Abdominal obesity, weight gain during adulthood and risk of liver and biliary tract cancer in a European cohort



General obesity has been positively associated with risk of liver and probably with biliary tract cancer, but little is known about abdominal obesity or weight gain during adulthood. We used multivariable Cox proportional hazard models to investigate associations between weight, body mass index, waist and hip circumference, waist-to-hip and waist-to-height ratio (WHtR), weight change during adulthood and risk of hepatocellular carcinoma (HCC), intrahepatic (IBDC) and extrahepatic bile duct system cancer [EBDSC including gallbladder cancer (GBC)] among 359,525 men and women in the European Prospective Investigation into Cancer and Nutrition study. Hepatitis B and C virus status was measured in a nested case–control subset. During a mean follow-up of 8.6 years, 177 cases of HCC, 58 cases of IBDC and 210 cases of EBDSC, including 76 cases of GBC, occurred. All anthropometric measures were positively associated with risk of HCC and GBC. WHtR showed the strongest association with HCC [relative risk (RR) comparing extreme tertiles 3.51, 95% confidence interval (95% CI): 2.09–5.87; ptrend < 0.0001] and with GBC (RR: 1.56, 95% CI: 1.12–2.16 for an increment of one unit in WHtR). Weight gain during adulthood was also positively associated with HCC when comparing extreme tertiles (RR: 2.48, 95% CI: 1.49–4.13; <0.001). No statistically significant association was observed between obesity and risk of IBDC and EBDSC. Our results provide evidence of an association between obesity, particularly abdominal obesity, and risk of HCC and GBC. Our findings support public health recommendations to reduce the prevalence of obesity and weight gain in adulthood for HCC and GBC prevention in Western populations.

Primary liver cancer is the sixth most common cancer in the world and the third most common cause of cancer-related death.1 Most liver cancer cases occur in developing countries, but the incidence is rising in developed countries as well.1 Hepatocellular carcinoma (HCC) accounts for 85–95% of primary liver cancer worldwide,1 followed by biliary tract cancer, which includes cancer of the bile ducts and gallbladder.2 Tumors of the bile ducts can be classified as extrahepatic bile duct cancer (EBDC) and intrahepatic bile duct cancer (IBDC).

Relative frequencies of primary liver cancer differ among countries. In areas with high HCC incidence, biliary tract cancer accounts for a small fraction, and in areas with low HCC incidence the number of biliary tract cancer cases exceeds that of HCC.3 In Europe, the incidence of HCC is higher in men and biliary tract cancer is more common in women.4

Established risk factors for HCC are chronic liver diseases and cirrhosis.1 Well-known risk factors for biliary tract cancer are primary sclerosing cholangitis, infection by liver flukes, cholelithiasis and cirrhosis.5 The major causes of cirrhosis include viral [e.g., hepatitis C (HCV) and B (HBV) virus infection], toxic (e.g., aflatoxin and alcoholic liver disease) and immune-related factors (e.g., primary biliary cirrhosis).6

The relative absence of the previously identified risk factors in Western countries points to other factors that may be relevant for the development of liver and biliary tract cancer.1 Previous studies suggested that obesity is a risk factor for HCC.7, 8 This observation is strengthened by the fact that higher body fatness is associated with nonalcoholic fatty liver disease (NAFLD), which has been hypothesized to be the link between obesity and risk of HCC.9 However, studies thus far have concentrated on general obesity characterized by a high body mass index (BMI) without much insight into factors pertaining to abdominal obesity, such as large waist circumference and high waist-to-hip (WHR) or waist-to-height ratio (WHtR),10 which may also be potential HCC risk factors. Indeed, it has been shown that abdominal, particularly visceral fat has a high metabolic activity11 and was associated with NAFLD, even after controlling for BMI.12 In addition, weight gain during adulthood could possibly be positively associated to NAFLD, even in nonobese individuals.13 Studies thus far concentrated on obesity usually among middle-aged or elderly individuals, and little is known about weight gain during adulthood and risk of HCC.

