Anthropometry and Esophageal Cancer Risk in the European Prospective Investigation into Cancer and Nutrition

During the past 2 decades, the incidence of esophageal adenocarcinoma (EAC) increased rapidly in Europe and North America, whereas the incidence of esophageal squamous cell carcinoma (ESCC) generally declined (1, 2). Formerly known as an uncommon form of esophageal cancer, EAC now represents the fastest growing malignancy of the last 25 years in the United States, with incidence rates among white males finally having surpassed those of ESCC (2).

This increasing trend in EAC incidence has been mainly attributed to the worldwide increase in obesity prevalence (1). Previous case-control and prospective studies have consistently shown a positive association between body mass index (BMI) and risk of EAC (3, 4). In addition, the World Cancer Research Fund/American Institute for Cancer Research has recently rated the evidence for a higher risk of EAC due to greater body fatness as “convincing” (5). In contrast, numerous epidemiologic studies found that BMI is inversely associated with risk of ESCC (3, 6, 7).

Because BMI is an indicator of general obesity, it remains unclear whether body fat distribution, particularly abdominal obesity, is associated with the development of esophageal cancer. For certain diseases, mainly type 2 diabetes and cardiovascular disease, it has been shown that abdominal obesity [reflected by a large waist circumference or waist-to-hip ratio (WHR)] is more closely related to morbidity than overall obesity and several studies underline the importance of assessing waist circumference or WHR in addition to BMI for predicting disease risk (8-11). However, to our knowledge, only two prospective studies have yet evaluated whether body fat distribution is related to risk of esophageal cancer (12, 13), with one of them exclusively considering abdominal diameter as indicator of body fat distribution (13) and the other one solely focusing on EAC combined with gastric adenocarcinoma as end point (12).

Case-control studies further suggest that the relation between BMI and esophageal cancer may depend on smoking status in a way that the inverse association of BMI with ESCC may be only present among (heavy) smokers (14) or that increases in EAC risk with BMI may be greater among nonsmokers than among smokers (15). However, these findings have rarely been evaluated in prospective studies. Moreover, many previous epidemiologic studies (7, 13, 16, 17) had only limited data on tobacco smoking, and therefore, residual confounding might have distorted the observed associations.

In a large prospective cohort study, the European Prospective Investigation into Cancer and Nutrition (EPIC), with detailed directly measured anthropometric characteristics and comprehensive information on smoking, alcohol consumption, and other lifestyle habits, we therefore evaluated the relation of body height and general and abdominal obesity to the incidence of EAC and ESCC.

The EPIC study is a multicenter prospective cohort study designed primarily to investigate the relation between nutrition and the incidence of cancer and other chronic diseases (18, 19). Subcohorts were recruited at 23 centers in 10 European countries: Denmark, France, Germany, Greece, Italy, the Netherlands, Norway, Spain, Sweden, and the United Kingdom. The 519,978 eligible men and women were mostly aged 25 to 70 years and recruited from the general population residing in a given geographic area, a town, or a province. Exceptions were the French cohorts (based on female members of the health insurance for school employees), the Oxford cohort in the United Kingdom (based on vegetarian volunteers and healthy eaters), parts of the Italian and Spanish cohorts (based on blood donors and their spouses), and the cohorts in Utrecht (the Netherlands) and Florence (Italy), which were based on women attending breast cancer screening. Eligible subjects were invited to participate in the study, and those who accepted and gave informed consent completed questionnaires on their diet, lifestyle, and medical history. Participants were then invited to a center to provide a blood sample and to have anthropometric measurements taken by trained staff. The baseline enrollment was carried out between 1992 and 2000.

For the present analysis, participants who reported cancer at any site at baseline examination were excluded from the original study population, among them 212 prevalent esophageal cancer cases. Further exclusions refer to subjects with missing anthropometric characteristics, missing questionnaire data on alcohol intake and diet, or missing dates of cancer diagnosis or incomplete follow-up data. We excluded participants from Norway for whom only self-reported data on anthropometry were available. Participants from the cohorts in Greece and France were excluded due to incomplete case identification routines at that time for this cancer site. After exclusions, 130,921 men and 215,633 women with complete information on height and weight remained for analyses, whereas analyses involving waist and hip circumferences were restricted to 115,390 men and 200,507 women.

