Dental X-rays and Risk of Meningioma: Anatomy of a Case-Control Study

Transcription

1 research-article2013 JDRXXX / PERSPECTIVE D. Dirksen*, C. Runte, L. Berghoff, P. Scheutzel, and L. Figgener Deartment of Prosthetic Dentistry and Biomaterials, University of Muenster, Waldeyerstr. 30 D Muenster, Germany; *corresonding author, Dental X-rays and Risk of Meningioma: Anatomy of a Case-Control Study J Dent Res XX(X):1-2, 2013 KEY WORDS: eidemiology, radiology, statistics, cancer, ionizing radiation, risk factors. The urose of comuting is insight, not numbers. (Hamming, 1973) If there were a list of the most feared diseases, cancer would robably reside among the to ositions. Thus, if a study ostulates that a factor x substantially increases the tumor risk (even if it is a benign one), a strong echo can be exected in the media. This haened when Claus et al. (2012) reorted a significantly increased risk for meningioma induced by dental x-ray diagnostics. Even rofessional journals ublished the results without critical assessment. There is strong evidence that ionizing radiation may cause cancer. This evidence is most thoroughly backed by eidemiological investigations on large collectives of survivors of the nuclear bombs droed on Jaan at the end of WW2 (Preston et al., 2003). However, the question remains whether low-dose exosure to ionizing radiation (Aendix 1) leads to a detectable increase in risk of tumors. There have been several similar studies investigating a ossible association between cases of meningioma and various factors, including x-ray diagnostics and amalgam, the latest by Longstreth et al. (2004) and Ryan et al. (1992). They all have at least two things in common: Their design is that of a casecontrol study, and data acquisition is erformed via interviews. Furthermore, most studies use the odds ratio as a measure of effect. However, the one resented by Claus et al. is the largest, and if any study could rovide evidence for an association between dental x-ray diagnostics and intracranial meningioma, it should be this one. But does it? The chosen case-control study design is the only reasonable aroach for study, given the small incidence of this disease (around 2/100,000 annually). This means that a samle of atients (cases) is comared with a samle of healthy control DOI: / Received February 5, 2013; Last revision March 5, 2013; Acceted March 7, 2013 A sulemental aendix to this article is ublished electronically only at htt://jdr.sageub.com/sulemental. individuals with resect to the relative frequencies of different tyes of exosures. In other words, the emirical robability of a atient having been subjected to a certain exosure is calculated, not the robability that an exosure is related to a disease. This is an inherent weakness of the case-control design besides the fact that it is rone to samling and recall bias (Mann, 2003). But let us look at the results. Central findings of the authors include (Aendix 2): (1) Over a lifetime, cases were more than twice as likely as controls [odds ratio (OR), 2.0 [ ]] to reort having ever had a bitewing examination. (2) individuals who reorted receiving such films [anorex] at ages < 10 years had a 4.9 times increased risk [ ] of meningioma. (3) Risk estimates for full-mouth films, although not statistically significant, were consistently in the same direction as for the other 2 film tyes. Finding #1 indicates a two-fold increase in risk, if the individual had at least one bitewing in his/her entire life. This is somewhat surrising, considering the small radiation dose associated with a bitewing, esecially when comared with the other diagnostic methods. A look at the actual relative frequencies unveils the reason: Both cases and controls reort this case with nearly the same (high) relative frequencies: 95.8% and 92.2%, resectively. As is demonstrated in the Aendix, the odds ratio may not simly be translated into a robability ratio (relative risk) if high robabilities are regarded (Davies et al., 1998) an imortant fact that is too often overlooked (Tetradis et al., 2012). Finding #2 ostulates a nearly five-fold increase in risk for individuals who had at least one anorex film at an age < 10 yrs. In absolute numbers, this means that 22 