Dental service areas: methodologies and applications for evaluation of access to care

Transcription

1 University of Iowa Iowa Research Online Theses and Dissertations 2012 Dental service areas: methodologies and applications for evaluation of access to care Susan Christine McKernan University of Iowa Copyright 2012 Susan C. McKernan This dissertation is available at Iowa Research Online: Recommended Citation McKernan, Susan Christine. "Dental service areas: methodologies and applications for evaluation of access to care." PhD (Doctor of Philosophy) thesis, University of Iowa, Follow this and additional works at: Part of the Oral Biology and Oral Pathology Commons

2 DENTAL SERVICE AREAS: METHODOLOGIES AND APPLICATIONS FOR EVALUATION OF ACCESS TO CARE by Susan Christine McKernan An Abstract Of a thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Oral Science in the Graduate College of The University of Iowa July 2012 Thesis Supervisor: Professor Raymond A. Kuthy

3 1 ABSTRACT Significant efforts have been undertaken in medicine to identify hospital and primary care service areas (eg, the Dartmouth Atlas of Health Care) using patient origin information. Similar research in dentistry is nonexistent. The goal of this dissertation was to develop and refine methods of defining dentist service areas (DSAs) using dental insurance claims. These service areas were then used as spatial units of analysis in studies that examined relationships between utilization of oral health services, dentist workforce supply, and service area characteristics. Enrollment and claims data were obtained from the Iowa Medicaid program for children and adolescents ages 3-18 years during calendar years 2008 through The first study described rates of treatment by orthodontists in children ages 6-18 years. Orthodontic DSAs were identified by small area analysis in order to examine regional variability in utilization. The overall rate of utilization was approximately 3%; 19 DSAs were delineated. Interestingly, children living in small towns and rural areas were significantly more likely to have received orthodontic services than those living in metropolitan and micropolitan areas. The second study identified 113 DSAs using claims submitted by primary care dentists (ie, general and pediatric dentists). Characteristics of these primary care DSAs were then compared with counties. Localization of care was used as a measure of how well each region approximated a dental

4 2 market area. Approximately 59% of care received by Medicaid-enrolled children took place within their assigned service area versus 52% of care within their county of residence. Hierarchical logistic regression was used in the final study to examine the influence of spatial accessibility and the importance of place on the receipt of preventive dental visits among Medicaid-enrolled children. Children living in urban areas were more likely to have received a visit than those living in more rural areas. Spatial accessibility assessed using measures of dentist workforce supply and travel cost did not appear to be a major barrier to care in this population. More studies are needed to explore the importance of spatial accessibility and other geographic barriers on access to oral health services. The methods used in this dissertation to identify service areas can be applied to other populations and offer an appropriate method for examining revealed patient preferences for oral health care. Abstract Approved: Thesis Supervisor Title and Department Date

5 DENTAL SERVICE AREAS: METHODOLOGIES AND APPLICATIONS FOR EVALUATION OF ACCESS TO CARE by Susan Christine McKernan A thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Oral Science in the Graduate College of The University of Iowa July 2012 Thesis Supervisor: Professor Raymond Kuthy

6 Copyright by SUSAN CHRISTINE McKERNAN 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 Susan Christine McKernan has been approved by the Examining Committee for the thesis requirement for the Doctor of Philosophy degree in Oral Science at the July 2012 graduation. Thesis Committee: Raymond Kuthy, Thesis Supervisor Peter Damiano Paul Hanley Michael Jones Michelle McQuistan Elizabeth Momany

8 ACKNOWLEDGMENTS This dissertation would not have been possible without the guidance and encouragement of my advisor Dr. Raymond Kuthy, who has provided an excess of support over the past several years. The other members of my committee Drs. Peter Damiano, Elizabeth Momany, Paul Hanley, Michelle McQuistan, and Michael Jones all deserve thanks for their assistance during this project and their patience in responding to my constant stream of s (and, in Dr. Momany s case, my constant presence in her doorway). I would like to express my gratitude to Dr. Peter Damiano and the University of Iowa Public Policy Center for housing me during this research and for granting me access to the Iowa Medicaid data. Many thanks to Ms. Adweta Joshi at the Public Policy Center for her technical assistance with the GIS software used with this research. Also, thanks to the OSCEP staff with the Iowa Health Professions Tracking Systems for their help with these datasets. I would like to thank Dr. Christopher Squier and the NIH/NIDCR T32 training grant for the financial support of my doctoral coursework and dissertation. Finally, I would like to thank my parents Peter and Elizabeth and all of my brothers and sisters for their emotional support during this arduous task. ii

9 TABLE OF CONTENTS LIST OF TABLES LIST OF FIGURES LIST OF ABBREVIATIONS v viii x CHAPTER I. INTRODUCTION Introduction Goal Objectives Significance II. III. IV. REVIEW OF THE LITERATURE Introduction Dental Caries in Children Preventive Dental Services Access to Care Barriers to Dental Care Service Area Analysis Dentist Workforce Evaluation Location Research Environmental Justice Use of Geocoded Data Review of the Literature: Summary GENERAL METHODS AND MATERIALS Overview Data Study Population Study Design STUDY 1 GEOGRAPHIC ACCESSIBILITY AND UTILIZATION OF ORTHODONTIC SERVICES AMONG MEDICAID CHILDREN AND ADOLESCENTS iii

10 Study 1 Abstract Introduction Methods Results Discussion V. STUDY 2 IDENTIFYING MARKET AREAS BASED ON UTILIZATION OF PRIMARY CARE DENTAL SERVICES BY MEDICAID-ENROLLED CHILDREN AND ADOLESCENTS Study 2 Abstract Introduction Methods Results Discussion VI. VII. STUDY 3 INFLUENCE OF SPATIAL ACCESSIBILITY ON RECEIPT OF PREVENTIVE DENTAL CARE Study 3 Abstract Introduction Methods Results Discussion DISCUSSION Study 1 Study 2 Study 3 Limitations and Future Directions Conclusions APPENDIX A. SUPPLEMENTAL MATERIAL STUDY 1 APPENDIX B. SUPPLEMENTAL MATERIAL STUDY 2 APPENDIX C. SUPPLEMENTAL MATERIAL STUDY 3 REFERENCES iv

11 LIST OF TABLES Table II.1 Community and child level influences on children s oral health tested by Bramlett et al. 32 Table IV.1 Table IV.2 Description of continuously enrolled Iowa Medicaid children ages 6 through 18 years ( ) and significance testing results between orthodontic utilizers and non-utilizers by Chi-square analysis 85 Multivariable logistic regression model for orthodontic utilization by Medicaid-enrolled children and adolescents, ages 6-18, during CY Table IV.3 Descriptive characteristics of orthodontic DSAs (N=19) 87 Table IV.4 Descriptive characteristics of orthodontic DSAs grouped by quartiles of utilization rates 88 Table V.1 Descriptive characteristics of primary care DSAs 108 Table VI.1 Table VI.2 Table VI.3 Table A.1 Table A.2 Description of continuously enrolled Iowa Medicaid children ages 3 through 18 years ( ) and significance testing results between children based on access by Chi-square analysis 128 Descriptive characteristics of primary care dental service areas (N=111) 129 Multilevel logistic regression for preventive dental visit by Medicaid-enrolled children and adolescents, ages 3-18, during CY Description of continuously enrolled Iowa Medicaid children ages 3 through 18 years ( ) and significance testing results between orthodontic utilizers and non-utilizers by Chi-square analysis 153 Description of continuously enrolled Iowa Medicaid children ages 6 through 18 years ( ) and significance testing results between orthodontic utilizers and non-utilizers by Chi-square analysis 154 v

12 Table A.3 Distribution of Medicaid enrollees by residential urbanicity and race/ethnicity 155 Table A.4 Table A.5 Table A.6 Table A.7 Odds of orthodontic utilization and 95% confidence intervals by residential urbanicity, stratified by race/ethnicity 155 Odds of orthodontic utilization and 95% confidence intervals by sex, stratified by race/ethnicity 155 Odds of orthodontic utilization and 95% confidence intervals by age, stratified by race/ethnicity 156 Aggregating claims to origin/destination level and assigning origins to service areas (example) 156 Table A.8 Descriptive characteristics of orthodontic DSAs 157 Table A.9 Table A.10 Table B.1 Summary statistics for characteristics of DSA Medicaid enrollees and bivariate correlations with utilization rates 160 Summary statistics for characteristics of DSAs and bivariate correlations with utilization rates 161 State origin of claims submitted by primary care dentists on behalf of Medicaid-enrolled individuals, ages 3-18 years, CY Table B.2 Localization of care for primary care DSAs and counties 176 Table B.3 Table B.4 Descriptive characteristics of primary care DSAs (final boundaries) 177 Geographic distribution by state of primary care dentists treating 1 Iowa Medicaid enrollee 177 Table C.1 Multilevel predictors of children s utilization of preventive dental care 198 Table C.2 Descriptive statistics for characteristics of primary care DSAs 199 vi

13 Table C.3 Table C.4 Table C.5 Table C.6 Table C.7 Table C.8 Description of continuously enrolled Iowa Medicaid children ages 3 through 18 years ( ) and significance testing results between children based on access by Chi-square analysis 200 Univariate logistic regression for dental utilization by Medicaid-enrolled children and adolescents, ages 3-18, during CY (N=146,055) 201 Association of utilization with individual and community level variables: comparison of model estimates produced by SAS PROC NLMIXED versus GLIMMIX 202 Multilevel logistic regression for preventive dental visit by Medicaid-enrolled children and adolescents, ages 3-18, during CY Multilevel parameter estimates: Des Moines metro population versus state without Des Moines 204 Solutions for random effects: DSA estimates from NLMIXED versus GLIMMIX 204 vii

14 LIST OF FIGURES Figure IV.1 Orthodontic DSAs for Medicaid-enrolled children in Iowa (n=19) 89 Figure V.1 Primary care DSAs for Medicaid-enrolled children in Iowa (N = 113) 109 Figure V.2 Localization of care by county and DSA 110 Figure V.3 Cumulative distribution function for localization of care 111 Figure VI.1 Medicaid enrollees per DSA with a preventive dental visit during CY Figure A.1 Age distribution of Medicaid-enrolled orthodontic utilizers, 6-18 years old during CY Figure A.2 Figure A.3 Assignment of patient origin zip codes to provider destinations 163 Unassigned zip codes assigned to the nearest provider destination 164 Figure A.4 Detail of adjustments to orthodontic service area boundaries 165 Figure A.5 Final orthodontic DSA boundaries 166 Figure A.6 Medicaid-enrolled youth (ages 3-18 years) treated per orthodontist in Iowa, Figure B.1 Crude primary care DSAs 178 Figure B.2 Adjusted primary care DSAs 179 Figure B.3 Final primary care DSAs 180 Figure B.4 LOC per crude PC-DSA (Step 1) 181 Figure B.5 LOC per adjusted PC-DSA (Step 2) 181 Figure B.6 LOC per final PC-DSA (step 3) 182 viii

15 FigureB.7 LOC per county 182 Figure B.8 Provider destinations (zip codes) per final primary care 183 DSA Figure B.9 ZCTAs per final primary care DSA 183 Figure B.10 Rate of primary care service utilization per 100 Medicaid enrollees per DSA (N=113) 184 Figure B.11 Rate of primary care service utilization per 100 Medicaid enrollees per county (N=99) 184 Figure B.12 Practice locations per DSA of primary care dentists who submitted 1 claim to Iowa Medicaid during the study period 185 Figure C.1 Conceptualizing the effects of barriers on different types of dental utilization 209 ix

16 LIST OF ABBREVIATIONS AI/AN ACSC BRFSS CDT CHC CHIP DSA EPSDT FPL FQHC GIS HEDIS HPSA IDTS IOM LOC NHANES PC-DSA PCSA RUCA SES American Indian/Alaska Natives Ambulatory Care Sensitive Condition Behavioral and Risk Factor Surveillance System Current Dental Terminology Community Health Center Children s Health Insurance Program Dental Service Area Early Periodic Screening, Diagnosis, and Treatment Federal Poverty Level Federally Qualified Health Center Geographic Information System Healthcare Effectiveness Data and Information Set Health Professional Shortage Area Iowa Dentist Tracking System Institute of Medicine Localization of Care National Health and Nutrition Examination Survey Primary Care Dental Service Area Primary Care Service Area (medical) Rural-Urban Commuting Area Socioeconomic Status x

17 SSI USPSTF ZCTA Supplemental Security Income U.S. Preventive Services Task Force ZIP Code Tabulation Area xi

18 1 CHAPTER I INTRODUCTION Introduction In 2010, 62 out of 99 counties in Iowa were designated as dental Health Professional Shortage Areas (HPSAs) (Iowa Department of Public Health 2011). These shortage areas are eligible for funding from the federal National Health Service Corps and the state PRIMECARRE (Primary Care Recruitment and Retention Endeavor) Loan Repayment Program. HPSA designation criteria are currently under review by the federal government due to the general consensus that the designation process is flawed. Counties are the most common geographic areas that receive HPSA designation. However, counties are used as a matter of convenience; it is unknown how well these approximate actual service areas. Significant efforts have been undertaken in medicine to identify hospital and primary care service areas (eg, the Dartmouth Atlas of Health Care) using patient origin information. Similar research in dentistry is nonexistent. Service area analysis using dental insurance claims would facilitate critical evaluation of HPSA designation procedures and the relationships between dentist distribution and access to care. It is unknown how service area boundaries based on patient utilization patterns compare with existing geopolitical units such as counties.

19 2 Publicly insured children are recognized as some of the most vulnerable and at-risk for oral health problems. In 2010, over 250,000 children were eligible for dental services through Iowa Medicaid; however, less than half received any type of preventive dental services (Iowa Department of Public Health 2010). The Institute of Medicine has defined preventive health services as any measure shown to improve well-being, and/or decrease the likelihood or delay the onset of a targeted disease or condition (Institute of Medicine Committee on Preventive Services for Women 2011). Healthy People 2020 calls for an increase in the proportion of low income children who receive preventive dental care (U.S. Department of Health and Human Services 2011). Dental care is an entitlement to Medicaid children through the federal Early and Periodic Screening, Diagnosis, and Treatment (EPSDT) program. However, limited availability of dentists willing to treat this population is a well-documented barrier to access. The extent to which dentists in Iowa treat Medicaid-enrolled children and the geographic distribution of these dentists are not well-studied. This project will identify service areas in Iowa using patient utilization data and examine associations with resident population characteristics.

20 3 Goal The goal of this dissertation is to develop a method of defining dentist service areas and examine the relationship between utilization of care, workforce availability, and service area characteristics in order to identify geographic barriers to care. Objectives The main objectives for this dissertation are: 1. To define dentist service area (DSA) boundaries using patient origin data. This project will develop and validate methods of identifying DSAs based on small area analysis techniques. a. The first project in this dissertation will identify DSAs for orthodontists in Iowa based on the services they provide to Medicaid enrolled children and adolescents. This preliminary project will develop the methods necessary for the second project, described below. b. The second project in this dissertation will identify DSAs for primary care providers (eg, general dentists and pediatric dentists) in Iowa based on the services they provide to Medicaid enrolled children and adolescents. 2. To identify predictors of the utilization of preventive dental services.

21 4 a. The third project in this dissertation will describe multilevel influences of children s utilization of dental services. Service area characteristics of workforce supply, travel cost, and population demographics will provide information about regional influences of utilization. Significance This dissertation provides researchers with the first service area boundaries identified using patient origin data at a level smaller than the county. It also contributes to our understanding of the access difficulties experienced by children insured by Medicaid. This research describes how spatial accessibility affects utilization of dental care for publicly insured children in Iowa, including the influences of travel distance, and urban or rural residency. If publicly insured children living in rural areas are found to utilize dental services less than urban children, policies and interventions can be designed to focus on these groups. Alternately, urban dwelling children from low income families may face greater barriers to dental care than their rural counterparts. The proposed research will enable policymakers to better identify access barriers faced by publicly insured children. This dissertation will compare service areas with counties as representations of dental market areas. Future research can evaluate whether DSAs in Iowa correspond to other existing regionalization schemes.

22 5 The service area boundaries identified here will provide a framework for future investigation of service area characteristics among different populations, including children versus adults, and privately versus publicly insured individuals.

23 6 CHAPTER II REVIEW OF THE LITERATURE Introduction The goal of this research is to identify dentist service areas for children and adolescents in Iowa, identify geographic barriers to access, and evaluate the relationship between geographic barriers and the receipt of dental services. Geographic barriers will be measured at the service area level. These projects will use dental claims data for Medicaid-insured children in Iowa to identify patient origin service areas. Service areas will be defined using the methods of small area analysis described by Wennberg (Wennberg and Gittelsohn 1973) that have been used in establishing health service area boundaries by the Dartmouth Atlas of Health Care (The Center for the Evaluative Clinical Sciences 1998) and others (Briggs et al. 1995; Klauss et al. 2005; Maine Quality Forum 2005) Dental Caries in Children Introduction In general, the primary pediatric dental disease of concern is caries due to its widespread prevalence and potential for negative health impacts. It is the most common chronic disease of childhood with 38% of American children experiencing dental decay by age eight (Dye, Arevalo, and Vargas 2010; U.S. Department of Health and Human Services 2000).

24 7 The National Health and Nutrition Examination Survey (NHANES) provides information about oral health conditions in the U.S. (Dye et al. 2007). In addition to information about caries prevalence and severity in American children, it also provides data about the incidence of trauma in children. The outcomes associated with dental trauma can be improved by encouraging children to have a regular source of dental care. However, trauma prevalence by definition cannot be prevented with improved access to care, unlike caries. Risk Factors and Prevalence The etiology of dental caries is traditionally described as being influenced by a triad of factors: host susceptibility, diet, and oral microflora (Keyes 1962). Each of these factors is affected in turn by complex interactions with other biological, environmental, and cultural conditions. This dissertation is primarily concerned with socioeconomic risk factors and other environmental conditions. The microbiological risk factors for dental disease are not discussed herein, other than to note that the preventive dental treatments administered by health care professionals act to directly limit the negative impact of these. Socioeconomic Status Socioeconomic status is a major influence on oral health for adults and children. Family socioeconomic status can be assessed primarily based on parents education, occupation, and income levels. NHANES data offer

25 8 information about family income according to the Federal Poverty Level (FPL) (Dye et al. 2007). Children whose family incomes are less than 100% of the FPL are more likely to have caries experience than children with higher income levels. Data from the NHANES show that 28% percent of children ages 6-11 years living at less than 100% of the FPL have caries experience in their permanent dentition, compared to 16% of children living at greater than 200% of the FPL (Dye et al. 2007). Data from NHANES also show that adults with less than a high school education are more likely to have untreated tooth decay (45%) compared with adults with more than a high school education (16%) (Dye et al. 2007). Educational level of parents has also been shown to be associated with caries experience in young children (Hallett and O'Rourke 2006; Oliveira, Sheiham, and Bonecker 2008). Race and Ethnicity Race and ethnicity are also known risk markers for dental caries. Blacks and Hispanics have greater prevalence of primary and permanent tooth decay than white American children (Dye et al. 2007). Data from the NHANES show that approximately 31% of Mexican American children ages 6-11 years have caries experience in their permanent dentition versus approximately 19% of Black and White children (Dye and others 2007). American Indians and Alaska Natives (AI/AN) show even more striking disease rates: a 1999 oral health survey found that 79% AI/AN

26 9 children ages 2-4 had experienced dental decay (Indian Health Service 2002) compared with 28% of all American children in the same age group ( ) (Dye et al. 2007). While national trends have shown overall improvements in the prevalence and severity of caries in children, certain high risk groups are not enjoying these benefits (Dye et al. 2007). Data from NHANES for children 2-11 years of age from and showed that caries prevalence in primary teeth remained relatively stable at approximately 40% among children 2-11 years of age (Dye et al. 2007). However, caries experience showed an increase among poor whites (ages 6-8) and poor Mexican Americans (ages 9-11) (Dye, Arevalo, and Vargas 2010). Overall, prevalence increased among all children ages 2-5 years over this time (Dye et al. 2007). This trend is especially concerning given evidence that children with caries in their primary teeth are more likely to develop caries in their permanent dentition (Li and Wang 2002). Community Water Fluoridation Community water fluoridation has wide evidentiary support for preventing dental caries and is strongly recommended by the U.S. Task Force on Community Preventive Services as a cost-effective intervention (Task Force on Community Preventive Services 2002). Kumar, Adekugbe, and Melnik evaluated the relationship between dental claims submitted to New York State s Medicaid program in 2006 and the presence of county water

27 10 fluoridation (2010). They found that the mean number of caries-related claims per individual submitted for children covered by the EPSDT program decreased as the availability of community water fluoridation increased in a county. Similar results linking community water fluoridation with reduced dental expenditures by public insurance programs have been demonstrated in other regions (Armfield 2010; Centers for Disease Control and Prevention 1999). The preventive effects of fluoridated water operate on a community level, providing benefits to all individuals on public water systems. As such, it is promoted as a means of reducing disparities in caries experience across all populations. Age For individuals of all socioeconomic and ethnic backgrounds, increasing age is one of the most important risk factors for dental caries. Based on crosssectional survey data (NHANES ), 42% of children and adolescents between 2 and 20 years of age have caries experience in their permanent teeth. Among adults over age 20, 90% of Americans have a history of dental decay (Beltran-Aguilar et al. 2005). Dental Caries in Children: Summary Despite an improved understanding of the disease etiology and process, dental caries in permanent teeth is estimated to affect 56% of 15- year old Americans (Dye et al. 2007). Experiencing dental caries, especially

28 11 significant disease early in life, is associated with adverse long term consequences including abnormal growth patterns, missed school days, learning difficulties, and increased risk of dental disease later in life (Filstrup et al. 2003). The impact of poor oral health in children can be compounded by a lack of care; 23% of children 2 to 11 years old have untreated dental decay (Dye et al. 2007). Barriers to dental care share common risk factors with caries including low socioeconomic status and belonging to an ethnic or racial minority. Preventive Dental Services Introduction Optimal oral health care is supported by routine dental visits and early interventions to prevent and treat disease (American Academy of Pediatric Dentistry 2010). Preventing dental disease can save public programs the expense of costly restorative care by minimizing the effects of dental disease or preventing its occurrence entirely. To support this, Healthy People 2020 called for an increase in the proportion of low income children receiving any preventive dental services (U.S. Department of Health and Human Services 2011). Without routine preventive visits, children s dental care focuses on therapeutic interventions, including treatment of urgent care needs. In 1998, a study of Iowa adults found that Medicaid adults were four times more

29 12 likely to have a tooth extracted and two times more likely to receive endodontic therapy than privately insured adults (Sweet et al. 2005). If children demonstrate similar patterns of utilization, then public programs should reconsider how they provide preventive services. Recommendations for Prevention Healthy People 2020 Healthy People 2020 states that lack of access to dental care for all ages remains a public health challenge (U.S. Department of Health and Human Services 2011). Several Healthy People objectives call for improved oral health and access to care in children and adolescents. Objectives that specifically address access to care and prevention of dental disease include: Oral Health Objective 7, which calls for an increase in the proportion of children and adolescents who received oral health services of any type in the past year, and Oral Health Objective 8, which calls for an increase in the proportion of lowincome children and adolescents who received preventive dental services in the past year (U.S. Department of Health and Human Services 2011). Report of the Surgeon General In 2000, the U.S. Department of Health and Human Services released the landmark report Oral Health in America: A Report of the Surgeon General (U.S. Department of Health and Human Services 2000). The importance of oral health and its relationship with overall systemic health

30 13 were stressed. This report supported recommendations based on systematic reviews, like those utilized by the U.S. Preventive Services Task Force. The report called for a focus on prevention of oral disease, stating that most common oral diseases can be prevented through a combination of community, professional, and individual strategies (U.S. Department of Health and Human Services 2000). Recommended preventive strategies include the use of dental sealants and application of fluoride. Community water fluoridation was also supported as a cost saving community-based intervention. It was noted that substantial portions of underserved Americans are unable to receive recommended preventive services (U.S. Department of Health and Human Services 2000). Institute of Medicine In 2011, the Institute of Medicine (IOM) released recommendations for a New Oral Health Initiative in the report Advancing Oral Health in America (Institute of Medicine 2011). The IOM cited increased evidence for links between oral and systemic health, including relationships between periodontal disease and heart disease, diabetes, and adverse pregnancy outcomes. In their call for an oral health initiative, one organizing principle is the emphasis on disease prevention. Prevention is encouraged in order to reduce or eliminate the need for future treatment, the associated costs of treatment, and allow for more efficient use of the existing dental workforce in meeting population needs (Institute of Medicine 2011).

31 14 This report supported the Surgeon General s call for the use of evidence-based services to prevent dental disease. Community water fluoridation was supported as a powerful preventive service. Other recommended services include professionally applied fluorides (eg, varnish or gel) and placement of pit and fissure sealants (Institute of Medicine 2011). U.S. Preventive Services Task Force The U.S. Preventive Services Task Force (USPSTF) is an independent group that reviews scientific research to recommend evidence based preventive interventions (Task Force on Community Preventive Services 2002). Besides recommendations related to the prevention of dental caries, they also have issued recommendations related to oropharyngeal cancer and craniofacial injuries. The USPSTF supports one intervention for the prevention of dental caries in young children: oral fluoride supplementation for children over 6 months of age prescribed by primary care providers according to current guidelines (Task Force on Community Preventive Services 2002). Preventive Dental Services: Summary Federal agencies, work groups, and professional organizations recognize the importance of access to dental care and the role of preventive services in improving children s oral health. Preventive dental care can reduce the need for more extensive, expensive treatment late in life and reduce other adverse health outcomes, including quality of life effects and

32 15 possible systemic effects. Recommended preventive dental interventions include community water fluoridation and professional services such as topically applied fluorides and pit and fissure sealants. While water fluoridation acts at a community level to prevent disease, dentists services must be sought out and depend on peoples ability to access care. Access to Dental Care Introduction An Institute of Medicine (IOM) committee defined access to health care services as the timely use of personal health services to achieve the best possible health outcomes (Millman 1993). Based on this definition, both utilization of services and the related outcomes are important indicators of access. The Institute of Medicine s recent call for a New Oral Health Initiative advocates for improved access to dental care for vulnerable populations and increased efforts in preventing dental disease (Institute of Medicine 2011). Defining Access Penchansky and Thomas described access to health care as having five dimensions: availability, accessibility, affordability, acceptability, and accommodation (Penchansky and Thomas 1981). Availability refers to workforce supply. Accessibility describes the location of the workforce supply and includes the related factors of travel time and distance. Affordability is

33 16 determined by fees, patient insurance coverage, and provider acceptance of insurance. Acceptability can refer to the patient s or the provider s tolerance and priority for personal characteristics. Accommodation is the aspect of accessibility that is created by the provider s office hours and staff assistance. Penchansky and Thomas assert that when researchers attempt to measure access, outcomes typically fall into one or more of these dimensions. For example, personal income as an access-related factor is a key component of affordability. In general, this framework describes access in terms related to Andersen s enabling factors (Penchansky and Thomas 1981). Based on existing scientific literature, Penchansky and Thomas state that acceptability where patients can and will be seen may be more important than accessibility (ie, provider locations) (Penchansky and Thomas 1981). This may be especially important for Medicaid-insured children who have well-documented difficulties finding a dentist due to insurance type or young age. Components of Access Guay identified three elements critical for access to occur: demand for services, an adequate workforce to meet demand, and an economic environment conducive to the receipt of services (Guay 2004). Within this framework, access is a function of supply and demand. If the dental workforce is unable or unwilling to meet the population s demand for services, barriers to access are created.

34 17 Demand for Dental Care Perceived need and demand for dental care are patient-based (Guay 2004). Internal factors such as health beliefs and cultural preferences contribute to the perception of need (Guay 2004). Effective demand occurs when patients have the perception of need and the abilities to seek and receive dental care. Along with perceived need, individuals must possess the financial means to seek and obtain dental care in order to generate demand. The financial ability to seek and obtain dental care is affected by income, insurance coverage, and support from employers and family. If an individual has dental insurance but cannot find a dentist to accept it, demand may not occur if that person cannot afford to pay for dental care out of pocket. Dental Workforce Factors related to the supply of dental services are external to the patient s demand for dental care (Guay 2004). Characteristics of the dental workforce that modify access include the number and geographic distribution of dentists, specialization of practice, and the restriction of new patients into practice based on age, insurance coverage, etc. People living in rural areas commonly encounter restricted access due to the geographic distribution of dentists. Low-income inner city residents, non-english speakers, publicly insured individuals, people with special needs, and young children may face access barriers that are created by practice limitations imposed by dentists (Guay 2004).

35 18 An economic environment supportive of access comes when patients possess the financial means to purchase dental services and when dentists offer their services for prices that patients can afford. Access is limited when dentists do not accept the lower reimbursement rates offered by public insurance programs. Low reimbursement rates are cited commonly as a primary reason why dentists do not accept or limit Medicaid patients in their practices. A 2003 survey of Medicaid dentists in Alabama found that the extent to which dentists participated in Medicaid was predicted by how generous they believed reimbursement to be relative to private insurance (Agili et al. 2007). A 2004 survey of pediatric dentists in Texas found similar results: low reimbursements were cited as the second most important issue pertaining to the state Medicaid program after broken appointments (Blackwelder and Shulman 2007). Measuring Access Access to dental care can be categorized as potential or revealed (Thouez, Bodson, and Joseph 1988). Potential access refers to the availability of health services, along with an individual s motivation to seek care. Once care is sought, revealed access occurs when an individual receives health services. Potential Access Potential access is affected by structural characteristics of local health care infrastructure, including availability of dentists, physicians, and

36 19 hospitals (Andersen et al. 1983). One method of measuring potential access is through workforce analysis. The least complicated and most commonly used method of measuring available dental workforce is the calculation of dentistto-population ratios. By calculating the number of dentists available to the local population, ratios can be used to describe the workforce supply. Using an arbitrary ideal ratio, such as one dentist per 1,000 people, and comparing this to actual ratios, the adequacy of the dentist supply can be evaluated. A second common method of workforce analysis involves the use of busyness surveys that gather information from dentists about how busy they are and whether they could treat additional patients to determine whether local dentists are over- or under- employed (DeFriese and Barker 1982; Henderson 1976). Based on responses from these surveys, policy-makers can identify areas in need of additional dentists. More complicated models of measuring workforce include demandbased workforce evaluations and econometric workforce studies (DeFriese and Barker 1982). Demand-based studies include information from dentists regarding the number of services provided and use estimates of population demand for dental services. If demand exceeds the existing supply of services, additional dentists are warranted. In contrast, econometric workforce studies use estimates of the supply of dental services to calculate demand.

37 20 By considering factors associated with practice productivity, such as numbers of patient visits or use of auxiliaries, these factors are correlated with the utilization of services. This method assumes that the supply of services determines utilization: as dentist supply increases, population demand increases. This type of econometric model is the basis for the American Dental Association s Dental Planning Information System (DPIS) (DeFriese and Barker 1982). Realized Access Realized access occurs when dental services are received. Objective factors that motivate realized access include convenience factors (eg, travel time, costs, waiting time) and the perception of dental need and want (Andersen et al. 1983). Measuring realized access can be accomplished using utilization rates. Number of dental visits in a given year, number of dental visits for emergency or preventive services, and proportion of the population receiving or not receiving care describe a population s use of dental services. The 2008 National Health Interview Survey demonstrated that 75% of publicly insured children in the U.S. had seen a dentist within the past year (Lewis et al. 2007b). In 2006, 73% of adult Iowans surveyed by the Behavioral and Risk Factor Surveillance System (BRFSS) had seen a dentist within the past year (Iowa Department of Public Health 2006). Another method of measuring realized access is to measure the level of unmet need in a population. The National Survey of America s Families

38 21 examines self-reports of unmet dental need among children age three and older. In a 2000 survey, there was no significant difference in unmet need between children with public versus private health insurance. However, children who were uninsured were more likely to report unmet dental need than those with private or public insurance (Kenney, Ko, and Ormond 2000). Different findings come from the 2007 National Health Interview Survey, where over four million children ages 2-17 years had unmet dental need because of an inability to afford care (Bloom and Cohen 2009). Publicly insured children were 50% more likely than privately insured children (6% versus 4%) to report unmet dental need. Uninsured children were even more likely to have unmet dental need, with 24% they were unable to afford care (Bloom and Cohen 2009). Finally, realized access can be assessed in terms of self-reported health status. A survey of Alabama s Medicaid patients found that parents perception of their child s oral health was significantly related to the receipt of dental care over the past year (Agili, Bronstein, and Greene-McIntyre 2005). Parents who perceived dental care as not needed were more likely to rate their child s oral health as excellent, compared with parents who tried to get care but could not get it. Access to Dental Care: Summary Potential access to dental care can be improved through professional monitoring and activities that adjust workforce availability to meet demand

39 22 for services. The federal government encourages dentists to practice in underserved areas with funding and loan repayments. State agencies also work to identify areas with access barriers. Research that can effectively identify the most at risk populations will support existing initiatives and provide a strong foundation for future activities. An evaluation of patient origin service areas would reveal whether and to what extent accessibility is more important than availability for Medicaid-insured children in Iowa. Barriers to Dental Care Introduction The IOM noted that, in general, barriers prevent vulnerable populations from obtaining access to needed health care services (Millman 1993). These factors have been characterized and evaluated within many different frameworks. The development and use of conceptual frameworks facilitates research and the translation of findings. Several of these frameworks and their application to this dissertation are reviewed. Conceptualizing Barriers IOM Barriers Barriers have been categorized as structural, financial, and personal or cultural by the IOM (Millman 1993). Structural barriers are related to the availability of dental providers: specialty and practice characteristics, geographic distribution, and participation with insurance programs

40 23 contribute to structural barriers. Financial barriers impede an individual s ability to pay for wanted services. Personal or cultural barriers can arise from limited health literacy or health beliefs (Millman 1993). In Access to Health Care in America (1993), the IOM reports that a sentinel health objective is the reduction of pain and morbidity through the provision of timely and appropriate health services. Access to dental care is highlighted as an indicator of how well this objective is met. Access can be evaluated in terms of utilization and outcomes, with the goal of equitable access for all populations. Utilization can be measured in terms of the number of visits to the dentist and outcomes can include physical indicators such as disease and survival rates (Millman 1993). Andersen et al Behavioral Model One of the most widely utilized conceptual frameworks in health services research was developed by Ronald Andersen in 1968 in order to explain families use of medical services (Andersen 1995). Initially, Andersen categorized barriers to the use of health services as predisposing, enabling, and need-related. Subsequent revisions to the behavioral model added domains representing characteristics of the health care system and individuals health behaviors. Outcomes have been expanded from simple utilization of services to include resulting changes in health status and quality of care measures (Andersen 1995).

41 24 The three original domains (predisposing, enabling, and need-related factors) have been retained in all iterations of this model (Andersen 1995). These core domains have been incorporated by numerous other researchers seeking to conceptualize the utilization of health care services (Beil and Rozier 2010; Borders 2006; Grembowski, Andersen, and Chen 1989; Hall, Lemak, and Steingraber 2008; Milgrom et al. 1998). Predisposing factors typically include demographic characteristics (eg, age, sex, race/ethnicity). Resources that enable individuals to access health services include insurance coverage, income levels, and having a regular source of care. Need-related characteristics refer to both the patient s perception of need as well as a professionally evaluated level of need. While predisposing and need-related characteristics tend to be related to biological imperatives, enabling resources come from family and social support. Other researchers have identified other early childhood oral health enabling factors, including transportation resources, healthy food options, and an effective health care system (Mattheus 2010). Limited dentist participation has been reported as a primary barrier for Medicaid children seeking care (Agili, Bronstein, and Greene-McIntyre 2005; Fisher and Mascarenhas 2007). Enabling resources tend to be the easiest to modify and are the focus of many public health initiatives. For example, states are required to provide transportation assistance for children to medical and dental appointments as

42 25 a component of the EPSDT program (U.S. Health Resources and Service Administration 2012). Geographic barriers have been reported to vary with urban or rural residence and the number of children per family (Agili, Bronstein, and Greene-McIntyre 2005; Borders 2006). Barriers frequently faced by the Medicaid population include extended travel times, travel costs, lack of reliable transportation, and difficulties managing large families (Borders 2006; Mattheus 2010). While national surveys and focus groups have compared reported access barriers to dental care (Kenney, Ko, and Ormond 2000; Lewis et al. 2007a), a study that compares these populations ideally will also take into consideration geographic variation in utilization and dental practice patterns. Physician-Enabling Characteristics Insurance coverage is identified as a major enabling factor that supports an individual s utilization of health services. However, insurance coverage and local availability of physicians are not sufficient to guarantee access. The difficulties that publicly insured individuals face when seeking medical or dental care are well-documented. Hall, Lemak, and Steingraber argue that the unique obstacles publicly insured individuals face should be considered within the enabling domain of the Andersen model (2008). These factors, or physician-enabling characteristics, refer to the extent to which physicians make themselves accessible to patients (Hall, Lemak, and Steingraber 2008). Hall et al conceptualize four components of accessibility

43 26 related to physician-enabling characteristics: overall, contact, appointment, and geographic accessibility. Overall accessibility refers to the extent to which physicians limit new patients in their practice or whether physicians limit patients based on insurance coverage. Contact accessibility describes the difficulties experienced by patients when calling medical offices. Appointment accessibility represents the length of time patients must wait to be seen or the access difficulties that arise when physicians do not offer evening or weekend appointments. Geographic accessibility incorporates both physical distances to health care providers and patients abilities to obtain transportation (Hall, Lemak, and Steingraber 2008). Hall et al. conducted a survey in 2003 of Florida MediPass providers (Hall, Lemak, and Steingraber 2008). At the time, MediPass provided primary care case management services to enrolled Medicaid recipients in all counties in the state. Interviewers contacted over 3,000 providers and inquired whether offices were accepting new Medicaid patients and attempted to identify common barriers to obtaining appointments. While 87% of providers said they were taking new patients, only 68% of MediPass providers would accept new Medicaid patients (Hall, Lemak, and Steingraber 2008). Other studies have shown similar results. Findings from a national survey showed that children with public health insurance were more likely

44 27 than privately insured children to report difficulty with access as a reason for unmet dental need (Kenney, Ko, and Ormond 2000). A state survey revealed that approximately half (48%) of Alabama parents reported that they had difficulty finding a dentist who accepted Medicaid for their children (Agili, Bronstein, and Greene-McIntyre 2005). Grembowski, Andersen, and Chen Public Health Model Grembowski, Andersen, and Chen reviewed existing models explaining the utilization of health care services in order to develop a conceptual model that describes the utilization of dental services as a decision-making process (1989). The foundation for this conceptual model came from social exchange theory. Utilization was to be measured in terms of an episode of care, which was initiated by a mother when she perceived that the rewards from a dental visit would exceed the costs. Important barriers to dental care identified by this model include public (ie, Medicaid) insurance coverage. Rates of dental utilization by Medicaid insured children were noted to be below national levels compared to children with private insurance (Grembowski, Andersen, and Chen 1989). Reasons cited for this include poor dentist participation and poor compliance by Medicaid patients (ie, failure to show for visits and failure to comply with treatment recommendations). Because of this, Grembowski and co-authors note that the supply of dentists available to publicly insured children may be substantially less than the supply available to other socioeconomic groups.

45 28 Given these conditions, the total cost barriers associated with seeking care may become insurmountable for some publicly insured individuals. Milgrom et al Explanatory Model Milgrom et al. used the Grembowski, Andersen, and Chen public health model as a framework to identify factors that were predictive of lowincome mothers initiating an episode of dental care for their child (Milgrom et al. 1998). Using interviews of mother-child dyads, the authors sought empirical support for the Grembowski model, which predicts that expected rewards greater than expected costs should be predictive of dental utilization. Costs were assessed in both economic and non-economic terms. Utilization of dental services was categorized in two ways in this study: a dental visit for any reason or a dental visit for a preventive reason. Interestingly, the authors identified preventive care as any non-emergent visit to the dentist (Milgrom et al. 1998). Based on the social exchange theory, Milgrom et al calculated the time costs of dental care. Time costs were quantified in terms of the length of time an individual had to wait for an appointment, travel time to the appointment, and waiting time in the office. Ultimately, family income was not found to be a significant predictor of utilization. However, race and years lived in the US were found to be associated with utilization of dental care. Cognition and expectation variables were positively associated with dental utilization in this population. Examples of these include the belief that dental check-ups were important,

46 29 absence of maternal dental fear, and maternal rating of child s oral health as excellent (Milgrom et al. 1998). Within this model, the authors were able to identify which factors affecting the utilization of dental care services were the most mutable (Milgrom et al. 1998). These factors were considered to be the most appropriate to target with policy and delivery interventions. Specifically, insurance coverage and the utilization of preventive medical visits were found to be positively associated with dental utilization. Fisher-Owens et al Conceptual Model Fisher-Owens et al have developed a conceptual model of determinants of children's oral health (Fisher-Owens et al. 2007). This multilevel model approaches oral health from a population perspective in order to facilitate policy and planning. Within this model, oral health is affected by influences from the community, the family, and at the individual level. Community factors include characteristics of the health care system and public infrastructure. Family-level factors include socioeconomic status, parental health status, and cultural values of the family. Individual factors that can influence oral health include diet, genetics, insurance coverage, and demographic characteristics. The presence of special health care needs is also an individual risk factor for dental disease. Within the community level, characteristics of the dental workforce can potentially act as influences on children s oral health. The availability of

47 30 dentists and other providers can facilitate or impede access to care, and this availability can be moderated by numerous factors. Major moderating factors include hours worked per week, acceptance of Medicaid, acceptance of young children as patients, and availability of evening or weekend appointments to accommodate working parents (Agili, Bronstein, and Greene-McIntyre 2005; Borders 2006; Fisher and Mascarenhas 2007). Other community level influences arise from the physical domain: the availability of public transportation, community water fluoridation, and the general health conditions of a community (eg, crowded housing, lead exposure, crime rates) (Bramlett et al. 2010; Fisher-Owens et al. 2007). Two additional factors are considered within this framework independent of the three major levels: time and vulnerability. The authors consider time to modify the relationships across levels at different points in children's lives and call for this to be incorporated into the study of determinants of oral health. The authors also include the concept of vulnerability, recognizing that factors from different levels can interact to create a heightened level of risk in an individual (Fisher-Owens et al. 2007). Testing the Fisher-Owens et al Conceptual Model In order to empirically test this model, Fisher-Owens et al assert that multilevel modeling would be more appropriate to evaluate the relationships between the determinants of oral health than traditional linear regression techniques (2007). Bramlett et al used data from the 2003 National Survey of

48 31 Children s Health (NSCH) to empirically test the relationships hypothesized by the Fisher-Owens et al model (Bramlett et al. 2010). Interviews with over 100,000 American children provided data from two model levels: children considered within states. The authors used exploratory modeling to identify characteristics that were significant predictors of parent-reported child oral health. A final multilevel model identified physical attributes, developmental influences, insurance coverage, and health behaviors that were significant predictors of oral health at the child level. Several characteristics describing family socioeconomic status remained significant throughout the modeling. From the state level, characteristics of the social and physical environment were significant predictors of oral health. Selected predictors identified by Bramlett et al are categorized within Table II.1. Interestingly, the findings from Bramlett et al did not identify any characteristics of the dental or medical care system as being significant predictors of parent-reported child oral health (Bramlett et al. 2010). The authors considered variables including the supply of dentists and physicians, the percent of children uninsured at the state level, Medicaid and CHIP funding at the state level, and other public health indicators (eg, vaccination and low-birth-weight rates). Relatively few variables from the state level remained significant in the final model. Service area characteristics may be more appropriate for multilevel modeling if they could accurately profile the regional variations in culture, social capital, and health care system.

49 32 Table II.1 Community and child level influences on children s oral health tested by Bramlett et al Domain Determinant Independent variables Community-level Physical environment Gini Index of income inequality Population with access to water fluoridation Child-level Physical attributes Race/ethnicity Sex Development Age Dental insurance Child has dental insurance Socioeconomic status Education level of parents Household income relative to Federal Poverty Level Neighborhood structure Child lives in a Metropolitan Statistical Area Health Behaviors Special health care needs Family structure Primary language spoken at home Country of birth Dependent variables Oral health outcome Oral health status Parental report of child s oral health condition (excellent/good vs. fair/poor) Note: Bramlett et al identify several Family-level domains in their final model that were actually considered within the child-level during the statistical analysis. They are included within the child-level characteristics here to reflect the actual analyses that demonstrated their statistical significance as influences on oral health.

50 33 Vulnerability Vulnerable populations are those groups of people at increased risk for adverse health outcomes (Flaskerud and Winslow 1998) and have been noted to develop due to lack of protection from risk factors (Mattheus 2010). Flaskerud and Winslow (1998) developed a conceptual framework that relates the availability of community resources to relative risk for adverse health outcomes and overall health status (eg, morbidity and mortality). A confluence of factors within these three levels (ie, resources, risk, and health status) can create vulnerable population groups. Resource availability encompasses the socioeconomic variables traditionally considered in health outcomes research: income, employment, education, housing, and health insurance coverage. Risk factors for disease include the use of health services such as immunizations, disease screenings, preventive services and lifestyle behaviors such as diet, physical fitness, and high risk activities (eg, not wearing a safety belt, drinking and driving, and drug use). According to the authors, vulnerability can be alleviated through research, policy action, and clinical practice (Flaskerud and Winslow 1998). These modifying professional activities can occur at multiple levels including health care organizations, educational institutions, state and federal governments, and within the health care professions.

51 34 Shi et al. (2008) argue that predisposing, enabling, and need characteristics interact to create an overall state of vulnerability for population groups. This vulnerability interacts with access and quality of care to produce health outcomes for individuals and communities. Because it is created by multiple risk factors that interact synergistically, research should assess vulnerability within this framework rather than factor-byfactor as is commonly done in access research. According to this framework, statistical interaction terms are not sufficient to capture the relationships between multiple risk factors for vulnerability. The authors propose risk profiles that can assess the "number and types of risk factors experienced by individuals" (Shi et al. 2008). Empirical support for this framework was gathered from several national datasets (eg, 1996 MEPS, 2000 National Survey of Early Childhood Health, and 2000 National Health Interview Survey). Overall, the authors found a "dose-response" relationship between the number and unique combination of risk factors and the likelihood of reported unmet need (Shi et al. 2008). Mattheus (2010) produced a concept analysis of vulnerability as it is specifically related to pediatric oral health. According to this concept map, factors that affect oral health are categorized into three levels: community, family, and individual. The specific combination of factors affecting a child can act synergistically to create vulnerability or to support optimal oral

52 35 health. Community level factors that can affect the oral health of an individual child include community water fluoridation, transportation resources, community dental programs for children, and average income levels (Mattheus 2010). Family level factors include socioeconomic status, social support, parental health status, and cultural beliefs. At the individual level, genetics and physical traits play important roles in health. Low birth weight, developmental disorders, poor nutrition, and insurance coverage can affect pediatric oral health. Barriers to Dental Care: Summary Numerous conceptual models exist in the health services literature; those reviewed here all seek to identify barriers to optimal health or to utilization of services. Overall, models have gradually become more complex moving from simple cause and effect models to multilevel models that incorporate interactions and an overarching concept of vulnerability. These models call for more sophisticated analysis beyond simple linear regression. As multilevel models are developed, the importance of community resources and supportive social structures are emphasized. Beyond a simple measure of the number of dentists in a county, the literature indicates that important interactions occur between insurance coverage, insurance acceptance by providers, transportation options, and family obligations. The concept of heightened vulnerability among groups with multiple risk factors should accompany any comprehensive discussion of barriers to dental care.

53 36 The major risk markers for dental disease in children poverty and lack of adequate dental insurance, ethnic or cultural minority status, rural residence, and special health care needs are all recognized as access barriers to dental care. Supported by research designed to understand these barriers, public health activities offer vulnerable populations opportunities for improved quality of life and health outcomes. Service Area Analysis Introduction Current health professional shortage area (HPSA) designation guidelines from the Bureau of Health Professions within the U.S. Department of Health and Human Services require any designation that is based on geographic area rather than population groups or facilities must be a rational area for the delivery of services (U.S. Department of Health and Human Services 1993). Rational service areas include single counties, adjacent parts of contiguous counties, or urban neighborhoods no smaller than a census tract (U.S. Department of Health and Human Services 1993). Most states choose individual counties as the unit of designation in rural areas and selected neighborhoods in urban areas. Urban neighborhoods receiving shortage designation by census tracts are typically low income areas with high levels of need for primary care medical, dental, or mental health services.

54 37 Regional variation in health care outcomes has been studied in medicine for decades using small area analysis (Goodman et al. 2003; The Center for the Evaluative Clinical Sciences 1998; Wennberg and Gittelsohn 1973). As an analytic method, small area analysis can evaluate workforce supply, compare regional utilization rates and provider practice patterns, and analyze cost of care issues. These methods have not been applied in dental service research. To date, market area studies in dentistry have been limited to the county level (Mayer 1999). Few studies have examined access in terms of workforce supply and patient utilization at more local levels (Mertz and Grumbach 2001; Siegal and Marx 2005). Small Area Analysis Wennberg and Gittelsohn defined hospital service areas in Vermont using 1972 Medicare Part B records and records from a 1963 state Health Department hospital expenditure study (Wennberg and Gittelsohn 1973). The study identified 13 hospital service areas in the state; however, methods are not described in detail. The authors only state they that grouped towns into hospital service areas surrounding the hospital used most frequently by the town (Wennberg and Gittelsohn 1973). These hospital services areas were then used by Wennberg and Gittelsohn to examine variation in health care expenditures and utilization rates, including age-adjusted surgical procedure rates, lengths of hospitalizations, and physician availability per 10,000 residents (1973). They

55 38 note that lack of information about the prevalence of various diseases in service areas limited their ability to measure the relationship between need and provision of care. The authors found little overlap in service areas: 85% of patients received hospital care in their area of residence. Wennberg and Gittelsohn examined variation in utilization and expenditures across service areas in order to highlight the need for resource planning that focused on areas of true population need. Their service area analysis demonstrated that supply-driven provision of services existed even after adjusting for population need. By 1982, Wennberg and Gittelsohn had identified service areas within six states in New England and begun to study variations in the provision of health care services in these areas (Wennberg and Gittelsohn 1982). They found that people were generally treated at their local hospital. Based on their studies, Wennberg and Gittelsohn noted that small area analysis can reveal variations in health care that are masked by aggregating data to larger geographic regions such as counties or states (Wennberg and Gittelsohn 1982). They asserted that this type of aggregation and reliance on large scale population utilization data contribute to inefficiencies in planning and resource allocation (Wennberg and Gittelsohn 1973; Wennberg and Gittelsohn 1982).

56 39 Physician Shortage Areas The effects of aggregation to mask shortages are noted by Luo and Qi in their development of a floating catchment method for identifying physician shortage areas (Luo 2004; Luo and Qi 2009). While their methods did not include information about patient utilization or the specific provision of services, they did quantitatively describe how changes in spatial scale affect the identification of physician shortages measured in terms of physician to population ratios. Luo notes that the use of a floating catchment method is hampered by the lack of research into what constitutes reasonable travel times from patient to provider (Luo 2004). More intelligent HPSA designation methods have been proposed by Wang and Luo (2005) using GIS to incorporate patient travel times to primary care physicians (representing spatial accessibility to providers) and nonspatial population characteristics that are indicative of need for these services. Population characteristics included socioeconomic and demographic factors, as well as household characteristics; all data used were obtained from the US Census data files (Wang and Luo 2005). A two-step floating catchment area method was used to quantify spatial access to primary care physicians (Wang and Luo 2005). This two-step approach identifying where providers get their patients and where local residents go to obtain care has been used in previous research. However, Wang and Luo hypothetically

57 40 identified local demand for care as a function of aggregated population characteristics. Primary Care Service Areas Small area analysis has focused largely on variation in utilization of services. However, it also provides a means by which researchers may evaluate how effectively a region manages its health care resources (Paul- Shaheen, Clark, and Williams 1987). Primary care service areas (PCSAs) were identified by Goodman et al. (2003) using Medicare claims for primary care clinicians. The authors reference Wennberg and Gittelsohn s methods of defining hospital service areas (Wennberg and Gittelsohn 1973) as the template for their activities. However, Goodman et al. describe their methods in significantly more detail than Wennberg and Gittelsohn, including considerations required to maintain geographic continuity of service areas, weighting patients based on the amount of individual utilization, and establishing minimum population thresholds. Adjustments for geographic continuity are mentioned, but how these are handled is not described. Other adjustments are made for service areas with excessive travel by residents into other service areas for care. Mobley et al. (2006) used PCSAs to aggregate information about hospital admissions for ambulatory care sensitive conditions (ACSCs) in the elderly. ACSCs were studied because hospital admissions for these are considered preventable with good access to ambulatory care, including

58 41 preventive care and disease management. Examples of ACSCs include chronic diseases such as asthma, hypertension, diabetes, and dental conditions (Agency for Healthcare Research and Quality 2001). The authors hypothesized that adequate access to primary medical care services should be related to reduced hospitalizations for ACSCs (Mobley et al. 2006). PCSA boundaries have also been used to evaluate the availability of primary care providers for children. Shipman et al calculated pediatric physician supply at the PCSA-level for the United States using 2006 physician and population data (Shipman et al. 2011). Small Area Analysis: Summary Studies that incorporate information about health service utilization describe realized access (or revealed) for a population of interest versus studies of potential access that evaluate the relationship between provider distribution and population characteristics that are associated with supply and demand for services (Thouez, Bodson, and Joseph 1988). Small area analysis offers researchers the ability to critically evaluate populations that are served by local health care providers: where patients come from, how they get to their appointments, and how utilization is related to level of need (Borrell et al. 2006; Phillips et al. 2000; The Center for the Evaluative Clinical Sciences 1998).

59 42 Service Areas in Dentistry In a 1999 study, Mayer generated dental market areas based on actual patient travel time using North Carolina Medicaid claims data (1999). Market areas are geographic regions that include all buyers and sellers whose interactions determine price (Mayer 1999) and are conceptually equivalent to service areas. Meyer s study dataset included dental Medicaid claims for over one million beneficiaries who received care in North Carolina from 1985 through Patients and providers were identified by their county of origin. While Mayer was able to evaluate patient travel patterns, she was only able to do so at the county level. Rather than examine true market areas, this study primarily identified the extent to which patients travel between counties for dental care. There were major limitations to Mayer s approach to defining dentist market areas, including the fact that market areas were defined using medical claims from a low income population under age 21. The author acknowledges that few sources of data on private dental insurance claims exist (Mayer 1999), making the generalizability of these market areas limited. Additionally, data was available only for the patient s county of origin, instead of for more localized levels such as ZIP code or street address. By not including street addresses, Mayer was not able to analyze travel times between patients and providers. Without this level of detail, it is not unexpected that Mayer found counties to be the basic unit of service area;

60 43 there is no alternative conclusion that she could have reached using this data. Despite its limitations, Mayer s study is the only published research that defines dentist service areas using patient origin data. Service Area Analysis: Summary To our knowledge, no research has been conducted to empirically define dentist service areas (DSAs) in the U.S. besides Mayer s county-based analysis (1999). Patient origin analysis could be used in dentistry similar to its previous applications in medicine to identify areas where access to care is impeded by physical or system barriers including long travel times to the nearest available dentist and limited provider options. Dentist Workforce Evaluation Introduction As dentist locations change over time, policy-makers including dental school administrators and state safety-net providers must make decisions about where to invest their efforts at recruitment and placement of new dentists and dental offices. Consideration must be given to existing dentist locations while evaluating the need and demand characteristics of the population who must be served. Workforce planning aims to monitor and adapt the distribution of health care professionals in a geographic area. Workforce efforts in dentistry often focus on the state level since the profession is regulated primarily at

61 44 this level in the U.S. by licensing boards and because state-affiliated dental schools commonly are the major sources of local dentists. In 2007, 64% of licensed, active practice dentists in Iowa were born and professionally trained within the state (Kuthy et al. 2009). Within the constraints placed by state licensing regulations, dentists in the U.S. are free to establish their practices where they choose. Dentists are concerned with maximizing profit; it is advantageous for them to select practice locations with few nearby competitors and high demand for dental services from the local population. Most dentists choose their practice locations with little input from outside organizations or regional planners. However, Federally Qualified Health Centers (FQHCs) and other Community Health Centers (CHCs) commonly employ dentists to provide care to locally underserved populations. Therefore, it is of interest for planners to identify dental shortage areas in order to assist in optimizing the location of new dentists in such a way that unmet demand can be met while minimizing local competition. GIS in Dental Research To date, dental workforce research that incorporates the use of Geographic Information Systems (GIS) is not common in the U.S. literature. This likely is due in part to the difficulty in obtaining complete and accurate records for licensed dentists and hygienists in most states. U.S. studies that

62 45 have been published are generally of basic design and fail to capitalize on the advantages of using GIS over traditional methods of analysis. Prevously, Susi and Mascarenhas (2002) used GIS to map the geographic distribution of dentists in Ohio. Their research examined the distribution of dentists at the county and ZIP code level, dentists that billed Medicaid in a given year and safety net clinics that offered free or reducedcost dental services. The maps produced by this study displayed the number of dentists per county and per ZIP code, and displayed the location of Medicaid dentists and safety net clinics in a dot density map. This study highlighted the low dentist-to-population ratios found in Appalachian counties and the need to target interventions here. However, this study did not utilize the ability of GIS to merge multiple datasets and incorporate sociodemographic information beyond population size in analyses. The study would have been enhanced if the authors could have included information about the volume of patients or Medicaid services provided by dentists. For this study, a dentist who submitted a single Medicaid claim in 1998 was considered equivalent to a dentist who submitted thousands of claims. Krause et al. (2005) used GIS to merge workforce and census data to generate maps to visually examine Mississippi workforce trends and patterns. Analyses were conducted at the county level and used secondary data sets to calculate dentist-to-population ratios for each county (number of dentists per 100,000 people) in ten year increments from 1970 until 2000.

63 46 The percentages of dentists located within each county and the corresponding percentages of the population found within each county were calculated and compared with the median family income per county. Changes in the number of dentists per county from 1970 through 2000 were calculated. The authors divided counties into groups having low, medium, and high levels of dentists in order to evaluate any changes in conditions (ie, which counties decreased from a high to low level of dentists) (Krause, Frate, and May 2005). Calculated z scores were used to identify counties with a surplus or shortage of dentists; counties with a significantly higher or lower number of dentists than the state mean were considered to be locations of dentist maldistribution. This evaluation of Mississippi s dentist workforce is significant in that it used population and dentist information to evaluate changes over time in order to identify areas of need. However, no maps were produced that correlated changes in county income and dentist levels, although the authors had the data and capability to do so. Siegal and Marx (2005) surveyed general and pediatric dentists and safety-net clinics in Ohio about their treatment of young children. This survey attempted to correlate dentist attitudes and behaviors about treating vulnerable populations with dentist and practice characteristics (eg, years since graduation from dental school). GIS was used only to geocode practice locations and to identify the urban or rural character of each practice s

64 47 surrounding census tract. However, GIS software was not used in the final analysis to produce any maps or to link the geocoded data with other spatially referenced data. Identifying the urbanicity of dentist practice locations may provide some information about the service population, but without corresponding data that describes where patients come from and where neighboring dentists can be found, this study does little to explain interactions between geography and access to care. Dental Workforce Evaluation: Summary A major function of workforce evaluation in dentistry is to describe availability of providers once dimension of access within Penchansky and Thomas framework. GIS technology allows researchers to incorporate other dimensions into access research. This dissertation will specifically address aspects related to accessibility, accommodation, and acceptability by considering local characteristics of the dental care system at the service area level. Location Research Introduction Brandeau and Chiu (1989) designed a classification scheme that can be used to create a framework for analyzing location problems. Using this scheme, the objective decision variables involved in the problem and system parameters must be specified. One way to identify dentist shortage areas

65 48 using this framework would be to set as an objective the goal of minimizing travel time to the closest dentist and to set a parameter that this travel time must be less than 40 minutes. Forty minutes is the upper limit of travel time established by HRSA to designate a rational service area for dentistry (U.S. Department of Health and Human Services 1993). The decision variables for this location problem would be the number of and sites for practice locations. Further constraints can be placed on the framework by establishing the limitation that each dentist may serve a certain maximum population (eg, no more than 4,000 people). Because there are already hundreds of dentists in Iowa 1,455 in 2007 (Kuthy et al. 2009) the locations of existing dentists are used as a system parameter. The number of additional dentists for which planners wished to find optimal practice locations can then be specified. Dentist Equivalencies Identifying dentist shortage areas and optimal locations to add dentists within a geographic region presents several complications. First, demand for dental services generally is unknown; this can be approximated using national or regional estimates for population subgroups, if such data exist. Second, not all dentists are equivalent in terms of quantity and types of services provided. Some dentists may limit their practice to treat only adults or practice less than full-time. These variations in supply and demand represent uncertainty in system parameters. A third complication arises

66 49 because individuals may choose which dentist to visit based on personal preferences, instead of selecting the closest available practitioner. If planners wished to identify areas that lack dentists and have unusually high levels of unmet demand combined with known barriers to access (eg, high poverty levels), the system may focus only on FQHCs, CHCs, and those few private practitioners that treat high volumes of Medicaid patients. Within this system, the demand population could be defined as individuals with incomes at 100% of the federal poverty level (FPL), or as the Medicaid eligible and uninsured population. The objective for this location problem would remain the same as in the above example (minimize travel distance to the nearest dentist), but would have different constraints. Maximal Coverage Models Polgreen et al (2009) used a similar scheme to identify optimal locations for influenza sentinel surveillance sites in Iowa. Sentinels include individual primary care physicians along with hospitals and clinics; all are voluntary and receive no compensation for their services. Sentinel distribution is not geographically coordinated and potentially contributes to bias in influenza reporting. This lack of coordination creates similar distribution problems to those found in the dental workforce. Polgreen et al. used two versions of Maximal Coverage Models (MCMs) to calculate how many sentinel sites are needed for a hypothetical model that covers the same state population as existing resources and how much of the state population

67 50 could be covered if existing resources were distributed according to the MCM algorithm (Polgreen et al. 2009). MCMs function to maximize the population within a fixed distance given a set number of facilities. The authors included strategies for dealing with populations living just out of the state (edge effects), near Iowa borders, and to account for potential travel into Iowa for medical care which effectively increased the total population for the state. Polgreen et al. (2009) assumed that maximal coverage with a standard distance (ie, 20 miles) to a sentinel was ideal. For dental safety-net planners, this type of spatial modeling could be used to choose optimal locations for new clinics or satellite offices. However, research is lacking that examines the distribution of safety net providers and their service areas. Sentinel locations should be distributed evenly with respect to the population that they serve; all residents are vulnerable to influenza infection. This contrasts with the need for dental safety net providers who should be distributed to maximize access to underserved populations, including Medicaid patients especially children. Location Research: Summary By considering dentist shortage areas as a location problem, we are concerned primarily with spatial accessibility of providers (Guagliardo 2004). Within Andersen and Aday s behavioral model of health services use (Andersen 1995), spatial accessibility is one of many enabling characteristics. By failing to account for other factors that affect an individual s ability to

68 51 obtain health care services, identifying shortage areas based on workforce supply over-simplifies the objective of workforce planning equitable access to needed health care services. Andersen and Aday s behavioral model includes two other major domains that affect access to health care predisposing characteristics and need characteristics (Andersen 1995). The interactions among these three domains and how they lead to utilization are not well-studied in dentistry. Due to the complicated interactions between supply, demand, and need for dental services, empirical research is needed to identify dentist service areas (DSAs). Once research has identified general characteristics of DSAs, including geographical size, population size and associated socioeconomic characteristics, and overlap of neighboring dentists, future endeavors can examine changes in service areas over time. Environmental Justice Introduction Similar to activities aimed at steering the dental workforce towards a favorable distribution, urban planners are often concerned with the equitable distribution of facilities and services. An equitable distribution of resources requires that planners ensure that all members of a community, especially disadvantaged groups, have equal access to publicly-supported services such as parks, schools, and libraries (Sister, Wolch, and Wilson 2010; Talen and

69 52 Anselin 1998). It is worth noting that, while this phrase is not often explicitly used by dental workforce planners, environmental justice is a concern for those working to improve oral health and access to care on the national and local levels. A critical review of environmental justice is not presented here; instead, brief mention will be made of how spatial analysis has been utilized in other areas of equity planning research. Park Service Areas A recent study by Sister et al. used ESRI software to calculate Los Angeles park service areas based on surrounding population size (Sister, Wolch, and Wilson 2010). The authors goal was to determine whether parks were equitably accessible to residents based on an examination of population race, income, and age, and then identify areas lacking in access with the expectation that disadvantaged communities could use these results to enhance their funding applications. Park service areas (PSAs) were identified by assigning all metropolitan area residents to their closest park. The number of residents per park, or park pressure, was considered to be a measure of potential demand placed on each location. The study identified the locations of all public parks within the Los Angeles metropolitan area and randomly sampled these parks to obtain site information (eg, presence of litter and park maintenance) (Sister, Wolch, and Wilson 2010). PSAs were identified by creating Thiessen polygons around each location and then assigning all residents within the polygon to that PSA.

70 53 Pearson s correlation and Spearman s rank coefficients and were calculated to examine the relationship between park pressures (ie, population within a PSA) and population characteristics. Once PSAs were identified, Sister et al then simulated the effects that establishing new parks in two available sites would have on the overall patterns of park pressure found in Los Angeles. Sister et al. identified important limitations with their methods: residents are not obliged to visit their closest park and park pressure is an imprecise estimate of the actual demand for local services. However, the concept of distance decay presumes that residents will prefer to seek services that are closer to home (Sister, Wolch, and Wilson 2010). Planning for the location of urban parks in this method relies on the concept that an equitable distribution of parks will be one in which these are evenly distributed with respect to the population distribution across the geographic landscape. In Sister s study, adjacent parks were clustered together and assigned to a common PSA; this resulted in a decrease from approximately 1,800 parks to approximately 1,700 PSAs. Environmental Justice: Summary Talen and Anselin (1998) describes four ways that equitable distribution of services can be obtained. The first method is for all members of a community to receive equal benefits, regardless of need, demand, or ability to pay. Equity can also be obtained by distributing services based on need or in accordance with demand. Finally, market criteria can work to

71 54 establish equity; willingness of consumers to pay for and travel to obtain services can define whether resources are equitably distributed (Talen and Anselin 1998). When researchers attempt to identify health service areas, they must specify whether they are attempting to describe existing utilization patterns (eg, through the use of insurance claims) or to identify areas based on the principle of equitable distribution. This, in turn, will depend on research goals. If the goal is to improve access to dental services through the addition of a Federally-Qualified Health Center (FQHC) or other public facility, then we are concerned with establishing an equitable distribution of dentists. However, if the goal is to encourage a private practitioner to relocate into an area, steps should be taken to ensure that local demand is sufficient to support this as an independently functioning business endeavor. Use of Geocoded Data Introduction Spatial analysis of the dentist workforce requires that observations be geocoded; dentists must be linked with geographic coordinates that define their practice locations. Patient locations can also be geocoded to identify where patients come from and how far they travel for care. Methods

72 55 There are three main methods of geocoding: assigning observations to a geographic unit, interpolation, and parcel matching (Rushton et al. 2006). Assigning observations to a geographic unit results in aggregated data (eg, total number of dentists found within a county) that can be used for ecological analyses, but lacks accurate positional information. How the data will be used should be considered when determining which type of geocoding will be performed; if an epidemiological relationship is to be established, data must often be geocoded with a high level of positional accuracy (Rushton et al. 2006). Several issues must be considered before data are geocoded. Researchers must use standardized methods to specify addresses, including street prefixes and suffixes, and a rule must be established to assign coordinates to post office box addresses. Confirmation match thresholds must be set to determine how perfectly an address must match with a geographic base file s data; lower thresholds allow more matching, but decrease geocoding accuracy. Use of ZIP Codes Geocoding based on ZIP codes is not recommended since these boundaries change significantly over time and these changes are not welldocumented (Krieger et al. 2002). Alternatively, U.S. census data have been aggregated for ZIP Code Tabulation Areas (ZCTAs) since the year ZCTAs are aggregations of census blocks that attempt to correspond to ZIP

73 56 code areas. However, these are a poor representation of ZIP code areas and geocoding population data based on ZCTAs is not recommended for linking data geocoded by ZIP code (Krieger et al. 2002; Rushton et al. 2006). The limitations associated with using ZIP codes as units of analysis have not prevented researchers from using them. The Dartmouth Atlas Project based their definitions of Hospital Service Areas and Hospital Referral Regions on U.S. ZIP codes. Data Confidentiality One major concern for researchers working with geocoded data is the issue of maintaining confidentiality of sensitive information. Individual privacy can be protected prior to analysis if more than one party will have access to data or it can be protected when results are depicted. In the case of data sharing, geocoding is often performed by an outside party. In this scenario, protected health information can be removed from the records that will be geocoded (Christen and Churches 2006). Elaborate protocols for removing identifiers and linking data can be developed based on the research needs. If maps accurately depict individual locations, names and addresses could be re-identified via inverse geocoding (Rushton et al. 2006). Published health data is commonly aggregated to an area of minimum population with a minimum number of cases to protect patient privacy. Masking techniques may be used when individual health data is used in analysis. Examples of masking include aggregating data and random perturbations in locations.

74 57 Ideally, masking should prevent the identification of individuals but still retain enough spatial accuracy for epidemiological analysis (Rushton et al. 2006). Use of Geocoded Data: Summary Optimal use of GIS technology in health services research requires detailed data: patient location, provider location, types of services provided, transportation networks, and demographic data all may be incorporated. Detailed data translates into more informative analyses: street addresses of patients and providers offer a higher level of spatial resolution than just ZIP code or county. Review of the Literature: Summary Recommended preventive dental interventions include the placement of dental sealants and topical fluoride, water fluoridation, and regular visits to a dentist. Despite these evidence-supported preventive strategies, dental caries maintains a tenacious presence among American children. Children who suffer from untreated decay are at risk for adverse long term consequences, including increased disease later in life. Accessing dental care is known to be difficult or impossible for many vulnerable pediatric populations, including very young children, Medicaid or uninsured individuals, individuals with chronic health conditions, and those living in rural areas. The persistent disparities between the urban and rural

75 58 distribution of health care providers calls for innovative research that investigates how this situation is maintained and to what degree it acts as a barrier to care. Research that can identify populations experiencing the greatest barriers to the receipt of dental care will support federal and state initiatives that are designed to alleviate workforce shortages and other access barriers. Access to care research is facilitated through the use of a model that conceptualizes the multilevel factors that influence children s oral health outcomes. This dissertation will examine the relationships between workforce supply and distribution, geographic barriers to access and the receipt of preventive dental services by Medicaid children in Iowa within the Fisher- Owens et al conceptual model. The two levels of interest for this dissertation include the child and DSA levels. Within the child level, physical attributes, developmental influences, insurance coverage, and health behaviors will be considered. The DSA level will function in place of the community level specified in the original Fisher-Owens model. Within the DSA level, oral health influences for consideration will include the cultural and social environment, and attributes of the dental care system. The three major objectives of this dissertation will evaluate characteristics of the dental care system in Iowa through a workforce evaluation, identify DSAs for the pediatric Medicaid population, and assess multi-level influences to access of preventive dental services. Service area

76 59 analysis will offer enhanced analysis of the interactions among the influences of oral health by controlling for regional variations in supply and demandrelated factors that are not otherwise measured.

77 60 CHAPTER III GENERAL METHODS AND MATERIALS Overview This dissertation examines the relationships between workforce supply and distribution, geographic barriers to access, and the receipt of preventive dental services by Medicaid children in Iowa. A service area analysis will identify local dental market areas and facilitate analysis of barriers by controlling for regional variations in supply and demand related factors that are not otherwise measured. Data This study relies on data from three major sources: the Iowa Dentist Tracking System (IDTS), Iowa Medicaid claims and enrollment databases, and the U.S. Census. Iowa Dentist Tracking System The IDTS attempts to gather and maintain information about all licensed dentists in the state. The IDTS is administratively located within the University of Iowa s Carver College of Medicine (Kuthy and others 2009). Staff members contact dentists semiannually to monitor changes in practice location, work history, and practice arrangements. The 12/2009 and 12/2010 data releases are used for this dissertation and includes the following variables:

78 61 1. Dentist name 2. Provider identifier - a unique, numeric code assigned to each dentist by the IDTS used to link providers across multiple years 3. Dentist sex male or female 4. Dentist age 5. Primary practice location street address, city, and ZIP code 6. Full-time equivalent average hours worked per week over the previous 12 months 7. Specialty primary practice specialty, including: general practice, endodontics, oral pathology, oral radiology, oral surgery, orthodontics, pediatric dentistry, periodontics, prosthodontics, and public health dentistry 8. Primary practice activity including private practice, administration, hospital staff, community health or local government, dental school faculty, hospital staff, state or federal government, and veteran s administration Dentists used in this study include those whose primary practice specialty is reported as general practice or pediatric dentistry and those whose primary practice activity is reported as private practice, community health, and hospital staff. These practice activities were selected because they represent the dentists who provide direct patient care. General practice and pediatric dentists are of interest due to their role in providing primary care to children.

79 62 Iowa Medicaid Program Dental claims from the Iowa Medicaid Program for dates of service during calendar years (CYs) 2008 through 2010 will be used to identify dentist service area boundaries. The claims files include all claims submitted by a dentist from 01/01/2008 to 12/31/2010 on behalf of any Iowa Medicaid enrollee who was between the ages of 3 and 18 years old during CYs 2009 and Claims data include the following variables: 1. Enrollee identification number used to link enrollees between claims and enrollment files 2. Enrollee address at time of service ZIP code and county 3. Date of birth 4. Age at time of service 5. Enrollee sex male or female 6. Race/ethnicity - designated by the Iowa Medicaid Program into the following categories: a. White b. Black c. American Indian d. Asian e. Hispanic f. Pacific Islander g. Multiethnic Hispanic

80 63 h. Multiethnic Unknown i. Other j. Missing 7. Provider name 8. Billed procedures identified using Current Dental Terminology (CDT) codes Previous analyses have found race/ethnicity to be missing from approximately 30% of Medicaid claims (Chi and others 2010; Momany, Damiano, Carter 2009). Because of this and Iowa s relatively racially homogeneous population, race/ethnicity will be re-categorized as White, Black, Other (categories c through g, and i), and Unknown/Missing (categories h and j). Medicaid enrollment files will be used to provide additional information about enrollees. Individuals will be linked between claims files and eligibility files using the enrollee identification number. Variables obtained from this data source include: 1. Medicaid eligibility program identifies the program through which each enrollee was eligible for Medicaid 2. Monthly enrollment total number of months enrollee was enrolled during CYs 2008 through 2010 Eligibility program and total monthly enrollment during the study period will be used to adjust for analyses that examine factors associated with the

81 64 utilization of dental services. Previous research has demonstrated that rates of dental utilization vary substantially based on the length of time that children have been enrolled in public dental programs (Damiano, and Momany, Crall 2006). U.S. Census Dental Service Areas (DSAs) will be created by aggregating ZIP codes into groups based on patient origin data. The shapefiles used to identify the geographic boundaries of service areas will be obtained from the U.S. Census Bureau. ZIP code geographic boundaries will be approximated using ZIP Code Tabulation Areas (ZCTAs). Population characteristics at the ZCTA level from the Decennial Census for 2010 will be obtained from the U.S. Census. These will be used to adjust for analyses that examine factors associated with the utilization of dental services. An additional variable, population density, will be calculated using data describing geographic area of DSA boundaries and ZCTA population totals. Study Population This dissertation will focus on primary care dentists in Iowa and children ages 3 to 18 years old enrolled in Medicaid who received dental services from the dentists during CYs 2008 through The specific

82 65 inclusion and exclusion criteria for each study will be described in more detail in the relevant manuscripts. Study Design Objective 1 The objective of the first study is to identify DSAs for orthodontic treatment among Medicaid enrolled children and adolescents. Objective 1.a Delineate orthodontic DSAs for the Medicaid enrolled study population based on ZIP codes of patients and dentist practice locations. Assigning patient origin ZIP codes to provider locations based on where a plurality of patients received care will result in the creation of service areas. Service areas will be adjusted based on requirements for minimum population size and contiguity. Objective 1.b. Describes rates of orthodontic utilization, identify patient and family characteristics associated with utilization, and describe geographic variation of these characteristics. Our primary interest is the association between utilization and residential urbanicity. Objective 2 This project will identify primary care DSAs in Iowa using the methods developed for Objective 1. The central hypothesis of this project is that

83 66 methods of small area analysis, which have been developed and refined in previous hospital utilization and primary medical care studies (Goodman and others 2003; Wennberg and Gittelsohn 1973; Wennberg and Gittelsohn 1982), can be used to delineate service areas that approximate local market areas better than counties. Identifying dentist service areas will facilitate shortage area designation, outcomes research, and workforce planning efforts. Objective 2.a Identify DSAs based on ZIP codes of Medicaid patients and dentist practice locations. Assigning patient origin ZIP codes to provider locations based on where a plurality of patients received care will result in the creation of service areas. Objective 2.b Describe DSA populations, patients, and dentists. The general DSA population and the local Medicaid population within each service area will be described. General populations and DSA characteristics will be compared. Objective 3 This project will consider multilevel factors, including variables that describe regional spatial accessibility and the association between these and the receipt of a preventive dental care by individual children. Potential oral health correlates operating on the child and DSA levels will be considered in this analysis.

84 67 Dependent Variable Because preventive dental care is associated with improved oral health status and long-term outcomes for children, receipt of a preventive dental visit provided by a general or pediatric dentist will serve as the primary outcome of interest for this study. Independent Variables The independent variables considered for this study include several traditional predictors of oral health along with several measures of spatial accessibility. Analysis Descriptive statistics will be generated for the overall study population at the child and DSA levels. Hierarchical logistic regression with random effects will be used to model the effects of individual and service area characteristics on the receipt of preventive dental care.

85 68 CHAPTER IV STUDY 1 GEOGRAPHIC ACCESSIBILITY AND UTILIZATION OF ORTHODONTIC SERVICES AMONG MEDICAID CHILDREN AND ADOLESCENTS Study 1 Abstract Objectives To describe rates of Medicaid funded services provided by orthodontists in Iowa to children and adolescents, identify factors associated with utilization, and describe geographic barriers to care. Methods We analyzed enrollment and claims data from the Iowa Medicaid program for a three-year period, January 2008 through December Descriptive, bivariate, and multivariable logistic regression analyses were performed with utilization of orthodontic services as the main outcome variable. Service areas were identified by small area analysis in order to examine regional variability in utilization. Results The overall rate of orthodontic utilization was 3.1-percent. Medicaid enrollees living in small towns and rural areas were more likely to utilize orthodontic services than those living in urban areas. Children who had an oral evaluation by a primary care provider in the year prior to the study period were more likely to receive orthodontic services. Service areas with

86 69 lower population density and greater mean travel distance to participating orthodontists had higher utilization rates than smaller, more densely populated areas. Conclusions Rural residency and increased travel distances do not appear to act as barriers to orthodontic care for this population. The wide variability of utilization rates seen across service areas may be related to workforce supply in the form of orthodontists who accept Medicaid-insured patients. Referrals to orthodontists from primary care dentists may improve access to specialty care for Medicaid enrollees. Introduction Access to orthodontic treatment for low income children has been mandated by the Early and Periodic Screening, Diagnosis and Treatment (EPSDT) Program since 1967 (U.S. Health Resources and Service Administration 2012). Despite federally required coverage, minority and lowincome children are consistently found to utilize orthodontic services at lower rates than white, affluent children while demonstrating higher levels of treatment need (Anderson May 2010; Nelson et al. 2004; Okunseri et al. 2007; Proffit, Fields, and Moray 1998). Limited research has examined the effects of geographic barriers on the receipt of specialty dental care.

87 70 In 2004, approximately 6 percent of U.S. children and adolescents insured through Medicaid received orthodontic care, compared to 17 percent of privately insured youth (Okunseri et al. 2007). Rates of orthodontic utilization by publicly insured children and adolescents vary considerably by state. Less than 1 percent of Medicaid eligible children in Washington state received orthodontic treatment in 1999 (King et al. 2006) and less than.5 percent of Medicaid eligible children in North Carolina received orthodontic treatment from 2002 to 2003 (Murdock et al. 2010). These rates are especially low given estimates that approximately 14 percent of all children and 29 percent of adolescents have severe or very severe handicapping malocclusions (Murdock et al. 2010). Among all income groups, utilization rates vary substantially with race and ethnicity. Even after adjusting for income and insurance status, Black and Hispanic children are less likely to have a history of orthodontic treatment than Whites (Okunseri et al. 2007; Proffit, Fields, and Moray 1998; Wheeler et al. 1994) Despite these differences in the receipt of orthodontic treatment, professionally defined need for treatment has been found to be similar among all racial and ethnic groups, with black youth demonstrating a greater proportion of individuals categorized as having severe need for treatment (Shi et al. 2008). Patient gender and age also demonstrate associations with orthodontic service utilization rates. Females consistently receive treatment at higher

88 71 rates than males (Anderson 2010; Okunseri et al. 2007; Wheeler et al. 1994), even though clinical data suggest that need may be greater among males.(wheeler et al.1994) Nationally, active receipt of orthodontic services peaks at age 14 and is highest for the age group including 12 to 18 year olds (Guay, Brown, and Wall 2008; Okunseri et al. 2007). A component of orthodontic need depends heavily on the subjective perceptions of patients, parents, and providers. A child and his or her family must consider a malocclusion aesthetically or functionally severe enough to seek treatment, a dentist must concur, and then accept or refer the case for treatment. This relationship is further complicated for those with public insurance. Child, family, orthodontist, and the state agent for case review must agree on treatment need. Approval for treatment in Iowa requires case review based on the process described by Salzmann (1968) and is approved for the most handicapping malocclusions only (State of Iowa Department of Human Services 2002). Conditions are considered to be handicapping if they are potentially detrimental to oral health or an individual s well-being through temporomandibular joint dysfunction, mastication impairment, increased risk of periodontal disease or caries, or speech impairment. Recent changes in Iowa Medicaid have increased the case complexity required for approval of orthodontic treatment. These changes may limit the orthodontic care that can be provided by a general or pediatric dentist for Medicaid children. The Iowa Medicaid program covers orthodontic services

89 72 for individuals through age 20. In addition to strict guidelines for approval of orthodontic services, access to care is also limited through low Medicaid participation of dentists, especially specialists. One study found that fewer than 10 percent of North Carolina orthodontists submitted claims for at least ten Medicaid patients during a 3-month period during 2005 (Murdock et al. 2010) It is unknown how Medicaid participation by orthodontists varies with urban or rural practice location. While previous research has found utilization to be higher among children living in urban areas than those living in rural areas (Wheeler et al. 1994), it is unclear whether this is due to variation in patient demand or variation in the supply of services offered by providers. Variation in utilization of health services can be examined using small area analysis, which defines local health market areas by examining the population s use of common resources, such as hospitals and physicians (Paul-Shaheen, Clark, and Williams 1987). This approach to identifying market areas has been employed in medical research since the 1950s in the US and is the foundation for research that examines regional variation in medical expenditures and utilization (Paul-Shaheen, Clark, and Williams 1987; The Dartmouth Institute for Health Policy and Clinical Practice). Despite its longstanding use in medicine, the application of small area analysis to examine variation in oral health services is nearly nonexistent.

90 73 We hypothesize that variation in utilization of orthodontic services among Medicaid children and adolescents is a function of individual need, demand, and access to care. The goal of this study is to describe rates of Medicaid funded services provided by orthodontists in Iowa to children and adolescents, identify patient and family characteristics associated with utilization, and describe geographic variation in utilization. We will identify dentist service area boundaries at the zip code level in order to account for regional variation in orthodontic access and demand. Incorporating information about local market areas will allow us to control for regional characteristics that may affect perceived need and ability to access care. Our primary interests are the associations between urbanicity, travel distance, and the receipt of orthodontic services. Methods This study used a retrospective cohort design to examine access to orthodontic services among Medicaid enrolled children and adolescents in the Iowa Medicaid program during 2008 to Data We analyzed data from Iowa Medicaid enrollment and claims files from calendar years (CYs) to examine rates of orthodontic utilization. The claims files included all claims submitted to Iowa Medicaid by orthodontists for services rendered to Medicaid enrolled children. Dental

91 74 services were coded using Current Dental Terminology. Medicaid enrollment files were obtained for all individuals who met the inclusion criteria described below. After claims data and enrollment data were linked, personal identifiers were removed in order to protect confidentiality. This study was approved by the University of Iowa Institutional Review Board. Study Population This study focused on Medicaid enrolled children and adolescents who were ages 6 to 18 years (hereafter referred to as children ) during CY 2008 to Because the utilization rate for 3-5 year olds was so low (<.01%), this age group was not included in our analyses. The study population was limited to individuals eligible for Medicaid through Supplemental Security Income (SSI) or based on income eligibility requirements. Rates of orthodontic utilization were examined among individuals who were continuously enrolled in the Iowa Medicaid program for at least 11 months during the study period. Age was calculated at the beginning of this qualifying enrollment period. The period of continuous enrollment is designed to correspond to the HEDIS requirement for children to be continuously insured for at least 11 months during the study period, as specified in HEDIS protocols (Agency for Healthcare Research and Quality 2006). Dental Service Areas (DSAs) DSAs were identified using claims data submitted by orthodontists for care provided to all eligible Medicaid enrolled children. Zip code level data

92 75 provided on the first claim per individual were used to identify the Medicaid enrollees residential and orthodontists practice locations. DSA boundaries were generated using the methods of small area analysis described by Wennberg and Gittelsohn (1982). A minimum of ten Medicaid enrollees who utilized orthodontic services was set as a service area threshold in order to protect confidentiality and to produce reliable statistics. Zip codes containing no enrollees who utilized orthodontic services were assigned to the nearest provider destination as identified by road network analysis. The initial DSA assignments were adjusted to improve contiguity of service area boundaries. Zip code tabulation area (ZCTA) shapefiles from the U.S. Census Bureau were used to approximate zip code geographic boundaries. Study Variables The dependent variable of interest was whether a child utilized orthodontic services during the study period. Utilization was identified as submission of a claim by an orthodontist to Iowa Medicaid on behalf of a continuously enrolled individual for Current Dental Terminology (CDT) codes D8010 through D8693. We examined individual and family demographic and socioeconomic characteristics as predictors of orthodontic utilization, including: age, sex, race/ethnicity, Medicaid eligibility program, and family poverty level. Race/ethnicity is optionally reported by parents to Iowa Medicaid in the enrollment files. This was categorized as White, Black, Hispanic, other, and unknown or missing. Other individual predictors

93 76 included length of enrollment during the study period (36 months vs months), and whether a child received an oral evaluation by a primary care dentist during Primary care dental visits were identified by the submission of a Medicaid claim by a general or pediatric dentist for a routine or comprehensive oral exam in 2007 (D0120, D0145, or D0150). Receipt of a primary care oral evaluation was hypothesized to be positively associated with the likelihood of utilizing orthodontic services. Previous research has indicated that orthodontists are more likely to accept Medicaid patients if they are referred by another dentist (Im et al. 2007). The urbanicity of each Medicaid enrollee s place of residence was measured using rural-urban commuting area (RUCA) codes. RUCA codes identify census tracts or zip codes as metropolitan (urbanized area), micropolitan (non-metro, urban cluster of at least 10,000 people), small town (urban cluster of 2,500 10,000), and rural (no urban cluster, or urban cluster <2,500) (USDA Economic Research Service 2005). Zip code approximations of census tract-based RUCA codes were obtained from the University of Washington (WWAMI Rural Health Research Center 2005) and linked with residential zip codes. Service area level variables were created, including rates of orthodontic utilization, mean travel distance between patient zip codes of origin and provider zip code destination (weighted by the total Medicaid

94 77 enrolled population per zip code), primary care dentist-to-population ratios, and population density. Population density was used as a measure of service area urbanicity. Statistical Analysis We generated descriptive statistics of the study population and the comparison groups (utilizers versus non-utilizers). Chi-square tests and analyses of variance (ANOVA) were used to evaluate bivariate relationships between demographic and socioeconomic variables and orthodontic utilization. Multivariable logistic regression was used to examine the relationship between residential urbanicity and orthodontic utilization while adjusting for known predictors of utilization. Maps were generated to examine geographic variation in orthodontic utilization rates across service areas. DSAs were linked with zip code tabulation area (ZCTA) population data from the 2010 U.S. Census and workforce data from the 2010 Iowa Dentist Tracking System (IDTS). IDTS maintains current information about licensed dentists actively practicing in the state (Kuthy et al. 2009). Service areas were grouped into quartiles of increasing orthodontic utilization rates. Characteristics were summarized for each quartile using counts, means, and proportions of service area-level characteristics. Statistical analyses were performed using IBM SPSS Statistics Version 20. Road network analysis and the generation of service area

95 78 boundaries were conducted using ESRI ArcMap A significance level of.05 was used in all hypothesis tests. Results During 2008 through 2010, there were 116,330 children and adolescents who were continuously enrolled for at least 11 months in the Iowa Medicaid program. The overall rate of orthodontic utilization among the study population was 3.1 percent. The most commonly billed procedures, excluding radiographs, were for comprehensive orthodontic treatment (CDT D8070 and D8080). Fewer than 4 percent of the utilizers received interceptive treatment (N=137) and there were no claims submitted during 2008 through 2010 for limited orthodontic treatment. The majority of children in the study population were white (57.4 percent), although approximately 22 percent had missing data for race/ethnicity (Table 1). Mean length of enrollment for children who met the 11 month minimum enrollment requirement was 28 months (SD 8.9); 40 percent of children were enrolled for the entire 36-month period. Orthodontic utilizers differed significantly from non-utilizers by age, sex, race/ethnicity, length of enrollment, aid category, and poverty level. Utilizers were also significantly more likely to have had an oral evaluation by a primary care provider during the year before the study period. A significantly greater proportion of females received orthodontic services than males. White

96 79 children accounted for 67.6 percent of the population with a visit, even though they comprised 57.4 percent of the study population. Individuals who received orthodontic services were significantly more likely to have been enrolled for the entire year preceding the study period and more likely to have been enrolled for the duration of 2008 through 2010 (36 months). Children living in small towns and rural areas were significantly more likely to have received orthodontic services than those living in metropolitan and micropolitan areas. After adjusting for covariates in the logistic regression model, children living in small towns and rural areas of Iowa were significantly more likely to have received orthodontic services than those living in metropolitan or micropolitan areas (Table 2). Children with a primary care visit in 2007 were significantly more likely to have received orthodontic services during the study period than those with no visit (OR = 1.60). Children between the ages of 10 through 13 were the most likely age group to receive orthodontic services. In order to facilitate comparisons, the 2010 Medicaid claims data were compared with the Iowa Dentist Tracking System (IDTS) for that year. The IDTS maintains information about all licensed dentists actively practicing in the state. Forty-two of the 85 licensed orthodontists (49.4 percent) practicing in Iowa during 2010 submitted a claim to Medicaid during that year. Mean number of Medicaid enrolled children treated by these providers in 2010 was

97 (SD 41.8) and ranged from 1 to 185. Thirty-two orthodontists submitted claims to Medicaid for the treatment of ten or more individuals. After aggregating zip codes based on where a plurality of Medicaidenrolled children received orthodontic treatment, 19 DSAs were created (Figure 1). Descriptive characteristics of orthodontic DSAs are provided in Table 3. Service areas are named for the provider location within each DSA. Mean number of zip codes per DSA was 49.2 (SD 36.2). The Council Bluffs DSA included 145 zip codes the most of any service area. Council Bluffs also had the greatest total population (n=284,501). The smallest service area in terms of number of zip codes (n=3) and total population (n=63,726) was South Des Moines. Mean travel distance per DSA ranged from 5.1 miles to 49.2 miles. Three service areas in the Des Moines metro area had the shortest mean travel distances ( miles). The greatest mean travel distances were found in the Council Bluffs DSA (41.9 miles) and the Iowa City/Coralville DSA (49.2 miles). Grouping orthodontic DSAs into quartiles of increasing utilization rates allowed us to compare findings from the individual-level analysis with the geographic variation seen across the state (Table 4). The service area analysis supported the results of the logistic regression model that demonstrated increased utilization rates among children living in small towns and rural areas. Service areas with the lowest utilization rates had the greatest mean population density (566 people per square mile), while the

98 81 DSAs with the highest utilization rates were the least densely populated. Additionally, areas with the lowest utilization rates had the largest minority populations. A sensitivity analysis was conducted to examine the effects of combining the three smallest Des Moines metro service areas: West Des Moines, Des Moines, and South Des Moines (see enlarged area, Figure 1). A combined Des Moines DSA would have the largest total service area population (n=383,732) and the fourth lowest utilization rate (.70 percent). Substantial variation in utilization rates would be obscured by combining these three metro DSAs, which displayed utilization rates of.50 to 1.52 percent. While geographically large DSAs (eg, Council Bluffs) have the potential to obscure local variation in utilization rates, it should be noted that providers serving this population were located in only one zip code within the DSA. The high utilization rate found in Council Bluffs was due to several orthodontists located in close geographic proximity providing services to a large number of Medicaid-enrolled children during the study period. Discussion The Medicaid program in Iowa offers coverage to children from lowincome families for specialty dental care, including orthodontics. Our findings of increased orthodontic utilization among females and among White children enrolled in Medicaid compared to males and other racial and ethnic groups

99 82 are congruent with previous studies (Anderson 2010; Nelson et al. 2004; Okunseri et al. 2007; Proffit, Fields, and Moray 1998; Wheeler et al. 1994). Our findings that an oral examination by a primary care dentist prior to the study period was associated with orthodontic utilization supports the importance of an established dental home. Primary care dentists may facilitate access through referrals for specialty services, including orthodontics. The discovery that children living in rural areas of Iowa were more likely to receive services from an orthodontist than those living in metropolitan or micropolitan areas was contrary to our working hypothesis that geographic accessibility would be an important factor limiting access to care. This relationship between urbanicity and utilization differs from the conclusions of previous research (Borders 2006; Wheeler et al. 1994) Differences in care seeking behavior among urban and rural dwellers may arise from different priorities in time budgeting; low income families in rural communities may have learned to accommodate the increased time commitment related to routine travel patterns associated with activities such as work, school, and health care services. The high degree of variability in rates of orthodontic treatment across service areas indicates that variation in provider availability may pose a more significant barrier to care for publicly insured children and adolescents than travel distance. Future studies that investigate characteristics of orthodontists who provide high volumes of care

100 83 to this population, including how their offices come to accept Medicaid patients, are needed. For example, these providers may only accept referrals from preferred primary care colleagues. This information could be used to design effective policy aimed at improving access to specialty dental care. Our use of service areas is supported by findings from the individual-level evaluation of access to orthodontic treatment among Medicaid-enrolled children. The similarities seen at both levels of analysis validate the use of dentist service areas generated through small area analysis of patient origins as a research tool. This study has several limitations. First, the data are cross-sectional and limited to a single state, which may limit generalizability of our findings. Second, the relationship between rural residency and increased utilization may not apply to other types of dental care or other populations. Medicaid patients may have more severe malocclusions than privately insured patients who seek care. This treatment severity may confound the relationship between urbanicity and utilization. Finally, due to the small number of orthodontic service areas in Iowa, we were unable to conduct multivariable analyses to examine associations between regional characteristics and utilization rates. Future research is planned that will examine regional variation of access to other dental services, including preventive and emergency services. Incorporating information about market areas for those

101 84 services will allow us to examine how utilization is affected by workforce supply.

102 85 Table IV.1 Description of continuously enrolled Iowa Medicaid children ages 6 through 18 years ( ) and significance testing results between orthodontic utilizers and non-utilizers by Chi-square analysis Variable Age Mean ± SD 6-9 years Sex Female Male Race/ethnicity White Black Hispanic Other Unknown/missing Length of enrollment ( ) Mean ± SD (months) months 36 months Previous enrollment (2007) 0 months 1-11 months 12 months Medicaid aid category Income eligible SSI Federal Poverty Level 0% 1-133% > 133% Primary care oral evaluation (2007) Yes No Urbanicity of residential zip code Metropolitan Micropolitan Small town Rural Total Ortho nonutilizers Ortho utilizers population N=116,330 N=112,749 N=3, ± ,315 (37.2) 35,355 (30.4) 37,660 (32.4) (50.9) (49.1) (57.4) 9904 (8.5) (8.7) 3797 (3.3) (22.2) 28.0 ± (60.0) (40) (28.0) (23.5) (48.6) (95.5) 5258 (4.5) (26.9) (69.8) 3,866 (3.3) (32.0) (68.0) (50.1) (17.7) (17.9) (14.3) N(%) 11.7 ± ,593 (37.8) 33,223 (29.5) 36,933 (32.8) (50.6) (49.4) (57.1) 9760 (8.7) 9798 (8.7) 3699 (3.3) (22.3) 27.9 ± (60.4) (39.6) (28.2) (23.5) (48.3) (95.4) 5159 (4.6) (27.1) (69.6) 3,754 (3.3) (31.5) (68.5) (50.3) (17.8) (17.8) (14.1) 11.5 ± (20.2) 2,132 (59.5) 727 (20.3) 2113 (59.0) 1468 (41.1) 2420 (67.6) 144 (4.0) 280 (7.8) 98 (2.7) 639 (17.8) 30.7 ± (47.9) 1865 (52.1) 762 (21.3) 810 (22.6) 2009 (56.1) 3482 (97.2) 99 (2.8) 693 (19.4) 2776 (77.5) 112 (3.1) 1640 (45.8) 1941 (54.2) 1525 (42.6) 540 (15.1) 803 (22.4) 713 (19.9) Significance < < < < < < < <0.0001

103 86 Table IV.2 Multivariable logistic regression model for orthodontic utilization by Medicaid-enrolled children and adolescents, ages 6-18, during CY Odds Ratio 95% Confidence Interval Significance Age 6-9 years Sex Female Male Race/ethnicity White Black Hispanic Other Unknown/missing Length of enrollment ( ) months 36 months FPL 0% 1-133% > 133% Primary care oral evaluation (2007) Yes No Urbanicity of residential zip code Metropolitan Micropolitan Small town Rural < < < < < < < < <0.0001

104 87 Table IV.3 Descriptive characteristics of orthodontic DSAs (N=19) DSA Recipient zip codes (n) Medicaid enrollees (n) Ortho utilizers (% of enrollees) Weighted travel distance (miles) Total population (2010) Population density Land area (square miles) Medicaid orthodontis ts (2010) Ames Bettendorf Cedar Rapids Charles City Clinton Council Bluffs Des Moines Fort Dodge Indianola Iowa City/Coralville Marshalltown Muscatine Newton Sibley/Spencer Sioux City South Des Moines Urbandale Waterloo/Cedar Falls West Des Moines Weighted by the number of Medicaid enrollees ages 6-18 years per zip code.

105 88 Table IV.4 Descriptive characteristics of orthodontic DSAs grouped by quartiles of utilization rates Low utilization rates Medium low utilization rates Medium high utilization rates High utilization rates Mean utilization rate per 100 Medicaid enrollees Number of service areas in quartile Mean population density (per square mile) Mean geographic area (square miles) Mean total population, 2010 (n) 136, , , ,083 White population (%) Black population (%) Hispanic population (%) Mean travel distance to orthodontist (miles)

106 Figure IV.1 Orthodontic DSAs for Medicaid-enrolled children in Iowa (n=19) 89

107 90 CHAPTER V STUDY 2 IDENTIFYING MARKET AREAS BASED ON UTILIZATION OF PRIMARY CARE DENTAL SERVICES BY MEDICAID-ENROLLED CHILDREN AND ADOLESCENTS Study 2 Abstract Objective To compare dental service areas delineated based on patient origin with counties as representations of market areas. Data Source Enrollment and claims data ( ) for Medicaid-enrolled youth ages 3-18 in Iowa. Study Design Claims submitted by primary care dentists were used to identify service areas through methods of small area analysis. Initial service area boundaries were adjusted for geographic contiguity and to meet a minimum population threshold. Characteristics of these dental service areas were compared to county characteristics in order to evaluate the use of counties as representations of dental market areas. Data Collection/Extraction Methods Claims submitted to Iowa Medicaid by general and pediatric dentists on behalf of individuals meeting the study inclusion criteria were used to identify zip codes of patient residence and provider location.

108 91 Principal Findings Small area analytic methods identified 113 primary care dental service areas in Iowa. Approximately 59-percent of care received by Medicaidenrolled children took place within their assigned service area versus 52- percent of care within their county of residence. Residents of counties with no dentists available to serve Medicaid-enrolled youth frequently travel to neighboring counties for care. Conclusions Small area analysis of insurance administrative data can be used to successfully determine dental service areas and more accurately describe regional travel patterns than using county boundaries. Introduction Studies of health care markets allow researchers to examine competition among providers, allocation of resources, and variation in utilization. In economics, markets are defined as that "set of suppliers and demanders whose trading establishes the price of a good" (Stigler and Sherwin 1985). However, the inavailability of price data limits the relevance and application of this definition to health care markets (Zwanziger, Melnick, and Mann 1990). A more appropriate definition of a market for health care delivery is the location of the suppliers and purchasers of a product (Elzinga and Hogarty 1973). Hospital markets, or service areas, are of interest for

109 92 planners interested in expanding services or adding new facilities (Garnick and others 1987). These are also a matter of legal interest in antitrust cases where market analysis forms the basis for measuring competition among hospitals (Elzinga and Hogarty 1973). In terms of access to health care, service areas offer a framework for studying geographic variation in workforce supply and utilization of services. The geographic boundaries of health service areas have been delineated using various approaches (Baker 2001; Garnick et al. 1987; Zwanziger, Melnick, and Mann 1990). The most common approach used in dentistry is to define service areas based on fixed geopolitical boundaries, such as counties (Krause, Frate, and May 2005a; Mayer 1999; Saman, Arevalo, and Johnson 2010). Using geopolitical units to approximate service areas offers several advantages, such as the ready availability of data, including information from the U.S. Census Bureau. However, these units are arbitrary designations that may not adequately describe the actual market for dental services. In rural areas particularly, counties can be too small to capture the locations of providers and the populations that they serve. As an alternative to geopolitical units, service areas can be defined by the region encompassed by a circle of fixed radius around each provider (Baker 2001; Garnick et al. 1987; Zwanziger, Melnick, and Mann 1990). This method offers a straight-forward method of identifying service areas, but

110 93 offers only a measure of potential for access (Garnick et al. 1987). Applying a fixed radius to all service areas may be inappropriate, including rural areas mountainous regions, where populations often travel further for health care services (Baker 2001). A third option, variable market area approaches, can identify service areas that account for differences in market characteristics (Baker 2001; Garnick et al. 1987; Zwanziger, Melnick, and Mann 1990). Variable market approaches are prevalent in health services literature where methodological details differ based on research priorities. One method involves varying the radius around a provider until it encompasses a certain percentage of the actual market population. This is the method that was used by Mayer (Mayer 1999) in a study of dental service areas for Medicaid-enrolled children in North Carolina. However, that study relied on counties as the units for service area aggregation. One of the most common variable market approaches used to determine hospital and physician market areas, which has not yet been used in dentistry, is small area analysis (Goodman et al. 2003; Klauss et al. 2005). This technique relies on patient origin data to generate service areas that are aggregates of small geopolitical units, such as zip codes or census tracts. Early small area studies by Wennberg and Gittelsohn identified hospital service areas in New England using Medicare admissions data (Wennberg and Gittelsohn 1973; Wennberg and Gittelsohn 1982). The most common

111 94 criteria used to assemble health care markets using this method include plurality and geographic contiguity (Paul-Shaheen, Clark, and Williams 1987; Rohrer 1987; Wennberg and Gittelsohn 1973; Wennberg and Gittelsohn 1982). Plurality is the fundamental requirement of small area analysis, whereby geographic regions are assigned to service areas based on where the most residents receive care (Wennberg and Gittelsohn 1973). Contiguity requires that the units making up a service area be geographically continuous and acts to improve the interpretability of maps and analysis of data (Klauss et al. 2005). Additional criteria can be applied to adapt the methods of small area analysis to the service category of interest, including requirements for minimum population sizes or maximum geographic areas. Once service areas have been delineated, these can be evaluated on the basis of how well they describe regional patterns of care. A well-defined market will meet two criteria: few patients from outside the area will enter the service area for care, and few patients will leave the service area to receive care (Elzinga and Hogarty 1973). Previous studies have assessed these criteria with various measures. Garnick et al. (1987) used the concepts of relevance and commitment to describe service area characteristics. Relevance refers to the market-share within a service area captured by each facility and is calculated as the proportion of patients treated at a facility that came from a given service area (Garnick et al. 1987; Rohrer 1987). While relevance describes a service area in terms of all patients

112 95 treated by a specific facility, commitment describes the service area in terms of all patients who live in the region. It is calculated as the proportion of patients from a service area who are treated at a specific facility (Garnick et al. 1987) Though appropriate in hospital market analysis, these variables have limited application in dentistry, where several dentists commonly serve members of a single community rather than a single, dominant provider. This limitation was addressed by Goodman et al. (2003) in their delineation of primary care service areas (PCSAs) in the U.S. based on Wennberg and Gittelsohn's methods of small area analysis. Goodman et al. evaluated market area fit by measuring localization of primary care, defined as the proportion of care received by the residential population that was provided at locations within the service area. Klauss et al. also assessed localization of care, which they referred to as a localization index, and other related measures of patient flow in their examination of hospital discharges in Switzerland (Klauss et al. 2005). Despite the long-standing use of variable market analysis in health services research, applications in dentistry are lacking. Mayer's study of dental service areas in North Carolina is the only example of a variable market approach of which we are aware (Mayer 1999). As noted, a major limitation of that study was its use of counties as the minimal unit of aggregation to assemble service areas. Most studies of access to oral health services rely on counties or other geopolitical units. The goal of this study was

113 96 to identify dental service areas for the pediatric Medicaid population in Iowa through small area analysis using zip codes as the unit of aggregation. We then compared these with counties as representations of market areas for primary dental care. Methods Data We analyzed data from Iowa Medicaid enrollment and primary care claims files from calendar years (CYs) 2008 through The enrollment files contained information about each individual s date of birth, sex, race/ethnicity, Medicaid eligibility program, zip code, and county of residence. The claims dataset included all claims submitted by general and pediatric dentists during the three-year period. Dental claims data contained information about dates of service, provider specialty, providers locations at the zip code and county level, and procedures performed. Procedures were identified using Current Dental Terminology Codes (American Dental Association 2010). Patient and provider locations were geocoded to the zip code level. We used zip code tabulation areas (ZCTAs) produced by the U.S. Census Bureau as geographic approximations of zip codes (U.S. Census Bureau 2011).

114 97 Study population The inclusion criteria for this project required that an individual be continuously enrolled in the Iowa Medicaid program for at least 11 months during CYs , and be between the ages of 3 through 18. Age was calculated at the beginning of this qualifying enrollment period, which corresponds to the HEDIS protocol requirement for children to be continuously insured for at for at least 11 months during the study period (Agency for Healthcare Research and Quality 2006). The study population was further limited to individuals eligible for Medicaid based on income eligibility requirements or through Supplemental Security Income (SSI). We excluded individuals eligible for Medicaid through the foster care, those residing in medical institutions, and the medically needy population. The final enrollment dataset contained 153,996 individuals who met our inclusion criteria. All claims submitted to Iowa Medicaid by a general dentist or pediatric dentist on behalf of individuals who met the inclusion criteria were used to identify dental service areas. Medicaid enrollees (hereafter, children ) were required to have a residential zip code on file that was located in Iowa. Provider locations were not constrained to Iowa. The final claims dataset included 247,245 dental claims submitted on behalf of 76,874 individuals.

115 98 After claims data and enrollment data were linked, personal identifiers were removed in order to protect confidentiality. This study was approved by the University of Iowa Institutional Review Board. Delineating dental service areas Dental service areas (DSAs) were identified through small area analysis of patient origin and destination, adapted from the methods described by Wennberg and Gittelsohn (1982) and Goodman et al. (2003). In the initial step, we generated weights for each child to describe the proportion of dental visits from each residential zip code (origins) per child to each provider zip code (destinations). If a child had ten visits during the study period, with five claims submitted by Dentist A and five claims submitted by Dentist B, 0.5 of that patient s care would be assigned to each provider s destination. These weights were then applied to claims at the individual level to adjust for children that had more than one residential zip code during the study period, children who received care from providers located in multiple destinations, and to prevent children with heavy utilization of primary care services during the study period from driving plurality assignments. A similar system of weighting was described by Goodman et al. (2003) in the process of identifying primary medical care service areas, although that study did not account for patients with multiple residential zip codes.

116 99 The child-level dataset was aggregated to the zip code level to sum the proportion of weighted care that occurred between each origin/destination dyad. Dental service areas (DSAs) were then generated by assigning origin zip codes to destinations based on where the plurality of primary dental care was provided. In instances of ties, the closest provider destination was selected based on straight line distance, as calculated using nearest neighbor analysis in Esri ArcView DSAs are named for the dentist destination that attracted surrounding residential zip codes. In addition to plurality, we applied several additional criteria to delineate DSA boundaries. First, each dentist destination was required to serve a plurality of its own residential population. This prevented the creation of service areas with a provider destination located outside of its boundaries. In cases where this self-service failed to occur, the service area was dismantled and zip codes were reassigned to the DSA where the residents of the original provider destination received care. For example, zip code had been assigned to DSA However, the residents of zip code received a plurality of their primary care from another service area DSA Therefore, was reassigned to Second, we established a minimum population threshold of 50 Medicaid patients per service area in order to protect patient confidentiality and to produce more reliable statistics. Finally, we required that DSA boundaries form contiguous geographic areas, in keeping with previous variable market approaches to

117 100 service area analysis (Goodman et al. 2003; Klauss et al. 2005; Paul-Shaheen, Clark, and Williams 1987; Rohrer 1987). Identification of service area boundaries and geospatial calculations were performed in ArcView. Measures Primary care dentists We identified the number of dentists in each DSA or county who submitted one or more claims to Iowa Medicaid for care provided to our study population. Additionally, we measured the total number of pediatric and general dentists practicing in a service area or county during the study period, regardless of Medicaid participation. During 2008 through 2010, there were 876 primary care dentists in 929 practice locations serving this population. Of the 929 practice locations, 56 of these were located outside of Iowa and could not be included in our service area counts since out of state zip codes were not included in our DSA assignments. The exception is the single dentist who served the only service area without an Iowa-based provider; this DSA with an Illinois destination was included in analyses. Utilization rates We calculated utilization rates for each geographic region as the number of children who received any care from a primary dentist during the study period per 100 Medicaid enrollees. Utilization rates were calculated at the service area and county levels.

118 101 Localization of care In order to evaluate how well counties and DSAs approximated market areas for primary dental care, we calculated the percentage of care received by the Medicaid-enrolled population of each geographic region from dentists located within that region. Localization of health care services within a market area has been used alternately as a criterion for building service areas (Goodman et al. 2003; Mayer 1999) and as a tool to quantitatively evaluate service area characteristics (Goodman et al. 2003; Guagliardo et al. 2004; Klauss et al. 2005). When localization is used as a criterion for assembling service areas, a minimum value is established. For example, Mayer (1999) required that dental service areas captured at least 75% of care provided to the study population. For this study, we chose not to establish a minimum value for DSA localization of care in order to evaluate the performance of small area analysis as a methodological tool for identifying dental service areas. Statistical analysis Descriptive statistics were generated for general and demographic characteristics of primary care DSAs. Localization of care (LOC) was calculated at each step of identifying DSAs and for counties. Mean LOC for DSAs was compared with mean LOC for counties. Hypothesis testing to compare means was omitted since the populations that make up service areas and counties satisfy neither an assumption of independence nor of matching.

119 102 LOC distributions for counties and service areas were displayed in an overlay plot of their cumulative distribution function. Statistical analyses were performed using SAS version 9.3. Results Weights were assigned to children based on each of their residential zip codes and provider zip codes found in the dataset, and adjusted based on the number of visits that occurred between each zip code dyad. Approximately 90% of the population had only one residential zip code on file (N = 68,982); the remaining 10% of the study population had up to five residential zip codes during the study period. The number of provider zip code destinations per patient ranged from one to five, with 92% of patients receiving care from only one destination. The number of visits per patient ranged from 1 to 24, with a mean of 3.2 (SD 2.2). Only 25.4% of patients had just one visit during the study period. The initial zip code assignments based on plurality resulted in 139 crude service areas. Of these, 16 failed to meet the population requirement of at least 50 Medicaid recipients and the 23 residential zip codes involved were reassigned. After adjusting for geographic contiguity and self-service, we identified 113 primary care DSAs for the Medicaid population in Iowa, ages 3 to 18 (Figure 1). Approximately 12% of zip codes (N = 111) were reassigned

120 103 during these adjustments. General characteristics of DSAs are provided in Table 1. Fifty percent of DSAs were made up of seven or more zip codes, with a maximum of 53 zip codes in one service area. Only five DSAs were comprised of one zip code. Although the number of Medicaid-enrolled children per service area ranged from 85 to 14,829, 50% of DSAs had 580 or fewer children. Geographic size of DSAs ranged from 97.5 to square miles. Mean geographic area of DSAs was slightly smaller than mean county area (497.6 versus square miles, respectively). A total of 876 dentists in Iowa provided care to the pediatric Medicaid population during the study period. The median number of claims submitted per dentist over the three-year period was 78; median number of children treated per dentist was 32. The Iowa City service area had the highest number of primary care dentists treating the pediatric Medicaid population (N=64). However, 22 of these 64 dentists were affiliated with the University of Iowa. There were 247 unique provider destinations that served this study population. The number of primary care dentist destinations (ie, zip codes) per dental service area ranged from 1 to 14. Counties had between 0 and 21 destinations (median=2). Approximately 60% of DSAs were served by only one dentist destination within their boundaries versus 25% of counties with only one dentist destination.

121 104 Medicaid claims submitted during 2010 were examined in order to calculate one-year dentist participation rates. During 2010, 640 primary care dentists in the state submitted at least one claim to Iowa Medicaid. This is approximately 55% of the 1,160 primary care dentists in Iowa (Office of Statewide Clinical Education Programs 2011). This figure slightly underestimates the percentage of primary care dentists that participated in Medicaid, since the numerator does not include dentists who worked in clinics or institutional settings, while the denominator does include these. A total of 319 dentists (27.5%) treated at least 25 children during this one year period. Overall, 49.9% of the study population utilized primary care services during the three-year study period (N = 76,874). Utilization rates within DSAs ranged from 20.1% to 75.9% (mean 52.9, SD 11.1). Figure 1 displays service areas categorized by quartiles of increasing utilization rates. At the county level, utilization ranged from 6.5% to 41.5% (mean 24.6, SD 7.3). Figure 2 displays the LOC values for counties versus service areas. Three counties had zero-values for LOC, indicating that all of their Medicaidenrolled children had to travel outside of their county for primary dental care. The lowest LOC value for service areas was 25.1%. The highest LOC value for counties was 96.8% versus 94.6% for DSAs. Overall, mean localization of care (LOC) was greater for dental service areas compared to counties (58.5% versus 52.4%, respectively). The overlay

122 105 plots of the LOC cumulative distribution functions show that a major difference between the two market area schemas is the large number of counties with low values for this variable compared with DSAs; LOC was less than 25% in 16 counties versus zero service areas (Figure 3). Discussion Overall, using small area analysis was successful in delineating dental service areas, which more accurately captured market areas than counties. While the use of small area analysis may be too data-intensive for some policymakers to conduct routinely, this methodology may be particularly appropriate to evaluate access for populations with known geographic barriers to care, including children in Medicaid. Mean localization of care was greater for DSAs compared to counties, but the appreciable differences in market area performance were found in an examination of the distributions for this variable. It is important to note that several counties had zero-values for localization of care. For counties with no dentists available to serve the population, using these geopolitical units as representations of health service areas will fail to accurately describe the availability of care for the Medicaid population. A variable market approach improves this limitation by providing an improved framework for identifying areas with access difficulties, including regions with low Medicaid participation by dentists.

123 106 In this project, we did not establish a minimum localization of care value as a criterion to establish a service area. Other authors have established minimum values for this, ranging from 30% (Goodman et al. 2003) to 75% (Mayer 1999). Because this exploratory study is the first to identify dental service areas through a small area analysis based on plurality and contiguity, we chose not to set a minimum value for this. However, the overall rate of 59% localization of care compares favorably with values obtained by previous medical service studies. If we had chosen to set a minimum value for DSA localization of care, these market areas would have performed even better against counties. Visual inspection of Figure 2 revealed that several service areas roughly correspond to county boundaries, especially in the northeastern region of the state. This is probably the result of dentists being located frequently in county seats, which are situated in the geographic centers of most counties in Iowa and typically the largest city in rural counties. In states with fewer, larger counties, the methods described in this study may offer even better approximations of market areas than demonstrated here. There are several limitations to the methods we used to identify service areas. The first, the modifiable areal unit problem (MAUP), is a common challenge in geographic analysis. The MAUP is an extension of the ecological fallacy, whereby the characteristics of dental service areas could change if the data had been aggregated differently (Waller and Gotway 2004).

124 107 Our process of adjusting service area boundaries could have altered regional characteristics. However, mean localization of care values at each step of DSA adjustments remained fairly stable (ranging from 55 to 58.5%), indicating robust methodologies. An additional limitation is that patient travel patterns may change over time. For markets with high provider turnover or rapidly changing population demographics, variable market approaches to identifying service areas may be most appropriate for shortterm planning (Garnick et al. 1987). Our primary care dental service areas are defined based on the utilization patterns of a population that faces unique challenges in seeking and finding care. Children insured by Medicaid face well known access barriers to dental services, not the least of which is the relative difficulty of finding a dentist who will accept this payment method (Agili, Bronstein, and Greene-McIntyre 2005; Kenney, Ko, and Ormond 2000). These primary care dental service areas can be used to examine the variation in provider availability across the state and by policymakers to inform decisions about where to invest efforts in recruitment and retention of dentists. Previous research has applied small area analysis to examine variation in medical care, including rates of surgical procedures and hospital admissions. We have demonstrated the extension of this technique to oral health services; future studies can examine variation in specific dental procedures that serve as indicators of access, such as emergency visits and preventive services.

125 108 Table V.1 Descriptive characteristics of primary care DSAs N Mean Median SD Min Max Zip code origins Provider destinations Medicaid recipients 76, , ,688 Medicaid enrollees 153, , ,829 Utilization rates (per 100 enrollees) Primary care dentists treating 1 Medicaid enrollee Geographic size (mi 2 ) ,236.6

126 Figure V.1 Primary care DSAs for Medicaid-enrolled children in Iowa (N = 113) 109

127 Figure V.2 Localization of care by county and DSA 110

128 Figure V.3 Cumulative distribution function for localization of care 111

129 112 CHAPTER VI STUDY 3 INFLUENCE OF SPATIAL ACCESSIBILITY ON RECEIPT OF PREVENTIVE DENTAL CARE Study 3 Abstract Enrollment and claims data from the Iowa Medicaid program were used to examine the effects of place, including aspects of spatial accessibility, on the receipt of a preventive dental visit by children ages 3-18 years over a three-year period ( ). Using hierarchical logistic regression, individual and service area effects were considered. Children living in urban areas were more likely to have received a visit than those in more rural areas. Spatial accessibility, assessed using measures of dentist workforce supply and travel cost, did not appear to be a major barrier to care in this population. Introduction Publicly insured children are recognized as some of the most vulnerable and at-risk for oral health problems. In response, Healthy People 2020 has called for an increase in the proportion of low income children who receive preventive dental care (U.S. Department of Health and Human Services 2011). However, utilization of preventive services remains low despite federally mandated coverage of dental care for low-income children

130 113 through the Early and Periodic Screening, Diagnosis and Treatment (EPSDT) benefit (U.S. Health Resources and Service Administration 2012). In 2010, less than half of the approximately 250,000 children eligible for dental services through Iowa Medicaid received any type of preventive dental services (Iowa Department of Public Health 2011). One of the most widely utilized conceptual frameworks in health services research was developed by Ronald Andersen in 1968 to explain families use of medical services (Andersen 1968). Barriers to the use of health services were categorized as predisposing, enabling, and need-related. Subsequent revisions to the behavioral model have added domains representing characteristics of the health care system and individuals health behaviors. (Andersen 1995). The three original domains (predisposing, enabling, and need-related factors) have been retained in all iterations of this model (Andersen 1995). These core domains have been incorporated by numerous other researchers seeking to conceptualize the utilization of health care services (Beil and Rozier 2010; Borders 2006; Hall, Lemak, and Steingraber 2008; Milgrom et al. 1998). Predisposing factors typically include demographic characteristics (eg, age, sex, race/ethnicity). Resources that enable individuals to access health services include insurance coverage, income levels, and having a regular source of care. Need-related characteristics refer to both the patient s perception of need as well as a professionally evaluated level of need.

131 114 While predisposing and need-related characteristics tend to be related to biological imperatives, enabling resources come from family and social support. Resources that enable individuals to access health services include insurance coverage, income levels, and having a regular source of care. Other researchers have identified other early childhood oral health enabling factors, including transportation resources, healthy food options, and an effective health care system (Mattheus 2010). Enabling resources tend to be the easiest to modify and are targeted by numerous public health initiatives. For example, states are required to provide transportation assistance for children to medical and dental visits as a component of the federal EPSDT benefit (U.S. Health Resources and Service Administration). Several states have implemented community-based programs to alleviate barriers within the dental care system. Iowa instigated the I-Smile Dental Home Initiative in 2006 with a network of dental hygienists acting to coordinate oral health care for Medicaid-enrolled children (Iowa Department of Public Health 2012). The ABCD program in Washington (Access to Baby and Child Dentistry) works correspondingly to reduce access barriers for young Medicaid-enrolled children (Washington Dental Service Foundation 2011). Community characteristics also give rise to spatial accessibility of health care that dimension of access shaped by local availability of providers and travel costs associated with utilization (Guagliardo 2004).

132 115 Travel cost, or burden, is commonly studied among the Medicaid-enrolled population and has been reported to be highest among rural residents (Borders 2006). Travel barriers frequently faced by the Medicaid population include extended travel times, travel costs, and lack of reliable transportation (Borders 2006; Mattheus 2010). While previous research has examined relationships between spatial accessibility and medical care (Goodman et al. 1994; Peipins et al. 2011), we are unaware of any studies that directly examine the effect of this on receipt of oral health care. One recent study of dental expenditures among Medicaidenrolled children found lower rates of preventive visits in counties with higher dentist supplies (Beil et al. 2012)). While that study adjusted for county urbanicity, no other aspects of spatial accessibility were considered. Furthermore, counties may not offer an appropriate framework for evaluation of geographic influence on access. Geopolitical units, such as counties, are often used as a matter of convenience and may not accurately describe local variations in culture, social capital, and the health care system that contribute to access (Wennberg and Gittelsohn 1982). Dental market areas may be more appropriate than counties to examine the influence of regional characteristics on access, as these can more accurately profile regional variations in travel patterns and utilization of oral health services. One method of identifying health market areas, small area analysis, uses patient origin information to delineate service areas based on

133 116 where patients travel for care (Wennberg and Gittelsohn 1973). The use of small area analysis in conjunction with studies of medical utilization has revealed significant associations that were masked when counties were used as the unit of analysis (Mobley, Kuo, and Andrews 2008; Wennberg and Gittelsohn 1982). Previous research by these authors has delineated primary care dental service areas (DSAs) in Iowa for children enrolled in Medicaid. Overall, these DSAs captured market patterns more accurately than counties in Iowa in terms of describing where patients come from and where they travel to for dental care. The aim of this study was to examine the effect of place, assessed using DSAs, on utilization of preventive dental care among Medicaid-enrolled children and adolescents. If characteristics of service areas are found to contribute to utilization, this research would support the importance of community-wide initiatives designed to improve access to oral health services among this population. Methods Iowa Medicaid enrollment and claims files from January 1, 2008 through December 31, 2010 for children and adolescents were obtained through an agreement with the Iowa Department of Human Services. The claims dataset included information about all visits and procedures that were

134 117 submitted to Medicaid by a primary care dentist (ie, general or pediatric dentist). Children aged 3-18 years who were continuously enrolled in Iowa Medicaid for at least 11 months during 2008 through 2010 were included in our analyses. This requirement for length of enrollment corresponds to HEDIS protocols developed by the National Committee for Quality Assurance (Agency for Healthcare Research and Quality 2006). As an inclusion criterion, age was calculated at the beginning of each child s qualifying enrollment period. The population was further limited to those children eligible for Medicaid based on income eligibility requirements or through Supplemental Security Income (SSI). Individual covariates were measured at the time of the individual's first preventive dental visit during the study period. For example, a child may have been seen multiple times during the 3-year period; however, age, income level (FPL), and residential location were assessed from the first relevant claim on file. Age, length of enrollment during the study period, length of previous enrollment during 2007, and FPL were categorized for univariate analyses. For multivariate analysis, the categorized form of FPL was used since it demonstrated a nonlinear relationship with the dependent variable. The continuous forms of age and length of enrollment during the study period and during 2007 were used since they demonstrated linear relationships with the dependent variable.

135 118 Residential urbanicity was calculated based on the Rural-Urban Commuting Area (RUCA) codes for residential zip codes. RUCAs were obtained from the University of Washington s Rural Health Research Center (WWAMI Rural Health Research Center 2005). Zip codes were categorized as metropolitan, micropolitan, small town, and rural. Primary care dental service areas (DSAs) in Iowa were identified previously through small area analysis of claims submitted to Iowa Medicaid by primary care dentists on behalf of children and adolescents, ages 3-18 years. DSAs were delineated using methods adapted from the techniques described by Wennberg and Gittelsohn (1982) and Goodman et al. (2003), whereby zip codes were assigned to provider destinations based on where a plurality of residents received dental care. This study was limited to those 111 primary care DSAs contained wholly within Iowa. Two additional DSAs that extended into neighboring states and the Medicaid enrolled population within these were omitted from this study; data were not available to calculate workforce and other DSA-level variables for these out-of-state service areas. DSA-level variables were assessed during year Variables describing the physical environment and the dental care system were included in the analyses as two dimensions of spatial accessibility. Three variables were used to describe DSA physical environment: geographic area (square miles), perimeter-area ratio (PAR), and population density (per

136 119 square mile). Residents living in geographically large service areas will have potentially greater average travel distances to obtain care within the local service area. The perimeter/area ratio accounts for the compactness of service areas. DSAs with elongated shapes and highly irregular borders, indicating potentially long travel distances, will have a high PAR (Helzer and Jelinski 1999). We hypothesized that PAR would be negatively associated with utilization of preventive dental care. Population density is included here as a measure of service area urbanicity. Other DSA-level variables that described the dental care system included the localization of primary dental care, the overall primary dental care utilization rate, and several dentist-to-population ratios. Localization of care (LOC) was defined as the percentage of dental care received by the Medicaid-enrolled study population of each DSA from primary care dentists located within that region. Localization of health care services within a market area has been used to quantitatively evaluate service area characteristics (Goodman et al. 2003; Guagliardo et al. 2004; Klauss et al. 2005). As a component of spatial accessibility, LOC describes the extent to which patients can seek and receive care within their local market area and was hypothesized to be positively associated with utilization of preventive dental care. Several measures of dentist supply were calculated for each service area during CY 2010: number of dentists who saw any Medicaid patients

137 120 from the study population, number of dentists who saw 25 or more Medicaid patients from the study population, total number of dentists (all specialties), and number of primary care dentists (general and pediatric). All dentist supply variables were expressed as the number of dentists per 10,000 total DSA population. We decided to express dentist ratios as the number of dentists relative to the total population rather than the study population living in each service areas in order to describe the dentist workforce relative to the entire effective demand for dental care. Information about the number of Medicaid patients treated per dentist was extracted from the study claims dataset. Data about the total number of dentists per DSA were obtained from the Iowa Dentist Tracking System (Office of Statewide Clinical Education Programs 2012). The individual outcome was receipt of a preventive dental visit during the study period ( ). Preventive dental visits were defined by the receipt of a periodic or comprehensive oral evaluation (CDT D0120 or D0150), with or without preventive or therapeutic procedures. This definition was chosen to represent utilization of comprehensive dental care that is focused on primary and secondary prevention of disease. The first dental claim per child that met this definition was selected and independent variables were assessed from this date of service. We hypothesized that children with increased spatial accessibility to dentists will be more likely to utilize dental

138 121 care. Increased spatial accessibility occurs with greater regional workforce supply and lower travel costs. Descriptive statistics of individual and DSA characteristics were generated. Hierarchical logistic regression was used to model the effects of individual and service area characteristics on receipt of preventive dental care. Children (level 1, n=146,055) were nested within service areas (level 2, n=111). Modeling was conducted using the NLMIXED procedure in SAS. DSA was included in the model as a random effects variable in order to account for regional differences that could not otherwise be measured. In a secondary analysis for the purpose of examining the contribution of the Des Moines metro area population to the model, we removed the children from the four Des Moines DSAs (n=20,540) and compared the changes in parameter estimates. Multicollinearity was assessed in single level models by examining condition indices and eigenvalues. When multicollinearity was observed, variables were removed if they demonstrated high bivariate correlations with other variables (Pearson s r or Spearman s rho 0.7, p 0.05). All hypothesis tests used an alpha level of Statistical analyses were conducted using SAS version 9.3. This study was approved by the University of Iowa s Institutional Review Board.

139 122 Results A total of 67,394 (46.1%) Medicaid-enrolled children had at least one preventive visit during the study period. Approximately 50% (n=73,426) had a visit to a primary care dentist for any type of care. Mean age for the cohort was 9.6 years (SD 4.5). Characteristics of the study population and dental service areas are shown in Tables 1 and 2. The associations between individual variables and receipt of a preventive dental visit were significant for all variables. DSAs ranged in size from 97.5 to 3,236.6 square miles (Table 2). Dentist-to-population ratios are displayed in Table 2, however the actual number of dentists per service area treating 25 or more Medicaid enrollees during 2010 ranged from 0 to 21 (mean 1.0). Total number dentists per DSA ranged from 1 to 158 (median 5). The number of Medicaid enrollees in the study population per service area ranged from 85 to 14,829. Proportion of Medicaid children per DSA that received a preventive dental visit during the study period ranged from % (Figure 1). Adjusted for the other covariates in the multivariate model, White and Hispanic children were more likely than other children to receive a preventive dental visit (Table 3). Females were more likely than males to receive a preventive dental visit. Additionally, children living in small towns and rural areas were less likely than those living in more urbanized areas to have had a preventive dental visit.

140 123 Other significant variables in the multivariate model include age, length of enrollment during the study period, previous enrollment (2007), and poverty level. The likelihood of receiving a preventive dental visit decreased with age and increased with length of enrollment in the Medicaid program. Children from families with reported income between 1-199% of the FPL were more likely than those from families with no reported income to have received a preventive dental visit during the study period. After adjusting for individual characteristics, two DSA-level variables were significantly associated with the odds of receiving a preventive dental visit: DSA utilization rate and perimeter-area ratio (Table 3). Service area utilization rates had a positive association with the dependent variable, while perimeter-area ratio was negatively associated with a preventive dental visit. When children living in the Des Moines metro area were excluded from analyses, the resulting study population was comparable to the original population, with the exception of having a slightly higher proportion of white children (57.6% vs. 54.6%). The parameter estimates between the two models were comparable, except for changes to the effects of race/ethnicity (results not shown). Specifically, Hispanic children living in the Des Moines Area (OR = 1.87; 95% confidence interval = 1.61, 2.18) were more likely to have had a preventive dental visit than White children, compared with Hispanic children living in the rest of the state who were less likely to have had a visit (OR = 0.91; 95% confidence interval = 0.87, 0.95). Children of other race/ethnicity

141 124 (ie, American Indian, Asian, Pacific Islander, or multiethnic) showed a similar pattern of utilization: higher likelihood of utilization in Des Moines with lower likelihood of utilization in the rest of the state. Black children living in the Des Moines metro area had higher odds of utilization compared to those in the rest of the state and demonstrated no significant difference from White children living in Des Moines (OR = 0.91; 95% confidence interval = 0.79, 1.04). Discussion This study examined the associations between individual and service area characteristics and receipt of preventive dental care among Medicaidenrolled children. Assessing influences from the service area level, as opposed to counties, enhanced this study by providing a more accurate framework for evaluating health care behaviors. Multilevel modeling demonstrated the significance of regional characteristics and location on access to dental care among Medicaid-enrolled youth in Iowa. However, the major components of spatial accessibility in this study dental workforce supply and variables describing DSA travel cost failed to demonstrate significance. Two DSA characteristics DSA utilization rate and perimeter-area ratio demonstrated relationships with the dependent variable as hypothesized. Utilization rate measured the total number of Medicaid enrollees in each service area that had a visit to a primary care dentist for

142 125 any reason. Children living in service areas with higher overall utilization rates were more likely to have had a preventive dental visit. Perimeter-area ratio (PAR) was included in this study as a measure of service area compactness. Generally, service areas with highly irregular borders or shapes will have higher PARs (Helzer and Jelinski 1999). Even though PAR was negatively associated with the likelihood of having a preventive dental visit, the 95% confidence interval for this odds ratio was extremely narrow. This study was powered by a large sample size (N = 146,055) that allowed us to detect very small differences between groups. Despite demonstrating statistical significance (p = 0.038), the association between perimeter-area ratio and preventive dental visits appears to be of little practical significance (OR = 1.001; 95% confidence interval = 1.000, 1.001). Localization of care (LOC), a variable that measures the extent to which children within a DSA are able to receive dental care within that area, failed to demonstrate significance in the multivariate model (p = 0.063). This variable, and several of the other DSA characteristics included in the final model, do not appear to influence children s receipt of preventive dental care to the same extent that individual characteristics do. Further research is needed to examine whether other service area characteristics may be more appropriate for examining the effects of place on receipt of care. Specifically,

143 126 additional or refined workforce measures should be investigated to examine the effect of local dentist supply on access to care. Our findings that children living in urban areas were more likely to receive preventive dental care agree with previous research showing that rural-dwelling Medicaid children are less likely to receive preventive dental care (Beil et al. 2012; Chi et al. 2010). Here, urbanicity was assessed at the individual level by applying RUCA codes to residential zip codes and by adjusting for service area population density. The interesting associations seen between utilization of preventive dental care among this Medicaidenrolled population, urbanicity, and location hint at complex relationships that are worth further investigation. Future studies should investigate why Medicaid-enrolled children living in small towns and rural areas in Iowa are less likely than urban residents to receive preventive dental care. This relationship may the result of variation in populations demand for dental care or due to limited dentist supply in rural areas. Previous research has indicated that rural residents may demonstrate a lower preference for dental care than urban-dwellers (Allison and Manski 2007). There are several possible explanations for our overall findings. First, spatial accessibility does not pose a major barrier to the receipt of preventive dental care among Medicaid enrolled youth in Iowa. It is also possible that the variables we chose to measure spatial accessibility did not adequately capture this dimension of access to care. Additionally, the primary care DSAs

144 127 used as the spatial unit of analysis may not be appropriate to evaluate regional effects on patterns of utilization. However, previous analyses have suggested that these service areas describe the local patterns of utilization in Iowa comparably, if not more accurately, than counties. Increasingly, public programs aim to improve oral health through community efforts. While this study hinted at potential relationships between service area characteristics and the receipt of preventive dental care, further studies are needed to further our understanding of the effects of place on access to dental care.