Weight and height were measured to the nearest 0.1 kg and 0.1 or 0.5 cm, respectively, with subjects wearing no shoes (20). Waist circumference was measured either at the narrowest torso circumference (Italy; United Kingdom; Utrecht, the Netherlands; and Denmark) or at the midway between the lower ribs and iliac crest (Bilthoven, the Netherlands; Potsdam, Germany; and Malmö, Sweden). In Spain and Heidelberg (Germany), a combination of methods was used, although the majority of participants were measured at the narrowest circumference. Hip circumference was measured at the widest circumference (Italy; Spain; Bilthoven, the Netherlands; and Malmö, Sweden) or over the buttocks (United Kingdom; Utrecht, the Netherlands; Germany; and Denmark). In Umeå (Sweden), anthropometric data collection was restricted to measurement of weight and height.

Body weight and waist and hip circumference were adjusted for heterogeneity due to protocol differences in clothing worn during measurement (20). For the “health conscious group” based in Oxford (United Kingdom), linear regression models were used to predict sex- and age-specific values from participants with both measured and self-reported body measures as previously described (21). BMI was calculated as weight in kilogram divided by height in meters squared (kg/m²). The waist and hip circumferences were used to calculate the WHR.

Further lifestyle- and health-related information was obtained using nondietary questionnaires. These included questions on education, occupation, history of previous illnesses and disorders or surgical operations, smoking status and lifetime history of tobacco use, consumption of alcoholic beverages, and recreational and household activity (hours/week). The information on occupational activity (coded as sedentary occupation, standing occupation, manual work, heavy manual work, unemployed, or missing) and the sum of the recreational activities cycling and sports (hours/week, coded in four categories: none, ≤3.5, 3.5-7.0, and >7.0) were used to create a variable for total physical activity by cross-classifying participants into five categories (inactive, moderately inactive, moderately active, active, and missing). The information on alcoholic beverage consumption at ages 20, 30, and 40 and at baseline was converted into a variable representing lifelong alcohol intake (22).

Diet was measured by country-specific food frequency questionnaires designed to capture local dietary habits and to provide high compliance. Food intake in grams per day was calculated by multiplying reported frequency and portion size. The observed intake was corrected for systematic overestimation and underestimation by the dietary instruments used in the different subcohorts and for random measurement error by linear calibration using a standardized computerized 24-h diet recall method (23).

The follow-up was based on population cancer registries (Denmark, Italy, Netherlands, Spain, Sweden, and United Kingdom) or a combination of methods including regional and local cancer registries together with an active follow-up through participants and their next of kin (Germany). Mortality data were also collected from either the cancer registry or mortality registries at the regional or national level. By March 2007, the time period of last cancer update for this analysis, for all centers using cancer registry data, censoring dates for complete follow-up were at December 31, 2002 (Granada), December 31, 2003 (Aarhus, Bilthoven, Copenhagen, Florence, Murcia, Naples, and Varese), December 31, 2004 (Asturias, Cambridge, Malmö, Navarra, Oxford, Ragusa, Turin, and Utrecht), and December 31, 2005 (San Sebastian and Umeå). For countries using individually based follow-up, the end of follow-up was considered to be the date of the last known contact, or date of diagnosis, or date of death, whichever came first. Mortality data were coded following the rules of the 10th revision of the International Statistical Classification of Diseases, Injuries and Causes of Death, and cancer incidence data following the 2nd revision of the International Classification of Diseases for Oncology. We included incident primary carcinomas of the esophagus coded as C15. Morphology information was used to classify the malignant tumors into EAC and ESCC. Because distinction between adenocarcinoma of the esophagus and the gastric cardia is difficult (24), for most reported cases, diagnoses and classification of tumor site were validated and confirmed by a panel of pathologists, including a representative from each participating country (25). The panel reviewed material provided by the centers using the original histopathologic reports, histologic slides, and slices obtained from paraffin blocks, and only those cases that could have been identified unequivocally as EAC were included in the analysis.

Associations between anthropometric measures and esophageal cancer were analyzed by calculating relative risks (RR) as incidence rate ratios using Cox proportional hazards regression. Age at recruitment was taken as the underlying time variable in the counting process with entry and exit time defined as the subject's age at recruitment and age at cancer diagnosis or censoring, respectively. Participants were categorized into quintiles based on the anthropometric variables of the entire male or female cohorts, respectively. We also did analyses by including anthropometric measures as continuous variables in the models and by grouping subjects into predefined categories of BMI according to current definitions of underweight (<18.5 kg/m²), normal weight (18.5-24.9 kg/m²), overweight (25.0-29.9 kg/m²), and obesity (≥30.0 kg/m²; ref. 11). In Cox models, the variables center and age were used as stratum variables to control for differences in questionnaire design, follow-up procedures, and other nonmeasured center effects, and to be more robust against violation of the proportionality assumption. In model 1, RRs were adjusted for sex and education (no school or primary school degree, technical/professional school degree, secondary school degree, university degree, not specified). Model 2 additionally included established risk factors for esophageal cancer, namely, smoking habits (lifelong nonsmoker, former smoking ceased ≥10 y, former smoking ceased <10 y, current smoking with <15, current smoking with 15-24, current smoking with ≥25 cigarettes/day, and current smoking with unknown quantity or smoking other than cigarettes, missing), duration of smoking (none, ≤20 years, >20 years, unknown), lifelong alcohol consumption (g/d), and alcohol consumption at recruitment (g/d). The final model (model 3) was further adjusted for intake of fruits, vegetables, meat and meat products (g/d), and physical activity (inactive, moderately inactive, moderately active, active, and missing). To estimate whether body fat distribution is associated with cancer risk statistically independently of the association with general obesity, analyses of waist and hip circumference and WHR were further adjusted for BMI (continuously). Tests for trend across quintiles of anthropometric characteristics were evaluated by assigning each participant a score from 1 to 5 for the respective quintile and modeling this value as a continuous variable.

Effect modification by sex was assessed by evaluating interaction terms. Further, we evaluated interaction with smoking and alcohol consumption by doing stratified analyses and by evaluating interaction terms. Smokers were defined as current smokers or former smokers who had quit <10 years ago; nonsmokers were defined as lifelong nonsmokers or former smokers who had quit at least 10 years ago. Baseline alcohol intake was stratified according to alcohol content of one standard glass with 12 g ethanol as 0 to 6 g/d (up to half a glass per day) versus >6 g/d.

All P values presented are two tailed and P < 0.05 was considered statistically significant. Analyses were done using SAS version 9.1 (SAS Institute).

A total of 346,554 participants were followed for an average of 8.9 ± 1.8 years. During follow-up, we documented 88 new cases of EAC (73 men and 15 women) and 110 cases of ESCC (68 men and 42 women), of which the majority was reported in the United Kingdom (33 EAC and 18 ESCC), Sweden (16 EAC and 23 ESCC), and Denmark (26 EAC and 32 ESCC). Mean age at inclusion was 51.1 ± 10.5 years; the largest country-specific mean BMI was observed in Spain (men, 28.4 kg/m²; women, 28.1 kg/m²), whereas it was lowest in the United Kingdom (25.4 and 24.2 kg/m², respectively), where predominantly health-conscious participants had been recruited.

Participant characteristics for men and women across quintiles of BMI are shown in Table 1. Men with higher BMI tended to be older and were more likely to have a history of smoking. They had higher baseline and lifelong alcohol consumption and consumed more fruits, vegetables, and meat. Although women with higher BMI also tended to be older, they were less likely to have ever smoked. Women with higher BMI consumed more fruit and meat but similar amounts of vegetables compared with women with lower BMI. In both men and women, participants with higher BMI tended to be less educated.

The correlation coefficients for BMI with waist circumference, hip circumference, and WHR were 0.78, 0.83, and 0.43, respectively.

Table 2 presents the associations between anthropometric characteristics and risk of EAC and ESCC, which were homogenous across sexes (all P for interaction >0.05). Adjustment for the intensity and duration of smoking as well as lifetime and current alcohol consumption had generally minor influence on the observed associations (model 2). In addition, the additional adjustment for dietary factors and physical activity did not materially alter the observed RR estimates (model 3). Body height was not associated with risk of either EAC or ESCC. BMI was positively related to risk of EAC [highest compared with the lowest quintile of BMI; RR, 2.60; 95% confidence interval (95% CI), 1.23-5.51; P_trend = 0.01]. A higher waist circumference or a higher WHR was also associated with a significantly higher risk of EAC (P_trend = 0.003 and 0.004, respectively). After dividing participants into predefined groups of BMI, the fully adjusted RRs (95% CI) of EAC for overweight and obese compared with normal-weight participants were 1.21 (0.74-1.99) and 1.80 (0.97-3.33), respectively. On a continuous scale, a 1 kg/m² higher BMI, a 5-cm higher waist circumference, or a 0.1-unit higher WHR was related to a 1.08-fold (95% CI, 1.02-1.14), 1.16-fold (95% CI, 1.04-1.29), or 1.59-fold (95% CI, 1.12-2.26) higher risk for EAC, respectively.

In contrast, body weight and BMI were negatively associated with ESCC (highest versus lowest quintile; RR, 0.33; 95% CI, 0.18-0.60 and RR, 0.26; 95% CI, 0.14-0.51). The fully adjusted RRs (95% CI) of ESCC for underweight, overweight, and obese compared with normal-weight participants were 4.04 (1.57-10.4), 0.48 (0.30-0.75), and 0.46 (0.24-0.90), respectively. Across quintiles of waist circumference, a tendency toward an inverse relation with ESCC was observed, which was significant when waist circumference was investigated on a continuous scale (per 5 cm higher waist circumference RR, 0.89; 95% CI, 0.80-0.99). In addition, an inverse association was observed for hip circumference (highest versus lowest quintile; RR, 0.33; 95% CI, 0.16-0.66; P_trend = 0.0002), whereas WHR was not related to ESCC risk.

In analyses that also adjusted for BMI (Table 3), the relation of WHR and waist circumference to risk of EAC became borderline nonsignificant (P_trend = 0.10 and 0.05, respectively). Similarly, BMI was no longer significantly associated with EAC risk when WHR or waist circumference was included in the model (data not shown).

For ESCC, we now observed a positive association with waist circumference and WHR after further adjustment for BMI (highest versus lowest quintile; RR, 6.91; 95% CI, 2.54-18.8 and RR, 3.12; 95% CI, 1.48-6.54), whereas the inverse association with BMI was even strengthened (data not shown).

We further evaluated whether smoking modified the relation of BMI, waist and hip circumference, and WHR with risk of EAC and ESCC (Table 4A and B). BMI and waist circumference seemed to be more strongly associated with risk of EAC in smokers than in nonsmokers, but these stratified analyses were hampered by the small number of cancer cases observed in our cohort, resulting in wide CIs (model 2). Tests for interaction of anthropometric measures with smoking for EAC risk were not significant.

Higher BMI as well as larger waist and hip circumference were significantly associated with lower ESCC risk among smokers (highest versus lowest quintile; RR, 0.18; 95% CI, 0.08-0.40; P_trend < 0.0001; RR, 0.41; 95% CI, 0.19-0.91; P_trend = 0.01; and RR, 0.27; 95% CI, 0.12-0.62; P_trend < 0.0001), whereas they were not associated with ESCC risk among nonsmokers (P for interaction = 0.004, 0.02, and 0.03, respectively). In sensitivity analyses, we tested the robustness of these findings by dividing the cohort into lifelong nonsmokers versus current smokers, excluding former smokers from the analysis. Despite decreased statistical power, a similar picture emerged, for example, BMI was inversely related to ESCC among current smokers (per 1 kg/m²: RR, 0.82; 95% CI, 0.76-0.88), whereas it was not associated with ESCC risk among lifelong nonsmokers (per 1 kg/m²: RR, 0.93; 95% CI, 0.82-1.06; data not shown). However, it is important to note that absolute risks of ESCC were considerably higher in smokers than in nonsmokers (Fig. 1). After additional adjustment for BMI, waist circumference and WHR were positively associated with ESCC risk in smokers and in nonsmokers (Table 4B).

We did not detect significant differences in the relation of anthropometric measures to risk of EAC when we stratified our analysis by alcohol consumption (data not shown). For ESCC, BMI was significantly inversely associated with ESCC risk at any alcohol consumption level (P = 0.51 for interaction), whereas higher WHR was associated with a lower ESCC risk among no or low alcohol consumers, whereas a higher risk was observed among participants with moderate or high alcohol consumption (P = 0.04 for interaction). Differences in the association of waist and hip circumference with ESCC across strata of alcohol consumption were not observed.

The main analysis was repeated after exclusion of all cases occurring during the first 2 years of follow-up with 70 EAC and 82 ESCC cases remaining for analyses of BMI and 64 EAC and 79 ESCC cases remaining for analyses of circumference measures, but results remained similar or were even strengthened. For example, the RRs (95% CI) for the highest quintiles of BMI or waist circumference were 3.36 (1.40-8.07) or 3.24 (1.32-7.96), respectively, for EAC and 0.24 (0.11-0.52) or 0.66 (0.30-1.44), respectively, for ESCC.

In this prospective European cohort study, we found that higher BMI, as a measure of general obesity, was related to higher risk of EAC and also that higher waist circumference and WHR, as indicators of abdominal obesity, were associated with a higher risk of this cancer type. In contrast, we observed an inverse association for BMI and waist and hip circumference with risk of ESCC. However, after stratification, these inverse associations were present only among smokers, whereas no significant associations were observed for nonsmokers. Interestingly, when controlled for BMI, we found positive associations for waist circumference and WHR with risk of ESCC, and these associations were observed among smokers and among nonsmokers, supporting the hypothesis that abdominal fat accumulation may in fact increase the risk of ESCC.

Our study has several strengths and also some limitations. Among the strengths are its prospective design and the cohort size that includes study populations from several European countries with large variation in anthropometry and different cancer incidence rates. Misclassification of exposure has been minimized as in all study centers (except for part of the Oxford cohort) anthropometric measurements at baseline were made by trained personnel, in contrast to self-reported data used in several previous studies (26-28). Further, unlike many previous studies, we had comprehensive information on smoking behavior and use of alcohol, the main risk factors for esophageal (squamous cell) carcinoma.

The major limitation of our study is the relatively small number of incident cases, leading to imprecise estimates and limiting our ability to evaluate gender differences with regard to the observed associations. Because fat distribution differed considerably between men and women, sex-specific quintiles of exposure were used and risk estimates represent an average over men and women. The fact that median values in exposure differ for men and women within quintiles has to be taken into consideration when interpreting the RR estimates. In addition, there were slight differences in the method of assessment of circumference measures between centers in EPIC, which might have limited the precision of our study. Nevertheless, a recent systematic review indicated that the measurement protocol for waist circumference has no substantial influence on the association of waist circumference with health outcomes (29). The follow-up time in our study was relatively short, and thus, the observed associations might in part be explainable by reverse causation with a preexisting cancer affecting body weight. However, associations were similar after excluding cases diagnosed within the first 2 years of follow-up. Another potential source of bias is misclassification of the outcome with regard to EAC, as the distinction between adenocarcinoma of the esophagus and the gastric cardia is difficult. In our study, however, diagnosis and classification of most adenocarcinoma cases were validated by a European panel of pathologists ensuring a high specificity of the outcome. Finally, we had no information on past infection with Helicobacter pylori, which has been found to be inversely related to risk of EAC (30) and has been shown to suppress appetite and body weight (31).

Our observation, that waist circumference and WHR were directly associated with risk of EAC, extends previous studies that consistently reported that higher BMI is related to higher EAC risk with a dose-dependent relationship found in a meta-analysis of case-control studies (4). In addition, in a recently published meta-analysis of prospective observational studies, a 5-kg/m² higher BMI was strongly associated with a higher EAC risk (RR, 1.52; 95% CI, 1.33-1.74 and RR, 1.51; 95% CI, 1.31-1.74, in men and women, respectively; ref. 3). These results are further corroborated by the findings of three more recent prospective studies (13, 26, 27) and the results of our study. To our knowledge, only two prospective studies (12, 13) have yet investigated the association between abdominal obesity and EAC risk. In an Australian cohort study, a significant positive association was observed with waist circumference (12), which is concordant with the result of our study. Similarly, in a nested case-control study, larger abdominal diameter was associated with a higher risk of EAC with this association being statistically independent of BMI (13). Our findings, however, suggest that abdominal body fat distribution does not predict EAC risk beyond obesity in general. In fact, after mutual adjustment for BMI and waist circumference/WHR, both anthropometric measures were no longer significantly related to EAC risk, implying that they do convey the same information and that general adiposity seems to be more important for the development of EAC than (abdominal) body fat distribution.

The positive relation between obesity and EAC risk has been thus far mainly attributed to the higher frequency of gastroesophageal reflux in obese individuals, possibly due to higher intra-abdominal pressure (32-35). Gastroesophageal reflux in turn is thought to predispose to Barrett's esophagus, a premalignant lesion for EAC (36). However, some studies have also shown that BMI per se is a risk factor for EAC, independent of gastroesophageal reflux (15, 16, 37), or have suggested that there is even a multiplicative effect between the two (37, 38). In our study, we had no information on history of reflux disease and thus were not able to evaluate this proposed mechanism. A recently suggested alternative hypothesis proposes that the endocrine function of the adipose tissue may also play a role in the development of EAC (13, 37). Visceral body fat is highly metabolically active, and therefore, it has been suggested that metabolic products, such as insulin-like growth factors or adipokines, might play a role or act as markers of risk.

Similar to previous case-control and cohort studies (3, 6, 13-15, 26, 39-41), we observed an inverse association between BMI and risk of ESCC. Further, we observed a significant inverse association with waist and hip circumference, whereas WHR was not related to ESCC risk. After additional adjustment for BMI, we found a strong positive relationship of waist circumference and WHR with ESCC. This implies that general obesity may be associated with lower risk of ESCC on the one hand, but it also suggests that abdominal fat distribution seems to be a risk factor for ESCC on the other hand. Certainly, the subsequent inclusion of strongly correlated anthropometric measures in our risk model might have hampered us to obtain precise and stable RRs. Thus, the RR estimates of waist circumference and WHR with ESCC should be interpreted cautiously. To compensate for the problem of collinearity, we also modeled the residuals from the regression of waist circumference or WHR on BMI as the main exposure variable and additionally included BMI as a covariate in the model; results remained virtually the same. However, despite consistency of these results, replication of our findings in future studies is needed to draw definite conclusions.

Importantly, we observed a strong interaction of BMI and waist and hip circumference with smoking for ESCC risk, suggesting that the lower risk associated with higher BMI and larger circumferences might be only present among smokers who are at higher risk due to their heavy smoking. Similarly, an inverse association between BMI and lung cancer risk has recently been described to depend on smoking status (3). To our knowledge, only three studies have evaluated effect modification by smoking for the association between BMI and ESCC and yielded contradictory results. Whereas in two prospective studies the inverse association remained significant when analyses were restricted to never smokers (6, 28), in a case-control study higher BMI was associated with lower risk of ESCC among smokers but not among nonsmokers (14), which agrees with our findings. However, that study did not determine whether this effect modification was statistically significant. In our study, the evaluation of effect modification by smoking was limited by the small number of cases among nonsmokers and has to be interpreted cautiously, although the interaction terms were statistically significant and the results did not change when we analyzed BMI and waist circumference as continuous variables to increase statistical power. Currently, we can only speculate why higher overall body mass is associated with reduced risk of ESCC in smokers. A metabolic interplay of smoking and body mass may be involved. For example, it might be possible that higher estrogen levels due to higher BMI (42-44) are counterbalancing the antiestrogenic effect of smoking (45). Although it has not been shown thus far that estrogen levels relate to ESCC, higher BMI may therefore have a protective effect among smokers. Estrogen has been recently observed to be inversely associated with the prevalence of oral leukoplakia (46) and the incidence of gastric cancer in men who had been diagnosed with prostate cancer and exposed to therapy with high doses of estrogen (47).

Importantly, in both smokers and nonsmokers, we observed a positive relation between waist circumference and WHR and ESCC after adjustment for BMI, suggesting that the positive association observed in the full cohort is not due to residual confounding by smoking but rather reflects a true association of intra-abdominal fat with ESCC risk.

In conclusion, in this European study, measures of general and abdominal obesity were strongly related to a higher risk of EAC. Further, our study is the first one suggesting that particularly an abdominal body fat distribution might be a risk factor for ESCC; however, these findings need to be confirmed in future studies. In addition to smoking prevention, current recommendations for the prevention of esophageal cancer should also focus on maintaining a healthy body weight and a healthy waist size.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Bertrand Hemon, Ellen Kohlsdorf, and Wolfgang Bernigau for data coding as well as all participants in EPIC for their invaluable contribution to the study.