out of 1,433 atients reorted such examinations, while this was the case for five out of 1,350 control individuals. With this result, it should be ket in mind that case-control studies ossess inherent uncertainties (Mann, 2003), amlified by the method of data collection, which is based on recall of treatments which, in this case, occurred, on average, five decades reviously and is rone to ersonal bias (see also Jorgensen, 2013; White et al., 2013). As is shown in Aendix 3, these ratios are very sensitive to even Downloaded from jdr.sageub.com at Universitatsbibliothek on Aril 1, 2013 For ersonal use only. No other uses without ermission. 1

2 2 Dirksen et al. J Dent Res XX(X) 2013 small errors. In the resent case, an assumed error margin of ± 1% would suffice to reverse the ratio. It is highly questionable whether such accuracy can be assumed. Finding #3 looks inconsicuous at first glance, but it unfolds a severe inconsistency of the results. The authors seem to be unaware of the fact that a full-mouth series consists of from 16 to 20 single films and thus exoses the atient to a dose which is larger by one order of magnitude than that with a single film and is still considerably larger than that obtained with a anorex. Nevertheless, in their findings, the authors suggest that a relatively large dose leads to no significant increase in risk, while much smaller doses do. A more lausible exlanation would be (besides the fact that the OR in finding #1 overestimates the risk) that the figures simly lack reliability, esecially when they are based on small absolute numbers, as with finding #2. Since finding #3 corresonds to a much larger radiation dose as well as to higher statistical reliability, this result alone would lead to the conclusion that there is no (observable) effect. A remark should be made concerning the resentation of the results. Findings #1 and #2, which ostulate an increased risk of meningioma, are rominently laced in the abstract (which is robably what most eole will read). In contrast, finding #3, which describes a null result (with much larger biohysical and statistical imact), is somewhat hidden in the discussion section. Thus, an imression is roduced which does not stand a closer insection. A reliability analysis of the results in general reveals that the higher the reliability, the smaller the effect size (Aendix 4). This confirms our conclusion that the study rovides little to no evidence of an increased risk of meningioma for individuals exosed to low-dose dental x-ray diagnostics. Future studies on this toic should ay more attention to dosimetry as well as to error analysis beyond standard confidence intervals. ACKNOWLEDGMENT The authors received no financial suort and declare no otential conflicts of interest with resect to the authorshi and/or ublication of this article. REFERENCES Claus EB, Calvocoressi L, Bondy ML, Schildkraut JM, Wiemels JL, Wrensch M (2012). Dental x-rays and risk of meningioma. Cancer 118: Davies HT, Crombie IK, Tavakoli M (1998). When can odds ratios mislead? BMJ 316: Hamming RW (1973). Numerical methods for scientists and engineers. New York: McGraw Hill. Jorgensen TJ (2013). Dental x-rays and risk of meningioma. Cancer 119:463. Longstreth WT Jr, Phillis LE, Drangsholt M, Koesell TD, Custer BS, Gehrels JA, et al. (2004). Dental x-rays and the risk of intracranial meningioma: a oulation-based case-control study. Cancer 100: Mann CJ (2003). Observational research methods. Research design II: cohort, cross-sectional, and case-control studies. Emerg Med J 20: Preston DL, Shimizu Y, Pierce DA, Suyama A, Mabuchi K (2003). Studies of mortality of atomic bomb survivors. Reort 13: Solid cancer and noncancer disease mortality: Radiat Res 160: Ryan P, Lee MW, North B, McMichael AJ (1992). Amalgam fillings, diagnostic dental x-rays and tumours of the brain and meninges. Eur J Cancer B Oral Oncol 28B: Tetradis S, White SC, Service SK (2012). Dental x-rays and risk of meningioma; the jury is still out. J Evid Based Dent Pract 12: White SC, Hildebolt CF, Lurie AG (2013). Dental x-rays and risk of meningioma. Cancer 119:464. Downloaded from jdr.sageub.com at Universitatsbibliothek on Aril 1, 2013 For ersonal use only. No other uses without ermission.