Furthermore, studies that investigated the association between general obesity and biliary tract cancer have provided inconsistent results.14–19 Only one case–control study investigated associations between abdominal obesity, weight change and risk of EBDC and did not find an association.18 A meta-analysis of eight cohort and three case–control studies observed a positive association between risk of gallbladder cancer (GBC) and general obesity,20 which might be mediated by gallstone diseases. Abdominal adiposity has also been shown to be related to gallstones diseases,21 but little is known about abdominal fat distribution as a risk factor for GBC with only one case–control study reporting a positive association between GBC and WHR.18

We investigated the associations between general and abdominal obesity, weight change during adulthood and the risk of HCC, IBDC and EBDC, including GBC, in the European Prospective Investigation into Cancer and Nutrition (EPIC) study.

Material and Methods

Study design

The EPIC study is a prospective multicenter cohort study designed primarily to investigate associations between nutrition and cancer. The study design and methods have been previously described.22, 23 Briefly, between 1992 and 2000, more than 520,000 participants were recruited in 23 centers in ten European countries: Denmark, France, Germany, Greece, Italy, The Netherlands, Norway, Spain, Sweden and United Kingdom. The eligible male and female participants were mostly aged 25–70 at recruitment and were selected from the general population residing in a given geographic area, i.e., a town or a province with exception of France (members of a health insurance for school employees), subsample of Oxford cohort (vegetarians and healthy eaters), subsamples of the Italian and Spanish cohort (blood donors and their partners) and the cohorts in Utrecht and Florence (women attending breast cancer screening). Eligible subjects were invited to participate, and those who accepted and gave informed consent were asked to complete questionnaires about their diet, lifestyle and medical history. Participants were then invited to a center to provide a blood sample, and anthropometric measurements were taken by trained staff. Study approval was obtained from the ethical review boards of the International Agency for Research on Cancer and from the boards pertaining to the local institutions where participants had been recruited.

Study population

Exclusion criteria were any prevalent cancer at baseline (n = 23,818), missing information on date of diagnosis or length of follow-up (n = 4,385), cases with metastasis or with ineligible information on histology (n = 78), missing information on measured height or weight, which excluded the cohorts of Norway (n = 35,889), 48,612 participants from the French cohort and 9,301 participants of other cohorts and missing information on measured waist or hip circumferences (n = 31,375), dietary questionnaire information (n = 8,347) that included missing information on alcohol consumption at baseline. The present analysis was based on 359,525 eligible participants.

Assessment of anthropometric data, lifestyle factors and diet

In all centers, height, weight, waist and hip circumference were measured using similar study protocols.24 To reduce heterogeneity in different assessment methods between study centers, weight, waist and hip circumferences were corrected.25 BMI, a measure of general obesity, was calculated by weight (kg) divided by the square of height (m2). WHR, a measure of abdominal obesity, was calculated by waist circumference divided by hip circumference. In addition, WHtR was chosen as an indicator of abdominal obesity because it has been shown to be a promising marker of fat distribution among individuals with different statures.26 WHtR was defined as waist circumference divided by height. In the baseline questionnaire, study participants were inquired about their weight at age 20 (or age 25 in EPIC-Potsdam). Weight change per year was computed as the difference between measured weight at recruitment and recalled weight at age 20 divided by time between age at recalled weight and baseline. Data on recalled weight were available for cohorts in Italy (Varese), United Kingdom, Greece, Sweden (Malmö), Denmark and Germany (Potsdam) (n = 192,078). Dietary habits of the previous 12 months were assessed at baseline using country-specific, validated dietary questionnaires.22, 27 These included questions on alcohol consumption (g/day). Information on past alcohol consumption (at ages 20, 30, 40 and 50) was available for 305,904 participants and was missing for cohorts in Italy (Naples), Sweden (Malmö) and The Netherlands (Bilthoven). Information on health- and lifestyle-related characteristics was obtained by using standardized questionnaires. These included questions on level of education, life history of smoking, physical activity, self-reported diabetes and gallstones.23

Assessment of endpoints

Incident cases of liver and biliary tract cancer were determined through record linkage with regional cancer registries (Denmark, Italy, The Netherlands, Spain, Sweden and United Kingdom; complete up to June 2006) or by a combination of methods, including health insurance records, contacts to cancer and pathology registries together with an active follow-up through study participants and their relatives (France, Germany and Greece; complete up to June 2010). Mortality data were also requested from cancer or mortality registries at the regional and local level. Time of follow-up began with recruitment and ended with diagnosis of liver or biliary tract cancer, respectively, as first primary cancer, death, emigration or end of follow-up. Mortality data were coded following the rules of the 10th revision of the International Statistical Classification of Disease, Injuries and Cause of Death (ICD-10), whereby cancer incidence data following the 2nd revision of the International Classification of Disease for Oncology (ICD-O-2).

HCC was defined as tumor in the liver with ICD-10 code C22.0 and IBDC as tumor in the intrahepatic bile ducts (C22.1). Extrahepatic bile duct system cancers (EBDSCs) were defined as tumors in the gallbladder (C23.9), extrahepatic bile ducts (C24.0), ampulla of Vater (C24.1) and biliary tract (C24.8 and C24.9).

Nested case–control subset

A nested case–control subset of the study was conducted to investigate whether associations between anthropometric measures and risk of HCC were independent of chronic HBV/HCV infection. The design and methods have been previously described.28 Briefly, 125 incident HCC cases were identified between recruitment and 2006. For each case, two controls were selected by incidence density sampling from the cohort. Participants were eligible as control if they were alive at the time of diagnosis of the corresponding case and free of cancer. Controls and cases were matched on study center, sex, age, date, fasting status and time of the day at blood collection. Women were additionally matched according to menopausal status and exogenous hormone use. After exclusion of participants with missing information on measured anthropometry, unknown HBV/HCV infection status and smoking status, 115 HCC cases and 229 matched controls were available for this analysis. Of these, one caseset relied only on one matched control. HBV and HCV infection has been determined by using the ARCHITECT HBsAg and the anti-HCV chemiluminescent microparticle immunoassays from Abbott Diagnostics Division, Lyon, France. All laboratory analyses were performed at the Centre de Biologie République laboratory, Lyon, France.

Statistical analyses

Based on the cohort study, associations between anthropometric factors and risk of HCC, IBDC, EBDSC and GBC were analyzed by calculating relative risks (RRs) as hazard rate ratios and their 95% confidence intervals (95% CIs) in Cox proportional hazard models. Age was taken as the underlying time variable, with entry time defined as age at recruitment and exit time defined as the participant's age at cancer diagnosis or censoring, respectively. All models were stratified by age at recruitment in 1-year intervals, sex and study center to control for differences in data collection across centers and to be more robust against possible violations of the proportional hazard assumption. Incident HCC cases were excluded in analyses of IBDC and EBDSC and vice versa.

Weight, BMI, waist and hip circumference, WHR, WHtR, weight at age 20 and weight change per year were analyzed using sex-specific tertiles with the lowest tertile as reference category. To test for linear trend, the median of tertiles was used as a continuous variable in the respective regression model. In addition, we ran analyses including the anthropometric measures as continuous variables into the models to estimate the RR of liver and biliary tract cancer per increase in body size measures. For GBC and IBDC, we will only report the results on continuous anthropometric measures because of the low number of cases.

Restricted cubic spline regression models were used to investigate whether the association between anthropometric measures and risk of cancer was linear. Models were fitted with three knots (5th, 50th and 95th percentile), four knots (5th, 25th, 75th and 95th percentile) and five knots (5th, 25th, 50th, 75th and 95th percentile) of BMI, waist circumference, WHR and WHtR,29 and the Akaike information criterion (AIC) was used to select the best model. Nonlinearity was evaluated with a Wald chi-square test.

Predefined categories of BMI (normal weight <25, overweight 25–<30 or obesity ≥30 kg/m2), waist circumference (for men <102 or ≥102 cm and for women <88 or ≥88 cm, respectively) and WHR (for men <0.95 or ≥95 cm and for women <0.80 or ≥0.80, respectively) based on the WHO criteria were analyzed additionally.30

In multivariable analysis, we adjusted for the following potential confounders: education (none/unknown, primary school, technical/professional/secondary school and university degree), smoking status (never, former <10 and ≥10 years, current <15, 15–24 and ≥25 cigarettes/day, other than cigarettes and unknown), alcohol consumption (g/day) and drinking status at recruitment (drinker versus nondrinker). Models for weight, waist and hip circumference, WHR, weight at age 20 and weight change were further adjusted for height, and the model analyzing weight change was additionally adjusted for weight at age 20. Furthermore, we investigated other potential confounders such as physical activity, smoking duration, total energy intake and consumption of fruits, vegetables, red and processed meat, poultry, coffee and tea at baseline by removing each variable separately from the multivariable adjusted model. Because RRs did not change appreciably (by <10%), we excluded these variables from the final model.

In additional models, analyses for BMI were further adjusted for waist circumference, whereas analyses for waist, hip circumference and WHR were adjusted for BMI and analyses for WHtR were adjusted for weight to investigate if general or abdominal obesity determines the risk.

We conducted sensitivity analyses to examine the consistency of our findings. First, the main multivariable-adjusted analyses were repeated with exclusion of subjects with prevalent diabetes or unknown status because higher rates of diabetes have previously been associated with risk of HCC31 and may be an intermediate factor. In addition, we restricted the analyses on GBC to individuals not reporting a history of gallstones at recruitment. Second, we conducted sensitivity analyses focusing on lifetime alcohol intake because the information was not available for the whole cohort and lifetime alcohol has previously been associated with increased risk of liver cancer.32 Therefore, we restricted our cohort to individuals with information available and excluded heavy alcohol drinkers in the past (for men >60 and women >30 g/day at ages 20, 30, 40 or 50, respectively). Third, we performed analyses stratified for men and women to evaluate whether the association depended on sex. However, because of the low number of HCC cases among women (54 cases), only results in men were meaningful. Fourth, we restricted the analyses on HCC to histologically confirmed HCC cases (C22.0 “M-8170/3” or “M-8173/3”) to reduce potential misclassification of cases. Finally, we excluded the initial 2 years of follow-up from the analyses to evaluate whether potential preexisting disease influenced the association between anthropometric variables and risk of liver and biliary tract cancer.

We investigated whether the association between anthropometric measures and liver and biliary tract cancer risk differed across categories of smoking or alcohol intake at baseline by adding interaction terms to the models, because synergistic effects between obesity, alcohol intake and smoking have been reported previously.33 Smoking status was categorized as never, former and current smokers, and baseline alcohol intake was used as a continuous measure. In addition, we examined interaction terms of anthropometric measure with sex and diabetes status for liver and biliary tract cancer. Statistical interaction was evaluated with the likelihood-ratio test comparing regression models with and without interaction terms.

Multivariable-adjusted conditional logistic regression models were used to investigate the association between anthropometric measurements and HCC risk in the nested case–control study among all subjects with adjustment for HBV/HCV and among HBV/HCV negative individuals. Odds ratios derived by conditional logistic regression can be interpreted as RRs because controls were selected by incidence density sampling.34 All p-values presented were two tailed, and p < 0.05 was considered statistically significant. All analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC).


During a mean follow-up time of 8.6 years, 177 cases of HCC (123 in men and 54 in women), 58 cases of IBDC (31 in men and 27 in women) and 210 cases of EBDSC (84 in men and 126 in women) were diagnosed. HCC cases included 151 histologically confirmed cases and EBDSC included 76 GBC cases.

Table 1 presents the baseline characteristics for men and women according to tertiles of BMI. Older age, diabetes and gallstones were positively related to BMI, whereas a higher educational level and current smoking were inversely associated to BMI, among men and women. Men with higher BMI levels were less likely to be never smokers, but tended to more frequently have a history of smoking and higher alcohol consumption. Conversely, women with higher BMI more often were never smokers, less often had a history of smoking and consumed less alcohol compared to women with lower BMI values.

Table 1. Characteristics across tertiles of BMI among men and women in EPIC (n = 359,525)
inline image

Weight, BMI, waist and hip circumference, WHR, WHtR and weight change per year were all associated with higher risk of HCC (Table 2). The multivariable-adjusted RRs were generally about twofold or higher in subjects within the highest tertile of each anthropometric measure compared to subjects within the lowest. The highest RR was observed for WHtR [multivariable-adjusted RR (95% CI): 3.51 (2.09–5.87); ptrend < 0.0001]. Participants within the highest tertile of weight gain had an about 2.5-fold higher risk of HCC compared to individuals in the reference group (Table 2). For weight at age 20, no significant association could be observed [multivariable RR (95% CI) comparing extreme tertiles = 0.99 (0.59–1.68); ptrend = 0.95].

Table 2. Relative risks (RRs) and 95% confidence intervals (95% CIs) of HCC according to anthropometric measures in EPIC
inline image

Furthermore, we grouped subjects into predefined categories of BMI, waist circumference and WHR, respectively. The multivariable RRs (95% CI) for HCC were 1.28 (0.88–1.86) for overweight and 2.19 (1.44–3.34) for obese subjects compared to normal weight subjects. For categories of waist circumference, the RR (95% CI) was 2.03 (1.43–2.88) and for WHR 1.96 (1.41–2.73).

The results of restricted cubic spline regression using three knots did not indicate evidence for a nonlinear shape of anthropometric measures and risk of HCC (p = 0.43 for BMI, p = 0.08 for waist circumference, p = 0.59 for WHR and p = 0.80 for WHtR). Goodness of fit was very slightly improved for models of waist circumference and WHR using four knots, but results did not change substantially (data not shown).

In analyses that mutually adjusted for other anthropometric measures, BMI and hip circumference were no longer associated with risk of HCC, whereas a positive association between WHtR and HCC remained statistically significant [multivariable-adjusted RR (95% CI) for upper tertile 2.96 (1.65–5.32); ptrend < 0.001]. After adjustment for BMI, the RRs (95% CI) for waist circumference and WHR were still increased but were more imprecise comparing extreme tertiles. However, the RR (95% CI) for an increment of 5 cm in waist circumference was 1.29 (1.13–1.47) and for an increment of 0.1 unit in WHR 1.45 (1.16–1.83) (Table 2).

When we restricted the analyses on HCC to subjects without diabetes, results did not change appreciably. Moreover, after exclusion of subjects with heavy alcohol consumption in the past, associations remained comparable (Table 3).

Table 3. Relative risks (RRs) and 95% confidence intervals (95% CIs) of hepatocellular carcinoma across tertiles of anthropometric measures, for nondiabetics and non-heavy alcohol consumers in the past
inline image

We also examined the associations of anthropometric measures and risk of HCC only in men (123 cases). The multivariable-adjusted RR (95% CIs) for the highest versus the lowest tertile of WHtR was 4.04 (2.22–7.33); ptrend ≤ 0.0001. The risk associated with weight change in men (81 cases) was also slightly higher compared to our main analyses [RR and (95% CI) 2.88 (1.60–5.20); ptrend = 0.0002].

In analyses that only included cases with histologically confirmed HCC (151 cases), the results remained similar when compared to our main analyses. RRs (95% CIs) for the highest versus the lowest tertiles of WHtR and weight gain were 3.29 (1.90–5.69); ptrend ≤ 0.0001 and 2.29 (1.35–3.90); ptrend = 0.0014, respectively.

After exclusion of participants with HBV/HCV infection in the nested case–control study, the results were more imprecise, but similar RRs were observed: When we compared extreme tertiles in 77 HCC cases and 148 controls, the RR for BMI was 2.27 (0.89–5.75); ptrend = 0.09, for WHR 5.24 (1.79–15.34); ptrend = 0.03 and for WHtR 3.45 (1.24–9.62); ptrend = 0.08. When we considered all nested case–control subjects and adjusted for HBV/HCV, the RR for BMI was 1.44 (0.64–3.23); ptrend = 0.42, for WHR 2.63 (1.10–6.29); ptrend = 0.25 and for WhtR 2.08 (0.88–4.92); ptrend = 0.46, comparing extreme tertiles. Tests for interactions were not statistically significant between anthropometric measures and HBV or HCV infection (p-values for interaction > 0.05).

None of the anthropometric measures were statistically significantly associated with risk of EBDSC (Table 4). Measures of general obesity tended to be associated with EBDSC risk; however, the RRs (95% CI) were not significant. Regarding IBDC, continuous measures of anthropometry, particularly of abdominal obesity, indicated an increased risk, but RRs were not significant (Table 5). When the analyses on EBDSC were restricted to GBC only, a positive association for general and abdominal obesity was observed (Table 5). The risk of GBC was no longer associated with BMI after adjustment for waist circumference [multivariable-adjusted RR (95% CI) for an increment of 5 kg/m2 in BMI 0.70 (0.43–1.15)]. The RR (95% CI) for an increment of 5 cm in waist circumference was 1.33 (1.10–1.62) after additional adjustment for BMI, whereas the RRs (95% CI) for other measures of abdominal obesity were more imprecise; e.g., the RR (95% CI) for a 0.1 unit higher WHtR was 1.33 (0.82–2.17) after adjustment for weight.

Table 4. Relative risks (RRs) and 95% confidence intervals (95% CIs) of EBDSC according to anthropometric measures in EPIC
inline image
Table 5. Relative risks (RRs) and 95% confidence intervals (95% CIs) of IBDC and GBC according to anthropometric measures in EPIC
inline image

When we excluded individuals with gallstone diseases or unknown status in analyses on GBC (35 cases remaining), the RR (95% CI) was 1.50 (1.01–2.22) for an increment of 5 kg/m2 in BMI, 1.24 (1.06–1.46) for a 5 cm higher waist circumference, 1.89 (1.16–3.09) for a 0.1 unit higher WHtR and 1.15 (0.98–1.36) for a weight change by 1 kg/year.

Analyses were repeated after excluding cases having occurred during the first 2 years of follow-up. For these analyses, 153 HCC cases, 53 IBD cases and 183 EBDSC cases, including 70 GBC cases, were available, and the results did not change substantially (data not shown).

Tests for interactions between anthropometric measures and smoking or alcohol consumption were not statistically significant (p-values for interaction > 0.05). Interaction analyses pointed only to some differences by sex and body weight or hip circumferences for HCC (p-value for interaction = 0.03 and 0.01, respectively) and BMI for EBDSC (p-value for interaction = 0.03). In addition, only for HCC borderline significant interaction terms could be observed between body weight or BMI with diabetes (p-value for interaction = 0.05).


In this prospective cohort study, general and abdominal obesity were associated with risk of HCC and GBC. In a nested case–control subset of our study, associations remained positive for anthropometric measures and risk of HCC in participants free of HBV/HCV infections. Across all anthropometric indices, WHtR, an indicator for abdominal obesity, showed the strongest association independent of general body weight, whereas BMI as an indicator for general obesity was not independently associated with HCC and GBC. Adult weight change was related to higher risk of HCC, independently of body weight in early adulthood. None of the anthropometric measurements were statistically significantly associated with risk of IBDC and EBDSC.

A number of studies reported positive associations between a high BMI and risk of HCC.7, 8 A systematic review showed that out of ten cohort studies, seven reported a statistically significant association between overweight, obesity and risk of HCC. The effect size ranged from 1.44 (95% CI: 1.28–1.61) to 4.13 (95% CI: 1.38–12.40) when comparing obese to normal weight subjects.7 These findings were further confirmed in a meta-analysis, in which summary RRs were 2.42 (95% CI: 1.83–3.20) and 1.67 (95% CI: 1.37–2.03) when comparing obese to normal weight men and women, respectively,8 which is comparable to our result for predefined BMI categories. In our study, all anthropometric factors were associated with risk of HCC independently of diabetes status, indicating that diabetes is not the only mechanism that linked the association between obesity and HCC.

The majority of studies investigating obesity and risk of HCC focused on general obesity and did not examine if abdominal obesity as indicated by a large waist circumference and high WHR or WHtR10 is an independent risk factor. It has been shown that abdominal fat distribution was associated with risk of certain cancers, independently of general obesity.35–37 One study from Taiwan reported a synergistic effect between waist circumference and hepatitis C infection with regard to risk of HCC, but no association was found in individuals without hepatitis infection.38 The latter was based on few cases (n = 53), and cutoff values for waist circumferences relied on criteria used in the Asian population. In our study, WHtR showed the strongest association with HCC risk, which was independent of general body weight, indicating that it is particularly abdominal obesity that may promote risk of HCC. The results on waist circumference and WHR were attenuated after adjustment for BMI.

Weight change during adulthood has been shown to be associated with risk of certain cancers beyond BMI achieved during childhood and adolescence.39–41 To the best of our knowledge, little is known whether body weight in childhood, early or later adulthood is relevant to liver carcinogenesis. Only one study investigated weight change as a risk factor for liver cancer in men.19 This study, based on very small numbers, reported a nonsignificant RR of 2.26 (95% CI: 0.53–9.59), ptrend > 0.5 for individuals with a weight gain in the highest category compared to individuals within the lowest category. Another prospective study showed that even when subjects were not considered as overweight, slight to moderate weight gain was associated with increased serum liver enzyme levels,42 which may indicate the presence of NAFLD.43 In our analysis, weight at age 20 was not associated, whereas weight at baseline and weight change during adulthood were positively associated with HCC. However, such comparison needs to be interpreted with caution because assessment for weight at age 20 in our study relied on recalled self-reported weight, and weight at baseline was measured by trained staff.

The elevated risk of HCC in participants with obesity or weight gain may be linked via NAFLD.9 NAFLD is a chronic liver disease that ranges from fatty liver alone to steatohepatitis, steatonecrosis and nonalcoholic steatohepatitis,13 which is defined as presence of hepatocyte injury, inflammation and fibrosis. The latter can lead to cirrhosis and possibly to HCC.9 The prevalence of NAFLD is higher in subjects with obesity,13 but the underlying pathophysiology is not clear. It has been suggested that energy excess associated with obesity or weight gain44 is linked to hyperinsulinemia, insulin resistance and cytokine production, which leads to fat accumulation in hepatocytes, oxidative stress and progression of liver damage.45 Abdominal fat, particularly visceral fat accumulation, has a high metabolic activity11 and is strongly related to liver fat content.46 Previously, it has been shown that abdominal fat distribution was associated with NAFLD,47 even when adjusted for BMI.12

Obesity was not associated with IBDC and EBDSC in our study. Previous studies have shown inconsistent results.14–19, 48 Two cohort studies, one Western and one Asian, reported positive associations between BMI and biliary tract cancer.19, 48 Regarding EBDC separately, two Western case–control studies have shown an association with BMI only in men,14, 17 whereas two case–control studies, one Western and one Asian, did not find an association.15, 18 BMI was also positively associated with risk of IBDC in one case–control study,15 but not in another.16

When we analyzed GBC alone, we found an association with general and abdominal obesity. A meta-analysis showed that general obesity was associated with risk of GBC,20 but little is known about abdominal obesity. One Chinese case–control study observed a positive association between WHR and risk of GBC.18 The mechanism how obesity may affect the risk of GBC is not clear as yet. One possible link could be the presence of gallstones. Obesity is strongly associated with gallstones, which are known risk factors for GBC.5 Abdominal adiposity has been shown to be related to gallstones diseases as well even after controlling for BMI.21 When we excluded individuals with self-reported gallstone diseases, results did not attenuate, suggesting that there might be different pathways that linked the association between obesity and GBC. However, for 24.5% of the participants, data on self-reported gallstone diseases were missing and many cases may have been undetected.

Our study has limitations that need to be considered when interpreting the results. First, the small number of incident cases limited the calculation of estimates separately for women. But, all tertiles of anthropometric exposure measures were calculated sex-specific and analyses were stratified by sex. Interaction terms pointed to some but not consistent differences by sex; however, numbers were small. Additionally, analyses were performed only in men and results did not change substantially. Second, calculations of weight gain were based on self-reported recalled weight at age 20 (25 in Potsdam, Germany). Misclassification due to underreporting of recalled weight by overweight and obese participants or overreporting by underweight participants might have led to lower or higher values of weight change, which might have influenced the results. If misclassification was present, true associations could have been underestimated or overestimated. Nevertheless, studies have shown that recalled past body weight is reproducible49 and accurate.50 Third, a synergistic effect between obesity, smoking, alcohol consumption and liver diseases has been shown previously.33 In our study, no statistically significant interaction between smoking, alcohol consumption and indicators of general and abdominal obesity was observed. As the number of cases was small in our study, more research is needed to elucidate this issue. Fourth, the duration of follow-up was relatively short and body measures may have been affected by preexisting cancers. We therefore excluded the first 2 years of follow-up and results remained the same. Finally, potential misclassification of liver cancer cases may be possible because histologically confirmed and probable HCC cases were included in the analyses. Nevertheless, we performed analyses only with histologically confirmed HCC cases and the results did not change. Additionally, because the distal part of the extrahepatic bile duct runs through the head of the pancreas, some of the cancers classified as EBDSC may in fact be cancers of the pancreas and vice versa.

One of the main strengths of our study is its prospective design and inclusion of several European countries, which enabled us to evaluate associations between obesity and risk of liver and biliary tract cancer in diverse European regions with different rates of cancer incidence and different prevalence of obesity. Furthermore, in our study, anthropometric factors at baseline were measured by trained personnel and did not rely on self-reported body measures, resulting in more accurate estimates than self-reports and precluding some misclassification, with exception of weight change that includes self-reported body weight. Also, detailed information on important risk factors for liver cancer such as smoking, alcohol consumption, diabetes and HBV/HCV infection was available for our analyses. Finally, we had the opportunity to investigate subcategories of endpoints rather than liver and biliary tract cancer by themselves.

In conclusion, our findings support the current evidence of a positive association between obesity and risk of HCC and GBC in European populations, even after controlling for important risk factors. The findings suggest that abdominal obesity is related stronger than general obesity. Furthermore, our study indicated that weight gain during adulthood was positively associated with risk of HCC. These results support public health recommendations to increase efforts to reduce the prevalence of obesity and weight gain in adulthood in Western populations for liver and GBC prevention.


SSchl was funded by a grant from Deutsche Forschungsgemeinschaft (DFG NO446/7-1) and Deutsche Forschungsgemeinschaft Excellence Cluster Inflammation at Interfaces (EXC306). Reagents for the hepatitis infection determinations were kindly provided by Abbott Diagnostics Division, Lyon, France. The funding sources had no influence on the design of the study; the collection, analysis and interpretation of data; the writing of the report or the decision to submit the article for publication.