiro-Wilks test to assess normality. If the test results supported normality, the distributions were described with an arithmetic mean and standard deviation. In the cases when the data did not follow a normal distribution, the data were transformed using a natural ln (log base e) and the ln-transformed values were tested to verify they were normally distributed prior to performing parametric statistics. If the ln-transformed values followed a normal distribution, they were described by a geometric mean and geometric standard deviation. The following analyses were conducted using the normalized data. A Student s paired t-test was performed to evaluate whether significant differences existed between the lowest livable short-term and the upper floor annual

92 74 radon concentrations as well as between the annual first floor and the average of the bedroom and living room annual radon concentrations. Paired differences between the radon measurements were examined to assess normality using Shapiro-Wilks test (an assumption of the Student s paired t-test). Radon measurements were paired for the t-test based on the site identification number where they were placed. The data were lntransformed in cases when they did not follow a normal distribution. A Shapiro-Wilks test was performed on the ln-transformed data to verify they were normally distributed. Data generated from the IRLCS was used to compute a temporally- and spatiallylinked cumulative radon exposure for each participant. The cumulative radon exposure assessment was based on the annual average radon concentrations on each floor of the residence, in the current bedroom, and in work area(s) within the home (e.g., office) weighted by the percent of time that the participants spent in these areas as well as outside the home (e.g., outdoors, inside other buildings, out of state on vacation). The outdoor radon exposure was estimated from the local average of the kriged grid values, which were mapped from annual average outdoor radon concentrations at 111 geographically dispersed sites throughout Iowa. The radon exposure inside other buildings was estimated based on the radon concentrations in bedrooms and inside other buildings for 100 women in Minnesota. The radon concentrations inside other buildings from the study in Minnesota were approximately 0.5 times the first floor annual radon concentrations in the IRLCS. The IRCLS estimated the radon concentrations inside other buildings for the participants in Iowa based on this relationship. Additional information concerning the estimation of outdoor radon exposure and exposure inside other buildings is presented elsewhere (55). A range of 5 to 19 years was selected as the period of interest to examine for the cumulative radon exposure assessment. The IRLCS investigators were not interested in a participant s radon exposure five years prior to lung cancer diagnosis (i.e., five-year period is considered the minimum latency period for lung cancer) (55), but were

93 75 interested in the protracted cumulative exposure period 5 to 19 years prior to diagnosis. The 5 to 19 year exposure assessment period was used to both eliminate the need to impute any radon measurement data and since this period prior to the development of lung cancer appeared to be a critical period of exposure for lung cancer development in the miner-based radon studies (55). If a 5 to 30 year exposure period would have been used, radon measurement data for 42 percent of the study homes would have required imputation. The time window of 5 to 19 years was a retrospective period for which a cumulative radon exposure was determined before a cancer diagnosis was made for cases and before controls were enrolled in the study. The cumulative radon exposure was computed by summing the duration of time spent in each area multiplied by the radon concentration in that area. This calculation used the working level, which is characterized as the concentration of the decay products of radon in one liter of air from the emission of million electron volts (MeV) of potential alpha radiation (4). The working level translates to 100 pci L -1 (3,700 Bq m -3 ) each of Rn-222, Po-218, Bi-214, and Po-214. For a given month, the cumulative radon exposure, denoted by the working level month (WLM), is equivalent to one working level (100 pci L -1 ) for a working month (170 hours = 8.5 hours /day 5days/week 4 weeks) (Equation 1). The mobility-linked WLM exposure for year y was computed as follows: Equation 1. Working level cumulative radon exposure assessment for year y WLM y = λ h ly r l l λ = assumed equilibrium ratio of 50 percent h ly = total hours spent at location l during the yth year prior to enrollment

94 76 r l = radon concentration (pci L -1 ) at location l The cumulative radon exposure over a year was computed and divided by the product of the working level and the working month to obtain the WLM and summed for the years of exposure over the 5 to 19 time frame to obtain a WLM( 5-19) for each participant. A traditional working month was applied rather than an all-inclusive month (672 hours=24 hours /day 7days/week 4 weeks) to enable the IRLCS investigators to compare the results to other studies that used this same working month (86,87,88) cumulative measure of exposure. The WLM took into account the equilibrium ratio that estimates for a given amount of radon in a room s air space, the amount of decay products generated from the decay of radon that are actually present in the air space of the room and available for inhalation (19). This equilibrium ratio varies greatly between homes and depends on air exchange rates in the room and the size and amount of aerosols in the air. An equilibrium ratio of 0.50 was used as it is generally applied by the EPA based on studies conducted in the United States (19). An equilibrium ratio of 0.50 indicates that for a given amount of radon in a room s air space, half of the decay products of radon will deposit on room surfaces, thereby reducing the amount of the decay products of radon that are available for inhalation, and the other half will available for inhalation. Pearson s correlation coefficients were computed to determine the strength of the linear relationship between the WLM cumulative radon exposure estimates and: 1) the lowest livable short-term; 2) the lowest livable annual; 3) the first floor annual; and 4) the average of the bedroom and living room annual radon measurements. These comparisons were of interest to assess the tendency of the cumulative radon exposure estimates to vary with each specified radon measurement. If a high correlation was found between the cumulative radon exposure estimate and a specific radon measurement, it would provide evidence that as the cumulative radon exposure increases, this particular radon

95 77 measurement would increase at the same rate. For instance, if a larger correlation coefficient were found between the WLM radon exposure estimate and an upper floor annual radon concentration than the correlation coefficient between the WLM exposure estimate and the lowest livable radon concentration, this could be an indicator that a participant s actual radon exposure can be more closely approximated by the upper floor radon measurement compared to a lower floor radon measurement. The ability of the lowest livable short-term radon measurement to classify the upper floor annual measurement with respect to being below or equal to or above the EPA s radon action level of 148 Bq m -3 was used as an indication of the lowest livable short-term measurement s capability to categorize the annual radon concentrations in living areas where participants spent most of their time. Although the EPA recommends that short-term testing be performed in the lowest livable level when rapid results are needed, long-term testing is also recommended in living areas where occupancy is greatest (19). The radon action level, therefore, can be applied to assist with decisionmaking for further radon testing at any test location within a residence. This would provide information of how decision-making to pursue additional radon testing is appropriately made based on a radon measurement that approximates more closely the radon concentration for which occupants are actually exposed. These diagnostic rates provide an indication of the potential for misclassification of radon exposure and its effect on generating precise radon exposure estimates. This study has the potential to significantly influence public health policy in regard to exposure surrogate measures, specifically in regard to public policy to encourage testing in the home for radon in living areas where occupancy is greatest. Although numerous studies have found lower radon concentrations on upper floors compared to radon concentrations in the lower portion of a building (27,32), radon concentrations below 148 Bq m -3 also carries some risk for cancer (18). The short-term measurements were

96 78 classified by various reference level cutoffs according to rates previously described in Chapter II. The ability of the basement winter short-term radon measurement to predict the upper floor annual radon measurement (i.e., predictor performance) using simple linear regression was also evaluated. Model assumptions for simple linear regression were previously discussed in Chapter II. Pearson s correlation coefficients were computed to determine the strength of the linear relationship between the basement short-term and upper floor annual radon measurements. To examine whether the errors (i.e., residuals) from this model were normally distributed, the Shapiro-Wilks test was performed. If the errors did not follow a standard normal distribution, the measurements were lntransformed. Normality of the ln-transformed data was further verified using the Shapiro- Wilks test. All regression analyses were conducted using the normalized data. To check for constant variance of the model s errors (i.e., homoscedasticity), the residuals (i.e., differences between sample values and fitted values) versus the model s predicted responses were plotted and their spread was evaluated. Multiple regression models were also evaluated. These included: 1) the prediction of the first floor annual radon measurements given certain housing factors and occupant practices; 2) the prediction of the average of the bedroom and living room annual radon measurements given certain housing factors and occupant practices; 3) the prediction of the first floor annual radon measurements given certain housing factors and occupant practices as well as the lowest livable short-term radon measurements; 4) the prediction of the average of the bedroom and living room annual radon measurements given certain housing factors and occupant practices as well as the lowest livable short-term radon measurements; 5) the prediction of the absolute difference between the lowest livable short-term and first floor annual radon measurement given certain housing factors and occupant practices; 6) the prediction of the absolute difference between the lowest livable short-term and average of the bedroom and living room annual radon measurement given

97 79 certain housing factors and occupant practices; and 7) the prediction of the signed difference (comprised of positive and negative values) between the lowest livable shortterm and average of the bedroom and living room annual radon measurements given certain housing factors and occupant practices. This predictive model can be further explained by examining how much variability occurs with different occupant practices and the presence of housing factors and their effect on each type of radon measurement as well as their difference. Based on previous studies, important predictors of temporal radon variability have included type of heating system, presence of crawl space, fireplace usage, foundation wall material, presence of toilet, shower, washing machine or hot tub in the basement, presence of a sump pump, and proportion of first level below ground (26). Additional factors that were associated with home air exchange rates from the impact of occupant behaviors on the upper floors where occupancy is greatest were considered. These included the frequency of opening windows (months per year) in a non-basement area, presence of window air conditioning in a non-basement area, and the participant s status as a current smoker (e.g., opening doors and windows when smoking). A backward variable selection process was used to generate final regression models with predictor variables being retained given p-values less than 0.20 (26). Analyses were performed using SAS version 9.2 (SAS Institute, Inc., Cary, North Carolina). The methods for the quality assurance performed for this study and sources of error associated with the radon detectors can be found elsewhere (Chapter II and Appendix H). Results All lowest livable level measurements in this study were obtained in the basement. Seven hundred seventy two annual measurements (772 in the basement, 772 in

98 80 the first floor above the basement, 772 in the bedroom, and 772 in the living room) and 18 basement winter short-term radon measurements were excluded due to a lack of a corresponding basement short-term or upper floor annual measurement, respectively. The survey resulted in the inclusion of 158 basement winter short-term radon measurements, 158 first floor, 158 bedroom, and 158 living room annual radon measurements for each residence. Sample characteristics After the basement short-term and its respective upper floor annual radon measurements were ln-transformed, the Shapiro-Wilks test supported normality of the distributions for the basement short-term (p=0.14), the first floor annual (p=0.32), and the average of the bedroom and living room annual measurements (referred to annual average bed/living room hereafter) (p=0.52). As stated previously, the basement winter short-term radon concentrations ranged from 42 to 1,331 Bq m -3 (Table11, Figure 10). Its geometric mean (199 Bq m -3, GSD=2.0) was found to be twice the geometric mean (104 Bq m -3, GSD=2.0) of the first floor annual as well the annual average bed/living room concentrations. Both upper floor radon measurements had identical geometric means. In fact, as seen in the box plot in Figure 10, the first floor annual and the annual average bed/living room measurements had almost identical distributions. This is not surprising because 90 percent of the living rooms and 75 percent of the bedrooms were located on the first floor (not presented in Table 11). Forty-four percent of the living room measurements were larger than their matching bedroom measurements and the same percentage of the bedroom measurements were larger than the living room measurements. Eighteen percent of the living room and bedroom measurements had the same test value. Almost all of the basement short-term radon measurements (96%) were larger than their matching first floor annual or annual average bed/living room measurements. More than half (61%) of the annual average bed/living room radon measurements were larger than the first floor annual radon measurements and seven percent had the same test

99 81 value. Sixty-three percent of the basement short-term measurements and less than half (30%) of each of the upper floor annual measurements were equal to or larger than the EPA s radon action level of 148 Bq m -3. Not surprising, the arithmetic and geometric means of the absolute differences between the basement short-term and the annual first floor measurements and the means of the absolute differences between the basement short-term and the annual average bed/living room measurements were very similar (Shapiro-Wilks test did not support normality of the ln- and non-ln-transformed distributions of the absolute differences). Table 11. Characteristics of basement short-term and upper floor annual radon measurements (Bq m -3 ) as well as their paired differences from 158 residences in Iowa Basement First Floor Average of Bedroom and Living Room Winter Short-Term Annual Annual Characteristic Electret Ion Chamber Test Alpha Track Test Alpha Track Test Arithmetic Mean (SD) Geometric Mean (GSD) Range 25th Percentile Median 75th Percentile 257 (217) 135 (121) 136 (121) 199 (2.0) 104 (2.0) 104 (2.0) % 29% 28% 148 Bq m Difference Basement - First Floor Difference Basement - Bedroom/Living Room Arithmetic Mean (SD) Geometric Mean (GSD) Range 25 th Percentile Median 75 th Percentile 122 (146) 122 (145) 71 (2.0) 70 (3.3)

100 82 Figure 10. Box plot of radon concentration by location Agreement between measurements The basement winter short-term radon concentrations significantly differed from the first floor annual (p<0.0001) as well as from the average of the bedroom and living annual (p<0.0001) radon concentrations after all measurements were ln-transformed. The first floor annual and annual average bed/living room radon measurements were not significantly different (p=0.81 when both measurements were ln-transformed, although the Shapiro-Wilks test did not support the normality of their paired differences). The classification of measurements, by location, by test type, and by reference level cutoffs is presented in Table 12 and Figure 11. The agreement between the basement short-term test and both upper level annual tests were almost identical so only comparisons with the first floor annual radon measurements were presented. The

101 83 basement winter short-term measurement was able to correctly classify its corresponding first floor annual measurement as both larger than or equal to or less than the action level of 148 Bq m -3 at a rate of 65 percent. This rate increased to 75 percent when the reference level was lowered to 74 Bq m -3. Given the first floor annual radon concentration was equal to or larger than the action level of 148 Bq m -3, the probability of the basement short-term radon concentration also being equal to or larger than this level (i.e., sensitivity) was 98 percent and increased to 100 percent for a 74 Bq m -3 reference level. The false positive rate when the basement short-term test is equal to or larger than 148 Bq m -3 given the first floor annual test was less than this level occurred half the time and increased considerably to 82 percent for a lower reference level of 74 Bq m -3. The probability of the short-term test being less than 148 Bq m -3 when the first floor annual measurement was equal to or larger than 148 Bq m -3 (i.e., false negative rate) was two percent. At lower reference levels of 111 and 74 Bq m -3, the false negative rates dropped to zero. Table 12. Number of basement short-term and first floor annual radon measurements that were equal to or above or less than the respective reference level First floor annual alpha track measurement (Bq m -3 ) Basement winter short-term electret ion chamber measurement (Bq m -3 ) < < < <

102 84 Figure 11. Percentages of diagnostic indicators comparing basement short-term radon tests to first floor annual radon tests by reference level % Negative Predictive Value False Positive (1 - Specificity) Positive Predictive Value False Negative (1 - Sensitivity) 74 Bq m-3 (2 pci L-1) 111 Bq m-3 (3 pci L-1) 148 Bq m-3 (4 pci L-1) Rn-222 conc. reference level cutoff The working level month (WLM 5-19 ) cumulative radon exposure assessment, a measure of cumulative radon exposure obtained from the IRLCS, was very strongly correlated with both the annual first floor (r=0.98, p<0.0001) and the annual average bed/living room radon concentrations (r=0.98, p<0.0001) (Figure 12). Although still strongly correlated, there was a weaker correlation of the WLM 5-19 cumulative radon exposure assessment with the basement winter short-term (r=0.75, p<0.0001) and the basement annual radon concentrations (r=0.73, p<0.0001). After the removal of the observation with the largest basement annual radon concentration which deviated

103 85 strongly from the other radon concentrations, the correlation between the basement annual concentration and the WLM 5-19 became much stronger (r=0.86, p<0.0001) (Figure 13). Figure 12. Scatter plots of working level months versus basement and upper level radon measurements

104 86 Figure 13. Scatter plots of working level months versus basement winter short-term, basement annual, first floor annual, and average of bedroom and living room annual radon concentrations after removing a basement annual radon concentration outlier Simple linear regression Model assumptions were addressed in Appendix I. The best fitted line for predicting first floor annual radon concentrations based on basement winter short-term radon concentrations was almost identical to the best fitted line for predicting the annual average bed/living room based on basement short-term concentrations after all measurements were ln-transformed. Based on these findings, only the regression of the first floor model will be presented (Figure 14). A strong linear relationship was noted (r = 0.82, p<0.0001) between the basement short-term and the first floor annual radon concentrations. The r-squared value indicated that 67 percent of the variability in the first floor annual radon measurements could be explained by the basement short-term radon measurements. As indicated by the best fitting ling equation, a one log unit increase in the basement winter short-term radon concentration would yield a 0.83 log unit increase in the first floor annual radon concentration (i.e., a 10 log unit

105 87 increase in the basement winter short-term radon concentration would yield a 8.3 log unit increase in the first floor annual radon concentration). The prediction equation to relate the non-transformed basement short-term and first floor annual measurements given the prediction equations based on ln-transformed measurements with a constant e (approximate value of 2.72) is: first floor annual radon concentration = e 0.23 (basement short-term radon concentration) 0.83 To check for non-constant variance (another assumption of linear regression), the residuals versus the predicted responses were also plotted in Figure 14. The constant variance assumption appears reasonable in the ln-transformed data, as the spread is similar throughout the distribution.

106 88 Figure 14. Scatter plot of ln(basement winter short-term Rn-222 conc.) versus ln(first floor annual Rn-222 conc.) and its residuals Backward stepwise regression In addition to the housing factors (i.e., percentage of the basement underground, presence of a forced air heating system) that were reported previously (Table 5 from Chapter II), other housing factors and occupant practices were added only to the multiple regression models for the prediction of first floor annual radon concentrations and the prediction of differences between the first floor annual and basement short-term radon concentrations (Table 13). These additional factors were

107 89 duration of opening windows per year in a non-basement area, whether the participant was a current smoker, presence of window air conditioning in a non-basement area, and square feet of house. Findings for the models incorporating the average of the bed/living room annual measurements are not presented as they were almost identical to the findings from the models comprised of the first floor annual radon measurements. Overall, participants opened windows in a non-basement area on average for about six months (SD=2.6) during the year. Among the residences sampled, half of them had two floors above the ground. Approximately 14 percent of the participants were a current smoker and about 61 percent had a window air conditioner located in a non-basement area. The mean square feet of the homes was 2,056 (SD=682) with a median of 1,969. Table 13. Summary of additional regression predictor parameters from 158 homes in Iowa Characteristic N (%) Mean (SD), Median Duration of opening windows per year in non-basement 126 (79.7) 6.3 (2.6), 6.0 Missing 32 (20.3) Number of above ground floors 158 (100) 1.5 (0.55), 2.0 One 76 (48.1) Two 78 (49.4) Three 4 (2.5) Participant was a current smoker - Yes 22 (13.9) No 136 (86.1) Presence of window air conditioning in non-basement - Yes 97 (61.4) No 61 (38.6) Square feet of house 158 (100) 2056 (682), 1969

108 90 The factors that were positively associated with first floor annual radon concentrations were the case status of the participant, presence of central air conditioning, and presence of a sump in the basement. The factors that were negatively associated with first floor annual radon concentrations were having a poured concrete foundation wall in the basement, number of floors above ground, participant s status as a current smoker, presence of window air conditioning in a non-basement area, and volume of the basement (Table 14). The final model s R-squared value showed that 45 percent of the variability in the first floor annual radon concentrations can be explained by these factors. With all the other variables fixed in the model except for the variable of interest, a one floor increase in the number of floors above the ground would yield a 33 percent decrease in the first floor annual radon concentration. When adjusting for the number of above ground floors and volume of the basement and all the other variables are fixed in the model, the first floor annual radon concentrations were 21 percent smaller in homes with a poured concrete foundation wall in the basement compared to homes with a concrete block foundation wall, 30 percent larger in homes with a case participant compared to homes with a control participant, 86 percent larger in homes with central air conditioning compared to homes without central air conditioning, 55 percent smaller in homes when the participant was a current smoker versus homes when the participant was a non-smoker, 72 percent larger in homes with a sump pump compared to homes without one, and 27 percent smaller in homes with window air conditioning in a non-basement area compared to homes without window air conditioning.

109 91 Table 14. Backward regression parameters (final model) for predicting ln(first floor annual radon concentration) Characteristic VIF a β b Coeff. (SE c ) Student's t-test (p) Basement foundation wall material 1.0 Concrete block Ref. d - Poured concrete (0.15) -1.5 (0.13) Case/control status of participant 1.2 Case 0.26 (0.17) 1.6 (0.13) Control Ref. - Number of above ground floors (0.14) -2.9 (0.01) Presence of central air conditioning 1.3 Yes 0.62 (0.18) 3.5 (0.001) No Ref. d - Participant was a current smoker 1.1 Yes (0.24) -3.4 (0.001) No Ref. - Presence of sump in basement 1.1 Yes 0.54 (0.17) 3.2 (0.002) No Ref. - Presence of window air conditioning in non-basement 1.4 Yes (0.16) -1.9 (0.06) No Ref. - Volume of basement E-5 (3.3E-5) -1.3 (0.19) R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Reference category The factors that were positively associated with first floor annual radon concentrations given the basement short-term radon concentrations and housing/occupant factors were the basement short-term radon concentration, case status of the participant, number of major plumbing penetrations in the basement, presence of central air conditioning, presence of a clothes dryer in the basement, and presence of a forced air heating system (Table 15). The factors that were negatively associated were having a

110 92 poured concrete foundation wall in the basement, participant s status as a current smoker, presence of window air conditioning in a non-basement area, and square feet of the home. The final model s R-squared value showed that 80 percent of the variability in the first floor annual radon concentrations can be explained by these factors. With all the other variables fixed in the model except for the variable of interest, a one plumbing penetration increase in the number of major plumbing penetrations in the basement would yield a nine percent increase in the first floor annual radon concentration. With all the other variables fixed in the model except for the variable of interest and adjusting for the basement short-term radon concentrations, the number of major plumbing penetrations in the basement, and the square feet in the basement, the first floor annual radon concentrations were 17 percent smaller in homes with a poured concrete foundation wall in the basement compared to homes with a concrete block foundation wall, 39 percent larger in homes with a case participant compared to homes with a control participant, 32 percent larger in homes with central air conditioning compared to homes without central air conditioning, 27 percent larger in homes with a clothes dryer in the basement compared to homes without a dryer in the basement, 35 percent smaller in homes when the participant was a current smoker versus homes when the participant was a non-smoker, 43 percent larger in homes with a forced air heating system compared to homes without one, and 17 percent smaller in homes with window air conditioning in a non-basement area compared to homes without window air conditioning in a non-basement area. The final model s R-squared value showed that 80 percent of the variability in the first floor annual radon concentrations can be explained by these factors. In comparison to the previous model when not adjusting for the short-term radon measurement (Table 14), housing and occupant factors alone explained much less of the variability in the first floor annual measurements (R 2 =0.45).

111 93 Table 15. Backward regression parameters (final model) for predicting ln(first floor annual radon conc.) given ln(basement short-term radon concentration) and housing/occupant factors Characteristic VIF a β b Coeff. (SE c ) Student's t-test (p) Ln(basement short-term radon conc.) (0.06) 13 (<0.0001) Basement foundation wall material 1.1 Concrete block Ref. d - Poured concrete (0.10) -1.9 (0.06) Case/control status of participant 1.3 Case 0.33 (0.11) 3.1 (0.003) Control Ref. - Number of major plumbing penetrations in basement (0.06) 1.7 (0.09) Presence of central air conditioning 1.5 Yes 0.28 (0.12) 2.4 (0.02) No Ref. - Presence of a clothes dryer in basement 1.5 Yes 0.24 (0.11) 2.1 (0.04) No Ref. - Participant was a current smoker 1.2 Yes (0.15) -2.9 (0.01) No Ref. - Presence of forced air heating system 1.3 Yes 0.36 (0.14) 2.5 (0.01) No Ref. - Presence of window air conditioning in non-basement 1.5 Yes (0.10) -1.8 (0.07) No Ref. - Square feet of home E-4 (7.6E-5) -1.7 (0.09) R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Reference category

112 94 As discussed previously in Chapter II, the factors that were positively associated with basement winter short-term radon concentrations were the presence of central air conditioning and presence of a sump in the basement (Table 16). The factors that were negatively associated with basement winter short-term radon concentrations was the presence of a clothes dryer in the basement. The final model s R-squared value showed that 17 percent of the variability in the basement winter short-term radon concentrations can be explained by all these factors. With all the other variables adjusted for in the model except for the variable of interest, the basement short-term concentration was 62 percent larger in homes with central conditioning versus homes without central air conditioning, 32 percent smaller in homes with a clothes dryer in the basement compared to those without a dryer in the basement, and 63 percent larger in homes with a sump in the basement versus homes without a sump.

113 95 Table 16. Backward regression parameters (final model) for predicting ln(basement winter short-term radon conc.) Covariate VIF a β Coeff. b (SE c ) Student's t-test (p) Percent change d Presence of central air conditioning 1.02 Yes 0.48 (0.17) 2.9 (0.01) 62 No Ref. e - Presence of clothes dryer in basement 1.14 Yes (0.16) -2.5 (0.01) 32 No Ref. - Presence of sump in basement 1.01 Yes 0.49 (0.16) 3.0 (0.003) 63 No Ref. - R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Given all other variables were fixed in model, e Reference category The only factor that was positively associated with the absolute difference between the basement short-term and the first floor annual radon measurements was the presence of a sump in the basement. The factors that were negatively associated with this absolute difference were the case status of the participant, number of floors above ground, number of major plumbing penetrations in the basement, presence of a clothes dryer in the basement, presence of a forced air heating system, and presence of a sump in the basement (Table 17). The final model s R-squared value showed that 27 percent of the variability in the paired differences between the basement short-term and the first floor annual radon concentrations can be explained by these factors. With all the other variables fixed in the model except for the variable of interest, a one plumbing penetration increase in the number of major plumbing penetrations in the

114 96 basement and a one floor increase in the number of floors above the ground would yield a 24 percent decrease and a 47 percent decrease in the absolute difference, respectively. With all the other variables fixed in the model except for the variable of interest and adjusting for the number of floors above ground and number of major plumbing penetrations in the basement, the absolute paired difference was 41 percent smaller in homes with a case participant compared to homes with a control participant, 39 percent smaller in homes with a clothes dryer in the basement compared to homes without a dryer in the basement, 49 percent smaller in homes with a forced air heating system compared to homes without this system, and 73 percent larger in homes with a sump compared to homes without one.

115 97 Table 17. Backward regression parameters (final model) for predicting the absolute difference, ln( basement winter short-term radon conc. - first floor annual radon conc. ) Characteristic VIF a β b Coeff. (SE c ) Student's t-test (p) Case/control status of participant 1.1 Case (0.27) -2.0 (1.1) Control Ref. - Number of floors above ground (0.20) -3.1 (0.003) Number of major plumbing penetrations in basement (0.15) -1.9 (0.06) Presence of a clothes dryer in basement 1.2 Yes (0.28) -1.7 (0.09) No Ref. - Presence of forced air heating system 1.2 Yes (0.37) -1.8 (0.07) No Ref. - Presence of sump in basement 1.1 Yes 0.55 (0.29) 1.9 (0.06) No Ref. - R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Reference category The variables that were associated with a smaller signed difference between the basement short-term and the first floor annual radon measurements were the case status of the participant (p=0.01), number of major plumbing penetrations in the basement (p=0.09), presence of central air conditioning (p=0.02), presence of a clothes dryer in the basement (p=0.04), and presence of a forced air heating system (p=0.01) (Table 18). The basement short-term measurement underestimated the first floor annual radon measurement (i.e., negative difference), for instance, in homes with a concrete block foundation wall, central air conditioning, a clothes dryer in the basement, a forced air heating system, no window air conditioning in a non-basement area, when the participant

116 98 was a case and a non-smoker given the square feet of the home was less than 2,249 for any number of plumbing penetrations in the basement. The final model s R-squared value showed that 39 percent of the variability in the signed differences can be explained by these factors. The variance inflation factors (VIF) for each variable in all the final regression models were close to one, indicating that the predictor variables were not strongly correlated (i.e., multicollinearity).

117 99 Table 18. Backward regression parameters (final model) for predicting the signed difference, ln(basement short-term radon conc.) ln(first floor annual radon conc.) Characteristic VIF a β b Coeff. (SE c ) Student's t-test (p) Foundation wall material of basement 1.1 Concrete block Ref. d Poured concrete 0.18 (0.10) 1.7 (0.09) Case/control status of participant 1.3 Case (0.11) -2.9 (0.01) Control Ref. - Number of major plumbing penetrations in basement (0.06) -1.7 (0.09) Presence of central air conditioning 1.4 Yes (0.12) -2.4 (0.02) No Ref. - Presence of a clothes dryer in basement 1.3 Yes (0.12) -2.1 (0.04) No Ref. - Participant is a current smoker 1.1 Yes 0.38 (0.15) 2.4 (0.02) No Ref. - Presence of a forced air heating system Yes (0.15) -2.7 (0.01) No Ref. - Presence of window air conditioning in non-basement 1.5 Yes 0.18 (0.11) 1.7 (0.10) No Ref. - Square feet of house (0.0001) 1.9 (0.07) R Adjusted R a Variance inflation factor, b Slope parameter estimate, c Standard error, d Reference category The quality assurance performed for this study and sources of error associated with the radon detectors can be found elsewhere (Chapter II and Appendix H, respectively).

118 100 Discussion Among the 158 residences tested for radon in this study, the geometric mean of the basement winter short-term radon concentrations was double the geometric mean of the first floor annual as well as the average of the bedroom and living room annual radon measurements. Almost all of the basement short-term radon measurements (96%) were larger than their matching first floor annual or annual average bed/living room measurements. However, the findings demonstrate that the basement short-term measurement cannot be relied on as a worse case radon measurement, as it is intended, since even a small percentage of upper floor radon measurements, where radon concentrations are expected to be much lower than the basement measurements (60), were larger than the basement radon measurement. The EPA state radon survey for Iowa (23) observed a similar trend where the basement two-day screening radon measurements obtained during the winter was twice the non-basement first floor two-day screening measurements. Another study of basement winter screening measurements and first floor annual alpha track radon measurements in New York also exhibited similar findings for all foundations and heating types (37). A New Jersey survey by Klotz and colleagues (32), which was known to have very low radon concentrations, found the geometric mean of four-day winter basement radon concentrations to differ to a larger degree (i.e., three times as large) from the annual concentrations in non-basement living areas, usually a bedroom. The percentage of first floor winter screening radon concentrations above the EPA s radon action level of 148 Bq m -3 in the EPA state survey for Iowa (23) was twice (60%) the percentage of first floor annual radon concentrations equal to or above the action level (29%) for this study. There was better agreement between the percentages above this guidance level for basement radon concentrations for this study and the EPA state survey for Iowa.

119 101 A measure of radon that has often been compared to a screening test as a surrogate marker of radon exposure is an annual average radon concentration in a living area. The basement winter short-term radon measurements predicted correctly, when additional testing of radon was recommended, 98 percent of the time based on the EPA s action level of 148 Bq m -3 and increased to 100 percent for a 74 Bq m -3 reference level. A previous study (Chapter II), however, found a lower rate (88 %) when the basement winter short-term tests correctly classified its corresponding annual test on the same floor level at the 148 Bq m- 3. This rate of correct classification (i.e., sensitivity) was almost equivalent between both studies when the reference level was lowered to 74 Bq m -3. In making comparisons between the basement short-term and first floor annual radon measurements, the false positive and false negative rates are especially important, as they are a measure of the potential exposure misclassification when the basement short-term radon measurement incorrectly categorizes the first floor annual radon measurement. The probability of the basement short-term test as less than 148 Bq m -3 when the first floor annual measurement was equal to or larger than 148 Bq m -3 (i.e., false negative rate) was very low (2%). This rate of misclassification drops to zero at a lower reference level of 74 Bq m -3. In contrast, the likelihood of misclassifying the basement short-term test as being equal to or larger than 148 Bq m -3 when the first floor annual test was actually less than 148 Bq m -3 (i.e., false positive rate) was much larger (50%) and increased considerably to 82 percent when the reference level was lowered to 74 Bq m- 3. Compared to the false negative misclassification rates, the false positive misclassification rates were considerably larger at both reference levels. The false positive misclassification rates indicate that an individual s potential exposure to radon is greater than the individual s actual radon exposure 50 percent and 82 percent of the time at the reference levels of 148 and 74 Bq m -3, respectively. In other words, individuals would be falsely overestimating their potential exposure to radon half the time at the

120 102 current radon action level of 148 Bq m -3. If the reference level is reduced to 74 Bq m -3, however, individuals would incorrectly overestimate their potential radon exposure much more frequently (80 percent of the time). The findings from this study provide evidence that short-term radon tests collected on the lowest livable level within a residence should not be relied on exclusively to generate radon exposure assessments as they were found to incorrectly classify annual radon measurements at high rates in an upper floor area. This study has the potential to significantly influence public health policy in regard to exposure surrogate measures, specifically in regard to public policy to encourage testing in the home for radon in living areas and not relying solely on a screening measurement to estimate the concentration of radon in the entire home. A previous analysis of the retrospective temporal and spatial mobility of participants from the Iowa Radon Lung Cancer Study (64) found participants who lived in two- and three-level residences to report spending the majority of their occupancy in the first story with average percent of time spent of 93 percent and 73 percent, respectively. The observed average percent of time spent in the basement was only 6.8 percent for two-level homes and 3.5 percent for houses with three levels. An average of 7.4 hours (SD=1.2) per day was reported to be spent in the bedroom. These findings indicate that the non-basement first floor is where the participants tended to spend most of the occupancy. The cumulative radon exposure assessments (WLM) were very strongly correlated with the upper floor radon measurements (r=0.98, p<0.0001) and although still strongly correlated, less correlation was found between the WLM estimates and the basement short-term (r=0.78, p<0.0001) and the basement annual (r=0.86, p<0.0001) radon measurements. These findings indicate that a participant s actual radon exposure can be more closely approximated by the upper floor radon measurement compared to a lower floor radon measurement. Given the sample of study participants were all women,

121 103 these findings are not surprising since women tend to spend more time indoors. The correlation between the cumulative radon exposure estimates and the upper floor radon concentrations may not be as strong for men or other individuals who spend less time per day in the home. These findings are supported by a study by Harley et al. (65) that found first floor radon concentrations to be a better estimate of personal radon exposure than basement radon measurements. The authors compared personal and home measurements in 52 homes in Illinois. Participants wore personal radon monitors for the same duration as the radon detectors deployed in each level of the home (average time was 24 ± 6 days). The average ratio of the personal radon monitor measurements to the first floor stationary radon measurements was 0.71 (SD=0.03) and were strongly correlated (R 2 =0.85). The average ratio of personal to basement measurements, however, was 0.22 (SD=0.04) and were weakly correlated (R 2 =0.31). The wide differences encountered between first floor and basement radon concentrations show the importance of not relying exclusively on one radon measurement in the lowest livable level to generate radon exposure estimates. For this study, the first floor annual and annual average bed/living room radon measurements were not significantly different (p=0.81) so the personal radon estimates would not vary much more when compared to the average of the bedroom and living room annual measurements versus those differences noted between the personal radon exposure estimates and the first floor annual tests. In a study by Klotz and colleagues (32) that was previously described, the false negative rate of the basement winter charcoal measurements to classify living area annual measurements was almost identical to the false negative rates for this study at both reference levels. It should be noted, though, that in the Klotz et al. study, only eight of 709 living area alpha track measurements had radon concentrations larger than 150 Bq m - 3 (95% of the living area annual measurements were less than 74 Bq m -3 ). Among the 111 basement canister measurements larger than 148 Bq m -3, there were 103 living area alpha

122 104 track measurements less than 148 Bq m -3 (93% false positive rate). This rate is much larger than the false positive rate of 50 percent found for this study. The false positive rates dropped to 77 percent when a lower reference level of 74 Bq m -3 was substituted for 148 Bq m -3. The false positive rate at this reference level is much closer to the rate of 82 percent that was observed for this study despite the fact that Klotz. et al. employed a different false positive denominator using the total number of basement screening tests equal to or larger than the reference level whereas this study used the total number of first floor annual measurements less than the reference level. White (38) compared the living area annual average measurement to a two-day winter screening test in the lowest livable level with reference to the EPA s action level of 148 Bq m -3 for measurements available from 1,449 residences participating in the U.S. State Residential Radon Survey. The false negative error rate of four percent for the number of cases when the screening measurement was less than 148 Bq m -3, but the annual average living area test was equal to or larger than 148 Bq m -3 was very close to the false negative rate for this study (2%). In contrast, the false positive rate of the likelihood of the screening measurement being equal or larger than 148 Bq m -3 given the annual average living area measurement was less than 148 Bq m -3 was four percent and much lower than the false positive rate for this study (50%). White compared the lowest livable winter short-term radon tests to an annual measurement that averaged the annual radon measurement on each floor level of the home whereas this study made short-term comparisons to the first floor annual measurement. The common significant housing factors that were suggestive of influencing larger first floor annual and basement winter short-term radon concentrations were the presence of central conditioning and the presence of a sump in the basement. A study by Fisher et al. (27) investigated spatial variation of annual mean radon concentrations in 918 Iowan homes as part of the Iowa Radon Lung Cancer Study where multiple yearlong measurements were obtained on each house level. Fisher et al. identified larger first

123 105 floor annual radon measurements were significantly associated with larger basement annual radon concentrations, a forced air heating system, and a larger basement volume. This study also observed larger first floor annual radon concentrations to be significantly associated with a forced air heating system and larger basement radon concentrations although short-term radon measurements were collected whereas the volume of the basement was found not to be significant. Smaller first floor radon concentrations were significantly associated with the presence of a window air conditioning in a nonbasement area although this variable was not significant in the study by Fisher et al. The only factor that was significantly associated with a larger absolute difference between the basement short-term and the first floor annual radon measurements was the presence of a sump in the basement. In the previously described New Jersey study by Klotz et al. (32), the ratio between the geometric mean of the basement winter screening radon tests and the geometric mean of the living area annual tests decreased from 3.1 in all residences to 2.4 among residences with only a forced air heating system. This study found homes with a forced air heating system to have smaller absolute paired differences than homes without this type of heating system. A previously described survey in New York (37) identified the geometric mean ratio of the basement winter screening radon measurement to the first floor annual radon measurement to be greatest in houses with a brick foundation (ratio=3.6), although only six houses had this foundation type, and a hot water heating system (ratio=4.1). Whereas, the lowest ratios were observed in these New York residences with a fieldstone foundation (ratio=2.2) and a wood/coal heating system (ratio=1.9). Foundation wall material, however, was not significantly associated with the absolute paired differences between the basement short-term and first floor annual radon measurements for this study. Limitations A limitation of the study is the lack of climate-related data to assess the influence of climate conditions on local geological conditions (e.g., soil porosity) and

124 106 therefore, its effect on the variability in the agreement between basement winter shortterm and upper floor annual radon concentrations. It should be noted, however, that the findings from this study are most generalizable to residences in the Midwestern region of the United States with high average screening radon concentrations where similar climate patterns and construction practices are found. Strengths This study explores the relationship between basement short-term and annual radon concentrations in the upper floor within a residence level in a robust data set of measurements that included a stringent quality assurance/quality control plan. It was carried out in a manner to minimize the potential of measurement errors by a number of approaches. The first approach was to test residences for radon in the winter months in the Midwestern region of the U.S. when closed house conditions are generally maintained. The possibility of failure to assure closed house conditions during the other months, primarily because of the potential for warmer weather outdoors and the potential for window opening, precluded placement of short-term measurements during nonheating seasonal months. Previous studies that compared short-term radon measurements to long-term measurements on the same floor did not incorporate methods that maximized the reliability (e.g., accuracy, precision) of the short-term radon test (30-32). For example, the E-PERMs used in this study are not affected by humidity (58) and provide a true integrated mean radon measurement as opposed to charcoal canisters that weights the measured radon concentration toward the radon concentrations in the home at the end of the measurement period (20). In addition, the 7 to 10 day duration of the screening measurement period minimized the effect of fluctuations in radon concentrations due to weather-related events.

125 107 Conclusions The data generated from this study provide insight into the degree of misclassification of radon exposure that would occur if the participant s exposure assessment was based solely on a short-term radon measurement. This study found that when comparing basement winter short-term radon measurements to first floor annual radon tests, individuals falsely overestimated their potential exposure to radon half the time at the current radon action level of 148 Bq m -3 based on false positive diagnostic indicators. At a lower reference level of 74 Bq m -3, however, individuals incorrectly overestimated their potential radon exposure much more frequently (80 percent of the time). This study has the potential to significantly influence public health policy in regard to exposure surrogate measures, specifically to encourage testing in the home for radon in living areas and not relying solely on a screening measurement to estimate the concentration of radon in the entire home. Additional studies are needed to evaluate whether the significant housing and occupant factors identified in this study are important for influencing the temporal and spatial radon variation in other geographical regions outside of the Midwestern area of the United States.

126 108 CHAPTER IV COMPARATIVE SURVEY OF OUTDOOR, RESIDENTIAL, AND WORKPLACE RADON CONCENTRATIONS Abstract This study investigated occupational radon concentrations in above ground workplaces in Missouri and compared them to above ground radon concentrations in nearby homes and outdoor locations. Data were also evaluated to determine the utility of non-basement first floor annual home and outdoor concentrations to predict above ground radon concentrations at a nearby workplace. Employees at county extension agencies, schools, and businesses were recruited to participate in the study between 2002 and Both indoor alpha track radon gas detectors and outdoor radon detectors were deployed for a one-year exposure period. Annual above ground workplace radon concentrations were not statistically different from nearby annual above ground home radon concentrations, but were statistically different from nearby annual outdoor radon concentrations. Home and workplace above ground radon concentrations were poorly correlated and an even weaker correlation was detected between workplace and outdoor radon concentrations. Annual above ground workplace radon concentrations were similar to annual radon concentrations in the upper floor of homes. Overall, annual non-basement first floor home and outdoor radon concentrations were poor predictors of the annual above ground occurrence of radon at a nearby workplace. The study provides insights into the potential for above-ground radon exposures in the workplace and the potential agreement between workplace and residential radon concentrations. More systematic nationwide surveys are needed to assess above ground workplace radon to generate more accurate estimates of overall radon exposures.

127 109 Introduction Radon (Rn-222) is a radioactive gas formed from the decay of Radium-226, a member of the Uranium-238 decay chain. It is the predominant source of ionizing radiation in the United States (1) and protracted exposure to radon s decay products has been associated with an increased risk of lung cancer. Radon is the second leading cause of lung cancer after tobacco smoking and the primary cause of lung cancer among individuals who have never smoked (67). Numerous studies have described the distribution of radon concentrations that occur in the home (68-70). Because overall estimates of radon exposures should incorporate both home and away-from-home exposures, the distribution of radon concentrations and duration of exposure in other settings deserves further assessment. Since individuals spend part of their day outside the home (e.g., workplaces, outdoors), radon concentrations occurring in these other locations may contribute substantially to an individual s overall radon exposure. The National Human Activity Pattern Survey estimates that individuals (comprised of various age groups) spend on average, 11 percent of their day at another indoor location and 8 percent outdoors (71). The Iowa Radon Lung Cancer Study found adult women, between the ages of 50 and 59, spent the most time (~17%) in another building compared to any other age group. Women in the 20 to 29 age group occupied the most time outside (~10%) compared to all other age groups with this percentage decreasing as age increases (64). Data regarding radon exposure in above ground workplaces, especially in the United States, are generally lacking (48,72). In addition, even fewer studies have been performed that provide comparative analyses of radon concentrations in distinct environments (e.g., home, workplace, outdoors) within close proximity to each other. This study seeks to provide insights into the potential for radon concentrations in a sample of above ground workplaces. The primary aim of this study is to characterize radon concentrations in above ground workplaces in Missouri and compare them to

128 110 above ground radon concentrations in nearby homes as well as to nearby outdoor locations. A secondary aim is to determine whether outdoor or residential radon concentrations have utility in predicting radon concentrations in nearby workplaces. This survey was carried out as an independent study of radon concentrations in Missouri following the Missouri Residential Radon Study (70) and the Iowa Radon Lung Cancer Study (IRLCS) (59). These case-control epidemiologic studies evaluated the association between protracted residential exposure to radon and its decay products and incidence of lung cancer among women. This study was conducted initially to obtain pilot data on the distributions of home and workplace radon concentrations that was needed to assess the impact of various radon concentration distributions. In order to obtain insights into the differences between home and workplace radon concentrations, this study was necessary to provide insights for imputing workplace radon concentrations based on residential radon measurements. Missouri was selected for the study to collect preliminary information on the association between outdoor, workplace, and home radon concentrations that were needed for the potential pooling of glass-based retrospective radon detector results (73,74) obtained during the Iowa and Missouri Residential Radon Studies. The majority of Missouri is classified by the EPA as having Zone 2 radon potential indicating that the state has overall moderate radon potential (75). Zone 2 areas have a predicted average indoor lowest livable radon screening level between 74 and 148 Bq m -3. However, northwest Missouri has higher radon potential (Zone 1) (76). This higher radon potential area includes higher uranium content along the Missouri River Valley, areas of windblown glacial deposits east of the Missouri River Valley, and black shales closer to Kansas City (76). This study was conducted in a state where previous EPA surveys estimated that 21 percent of basement and 9 percent of first floor screening radon measurements exceed the EPA s action level of 148 Bq m -3 (23).

129 111 Part of the evaluation of radon exposure in the IRLCS involved the assessment of a participant s exposure when in another building (e.g., workplace) and linking it with the residential and outdoor radon concentrations to determine a spatial- and time-weighted average radon exposure for each participant. Workplace radon measurements were not collected for the IRLCS, but were imputed based on a previous study performed in Iowa and Minnesota that suggested workplace radon concentrations were approximately 50 percent of first floor residential radon concentrations (53,59 unpublished data). In order to perform a future pooled analyses of the Iowa and Missouri residential radon studies, an estimate of the ratio of residential to workplace radon concentrations was needed for Missouri. Methods Sample selection Initially, science teachers at public schools and private businesses in Missouri were randomly selected from lists of schools and businesses between 2002 and 2004 as a contact list for possible participation in the radon survey. Because of the poor response rate from schools (~5%) and businesses (~8%), the sampling strategy was abandoned. The workplaces that were recruited from the previous sampling strategy for this study were retained in the total sample of workplaces surveyed. In 2004, the strategy changed to recruiting county extension agencies (76). It was believed that the state county extension agents, especially the environmental specialists, would be more likely to agree to assist with the study since they were more experienced with issues related to environmental health. Phone numbers for environmental health specialists and other county extension workers in Missouri were obtained from the website (76) operated by the University of Missouri that maintains listings for Missouri s county extensions.

130 112 Radon measurements Respondents who agreed by phone to take part in the study were mailed two Radtrak alpha track radon gas detectors (Landauer, Inc., Glenwood, Illinois), instructions for placing them in the home and workplace (Appendix J), and a survey questionnaire (Appendix K) between 2002 and The instructions informed the county extension agent about how to place one alpha track detector (ATD) on the first floor, non-basement, of the home where they were residing and one ATD on the first floor, non-basement, of the extension agency where they worked. Placement instructions that accompanied the detectors adhered to EPA placement protocols (60). The extension agent was requested to place the detector in a high occupancy room, non-bathroom, 50.8 centimeters off the floor and at least five inches from other items. The Landauer ATD was selected to provide a yearlong radon concentration. It was selected based on its past performance of good accuracy and precision compared to other commercially available alpha track detectors (56) as well as its low cost, mailability, and simplicity of use. The outdoor radon detector (ORD) was mailed a few days after the ATDs were sent out for matched time periods and had detailed instructions (Appendix L) for proper placement. The outdoor detector was similar to the detector used in an outdoor radon survey of Iowa and Minnesota (53). The unit was constructed by Dr. Daniel Steck from the Minnesota Radon Project using Lantrak alpha track plastic (Landauer, Inc., Glenwood, Illinois) enclosed in a plastic PVC (polyvinyl chloride) pipe that was covered with Tyvek (DuPont, Wilmington, DE) material at the bottom allowing radon (not water or insects) to pass through. The pipe was attached to the top of a one meter long fiberglass pole. Participants were instructed to select a space to mount the detector module at least one to two meters from an occupied building, while avoiding areas near significant air flowing out of the building (e.g., near vents or appliance exhausts). Once an acceptable placement location was selected, the extension agent was instructed to drive the post into the ground approximately 23 centimeters deep and to leave the post for a one-year exposure period.

131 113 Survey questionnaire Once the devices were positioned, the participants were instructed to record the placement date both on the side of the detector and on the survey questionnaire. The questionnaire was used to collect information on the following: detector placement date, addresses of placement; detector identification number, participant s confirmation of location of detector s placement (i.e., on the first floor), distance from the workplace or the outdoor placement to the home detector placement, main activity or function of the workplace, and the total floors of the workplace. Participants were then instructed to return the questionnaire in a postage paid preaddressed envelope to the study office once completed. Sample handling After the one-year measurement period, participants were contacted and instructed to place the detector in a sealed envelope, record the ending date of the exposure period, and return the ATDs to the project office. Once received, the data collection forms were reviewed to determine whether they were completed with respect to the start and end dates of the exposure period. The detectors were wrapped in aluminum foil and sent, using overnight delivery, to Landauer via Federal Express. Sample detector security and chain of control documentation was maintained by requesting a signature when it was received by Landauer. Overnight delivery also minimized the chance that detectors could have picked up additional radon exposure while in transit. Alpha tracks were counted by Landauer using computer-assisted image analysis equipment, which was needed to calculate the average annual radon concentration (78). A similar process was followed for the return of the ORDs to the Minnesota Radon Project. Geocoding of home and workplace addresses In each of these counties, the home, outdoor, and workplace addresses were geocoded using ArcMap 8.3 (ESRI, Inc., Redlands, CA) and a 2009 TIGER/line file of Missouri as the reference address locator. The output was verified to ensure that the geocoded point s location matched that of the home or workplace. Some locations were not geocoded to the home, such as those with

132 114 post office boxes or rural routes, but were instead geocoded to the center of the applicable zip code. Data analysis Analyses were performed using R version (R Foundation for Statistical Computing). Descriptive statistics (e.g., geometric mean, geometric standard deviation) were computed for the indoor (home and workplace) and outdoor radon concentrations. Comparisons were made between home and workplace radon concentrations and by type of workplace. A Student s paired t-test was performed to evaluate whether significant differences exist between radon concentrations from each of the three types of environments. Paired differences between the radon measurements from the sampling environments were examined to assess normality using Shapiro- Wilks test (an assumption of the Student s paired t-test). Home and workplace and outdoor and workplace radon measurements were paired for the t-test based on the study participant identification number deploying the detectors. The data were transformed using a natural ln (log base e) in cases when they did not follow a standard normal distribution. A Shapiro-Wilks test was performed on the ln-transformed data to verify they were normally distributed. The following analyses were conducted using the normalized data. The utility for predicting the annual occurrence of radon in nearby workplaces was evaluated using simple linear regression. The prediction of annual workplace radon concentrations given nearby annual residential radon concentrations was compared to the regression of predicting annual workplace radon concentrations based on nearby annual residential radon concentrations as well as the distance between the home and work detector locations. The addition of this distance variable was evaluated to determine whether a change in the distance between home and workplace locations had any impact on the value in predicting workplace radon concentrations given annual residential radon concentrations.

133 115 For these regression models, Pearson s correlation coefficients were computed to determine the strength of the linear relationship between the annual workplace radon concentrations and the predictor variables. To examine whether the errors (i.e., residuals) from these models were normally distributed, the Shapiro-Wilks test was performed. If the errors did not follow a standard normal distribution, the measurement data were lntransformed. Normality of the ln-transformed data was further verified using the Shapiro- Wilks test. All regression analyses were conducted using the normalized data. To check for constant variance of the models errors (i.e., homoscedasticity), the residuals (i.e., differences between sample values and fitted values) versus each model s predicted responses were plotted and their spread was evaluated. Additional methods for diagnostics of the models are presented in Appendix M. Quality assurance This study followed the EPA s guidelines for quality assurance to measure radon including: 1) quality assurance objectives for measurement data; 2) calibration procedures and frequency; 3) analytical procedures; 4) data validation and reporting; 4) internal quality control; 4) quality assurance audits and reports; 5) preventative maintenance of monitoring equipment; and 6) procedures to assess measurement precision, accuracy, and completeness (59-63). Duplicate detectors (10%) were treated identically to the primary device and positioned ten centimeters from the primary detector. The ATDs were exposed to known radon concentrations (i.e., spikes) (5%) in the EPA s National Air and Radiation Environmental Laboratory and the ORDs in the lab of the Minnesota Radon Project to test for detector accuracy and precision. In addition, field and laboratory control detectors (blanks) (5%) were deployed to detect possible radon exposure encountered in the field and in transit to the laboratory. Two detectors (one detector designated as the blank and the other as the primary detector) were sent to participants. They were instructed to leave the blank detector closed and store it in a low radon environment. After the exposure period ended, they were contacted to return both detectors. Duplicate, spiked, and blank detectors were labeled in the same

134 116 manner as primary devices to ensure identical processing by the laboratory. All data were double entered into a database and verified to prevent data entry errors. Results The response rate of workers at county extension agencies was 70 percent (108 agencies were contacted). Annual radon measurements were obtained for 85 residences of 107 ATDs distributed that included 11 not returned and 11 used for quality assurance (e.g., duplicates, controls), 82 workplaces of 106 ATDs distributed that included 14 not returned and 10 used for quality assurance, and 82 outdoor locations of 105 ORDs distributed that included 15 not returned and 8 used for quality assurance. All the participants who agreed to be in this study reported they had not tested their residence or that their current workplace was not tested previously for radon. Two home and one workplace radon measurements were excluded because it was documented that these detectors were not placed on the first floor. This resulted in the inclusion of 82 outdoor, 83 home, and 81 workplace radon measurements in the study. Sample characteristics Additional results are presented in Appendix M. Radon measurements were collected from 80 out of 114 Missouri counties with some counties (Boone, Camden, Greene, and Lincoln) having more than one set of radon tests performed. In other words, these counties had radon measurements from two distinct home and outdoor locations, and in some cases two distinct workplaces. Figure 15 illustrates the spatial distribution of the study participants identifying their workplace, residence, and outdoor locations. Two homes, one workplace, and one outdoor location had post office box addresses so they were geocoded to their applicable zip code. Graduated symbols for each type of environment tested distinguish areas with respect to the EPA s radon action level of 148 Bq m -3.

135 117 Among the counties with workplace and home radon measurements, the few workplaces and homes with radon concentrations equal to or larger than 148 Bq m -3 tended to cluster towards the lower part of the state. The outdoor radon measurements were collected in areas situated near the home or workplace. Not surprisingly, no outdoor radon concentrations exceeded the EPA s indoor radon action level. The few outdoor radon concentrations equal to or above 37 Bq m -3 were found mostly in counties in the central eastern and upper western part of the state. Figure 15. Annual radon concentrations (Bq m -3 ) by type of test environment across counties in Missouri

136 118 Radon detectors were placed in 2002 through 2004 during the months of September through February (Table 19). About 70 percent of the radon tests began in 2002 with a little more than half of the tests starting in the month of September (58%) followed by about 25 percent in December. Among the 81 workplaces surveyed, 37 (46%) only had a first floor, 26 (32%) had a second floor, 14 (17%) had a third floor, and four (5%) had a fourth floor (not described in table). The placement of the workplace detectors averaged about 4.3 kilometers (0.40 and 24 km) and outdoor detectors averaged 10 feet (less than 0.30 m to more than 152 m) from the home measurement site. Table 19. Summary of annual radon detector placement counts by type of environment and date Indoor Home Workplace Outdoor Placement Date N (%) N (%) N (%) Year (68.7) 58 (71.6) 57 (69.5) (28.9) 22 (27.1) 22 (26.8) (2.4) 1 (1.2) 3 (3.7) Month September 47 (56.6) 48 (59.2) 47 (57.3) October 7 (8.4) 7 (8.6) 8 (9.8) November 3 (3.6) 3 (3.7) 1 (1.2) December 22 (26.5) 20 (24.7) 22 (26.8) January 3 (3.6) 2 (2.5) 3 (3.7) February 1 (1.2) 1 (1.2) 1 (1.2)

137 119 The indoor (home and workplace) radon concentrations ranged from 7.4 to 400 Bq m -3, with an arithmetic mean of 71 (SD=67) (Table 20, Figures 16-17). In comparison to the outdoor concentrations, the mean and median of the indoor radon concentrations was about two times larger. The indoor and outdoor geometric means were 53 (GSD=2.2) and 25 (GSD=1.5), respectively, with the indoor geometric mean being closer to its median than its arithmetic mean as expected. The median of the home and work radon concentrations were both 52 Bq m -3 and ranged from 11 to 400 Bq m -3 and 33 to 333 Bq m -3, respectively. About 10 percent of the home and 10 percent of the workplace radon measurements were equal to or exceeded the EPA s radon action level. Table 20. Characteristics of radon concentrations (Bq m -3 ) by sampling environment Percentage Environment N Arithmetic mean (SD) Geometric mean (GSD) Median Range 148 Bq m -3 Indoor (67) 53 (2.2) Home (73) 54 (2.2) Workplace (60) 51 (2.2) Type of Workplace County Office (62) 57(2.0) Retail 5 43 (36) 33 (2.3) School 6 34 (24) 27 (2.1) Service 8 69 (72) 43 (3.0) Private (59) 38 (2.6) Public (60) 54 (2.1) Outdoor (13) 25 (1.5)

138 120 Figure 16. Box plots of radon concentration by type of sampling environment Figure 17. Box plots of radon concentration by type of workplace

139 121 Among the type of workplaces surveyed, a little more than 75 percent were county offices. Retail workplaces, including a grocery, a florist/gift shop, and an antique business, and schools each made up about 7 percent. Service workplaces, including an auto repair shop, a heating/cooling contractor, a church, insurance sales, a real estate office, and a motel, comprised the remaining 10 percent. There were no retail work areas or schools that were equal to or exceeded the action level although they only made up 5 and 6 percent of the workplaces surveyed, respectively. However, approximately 10 percent of the county offices and 13 percent of the service work areas were equal to or exceeded the EPA s action level. There was a slightly larger percentage (10%) of public workplaces equal to or exceeding the action level than the private workplaces (7%). Agreement between measurements According to the Shapiro Wilks test, the paired differences between the ln-transformed workplace and ln-transformed home measurements followed a standard normal distribution (p=0.20) (an assumption of the Student s paired t-test) as opposed to the non-transformed paired differences. This was also the case for the paired differences between the ln-transformed workplace and lntransformed outdoor measurements (p=0.57). The residential and workplace radon concentrations were compared and found to be not significantly different (p=0.33) while the outdoor radon concentrations were found to be significantly different with the home (p<0.0001) and work (p<0.0001) radon concentrations. Overall, indoor and outdoor radon concentrations were found to be significantly different (p<0.0001). Information concerning Student s paired t-test assumptions are presented in Appendix N. Unmatched data were not included in these analyses. The range of differences of radon concentrations between the sampling environments is displayed in Figure 18 for those participants (N=72) having all three types of measurements (home, workplace, and outdoor). There was a similar percentage (48%) of the home radon measurements greater than their matching workplace measurements and vice versa while two percent of the matching home and workplace

140 122 measurements had the same test value. Among the matching outdoor and workplace radon measurements, 86 percent had a larger workplace radon concentration. Figure 18. Range of differences between annual home, outdoor, and work radon concentrations (Bq m -3 ) by increasing work concentrations Bq m Work-Home Work-Outdoor Home-Outdoor -350 Participant Number Note: Only participants with all three types of measurements (home, workplace, and outdoor) (N=72) were included in the figure. Simple linear regression Simple linear regression was used to examine the utility of the annual home or outdoor radon concentrations (ln-transformed) to predict the annual radon concentration in a nearby workplace. After all the variables underwent a ln-

141 123 transformation, the best fitted lines for both models were generated and plotted in Figures 19 and 20. Information concerning model assumptions are presented in Appendix N. A weak linear relationship was noted between home and workplace radon concentrations (r = 0.22, p = 0.05) and even a weaker relationship between outdoor and workplace radon concentrations (r = 0.11, p = 0.34). The Shapiro-Wilks test supported normality for the residuals of both models predicting workplace radon concentrations based on nearby home (p = 0.98) or outdoor (p = 0.98) radon measurement (after all the variables were lntransformed). The r-squared (represented by lower case r-squared to indicate a model comprised of only one predictor variable) coefficient indicated that only five percent of the variability in the annual workplace radon measurements can be explained by the annual home radon measurements and even less (1%) by the annual outdoor radon measurements. As indicated by the best fitted line equation, a one percent increase in the annual average home as well as the annual average outdoor radon concentration would each yield a 0.22 log unit increase in the annual average workplace radon concentration. The prediction equations to relate the non-transformed measurements given the prediction equations based on ln-transformed measurements with a constant of e (approximate value of 2.72) are: I. work radon concentration = e 3.0 (home radon concentration) 0.22 II. work radon concentration = e 3.2 (outdoor radon concentration) 0.22 To check for non-constant variance, an assumption of linear regression, the residuals (i.e., differences between sample values and fitted values) versus the predicted responses were plotted after the data were ln-transformed. The constant variance assumption appears reasonable in the ln-transformed data for the two models, as the spread in the residuals is similar throughout the distribution. When the distance between the home and workplace detectors was added as another predictor variable in the regression model predicting the annual workplace radon concentrations given the annual home concentrations, the R-squared value increased only slightly from 0.05 to 0.06.

142 Figure 19. Scatter plot of annual ln(home Rn-222 conc.) versus annual ln(work Rn-222 conc.) and its residuals 124

143 125 Figure 20. Scatter plot of annual ln(outdoor Rn-222 conc.) versus annual ln(work Rn-222 conc.) and its residuals Quality assurance Twelve percent of the ATDs and 10 percent of the ORDs had a duplicate (i.e., side by side or collocated) detector placement. The mean coefficient of variation (COV) (SD / mean of duplicate measurements) for the duplicate ATDs and ORDs was about seven percent. Five percent of the ATDs were exposed to known Rn-

144 concentrations (spikes) in the EPA s National Air and Radiation Environmental Laboratory. A mean COV of 8.3 percent for an exposure at 74 Bq m -3, a COV of 7.5 percent at 148 Bq m -3, and 5.7 percent at 222 Bq m -3 was found. The mean absolute relative error for the spiked ATDs (( MV i TV i TV i ) 100 where MV i is the measured value for the ATD i and TV i is the target radon chamber value for ATD i) was 12.8 percent at 74 Bq m -3, 11.4 percent at 148 Bq m -3, and 8.5 percent at 222 Bq m -3. About 10 percent of the ORDs were exposed to known radon concentrations (spikes) in the radon chamber at the Minnesota Radon Project. There was no change in the calibration constant from 2002 to The mean COV for the spiked ORDs was 7 percent. At least five percent of the ATDs were designated as field controls (i.e., blanks). About two percent of the ORDs were designated as field control detectors and three percent as laboratory control detectors (blanks). The blank detectors exhibited no extraneous radon exposure above Landauer s lowest level of detection of 4 Bq m -3. The spiked detector measurements were within the mean absolute relative error of 25 percent, which meets the EPA s guidelines of detector performance. The duplicate detector measurements also met the EPA s suggested a priori 10 percent precision criterion. All data underwent double entry with follow-up verification for entry errors. Detailed information concerning quality assurance is presented elsewhere (59). Discussion The findings from this study are most representative to areas of the United States with moderate radon potential (i.e., predicted average indoor lowest livable radon screening level between 74 and 148 Bq m -3 ) and offices similar to those surveyed as far as size. Compared to the geometric mean of the annual radon concentration in this study of 83 residences (54 Bq m -3, GSD=2.2), the EPA state radon survey for Missouri (23) found a smaller geometric mean (40 Bq m -3, GSD=2.7) for screening radon

145 127 concentrations in non-basement areas within residences. With respect to the EPA s action level of 148 Bq m -3, this survey found about 10 percent of the annual home radon concentrations to be equal to or larger than this level. The EPA state radon survey found similar rates (i.e., 9.2%) of winter screening radon concentrations in non-basement areas above this guidance level. In comparison to this study s geometric mean of annual radon concentrations of 51 Bq m -3 (GSD=1.4) for 81 workplaces (7% were schools), a national survey of radon levels in U.S. public schools by the EPA from 1990 through 1991 school year (38) found the arithmetic mean of the 5-month radon measurements to be smaller (30 Bq m -3 ), although it was slightly larger than the geometric mean of 27 Bq m -3 (GSD=2.1) for only schools in this study. A larger average concentration would have been expected for the school survey compared to this study because the school survey measured radon in frequently occupied rooms with ground contact that were closer to the soil source than the deployment of radon detectors for this study (above ground first floor). This study, on the other hand, had larger radon concentrations, but the radon measurements were collected above ground. Compared to the percentage (10%) of workplaces equal to or above the EPA s action level for this study, there was a smaller percentage (1.5%) of the rooms in the EPA school survey that were equal to or above this action level. The difference between the two studies may also be affected by air exchange rates or other factors (e.g., radon source potential) between buildings in this study and the EPA s school survey. The geometric mean of the annual outdoor radon concentrations of 25 Bq m -3 (GSD=1.5) was slightly larger than the mean (22 Bq m -3 ) of the annual average outdoor radon concentrations in a nationwide ambient study by the EPA for Missouri (79). Only three electret ion chamber radon detectors were placed at the EPA s Environmental Radiation Ambient Monitoring System (ERAMS) station in each state for 90-day periods and combined to attain the annual average radon concentration. This study conducted

146 128 multiple outdoor radon measurements in various regions throughout the state as compared to the EPA survey that relied on one outdoor radon measurement in one location in each state. Annual average workplace radon concentrations were not statistically different from nearby annual average home radon concentrations (p=0.33). Both types of annual measurements had geometric means close to 50 Bq m -3 (GSD=2.2). There was a similar percentage (48%) of the home radon measurements greater than their matching workplace measurements and vice versa while two percent of the matching home and workplace measurements had the same test value. Based on radon test measurements in 100 homes in Minnesota, the Minnesota Radon Project (Steck unpublished study, 53) estimated the radon concentrations in other buildings (e.g., workplaces, schools), however, to be half of first floor bedroom radon concentrations for the Iowa Radon Lung Cancer Study. In contrast, studies by Gaidolfi in Italy (49) and Iyogi in Japan (51) found schools and workplaces exhibit larger annual average radon concentrations than the annual radon concentrations measured in dwellings. Gaidolfi et al. found the differences between the annual radon concentrations in schools and homes to be smaller when the floor level where the measurements took place were taken into account. Iyogi and colleagues also noted that the workplace radon concentrations were smaller based on active detector measurements during working hours when air conditioning was activated compared to non-working hours. In contrast, Poffijn et al. (46) found a greater percentage of dwellings (24%) that exceeded 200 Bq m -3 compared to schools (12%) or public buildings (10%) in a Belgian study covering 79 houses (45), 421 schools, and 36 other buildings, although the distribution of the radon measurements in homes and schools were similar (both had medians of 90 Bq m -3 ). Charcoal measurements were mostly taken only over a weekend during closed conditions. Similar findings were found in a survey in Finland consisting of 2-month home and workplace alpha track measurements during the spring (47). A poor

147 129 correlation was observed (r=0.18, p=0.04) between 333 workplace (detectors deployed in actual area of work) and 939 home (447 living room and 492 bedroom) measurements. The geometric mean radon concentration in the dwelling (68 Bq m -3 ) was more than three times larger than the concentration in the workplace (20 Bq m -3 ). A survey in New Mexico (United States) also found the median radon concentration in the dwelling (55.5 Bq m -3 ) to be three times larger than the median concentration on the first floor in offices (18.5 Bq m -3 ) (48). The lower office radon concentrations may be attributed to high air exchange rates often encountered in office buildings. Alpha track detectors were placed in 65 offices (number of measurements used for analyses not found) and 47 homes for an exposure period of three months in the same county. The radon concentrations across floors in office areas were not statistically different. In other words, basement radon concentrations were not found to be larger than upper floor radon concentrations. Radon concentrations can vary between floors if a building s ventilation system is not activated. The location where the radon measurements were collected within the residence (comprised of only 39 measurements) was not indicated. The larger radon concentrations in the home compared to first floor office radon concentrations may be likely if measurements were obtained in a floor level closer to the ground. In this study, the geometric mean of the annual radon concentration at a workplace or at a nearby residence was double the annual radon concentration at a nearby outdoor location. As part of the Iowa Radon Lung Cancer Study, annual average indoor and outdoor radon concentrations were measured in Iowa and Minnesota (53) and differed to a greater degree. The geometric mean annual concentrations in the bedrooms were slightly larger than three times higher than outdoor levels in Iowa and fivefold larger in Minnesota. Limitations The representativeness of the findings is generalizable to a limited sample composed predominately of county extension offices. Other types of workplaces may have different ventilation practices, building structures, and additional floors that

148 130 have the potential to affect the variability in annual average radon concentrations. Another limitation is the inability to take into account the effect of air circulation patterns on the annual average radon concentration in workplaces where ventilation usage varies by work hours (e.g., evenings, nights, weekends). Strengths This survey presented a unique opportunity to compare and characterize annual average above ground (i.e., non-basement) measurements of radon in three distinct environments within close proximity to each other. The study represents one of the largest surveys of long-term occupational above ground radon concentrations in the U.S. and the largest study of occupational radon concentrations ever performed in Missouri. An advantage of enrolling extension agency workers who were experienced with issues of environmental health, and therefore, were equipped with technical expertise, increased the likelihood of proper placement and return of radon detectors. For this reason, they were more likely to participate and may be a useful resource in future self-monitoring studies as well as the potential to be a cost-effective approach to test multiple environments within close proximity to each other. Findings from this study demonstrate that some above ground office areas exhibit similar radon concentrations to upper floors in homes. In fact, the percentage of the home radon measurements greater than their matching workplace measurements and vice versa were comparable as well as the percentage of the workplace and home radon concentrations equal to or larger than the EPA s radon action level of 148 Bq m -3. Based on these findings, radon concentrations in the non-basement first floor of homes and workplaces present a similar level of concern indicating that attention needs to also be placed on the measuring of radon in workplaces since individuals spend part of their day inside these other buildings as well. A weak correlation was noted between home and workplace radon concentrations and even a weaker relationship between outdoor and workplace radon concentrations. These results indicate that annual non-basement home

149 131 and outdoor radon concentrations can be poor surrogate measures of the annual level of radon in nearby above ground workplaces. Conclusions This study provides insights into the potential for radon concentrations in above ground workplaces and the potential agreement between workplace and residential radon concentrations in a sample of predominately county extension offices who were more likely to participate. This group may be a useful resource for future self-monitoring studies when obtaining a geographically diverse and regional sample population. More systematic nationwide surveys are needed to assess above ground workplace radon to generate more accurate estimates of overall radon exposures.

150 132 CHAPTER V CONCLUSIONS In Chapter II, Evaluation of agreement of time-integrated basement residential radon measurements and correctness of further radon testing indicators, we investigated the temporal variability between short-term and annual residential radon measurements obtained on the lowest livable level. Another objective was to evaluate the utility of predicting the annual radon concentration in the lowest livable level based on a shortterm radon measurement at the same location and the types of housing factors and occupant practices that affect this predictor performance. Basement electret ion chamber (short-term) and basement alpha track (annual) radon measurements were collected in residences between 1992 and 1997 as part of the Iowa Radon Lung Cancer Study. The data generated from this study provide insight into the degree of how often a single short-term radon test appropriately classified the advised long-term annual test with regard to being below or equal to or above the EPA s radon action level of 148 Bq m -3 as as well at lower reference levels. The basement winter short-term tests predicted correctly, when additional measurements of radon were recommended 88 percent of the time based on the action level of 148 Bq m -3. Its performance improved greatly at a lower reference level of 74 Bq m -3 (98%). The false negative rate of how often the short-term test incorrectly indicated that further radon testing was unnecessary based on an annual measurement was 12 percent at the action level of 148 Bq m -3, but dropped considerably to 2 percent at a 74 Bq m -3 reference level. This study was performed in a state with the highest mean radon concentrations and the greatest percentage of screening radon measurements above the EPA s action level of 148 Bq m -3 compared to any other state surveyed in the U.S. This study has the potential to significantly influence public health policy concerning radon testing protocols, specifically concerning the need to re-assess the EPA s current radon

151 133 mitigation guidance level of 148 Bq m -3 by providing evidence of how basement winter short-term radon measurements approximate basement annual radon measurements at a lower reference level of 74 Bq m -3. Additional studies are needed to evaluate whether the significant housing factors influencing the temporal radon variation in our study affect temporal radon variation in other regions outside of the Midwestern area of the United States. In Chapter III, Temporal and spatial variation associated with residential airborne radon measurements, we investigated the temporal and spatial variability between basement winter short-term and annual radon measurements collected in upper levels (i.e., non-basement) of the home. This examined the ability of the basement winter short-term measurements to predict upper floor annual radon concentrations and identify housing factors and occupant practices that may affect this predictive value. Basement electret ion chamber (short-term) and upper floor alpha track (annual) radon measurements were collected in residences between 1992 and 1997 as part of the Iowa Radon Lung Cancer Study. The geometric mean of the basement winter short-term radon concentrations was double the geometric mean of the upper floor annual radon concentrations. About 60 percent of the basement winter short-term and 30 percent of the upper floor annual radon measurements were equal to or exceeded the EPA s action level of 148 Bq m -3. This study found that when comparing basement winter short-term radon measurements to first floor annual radon tests, individuals falsely overestimated their potential exposure to radon half the time at the current radon action level of 148 Bq m -3 based on false positive diagnostic indicators. At a lower reference level of 74 Bq m -3, however, individuals incorrectly overestimated their potential radon exposure much more frequently (80 percent of the time). The significant common factors that were suggestive of influencing the upper floor annual and basement short-term radon concentrations were the presence of central air conditioning and the presence of a sump in the basement. The presence of a

152 134 sump was the only factor that was significantly associated with a larger absolute difference between the basement short-term and the first floor annual radon concentrations. These studies have the potential to significantly influence public health policy in regard to exposure surrogate measures, specifically in regard to public policy to encourage testing in the home for radon in living areas and not relying solely on a screening measurement to estimate the concentration of radon in the entire home. Additional studies are needed to evaluate whether the significant housing and occupant factors identified in this study are important for influencing the temporal and spatial radon variation in other geographical regions outside of the Midwestern area of the United States. In Chapter IV, Comparative survey of outdoor, residential, and workplace radon concentrations, we investigated occupational radon concentrations in above ground workplaces in Missouri and compared them to above ground radon concentrations in nearby homes and outdoor locations. Data were also evaluated to determine the utility of above ground annual home and outdoor concentrations to predict above ground radon concentrations at a nearby workplace. Employees at county extension agencies, schools, and businesses were recruited to participate in the study between 2002 and Both alpha track radon gas detectors and outdoor radon detector modules were deployed for a one-year exposure period. Annual above ground radon concentrations in the workplaces were not statistically different from annual above ground radon concentrations at a nearby home, but were statistically different from the annual above ground radon concentrations at a nearby outdoor location. Home and workplace above ground radon concentrations were poorly correlated and an even weaker correlation was detected between workplace and outdoor above ground radon concentrations.

153 135 Findings from this study demonstrate that office areas exhibit similar radon concentrations to upper floors in the home. Also, annual non-basement and outdoor radon concentrations were found to be poor predictors of the annual occurrence of radon in nearby workplaces. The study provides insights into the potential for above-ground radon exposures in the workplace and the potential agreement between workplace and residential radon concentrations. More systematic nationwide surveys are needed to assess above ground workplace radon to generate more accurate estimates of overall radon exposures.

154 136 APPENDIX A HUMAN SUBJECTS OFFICE/INSTITUTIONAL REVIEW BOARD APPROVAL FOR DISSERTATION STUDIES

155 137 APPENDIX B IOWA RADON LUNG CANCER STUDY (IRLCS) QUESTIONNAIRE ABOUT THE HOME (CHAPTERS II AND III)

156 138

157 139

158 140

159 141

160 142

161 143

162 144

163 145

164 146

165 147

166 148

167 149

168 150

169 151 APPENDIX C IRLCS QUESTIONNAIRE ABOUT THE PARTICIPANT (CHAPTERS II AND III)

170 152

171 153

172 154

173 155

174 156

175 157

176 158

177 159

178 160

179 161

180 162

181 163

182 164

183 165

184 166

185 167

186 168 APPENDIX D IRLCS PLACEMENT OF ALPHA TRACK RADON DETECTORS (CHAPTERS II AND III)

187 169 APPENDIX E IRLCS MOBILITY FLOW CHART (CHAPTERS II AND III)

188 170 APPENDIX F IRLCS SHORT-TERM E-PERM RADON MEASUREMENT FORM (CHAPTERS II AND III)

189 171 APPENDIX G ADDITIONAL METHODS AND RESULTS FOR CHAPTER II Methods Radon concentrations for the basement winter short-term E-PERM measurements for an S E-PERM chamber loaded with a short-term (ST) electret were calculated as follows: Equation 1. Calculation of the average radon concentration using an E-PERM Concentration Rn 222 in Bq m 3 = 37 Voltage initial Voltage final (BG ECF) CF D Equation 2. Calculation of the E-PERM calibration factor I + F CF = ( ) 2 where the initial and final electret voltages were measured with an electret reader; CF is the calibration factor; D is the duration of the exposure period of the detector in days; BG is the background gamma exposure rate in microroentgens (μr) per hour; is the gamma conversion constant; and ECF is the elevation correct factor in which 1.0 is used for an elevation less than or equal to 4,000 feet (58). The CF can be determined given the constants of and as provided by the manufacturer for electrets made before Sources of error for measurements obtained from both types of radon detectors are presented elsewhere (Appendix G). The best fitted line for the model was plotted with model diagnostics performed. Diagnostics were conducted to determine points with high leverage (i.e., potential to greatly affect the fitted model) using the two times and three times the average hat value criterion ((2 (k+1)/n and 3 (k+1)/n) where k equals number of predictor variables in the model and n is the total number of observations) and standardizing the residuals (i.e., to

190 172 measure influence). Observations that deviate from the other observations were assessed and removed from the model to observe the resulting effect on the fitted model. Results Thirteen winter short-term basement measurements were excluded for the following reasons: 1) multiple subsequent short-term tests were collected at the same study site (N=4) so only the first short-term radon test was accepted since it was placed at the initial ATD placement; 2) the short-term test was not obtained on the lowest livable floor (N=4); 3) the top of the short-term radon detector was not screwed on tightly when returned allowing for the possibility of radon entry after the exposure period ended (N=1); 4) the bottom of the short-term detector was loose (N=1) and taken off (N=1) interfering with the electrostatic charge in the electret; 5) no date or time was documented when the detector was screwed down making it impossible to determine its exposure period (N=1); and 6) the documented date of screwing down the detector was the same as the placement date (N=1). The distributions of the basement winter short-term and annual basement radon concentrations as well as their differences are asymmetrical and skewed to the right, with a large percentage of the measurements concentrated on the leftmost side (Figure G-1). All three distributions have relatively few high values. Few measurements were larger than 500 Bq m -3 and a few of their differences were larger than 100 Bq m -3. To determine whether the basement winter short-term radon concentrations are different from the basement annual radon concentrations, a Student s paired t-test was performed. The paired differences follow a normal distribution (an assumption of the Student s paired t- test) after undergoing a natural log (ln log base e) transformation as reflected in the histogram and quantile-quantile (Q-Q) plots in Figure G-2, as opposed to the nontransformed paired differences in Figure G-1. The observations were assumed to be independent, a required assumption of the paired t-test, since there was only one pair of measurements per residence being compared.

191 Figure G-1. Histograms of radon concentration by type of measurement and their paired difference 173

192 174 Figure G-2. Histogram and quantile-quantile plot of the paired difference after natural log (ln) transformation Simple linear regression Another assumption of simple linear regression is normally distributed errors (i.e., residuals). The residuals of the model for predicting basement annual radon concentrations based on basement winter short-term radon concentrations after undergoing a ln- transformation resulted is a more symmetrical distribution compared to the residuals of the non-transformed model (Figure G-3). This provides preliminary evidence that the residuals for the model are normally distributed.

193 175 Figure G-3. Histograms of residuals for model predicting basement annual radon concentrations based on basement short-term radon concentrations To test whether the normality assumption holds, Q-Q plots were used to plot the residuals from the models against a normal distribution (Figure G-4). After the model is ln-transformed, the distribution of their residuals closely follows the line, which provides additional evidence that the residuals from the model are normally distributed although the Shapiro-Wilks test to assess normality fails (W = 0.97, p = 0.001) after all the variables were ln-transformed. The ln-transformation, as opposed to the log base 10 transformation, was applied due to its widespread use in the radon literature (69,71).

194 176 FigureG-4. Quantile-Quantile plots of residuals for model predicting basement annual radon concentrations based on basement short-term radon Diagnostics of the model were concentrations performed. The observation with the highest leverage and largest residual for the regression was examined further and removed from the model (Figure G-5). The r-squared coefficient and correlation coefficient decreased only slightly to 0.73 from 0.75and 0.86 from 0.87, respectively.

195 177 Figure G-5. Diagnostics for regression of ln(basement winter short-term radon conc.) on ln(basement annual radon conc.) Quality assurance All data were double entered into a database and verified for entry errors. Detailed information concerning quality assurance is presented elsewhere (55). The lowest reported radon concentration of the ATDs was 36 Bq m -3 (0.86 pci L -1 ), which was larger than the minimum detectable concentration of 3.0 Bq m -3 (0.08 pci L -1 ) for an annual exposure given a MDC of 1,110 Bq day m -3 (30 pci -day L -1 ). The minimum detectable concentration was determined by calculating the standard deviation of the alpha tracks from blank detectors and multiplying by 4.65 then converting to the total radon concentrations.

196 178 APPENDIX H SOURCES OF ERROR FOR RADON DETECTORS Sources of uncertainty of the radon detectors can be attributed to the counting of tracks associated with the ATDs and components of the system (e.g., measuring voltage of electret) for the E-PERM. The major source of uncertainty for the ATDs is associated with the counting of the alpha tracks (77) that is dependent on the total tracks counted and the radon concentration: Equation 3. Error from counting tracks for ATDs total tracks Error counting tracks = 100 total tracks There is also an uncertainty of 4 percent related to the calibration of the standard radon chambers (77). The sources of error associated with the E-PERM detectors are related to the system s components including the chamber volume, electret thickness, and other chamber parameters, voltage uncertainty related to the readings of the electret, and the adjustment of background gamma radiation (58,80-81). Equation 4. Error associated with the E-PERM s system components Error system = Concentration Rn 222 in pci L 1 ± 0.05 (V initial V final ) (CF D) where CF is the calibration factor and D is the duration of the exposure period of the detector.

197 179 Equation 5. Error from initial and final voltage readings for E-PERM Error voltage reading = Concentration Rn 222 in pci L 1 2 ± (V initial V final ) (V initial V final ) CF D Equation 6. Error associated with adjusting the background gamma radiation Error background gamma = Concentration Rn 222 in pci L 1 ± (0.10 BG G) where BG is the background gamma radiation expressed in µr and G is the gamma conversion constant. Equation 7. Overall sources of error Total error = E system + E voltage reading + E background gamma The mean error associated with the system of measurement for the E-PERMs is about 5.6 percent (SD=0.7) with a median of 5.5 percent and ranging from 4.8 to 9.2 percent (Equation 4). The error associated with the electret readings of E-EPERMs has a mean uncertainty of 1.8 percent (SD=1.2) with a median of 1.5 and ranging from 0.3 to 7.2 percent (Equation 5). The mean uncertainty of the background gamma radiation is 8.4 percent (SD=2.3) with a median of 8.7 and ranging from 2.6 to 14.8 percent (Equation 6). The average overall error of the E-PERMs is about 10.4 percent (SD=2.1) with a median of 10.1 and ranging from 5.8 to 17.4 percent (Equation 7).

198 180 APPENDIX I ADDITIONAL RESULTS FOR CHAPTER III The distributions of the basement winter short-term and both annual upper level radon concentrations as well as their differences with the basement short-term tests are asymmetrical and skewed to the right, with a large percentage of the measurements concentrated on the leftmost side (Figure I-1). All the distributions have relatively few values exceeding 500 Bq m -3 for short-term measurements and 200 Bq m -3 for annual measurements. To determine whether the basement winter short-term radon concentrations differ from the upper level annual radon concentrations, a Student s paired t-test was performed. The paired differences follow a normal distribution (an assumption of the Student s paired t-test) after undergoing a ln-transformation as reflected in the histograms and Q-Q plots in Figure I-2, as opposed to the non-transformed paired differences in Figure I-1. The observations were assumed to be independent, a required assumption of the paired t-test, since there was only one pair of measurements per residence being compared.

199 Figure I-1. Histograms of radon concentration by type of measurement and their paired differences 181

200 182 Figure I-2. Histograms and quantile-quantile plots of paired differences after a logarithmic (ln) data transformation A key assumption of the simple linear regression, which was used to examine the ability of the basement winter short-term concentration to predict the annual radon concentration on the first floor as well as the annual average radon concentration of the bedroom and living room, is independent errors. The assumption was met since only one pair of radon measurements per residence is being compared. Another assumption of simple linear regression is normally distributed errors (i.e., residuals). The residuals of the models for predicting the annual first floor and the annual average of the bedroom and living room radon concentrations based on basement winter short-term radon concentrations after undergoing ln-transformations resulted in a more symmetrical distribution compared to the residuals of the non-transformed models (Figure I-3). These

201 183 findings provide preliminary evidence that the residuals for the models are normally distributed. Figure I-3. Histograms of residuals for models predicting first floor and annual average of bed/living room annual radon concentrations based on basement short-term radon concentrations To test whether the normality assumption holds, Q-Q plots were used to plot the residuals from the models against a normal distribution (Figure I-4). After the models are ln-transformed, the distribution of their residuals closely follows the line, which provides additional evidence that the residuals from the models are normally distributed although the Shapiro-Wilks test to assess normality fails for predicting the annual first floor (W = 0.98, p = 0.02) and the annual average of the bedroom/living room (W = 0.98, p=0.01) radon concentration after all the variables were ln-transformed. The ln-transformation, as

202 184 opposed to the log base 10 transformation, was be applied due to its widespread use in the radon literature (68,81). Figure I-4. Quantile-Quantile plots of residuals for models predicting first floor and annual average of bed/living room radon concentrations based on basement short-term radon concentrations

203 185 APPENDIX J MISSOURI INDOOR RADON DETECTOR PLACEMENT (CHAPTER IV)

204 186 APPENDIX K MISSOURI RADON SURVEY QUESTIONNAIRE (CHAPTER IV)

205 187 APPENDIX L MISSOURI OUTDOOR RADON DETECTOR PLACEMENT (CHAPTER IV)

206 188