3 PERSPECTIVE D. Dirksen*, C. Runte, L. Berghoff, P. Scheutzel, and L. Figgener Deartment of Prosthetic Dentistry and Biomaterials, University of Muenster, Waldeyerstr. 30 D Muenster, Germany; *corresonding author, Dental X-rays and Risk of Meningioma: Anatomy of a Case-Control Study J Dent Res DOI: / APPENDIX (1) Effective Doses from Diagnostic X-ray and Natural Background Radiation As a motivation for their study, the authors claim that dental x-rays are the most common artificial source of ionizing radiation. This is true; however, the effective doses are [even in the case of higher doses in the ast (Claus et al., 2013) or multile diagnostics er year; see Aendix Table 1] far below the yearly exosure received from natural background radiation, which is aroximately 3000 μsv in the United States (Mettler et al., 2008; Health Physics Society, 2010). Even more questions arise because the cerebral region is not within the rimary radiation field and receives mainly scattered radiation. The main roblem is how such an effect could be statistically detectable. Natural background radiation also varies with geograhical location. Aendix Table 1. Effective Radiation Doses from Diagnostic X-ray Procedures (Mettler et al., 2008) Tye Effective Dose [ Sv] Dental bitewing 5 Dental (anoramic) 10 Dental (full-mouth) The dose for full-mouth examinations is calculated as 16 to 20 times the dose of a single film. (2) Values Determined in the Study by Claus et al. The case-control study under consideration (Claus et al., 2012) included 1,433 atients with intracranial meningioma and a control grou of 1,350 individuals. The mean age of both grous was aroximately 57.5 yrs. Data were collected by means of telehone interviews in which the articiants were asked for tye (bitewing, anorex, full-mouth) and frequency (none, less than yearly, yearly or more) of dental x-ray diagnostics, as well as age at time of examination (< 10, 10-19, 20-49, > 50 yrs). As the main statistical arameter, the authors comuted the odds ratio (OR, adjusted for age, gender, etc.) for atients and control individuals based on the number of exosed individuals. All values in Aendix Table 2 are taken from Table 2 in Claus et al. (2012). A roblem with regard to the data in that Table is that they are inconsistent: The ercentage figures are not equal to the number of exosed ersons divided by the total number of ersons in the grou. Furthermore, the listed odds ratios cannot be reroduced when calculated from the ercentage figures. The authors should have commented on this. Nevertheless, we use the given values, since this does not substantively affect results. (3) Statistical Methods and Error Proagation The concet of risk estimation in a case-control study is based on comarison of the estimated robabilities of exosure for both atients and control individuals. There are several statistical arameters available for this comarison (Breslow and Day, 1980), among which the relative risk (RR) and the odds ratio (OR) are the most commonly used. Since there frequently aear to be some misunderstandings, esecially in the interretation of the OR (Pearce, 1993; Davies et al., 1998; Langholz, 2010), we exlain and comare both arameters. Secial attention has to be aid to the imact of errors in the inut data on these quantities. Instead of treating this roblem in a formal analytical way as roagation of uncertainties (Bronshtein et al., 2007,.794), we illustrate the roblem with numerical examles. The basic arameter that is calculated in a case-control study is the emirical robability (i.e., the relative frequency) of a erson exosed to a risk factor, e.g., having received a certain diagnostic x-ray at a certain age. It is defined as number of exosed ersons n total number of ersons N This quantity may also be called risk, but this is just a naming convention. To comare the robabilities 1 and 2 in two grous, the relative risk RR is calculated as RR In Aendix Table 3, the relative risk (which is just the robability ratio) is calculated for two different cases: In the left air of (1) (2) DS1