145 128 Table VI.1 Description of continuously enrolled Iowa Medicaid children ages 3 through 18 years ( ) and significance testing results between children based on access by Chi-square analysis Total population Preventive Dental Visit NO YES N = 146,055 N = 78,661 N = 67,364 Significance Variable N (%) Age Mean ± SD (years) 3-5 years Sex Female Male Race/ethnicity White Black Hispanic Other Unknown/missing Length of enrollment ( ) Mean ± SD (months) months 36 months Previous enrollment (2007) 0 months 1-11 months 12 months Medicaid aid category Income eligible SSI Federal Poverty Level (FPL) 0% 1-133% % 200% Urbanicity of residential zip code (RUCA) Metropolitan Micropolitan Small town Rural 9.6 ± ,636 (24.4%) 58,566 (40.1) 51,853 (35.5) 73,557 (50.4) 72,498 (49.6) 79,760 (54.6) (8.2) 12,688 (8.7) 4,570 (3.1) 37,005 (25.3) 28.3 ± ,402 (58.5) 60,653 (41.5) 39,635 (27.1) (23.9) 71,578 (49.0) 140,324 (96.1) 5,731 (3.9) 39,554 (27.1) 101,684 (69.6) 4,212 (2.9) 582 (0.4) 69,965 (47.9) 27,224 (18.6) 27,421 (18.8) 21,445 (14.7) 10.0 ± ,628 (22.4%) 29,982 (38.1) 31,051 (39.5) 39,162 (49.8) 39,399 (50.2) 41,345 (52.6) 6,946 (8.8) 6,426 (8.8) 2,365 (3.0) 21,579 (27.4) 26.6 ± ,699 (65.7) 26,962 (34.3) 25,956 (33.0) 19,152 (24.3) 33,553 (42.7) 75,235 (95.6) 3,426 (4.4) 22,720 (28.9) 53,287 (67.8) 2,280 (2.9) 357 (0.5) 38,173 (48.4) 14,774 (18.8) 14,536 (18.5) 11,178 (14.2) 9.1 ± ,008 (26.7%) 28,584 (42.4) 20,802 (30.9) 34,395 (51.0) 32,999 (49.0) 38,415 (57.0) 5,086 (7.5) 6,262 (9.3) 2,205 (3.3) 15,426 (22.9) 30.2 ± ,703 (50.0) 33,691 (50.0) 13,679 (20.3) 15,690 (23.3) (56.4) 65,089 (96.6) 2,305 (3.4) 16,834 (25.0) 48,397 (71.8) 1,932 (2.9) 225 (0.3) 31,792 (47.2) 12,450 (18.5) 12,885 (19.1) 10,267 (15.2) <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001

146 129 Table VI.2 Descriptive characteristics of primary care dental service areas (N=111) Domain Variable Median Mean SD Min Max Physical environment Dental care system Geographic area (mi 2 ) ,236.6 Perimeter-area ratio Population density (per mi 2 ) ,177.4 Total Medicaid enrollees (study population) , , ,829 Total population 12, , , , ,931 Localization of care (%) Primary care dental utilization (%) Dentists who saw any Medicaid patients per 10k Dentists who saw 25+ Medicaid patients per 10k Dentists, all specialties, per 10k Dentists, primary care only, per 10k Dentist supply variables are expressed as the number of dentists per 10,000 total DSA population. Mean distance to primary destination was weighted by the number of Medicaid enrollees per zip code.

147 130 Table VI.3 Multilevel logistic regression for preventive dental visit by Medicaid-enrolled children and adolescents, ages 3-18, during CY β SE p OR 95% Confidence Interval L U Intercept < Child-level variables Sex Male (reference) Female < Race White (reference) Black < Hispanic Other Unknown/missing < Age < Length of enrollment < Previous enrollment (2007) < FPL 0% (reference) % < % % RUCA Metro (reference) Micro Small town Rural < DSA-level variables Population density Primary care dentists per 10k Dentists treating 25 Medicaid kids per 10k Utilization rate < Localization of care Geographic area Perimeter-area Ratio DSA (random intercept) -2LOGLIKE AIC

148 Figure VI.1 Medicaid enrollees per DSA with a preventive dental visit during CY

149 132 CHAPTER VII DISCUSSION This aim of this dissertation was to address the limited use of service area analysis in oral health service research. The three studies herein describe methods of identifying dental service areas (DSAs) using small area analysis. These service areas were then used as frameworks for evaluating utilization of dental services among Medicaid-enrolled youth in Iowa. Small area analysis, which has been used previously in medical services research, provided a variable market approach to delineate market areas based on where patients come from and where they travel to for care. Currently, most oral health service research uses counties or other geopolitical units as the spatial units of evaluation an approach fraught with shortcomings. It is our hope that the conclusions of this dissertation will provide initial support for small area analysis in dentistry that will contribute to further development of this field. Study 1 The first study emerged from a pilot test of the methods that were to be used to identify primary care dental service areas. When the methods proved to be more arduous than anticipated, this pilot test was expanded into the study described in Chapter IV. Orthodontic treatment is frequently considered to be elective; however, for individuals with handicapping or

150 133 disfiguring malocclusions, treatment can improve speech, mastication, and quality of life. Recent policy changes in Iowa Medicaid have increased the case complexity required for approval of orthodontic treatment, which may prevent the contribution by general dentists to providing care for this population. This study provided information about the current role that orthodontists play in providing care to Medicaid children. Several findings from this study bear commenting on. First, the results of this study are contrary to previous research that has found demand for orthodontic treatment to be higher among children living in urban areas than those living in rural areas (Borders 2006; Wheeler et al. 1994). After adjusting for other known predictors of utilization in the logistic regression model, children living in rural areas of Iowa were significantly more likely to have received orthodontic services than those living in metropolitan areas (OR = 1.46; 95% confidence interval = 1.33, 1.61). Small area analysis of claims submitted to Iowa Medicaid generated 19 orthodontic DSAs. Examination of service areas grouped by quartiles of utilization rates supported the urban-rural trend seen in the individual analysis. DSAs with the highest utilization rates (mean = 5.8 utilizers per 100 Medicaid-enrollees) had the lowest mean population density. Second, children who had received an oral evaluation by a primary care dentist during 2007 the year preceding the study period were more likely to receive orthodontic treatment than children without a visit

151 134 (OR=1.58; 95% confidence interval = 1.47, 1.69). This finding supports the important influence of the dental home, which may facilitate coordination of oral health care among multiple providers. This finding is in agreement with previous research (Hans et al. 2004; Okunseri et al. 2007). The high degree of variability in rates of orthodontic treatment across service areas indicates that variation in provider availability may pose a more significant barrier to care for publicly insured children and adolescents than travel distance. Study 2 The second study in this dissertation identified primary care DSAs for Medicaid-enrolled youth in Iowa. Characteristics of DSAs and counties were compared in order to evaluate these as representations of dental market areas. Using small area analysis of claims submitted by general and pediatric dentists, we identified 113 primary care DSAs in Iowa. Localization of care (LOC) per geographic unit was calculated as the percentage of care received by the Medicaid-enrolled population of each geographic region from dentists located within that region. This variable has been used in previous research as a measure of how well a region captures the health care market (Goodman et al. 2003; Guagliardo et al. 2004; Klauss et al. 2005). A high LOC value indicates that a high proportion of Medicaidenrolled children in a region receive their care from dentists within that area.

152 135 In this study, mean LOC for counties (N=99) and DSAs (N=113) was comparable (94.6% versus 96.8%, respectively). However, examination of the distribution of LOC values for counties and service areas revealed that a major difference between the two market area schemas was the large number of counties with low values for this variable compared with DSAs; LOC was less than 25% in 16 counties but less than 25% in no service areas. For counties with no dentists available to serve the population, using these geopolitical units as representations of health service areas will fail to accurately describe the local availability of care for the Medicaid population. These primary care DSAs identified in Study 2 then were used as the spatial unit of analysis in Study 3, whereby regional characteristics were considered as potential influences on the receipt of preventive dental care. Study 3 The goal of Study 3 was to examine the effect of place, using primary care DSAs, on utilization of preventive dental care among Medicaid-enrolled youth. Preventive dental visits were defined as any visit to a general or pediatric dentist with a charge for a comprehensive or periodic oral evaluation (CDT D0120 or D0150). Previous research has used a variety of definitions for a preventive dental visit. The most common components of these definitions are an oral evaluation and/or a preventive dental service (Beil et al. 2012; Chi et al. 2010; Savage 2004), with some researchers placing

153 136 additional restrictions, such as requiring that no treatment procedures (eg, restorations) occur for three months after a preventive visit (Beil et al. 2012). Of particular interest in this study was the influence of spatial accessibility, assessed at the DSA level, on utilization. Spatial accessibility increases with greater regional workforce supply and lower travel costs (Guagliardo 2004). We hypothesized that children with increased spatial accessibility to dentists would be more likely to have had a preventive dental visit. Two service area characteristics were significantly associated with the odds of receiving a preventive dental visit: overall primary care dental utilization rate and perimeter-area ratio. Utilization rate measured the total number of Medicaid enrollees in each service area that had a visit to a primary care dentist for any reason. It was hypothesized that children living in DSAs with higher overall utilization of primary dental services would be more likely to receive a preventive dental visit during the study. Overall utilization rate demonstrated a positive association with receipt of a preventive dental visit, while perimeter-area ratio had a negative association. The major components of spatial accessibility in this study dental workforce supply and variables describing DSA travel cost failed to demonstrate significance. Future studies should explore alternative measures of dentist workforce supply and travel burden in order to examine the influence of spatial accessibility on the receipt of care.

154 137 Overall, children in urban areas were more likely to have had a preventive dental visit than rural-dwelling children. This relationship has been demonstrated by previous research (Beil et al. 2012; Chi et al. 2010). However, children living in the Des Moines metro area showed different utilization patterns than children in the rest of the state based on race or ethnicity. Hispanic children living in Des Moines were more likely than White or Black children to have had a preventive dental visit. Elsewhere in the state, Hispanic children were less likely than White children to have had a visit. Future studies should explore whether these variations in utilization persist among other populations, including privately insured children. Rather than use county-level measures of urbanicity, Studies 1 and 3 measured urbanicity at the individual level using Rural-Urban Commuting Area (RUCA) codes for residential zip codes. RUCA codes integrate information from the U.S. Census Bureau with commuting information in order to describe the local urbanicity (WWAMI Rural Health Research Center 2005). These studies also examined service area population density as an additional measure of urbanicity. Limitations and Future Directions We examined administrative data from the Iowa Medicaid program to improve our understanding of factors affecting utilization of oral health services. It is assumed by policymakers and researchers that preventive

155 138 dental care improves oral health outcomes. However, we did not examine the outcomes associated with utilization of care. Future studies that examine the effects of preventive care, including cost-effectiveness, quality-of-life issues, and dental disease measures, would bridge this notable gap between utilization and outcomes. These studies were cross-sectional in design. Longitudinal analyses could examine whether preventive dental care improves these long term outcomes. The concept of a preventive dental visit is nebulous in dentistry, as evidenced by the wide variety of operational definitions used by other researchers. Oral health services research would benefit from a uniformly accepted definition of a preventive dental visit, which would facilitate comparisons between studies and lend policy credence to research findings. For Study 3, we used an operational definition that attempted to identify comprehensive dental care that was focused on disease prevention and maintenance of optimal oral health. Among children who had received a comprehensive or periodic oral evaluation thus meeting our definition of a preventive dental visit 89.6% also received a prophylaxis (CDT D1110 or D1120) at the time of service. If the definition had been expanded further to require any preventive dental procedures (CDT D1110 through D1206), we would have included 90.4% of the children who were captured by our original definition. It is doubtful that constricting our definition of a preventive dental visit would have improved the strength of our multivariate model in Study 3.

156 139 The methods used to identify service areas in Studies 1 and 2 are widely accepted and applied in medical services research. However, the lack of automation in the delineation process may be seen as a limitation of this method. There is a heuristic element to small area analysis; for example, when adjusting for contiguity, reassignment decisions must take into account proximity to neighboring service areas, as well as all of the other sources of care for a residential population. It is reasonable to suggest future research to develop an automated algorithm for assigning geographic units to service areas while taking into consideration issues such as geographic contiguity and compactness. However, any algorithm that includes criteria for contiguity and compactness will effectively eliminate an important source of variation that is revealed with the traditional methods of small area analysis. For example, if geographic compactness is a requirement of an algorithm, we will be unable to identify areas of interest where patients are forced to travel extreme or circuitous routes to obtain dental care. Conclusions These studies examined geographic variation in utilization of orthodontic and primary care dental services within service areas. While dentist workforce supply within DSAs was considered as a potentially mediating effect on utilization, it was not the focus of this dissertation. However, service areas offer an ideal framework for oral health workforce

157 140 studies since DSAs are generated based on revealed patient preferences for sources of care. Using Rural-Urban Commuting Area codes as measures of regional urbanicity revealed associations with utilization of orthodontic treatment that were contrary to other oral health research. However, because these codes provide a more precise measure of local urbanicity, future studies could be enhanced by using these zip code based codes rather than more traditionally used county-level measures. Whether our findings are supported will be of interest. While Study 3 hinted at potential relationships between regional characteristics, including aspects of spatial accessibility, and the receipt of preventive dental care, further studies are needed to further our understanding of the effects of place on access to dental care.

158 141 APPENDIX A SUPPLEMENTAL MATERIAL STUDY 1 The study described in Chapter IV evaluates access to orthodontic treatment among Medicaid enrolled children in Iowa. This appendix provides additional details about the methods used to create the Medicaid dental claims and enrollment datasets for the study, as well as a detailed description of process of generating dental service areas (DSAs). Medicaid Claims Datasets The initial dataset included all claims submitted by dentists to the Iowa Medicaid program during calendar years (CYs) In order to identify claims of interest, the following steps were taken: Claims were limited to those with a provider type of dentist (provtype=4) and a provider specialty of orthodontics (prospec=93). Claims: N = 22,977 Individuals: N = 6,475 Claims were limited to those where the patient s age at the date of service was 6-18 years (including age 18). Claims: N=22,660 Individuals: N=6,439 Only 2 claims were submitted for 2 children less than 3 years of age. For individuals 19 years, 315 claims were submitted on behalf of 106

159 142 individuals. However, only 34 of the 106 had not been seen previously (at a younger age during the study period). Only 14 children aged 3 through 5 were seen during the study period (out of 37,641 Medicaid enrolled children in this age group. Due to the extremely low utilization rate (<.01%), this age group was not included in the study. Preliminary descriptive analyses that include this age group are presented in Table 1. However, all subsequent analyses include ages 6 through 18 only (Table 2). Claims were limited to those submitted for orthodontic procedures. Orthodontic procedures were identified using Current Dental Terminology (CDT) codes D Once claims were limited by provider, patient s age, and procedure, the variables for provider zip code (provzip) and recipient zip code (reczip) were examined. I removed 3 children from the dataset who had no in-state residential address. I replaced the recipient zip code for 4 claims where a child had an Iowa residential zip code for previous orthodontic dates of service. Out of state provider zip codes were retained. The only non-iowa provider zip code was (Omaha, NE). Both zip code variables were inspected for missing values. Missing provider zip codes were replaced with values from the prozip02 variable. No recipients had a missing residential zip code.

160 143 Duplicate entries per individual were identified based on Medicaid ID number (ID). Claims: N = 6,8115 Individuals: N = 4,386 Medicaid Enrollment Dataset The study population for this project included all individuals who met the following inclusion criteria: 1. Continuously enrolled in Medicaid for at least 11 months during CYs 2008 to Ages 3 or over and under 19 years as of January 1, Eligible for Medicaid through Supplemental Security Income (SSI) or based on income eligibility requirements. Any individual who was eligible for Medicaid through the foster system, institutionalized status, or who was enrolled in IowaCare was excluded from the study population. Applying the inclusion criteria resulted in a dataset with 156,899 individuals. After applying the inclusion criteria described above, the following steps were taken: Patient state was limited to Iowa (N=155,983) Deleted cases with an invalid zip code (N=155,932)

161 144 The final study population included 155,932 individuals. Approximately 73% (N=113,741) individuals were eligible for some period of time during 2007, the year immediately preceding the study period. Mean length of eligibility during 2007 was 10.3 months (SD 3.1). Forty-nine percent of the total study population was eligible for the entire year during Length of eligibility during the study period ( ) ranged from 11 months (N=4,285) to 36 months (N=64,736). Mean length of eligibility for all individuals during was 28.3 months (SD 8.9). The majority of individuals were enrolled in Medicaid based on income eligibility (96%). Merging Claims and Enrollment Datasets The Medicaid recipient dataset was linked with the Medicaid enrollment dataset in order to identify individuals who were utilizers of orthodontic services during the study period. Out of 4,386 individual who received orthodontic services during , 742 failed to meet the study inclusion criteria. The claims dataset was linked with the enrollment dataset to identify claims submitted on behalf of only those individuals who met the study inclusion criteria.

162 145 Model Covariates The following covariates were included in the analysis: 1. Age: Measured as of January 1, Individuals age 6 through 18 were included in the enrollment and claims files. Age was categorized as 6-9 years, 10-13, and to capture information about the trends in utilization (Figure 1). 2. Race/ethnicity: Condensed into a 5-category variable from the original 9-categories used by the Iowa Medicaid Program. The 5 categories include: White, Black, Hispanic (combining Hispanic and Hispanic Multiethnic categories), Other (American Indian, Asian, Pacific Islander, or Multiethnic Unknown), and Unknown/Missing. 3. Length of enrollment: Due to the large proportion of children enrolled for the entire 36-month study period (40%), this variable was dichotomized as months versus 36 months. 4. Previous enrollment: 49% of the study population was enrolled for the entire 2007 calendar year. Previous enrollment was considered as a continuous variable (number of months) and in a categorized form. Individuals were categorized as 0 months, 1-11 months, or 12 months of enrollment in CY Medicaid aid category: The study population was limited to individuals eligible for Medicaid through SSI or based on income requirements.

163 Federal Poverty Level (FPL): This variable is contained in the Medicaid enrollment file and reflects the household income level. It is measured for each month that a child is enrolled in Medicaid. The value from the first month of enrollment was used to categorize an individual s poverty level. 7. Primary care dental visit: A dichotomous variable that indicated whether an individual had a primary care dental visit during CY Whether or not a child had a primary care dental visit was identified by linking the enrollment dataset with the 2007 dental Medicaid claims dataset. The following restrictions were used to identify these claims: Provider type (PROVTYPE): Any type provider. Provider specialty (PROSPEC): Limited to general practice (PROSPEC=1), clinic (70), general dentistry (91), and pediatric dentistry (94). Claim submitted for a comprehensive or periodic oral exam (D0150 or D0120). As referenced previously, Tables 1 and 2 provide descriptive statistics about the study population (ages 3-18 and 6-18, respectively). These tables offer a comparison of characteristics based on orthodontic utilization. Tables 3 through 6 provide additional stratified analyses of the study population based on residential urbanicity, race/ethnicity, sex, and age.

164 147 Dental Service Areas (DSAs) Boundaries of DSAs were generated using all claims submitted to Iowa Medicaid by orthodontists during CYs 2008 to The first claim on record per Medicaid recipient was used to identify patient origin. Iowa Medicaid allows orthodontists to bill one time at the beginning of treatment or quarterly. By using only the first claim on record, we reduced potential bias in service area assignment that could have resulted from over-representation by children whose claims were submitted quarterly. No individual eligibility requirements were applied to the dataset during the process of identifying DSAs. Claims were aggregated to the recipient/provider zip code level to determine the number of recipients from each zip code (origins) that received care from each provider zip code (destinations). Origins were assigned to destinations based on where a plurality of children received care (Figure 2). In instances of ties, the destination was assigned randomly. There were 100 cases of ties among 46 origin zip codes, all of which were tied with either 2 or 3 destination zip codes. A threshold of ten claims per DSA was set as the minimum number required to establish a service area; this requirement eliminated two service areas (providers located in zip codes and 50401). DSA had included three zip codes, and DSA included only one zip code. This minimum claims number was established to protect patient confidentiality in these areas and to facilitate statistical analysis.

165 148 In the example provided in Table 7, origin zip code was assigned to because a plurality of Medicaid recipients from this area received orthodontic services there. Origin was assigned to one of three zip codes at random to handle the ties between destinations. This process assigned 1,135 origin/destination dyads to 26 DSAs. Service areas were created through the aggregation of origin zip codes. Twenty-six DSAs were originally created; 25 of these were assembled around orthodontists practicing in Iowa and 1 was assembled around orthodontists practicing in Omaha, NE. Zip codes were linked to ZCTAs in order to generate DSA boundaries using ERSI ArcMap Through the process described below, these original 26 DSAs were reduced 19 orthodontic service areas. The Medicaid recipient and eligibility datasets were then linked with the DSA boundary files to identify which DSA each individual lived in. Cases where an individual s zip code failed to match with a 5-digit ZCTA were deleted from the dataset. These deletions included 49 orthodontic utilizers (1.3% of all utilizers) and 1,961 Medicaid enrollees (1.3% of all enrollees). There were 1,089 zip codes in the dataset. Of these, 935 (86%) had an exact match with a 5-digit ZCTA and 93 (8.5%) were non-existent zip codes according to the zip code look up at USPS.com. The remaining non-matching zip codes appeared to be post office boxes. DSA-level aggregated variables were generated as described in Chapter IV.

166 149 After assigning origins to destinations based on plurality, the remaining unassigned zip codes in the state were assigned to service areas based on the nearest provider destination (Figure 3). Note: provider destinations were identified based on utilization of orthodontic services rather than from the population of all orthodontists in the state. Once all zip codes had been assigned to service areas, manual adjustments were made to improve contiguity of DSA boundaries. Of the 935 zip codes in Iowa, 230 were reassigned during manual adjustments (25%). This number could have been minimized if the nearest provider zip code was selected in cases of ties, rather than assigning ties at random. Six cases accounted for 71% of all adjustments: 1. Two cases where the providers for one service area were located in another service area. 2. DSAs and were condensed into one DSA (51249); 32 zip codes were reassigned. 3. DSAs 52405, 52403, and were condensed into one DSA (52402); 32 zip codes were reassigned. The provider locations for these DSAs are identified in Figure Two cases where adjacent service areas had convoluted borders with providers located in adjacent zip codes. 5. DSAs and were combined into a single service area (50613); 41 zip codes were reassigned.

167 DSAs and were combined into a single service area (52241); 53 zip codes were reassigned. The original boundaries for and are shown in detail in Figure 4. The other major type of adjustment was for islands, where a single origin zip code was found entirely contained within another service area. Figure 4 shows several examples of this situation. Origin zip code was originally assigned to DSA To maintain service area contiguity, it was reassigned to DSA along with origin Similarly, origin was originally assigned to It was reassigned to the service area defined by provider destination A list of all reassignments is not included here, but this dataset was linked back to the Medicaid enrollment file, which had previously been linked to the dental claims data. From this merged data, service area level characteristics were produced by aggregating utilization and demographic data to the DSA level (Table 4). Note that DSAs are named by zip code/city of the provider destination that acted to aggregate the recipient origins. Primary care dentist-to-population ratios for each service area were calculated using 2010 data from the Iowa Dentist Tracking System. Primary care dentists were defined as the following specialties: general practice, pediatrics, operative, and public health dentistry. Ratios were expressed as the number of primary care dentists per 10,000 DSA population. The number of orthodontists per DSA was calculated using this same dataset (SPID=05).

168 151 The number of Medicaid orthodontists in Iowa was calculated using the Medicaid claims dataset (year 2010 claims for ages 3-18 years). After calculating the number of orthodontists per service area using data from IDTS and from Medicaid claims, it became apparent it is impractical to compare numbers (ie, proportion of orthodontists per DSA who treat Medicaid patients). IDTS maintains information about primary office location only, while the Medicaid claims dataset has information about each practice location. Specifically, three orthodontists have two locations that submit claims to the Iowa Medicaid program. Additionally, one orthodontist who submitted a single claim to Iowa Medicaid during 2010 was not found in the IDTS dataset. We are unable to obtain information about secondary practice locations for those orthodontists that did not participate in Medicaid during 2010; service area comparisons would be invalid. Therefore, in order to estimate Medicaid participation by orthodontists, the list of individuals who submitted claims to Medicaid during 2010 was compared with the list of orthodontists from the 2010 IDTS dataset (n=85). Forty-eight of these 85 (48.2%) submitted a claim to Medicaid for at least one individual during that year. Figure 5 displays information about the number of Medicaid youth treated by Iowa orthodontists during Orthodontistto-Medicaid population ratios were calculated as the number of Medicaid providers in 2010 divided by the enrolled population and expressed as the number of orthodontists per 10,000 enrollees.

169 152 Service area level descriptive characteristics are summarized in Table 8. Bivariate correlations between utilization rates and these characteristics were assessed using Pearson s r and a significant level of.05. Table 9 and 10 provide additional summary statistics at the service area level.

170 153 Table A.1 Description of continuously enrolled Iowa Medicaid children ages 3 through 18 years ( ) and significance testing results between orthodontic utilizers and non-utilizers by Chi-square analysis Total population Ortho nonutilizers Ortho utilizers N=153,971 N=150,376 N=3,595 P value Variable N(%) Age Mean ± SD 3-5 years Sex Female Male Race/ethnicity White Black Hispanic Other Unknown/missing Length of enrollment ( ) Mean ± SD (months) months 36 months Previous enrollment (2007) 0 months 1-11 months 12 months Medicaid aid category Income eligible SSI Federal Poverty Level (FPL) 0% 1-133% % 200% Primary care oral evaluation (2007) Yes No Urbanicity of residential zip code Metropolitan Micropolitan Small town Rural 9.6 ± ,641 (24.4) 61,848 (40.2) 54,482 (35.4) 77,548 (50.4) 76,423 (49.6) 83,745 (54.4) 12,409 (8.1) 14,669 (9.5) 5,311 (3.4) 37,837 (24.6) 28.3 ± ,192 (58.6) 63,770 (41.4) 41,817 (27.2) 36,858 (23.9) 75,296 (48.9) 147,967 (96.1) 6,004 (3.9) 41,844 (27.2) 107,077 (69.6) 4,420 (2.9) 606 (.4) 48,987 (31.8) 104,984 (68.2) 77,430 (50.3) 27,224 (17.7) 27,421 (17.8) 21,896 (14.2) 9.5 ± ,627 (25.0) 60,001 (39.9) 52,748 (35.1) 75,427 (50.2) 74,949 (49.8) 81,317 (54.1) 12,265 (8.2) 14,389 (9.6) 5,213 (3.5) 37,192 (24.7) 28.2 ± ,468 (58.8) 61,908 (41.2) 41,053 (27.3) 36,044 (24.0) 73,279 (48.7) 144,471 (96.1) 5,905 (3.9) 41,147 (27.4) 104,292 (69.4) 4,320 (2.9) 593 (.4) 47,340 (31.5) 103,036 (68.5) 75,898 (50.5) 26,684 (17.7) 26,614 (17.7) 21,180 (14.1) 11.5 ± (.4) 1,847 (51.4) 1,734 (48.2) 2,121 (59.0) 1,475 (41.0) 2,428 (67.5) 144 (4) 280 (7.8) 98 (2.7) 645 (17.9) 30.7 ± 7.6 1,724 (48.0) 1,871 (52.0) 764 (21.3) 814 (22.6) 2,017 (56.1) 3,496 (97.2) 99 (2.8) 697 (19.4) 2,785 (77.5) 100 (2.8) 13 (.4) 1,647 (45.8) 1,948 (54.2) 1,532 (42.6) 540 (15.0) 807 (22.4) 716 (19.9) <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001

171 154 Table A.2 Description of continuously enrolled Iowa Medicaid children ages 6 through 18 years ( ) and significance testing results between orthodontic utilizers and non-utilizers by Chi-square analysis Total population Ortho nonutilizers Ortho utilizers N=116,330 N=112,749 N=3,581 P value Variable N(%) Age Mean ± SD Sex Female Male Race/ethnicity White Black Hispanic Other Unknown/missing Length of enrollment ( ) Mean ± SD (months) months 36 months Previous enrollment (2007) 0 months 1-11 months 12 months Medicaid aid category Income eligible SSI Federal Poverty Level (FPL) 0% 1-133% % 200% Primary care oral evaluation (2007) Yes No Urbanicity of residential zip code Metropolitan Micropolitan Small town Rural 11.4 ± , 848 (53.2) 54,482 (46.8) (50.9) (49.1) (57.4) 9904 (8.5) (8.7) 3797 (3.3) (22.2) 28.0 ± (60.0) (40) (28.0) (23.5) (48.6) (95.5) 5258 (4.5) (26.9) (69.8) 3385 (2.9) 481 (.4) (32.0) (68.0) (50.1) (17.7) (17.9) (14.3) 11.7 ± (53.2) (46.8) (50.6) (49.4) (57.1) 9760 (8.7) 9798 (8.7) 3699 (3.3) (22.3) 27.9 ± (60.4) (39.6) (28.2) (23.5) (48.3) (95.4) 5159 (4.6) (27.1) (69.6) 3286 (2.9) 468 (.4) (31.5) (68.5) (50.3) (17.8) (17.8) (14.1) 11.5 ± (51.6) 1734 (48.4) 2113 (59.0) 1468 (41.1) 2420 (67.6) 144 (4.0) 280 (7.8) 98 (2.7) 639 (17.8) 30.7 ± (47.9) 1865 (52.1) 762 (21.3) 810 (22.6) 2009 (56.1) 3482 (97.2) 99 (2.8) 693 (19.4) 2776 (77.5) 99 (2.8) 13 (.4) 1640 (45.8) 1941 (54.2) 1525 (42.6) 540 (15.1) 803 ( (19.9).053 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001

172 155 Table A.3 Distribution of Medicaid enrollees by residential urbanicity and race/ethnicity White Black Hispanic Other Unknown Total Metropolitan 28,496 8,484 5,319 2,552 13,404 58,255 Micropolitan 12, , ,588 20,630 Small town 13, , ,530 20,826 Rural 12, ,249 16,619 Total 66,780 9,904 10,078 3,797 25, ,330 Table A.4 Odds of orthodontic utilization and 95% confidence intervals by residential urbanicity, stratified by race/ethnicity White Black Hispanic Other Unknown Crude OR Metropolitan Micropolitan.79 (.69,.89) Small town 1.20 (1.07, 1.33) Rural 1.36 (1.22, 1.51) Significant at p < (.52, 1.64) 1.31 (.57, 3.00) 1.76 (.55, 5.61) 1.60 (1.20, 2.12) 1.09 (.77, 1.56) 1.59 ( ).90 (.47, 1.72) 1.12 (.61, 2.05) 2.28 (1.23, 4.21) 1.20 (.95, 1.52) 2.24 (1.84, 2.72) 1.87 (1.49, 2.36) 1.0 (.91, 1.10) 1.49 (1.37, 1.63) 1.67 (1.52, 1.83) Table A.5 Odds of orthodontic utilization and 95% confidence intervals by sex, stratified by race/ethnicity Female White Black Hispanic Other Unknown Crude OR 1.45 (1.34, 4.58) 1.33 (.95, 1.85) 1.50 (1.18, 1.91) 1.16 (.77, 1.73) 1.37 (1.16, 1.61) 1.41 (1.31, 1.53) Male Significant at p <.05

173 156 Table A.6 Odds of orthodontic utilization and 95% confidence intervals by age, stratified by race/ethnicity White Black Hispanic Other Unknown Crude OR 6-9 years (3.55, 4.39) (1.07, 1.39) Significant at p < (2.07, 4.67).77 (.45, 1.32) 3.44 (2.51, 4.46) 1.63 (1.14, 2.34) 2.56 (1.63, 4.03).78 (.41, 1.48) 3.79 (3.10, 4.63) 1.05 (.83, 1.32) 3.79 (3.48, 4.12) 1.16 (1.05, 1.29) Table A.7 Aggregating claims to origin/destination level and assigning origins to service areas (example) Origin Destination Children (n)

174 157 Table A.8 Descriptive characteristics of orthodontic DSAs DSA Recipient zip codes (n) Medicaid enrollees (n) Ortho utilizers (% of enrollees) Primary care oral evaluation in 2007 (% of enrollees) Mean length of enrollee eligibility in 2007 (months) Mean length of enrollee eligibility in (months) Weighted travel distance (miles) Medicaid-enrolled refers to the total population of individuals meeting the study inclusion criteria. Weighted by the number of Medicaid enrollees ages 6-18 years per zip code.

175 158 Table A.8 Continued DSA Mean Age distribution of Race/ethnicity of Medicaid enrollees enrollee Medicaid enrollees age Unknown/missing White (%) Black (%) Hispanic (%) Other (%) (years) (%) (%) (%)

176 159 Table A.8 Continued DSA DSA name Total population (2010) Population density Land area (square miles) Primary care dentists (n) Primary care dentist to population ratio Medicaid Orthodontists (2010) Total Orthodontists (2010) Ames 116, , Indianola 102, , Marshalltown 70, , Newton 125, , West Des 109, Moines Des Moines 210, South Des 63,726 1, Moines Urbandale 132, Fort Dodge 104, , Charles City 115, , Waterloo/Cedar 377, , Falls Sioux City 190, , Sibley/Spencer 97, , Council Bluffs 284, , Iowa 306, , City/Coralville Cedar Rapids 29, , Bettendorf 156, Clinton 70, , Muscatine 101, ,

177 160 Table A.9 Summary statistics for characteristics of DSA Medicaid enrollees and bivariate correlations with utilization rates Orthodontic utilization (%) Mean Median Std Dev Min Max Correlation with utilization rate (p-value) Orthodontic utilizers (n) (<.0001) Medicaid enrollees (n) (.408) Primary care oral evaluation 2007 (%) (.951) White race/ethnicity (%) (.019) Black race/ethnicity (%) (.077) Hispanic race/ethnicity (%) (.976) Other race/ethnicity (%) (.730) Age 6-11 years (%) (.498) Age (%) (.498) Mean age of enrollees (years) (.322) Mean length of eligibility, 2007 (months) (.576) Mean length of eligibility, (months) (.164) Medicaid orthodontists (2010) Medicaid orthodontist to enrollee ratio (2010) Significant at p < (.326) (.989)

178 161 Table A.10 Summary statistics for characteristics of DSAs and bivariate correlations with utilization rates Mean Median Std Dev Min Max Correlation with utilization rate (p-value) Zip codes (n) (.006) Total population (2010) (.537) Population density (.062) Area (square miles) (.003) Primary care dentist to population ratio (2010) (.825) Weighted travel distance (.097) Significant at p <.05.

179 162 Figure A.1 Age distribution of Medicaid-enrolled orthodontic utilizers, 6-18 years old during CY

180 Figure A.2 Assignment of patient origin zip codes to provider destinations 163

181 Figure A.3 Unassigned zip codes assigned to the nearest provider destination 164

182 Figure A.4 Detail of adjustments to orthodontic service area boundaries 165

183 Figure A.5 Final orthodontic DSA boundaries 166

184 167 Figure A.6 Medicaid-enrolled youth (ages 3-18 years) treated per orthodontist in Iowa, 2010

185 168 APPENDIX B SUPPLEMENTAL MATERIAL STUDY 2 The study described in Chapter V details the process of generating dental service areas (DSAs) in Iowa based on claims submitted to Iowa Medicaid by primary care dentists. Primary care dentists are defined by this project as general dentists and pediatric dentists. This appendix provides additional details about the methods used to create these service areas. Medicaid Eligibility Dataset The initial inclusion criteria for this project included all individuals who met the following requirements: 1. Continuously enrolled in Medicaid for at least 11 months during CYs 2008 to Ages 3 through 18 years as of January 1, Eligible for Medicaid through Supplemental Security Income (SSI) or based on income eligibility requirements. Any individual who was eligible for Medicaid through the foster system, institutionalized status, or who was enrolled in IowaCare was excluded from the study population. After applying those requirements, the study population was limited to those enrollees whose state of residence was Iowa with a valid zip code that corresponded to a zip code tabulation area (ZCTA).

186 169 Medicaid Claims Dataset The claims dataset was limited to individuals ages 3 through 18 years who received any services from a primary care dentist (ie, general or pediatric dentist) (prospec = 1, 91, or 94) in a non-clinic setting (provtype = 4). The Medicaid eligibility dataset was linked with the claims dataset to identify individuals who received dental services during the study period and met the inclusion criteria described above. Claims submitted on behalf of individuals not meeting the inclusion criteria were deleted. Additionally, the provider zip code was matched to the ZCTA dataset, to ensure that providers could be successfully geocoded. Note: there were no cases where the provider zip code failed to match with a ZCTA. In order to identify the number of unique provider zip code locations where each child received primary dental care, the dataset above was aggregated to produce a count of unique provider zip codes per recipient ID number along with a variable that described how many claims were submitted per individual. This dataset was used to generate weights for each individual based on the proportion of visits from each patient origin to each provider destination. This procedure is similar to the one used by Goodman et al to generate patient preference fractions in defining primary medical care service areas(the Dartmouth Institute for Health Policy and Clinical Practice 2007).

187 170 The claims dataset represented 76,839 unique recipients of primary dental care services during CY 2008 through Mean number of primary care visits per individual was 3.2 (SD 2.2, median=3). Number of visits per recipient ranged from 1 to 24. Individuals received care from a mean of 1.2 provider destinations (SD 1.0, median=1). Number of provider destinations per individual ranged from 1 to 5. Over 96 percent of claims were submitted by dentists in Iowa. A complete list of states where Medicaid-enrolled individuals received primary dental care is provided in Table 1. The mean number of provider destinations serving each recipient origin was 6.9 (SD 5.7) and ranged from 1 to 42. The number of recipients per origin zip code ranged from 1 (n=47) to 2,175. Mean number of recipients per origin was 91.0 (SD 229.8); median number of recipients was 25. The individual-level dataset was aggregated to the origin/destination level to calculate the sum of patient preference weights per zip code dyad. Origins were assigned to destinations based on where the highest fraction of patient visits was provided. Out of 953 origin/destination dyads, only 22 had ties between either 2 (n=18) or 3 (n=4) provider destinations. In instances of ties, the closest destination was selected using straight line distance as calculated using ArcView After eliminating cases with ties, Medicaid recipients in 927 zip codes received a plurality of care from 168 unique provider zip codes. Of these 168 provider destinations, 163 were located in Iowa, 1 in South Dakota, 2 in

188 171 Missouri, 1 in Nebraska, and 1 in Wisconsin. These provider destinations formed the basis for crude service areas. Thirty-four of the provider destinations failed to serve a plurality of their own residential population. In these cases, the component residential zip codes (n=77) were reassigned to the DSA where the residents of each provider destination received care. For example, zip code had been assigned to DSA However, the residents of received a plurality of their primary care from DSA Therefore, was reassigned to This ensured that each DSA contained only one provider location, and that the provider location for each DSA s constituents was located within that service area s boundaries. Fiftyone zip codes were reassigned during this adjustment, which resulted in 139 crude DSAs; 135 were located in Iowa, 1 in Illinois, 2 in Missouri, and 1 in Nebraska. Creating primary care DSAs There were three major steps during the process of generating service area boundaries: Step 1: Initial zip code assignments based on plurality resulted in 139 crude DSAs (Figure 1). Step 2: Adjust crude DSAs based on minimum population requirement of 50 Medicaid recipients per service area - reduced to 123 adjusted DSAs (Figure 2).

189 172 Step 3: Finalize DSA boundaries by removing island zip codes (those areas separate from the major service area body) resulted in 113 final DSAs (Figure 3). The zip code to DSA assignments from each step in the process were linked in a dataset, in order to evaluate this process. Maps were generated to display the service area boundaries at each stage. Localization of care From this list of crude service area assignments, I calculated the proportion of primary care received within each DSA by the Medicaid enrolled population of that service area. This variable, localization of care, serves as a measure of how well each service area is capturing the local market exchange for primary dental care. Mayer (1999) used a similar measure in her county-level study of Medicaid service areas. In that study, service areas were expanded until they accounted for 75 percent of all dental care obtained by area residents. Goodman et al calculated a preference index per primary care service area to measure the proportion of care that was provided to residents by providers within the same area.( The Dartmouth Institute for Health Policy and Clinical Practice 2007). Goodman et al. (2003) established a minimum preference index per service area of 30 percent; they asserted that service areas with greater than 70 percent of care provided by providers outside of the area suffered from excessive border crossing.

190 173 Localization of care per DSA was calculated as the proportion of care received by children from all providers located within each service area. Provider locations were identified from the claims data and a key was created that identified where each provider zip code was located at each step during service area creation. Specifically, localization of care (LOC) was calculated as: Where, LOCr = localization of care in DSA r Vij = number of visits by child i to provider j Pr = number of providers in DSA r Pt = number of providers in all DSAs Cr = number of children in DSA r LOC was calculated for each DSA at the three stages of generating service area boundaries. LOC was then calculated per county as the percentage of care received by the residents of each county by providers within that county (eg, aggregated by recipient county). Descriptive statistics and frequency histograms for each scheme were created (Table 2, Figures 4-7). General characteristics of DSAs The number of primary care dentist destinations (zip code level) ranged from 1 to 14 (Table 3). Approximately 60% of DSAs had one dentist destination serving the study population within their boundaries (mean 2.2,

191 174 SD 2.5) (see Figure 8 for frequency histogram). The number of Medicaid enrolled children who received primary dental care within their DSA boundaries ranged from 55 to 7,688 (mean 680.0, SD 1,155.3). The number of ZCTAs per service area ranged from 1 to 53 (mean 8.3, SD 8.4) (see Figure 9 for frequency histogram). Utilization rates The distributions of utilization rates per county and per DSA are displayed in Figures 10 and 11. Dentists per DSA In order to calculate the number of dentists serving each DSA, the number of unique practice sites per dentist (identified by the PVNUM variable) was calculated. It should be noted that this variable differentiates between multiple practice locations per dentist. Therefore, the sum of dentists serving all service areas is greater than the sum of individual dentists serving the Medicaid population (929 vs. 876) since a single dentist may serve more than one DSA. In order to accurately reflect access within each service area, I allowed dentists to be counted once in every DSA where they had practice locations. There were 876 dentists practicing in 929 locations (Table 4) that submitted one or more claims to Iowa Medicaid on behalf of the individuals in our study population. The number of dentist locations per DSA is displayed in Figure 12. Approximately 20% of DSAs (N=23) were served by only one dentist. Twenty-five percent of DSAs were

192 175 served by 7 or more dentists. The Iowa City service area had the highest number of primary care dentists treating the pediatric Medicaid population, with 64. However, 22 of these 64 dentists were affiliated with the University of Iowa. The dentist supply measures above describe the primary care workforce during the three-year study period ( ). In order to provide a one-year description of the workforce, I calculated the Iowa dentist workforce that provided primary care to Medicaid enrolled children and adolescents during CY During 2010, 640 dentists in the state submitted at least one claim to Iowa Medicaid. A total of 319 dentists treated at least 25 children during this one year period; 197 dentists treated at least 50 children, and 97 treated at least 100.

193 176 Table B.1 State origin of claims submitted by primary care dentists on behalf of Medicaid-enrolled individuals, ages 3-18 years, CY State Claims (%) Iowa 237,971 (96.2%) South Dakota 6,273 (2.5) Nebraska 1,810 (0.7) Illinois 916 (0.4) Missouri 191 (0.1) Wisconsin 95 (0) Minnesota 3 (0) North Dakota 2 (0) Total 247,261 Table B.2 Localization of care for primary care DSAs and counties Step 1 Crude PC- DSAs Step 2 Adjusted PC- DSAs Step 3 Final PC-DSAs Counties N=139 N=123 N=113 N=99 Mean 55.0% 57.0% 58.5% 52.4 Median Std Dev Min Max

194 177 Table B.3 boundaries) Descriptive characteristics of primary care DSAs (final N Mean Median SD Min Max ZCTAs Provider destinations Medicaid recipients 76, , ,688 Medicaid enrollees 153, , ,829 Utilization rates (per 100 enrollees) Visits to providers within DSAs 247,052 2, , ,395 Visits to providers outside of DSA 64, , ,633 Primary care dentists treating 1 Medicaid enrollee Geographic size (mi 2 ) ,236.6 Table B.4 Geographic distribution by state of primary care dentists treating 1 Iowa Medicaid enrollee State Dentists (n) Percent Iowa % Nebraska South Dakota Wisconsin Missouri Illinois Minnesota North Dakota Total

195 Figure B.1 Crude primary care DSAs 178

196 Figure B.2 Adjusted primary care DSAs 179

197 Figure B.3 Final primary care DSAs 180

198 181 Figure B.4 LOC per crude PC-DSA (Step 1) Figure B.5 LOC per adjusted PC-DSA (Step 2)

199 182 Figure B.6 LOC per final PC-DSA (step 3) Figure B.7 LOC per county

200 183 Figure B.8 Provider destinations (zip codes) per final primary care DSA Figure B.9 ZCTAs per final primary care DSA

201 184 Figure B.10 Rate of primary care service utilization per 100 Medicaid enrollees per DSA (N=113) Figure B.11 Rate of primary care service utilization per 100 Medicaid enrollees per county (N=99)

202 Figure B.12 Practice locations per DSA of primary care dentists who submitted 1 claim to Iowa Medicaid during the study period 185