4 DS2 Dirksen et al. J Dent Res Aendix Table 2. Data Corresonding to the Main Findings Extracted from Table 2 in Claus et al. (2012) Cases, n = 1433 Controls, n = 1350 Variable No. % No. % OR (95% CI) Ever had bitewing Any age 1, ( ) Ever had Panorex Aged < 10 yrs ( ) Ever had full-mouth Aged < 10 yrs ( ) Ages yrs ( ) Ages yrs ( ) Aged > = 50 yrs ( ) Any age ( ) Aendix Table 3. Examles of Calculated Relative Risks in the Cases of Low and High Relative Frequencies (of exosed atients and controls) Low Relative Frequency High Relative Frequency Patients Controls Patients Controls Ex. cases, n 4 2 (-1) (-1) Total, N Risk, (.01).4.2 (.19) Rel. risk, RR 2.0 (4.0) 2.0 (2.11) The numbers in arentheses reresent a case where a data acquisition error of -1% for the no. of exosed controls is assumed. columns, a small robability of exosure, for both atients (4/100) and control individuals (2/100), is assumed, while, in the second case, 40 out of 100 atients and 20 out of 100 control individuals are affected. In both cases, the relative risk of being exosed is 2.0. However, acquisition error (e.g., by wrong diagnostics, false answers during interviews) can seriously affect results. To demonstrate this effect, we assume that 1% of control individuals give false-ositive answers. Thus, the true number of exosed control individuals is reduced by one (figures in brackets). Re-calculation of the relative risks shows that, in the case of high robabilities, this has little effect (RR = 2.11 instead of 2.0), while for low robabilities, the relative risk doubles (RR = 4.0 instead of 2.0). A roer analytical discussion would require consideration of artial derivatives (Bronshtein et al., 2007) of the RR function, but the numerical examles should be sufficient for two conclusions to be drawn: (1) An error estimation should be carried out to judge the reliability of the RR; and (2) the larger the relative frequency of exosed ersons, the more reliable is the calculated RR (or vice versa: For small robabilities, the RR is sensitive to errors.). In the study discussed, the authors use the odds ratio rather than the RR to comare the exosure robabilities. The odds (o) are defined as the ratio of the robability of an event divided by its converse robability: o =. (3) 1 The odds ratio OR of the odds o 1 and o 2 of two grous is then calculated as: OR o 1 1( 1 2) = =. (4) o ( 1 ) As before, 1 and 2 denote the robabilities of atients and control individuals being exosed, resectively. It should be evident from the definition of OR (eq. 4) that, in general, it cannot be interreted as a ratio of robabilities (e.g., OR = 2 does not mean two times as likely!). Again, numerical examles should hel to clarify this. In Aendix Table 4, two situations are comared (very similar to the receding examle). In the left columns, robabilities are low, i.e., 9 out of 100 atients and 8 out of 100 control individuals are exosed. This results in a relative risk RR = and an odds ratio OR = If, conversely, these arameters are calculated for 90 out of 100 atients and 80 out of 100 control individuals, the corresonding OR is 2.25 (RR remains 1.125). Thus, for small robabilities, RR and OR are close, while for large robabilities, this is not the case; moreover, the odds ratio is hard to interret (Davies et al., 1998). The argument that in a case-control study the risk cannot be comuted is misleading (Langholz, 2010), because it is the robabilities of exosure that are relevant.

5 J Dent Res Dental X-rays and Risk of Meningioma DS3 Aendix Table 4. Examles of Calculated Odds Ratios in the Cases of Low and High Relative Frequencies (of exosed atients and controls) Low Relative Frequency High Relative Frequency Patients Controls Patients Controls Ex. cases, n Total, N Risk, Odds ratio, OR Rel. risk, RR Odds ratio, OR effect size (rela ve risk) reliability (ercentage of exosed controls) Aendix Figure. Effect size vs. reliability (ercentage of exosed controls). (4) Reliability Analysis In Aendix 3, we concluded that, with larger exosure robabilities, the risk ratio becomes more reliable (i.e., less rone to errors). In the Aendix Fig., the relative risk (as a more solid estimator of effect size) is calculated from the exosure robability of control individuals (the ercentage figures). All figures from Table 2 in Claus et al. (2012) have been included, regardless of the underlying question. APPENDIX REFERENCES Breslow NE, Day NE (1980). Statistical methods in cancer research. Volume I - The analysis of case-control studies. IARC Sci Publ 32: Bronshtein IN, Semendyayev KA, Musiol G, Muehlig H (2007). Handbook of mathematics. Berlin: Sringer. Claus EB, Calvocoressi L, Bondy ML, Schildkraut JM, Wiemels JL, Wrensch M (2012). Dental x-rays and risk of meningioma. Cancer 118: Claus EB, Wiemels J, Wrensch M (2013). Dental x-rays and risk of meningioma: resonse to Drs. Calnon, Jorgensen, and White. Cancer 119: Davies HT, Crombie IK, Tavakoli M (1998). When can odds ratios mislead? BMJ 316: Health Physics Society (2010). Radiation exosure from medical exams and rocedures. URL accessed on 3/7/2013 at: htt://hs.org/documents/ Medical_Exosures_Fact_Sheet.df. Langholz B (2010). Case-control studies = odds ratios: blame the retrosective model. Eidemiology 21: Mettler FA Jr, Huda W, Yoshizumi TT, Mahesh M (2008). Effective doses in radiology and diagnostic nuclear medicine: a catalog. Radiology 248: Pearce N (1993). What does the odds ratio estimate in a case-control study? Int J Eidemiol 22: