Diabetes mellitus, insulin treatment, diabetes duration, and risk of biliary tract cancer and hepatocellular carcinoma in a European cohort

Abstract

Background

Evidence on associations between self-reported diabetes mellitus, diabetes duration, age at diabetes diagnosis, insulin treatment, and risk of biliary tract cancer (BTC) and hepatocellular carcinoma (HCC), independent of general and abdominal obesity is scarce.

Patients and methods

We conducted a prospective analysis in the EPIC-cohort study among 363 426 participants with self-reported diabetes data. Multivariable adjusted relative risks and 95% confidence intervals were estimated from Cox regression models. In a nested case–control subset, analyses were carried out in HCV/HBV-negative individuals.

Results

During 8.5 years of follow-up, 204 BTC cases [including 75 gallbladder cancer (GBC) cases], and 176 HCC cases were identified. Independent of body mass index and waist-to-height ratio diabetes status was associated with higher risk of BTC and HCC [1.77 (1.00–3.13) and 2.17 (1.36–3.47)]. For BTC, the risk seemed to be higher in participants with shorter diabetes duration and those not treated with insulin. Regarding cancer subsites, diabetes was only associated with GBC [2.72 (1.17–6.31)]. The risk for HCC was particularly higher in participants treated with insulin. The results were not appreciably different in HCV/HBV-negative individuals.

Conclusion(s)

This study supports the hypothesis that diabetes is a risk factor for BTC (particularly GBC) and HCC. Further research is required to establish whether diabetes treatment or duration is associated with these cancers.

introduction

Epidemiological studies suggest that individuals with diabetes mellitus are at higher risk of cancer [1]. Studies on biliary tract cancer (BTC), including extrahepatic bile duct (EBD) and gallbladder cancer (GBC), have shown inconsistent results regarding diabetes status [2], whereas the evidence for an association between diabetes and hepatocellular carcinoma (HCC) has been more straightforward [3, 4].

The possible mechanisms that may link diabetes with risk of BTC and HCC include hyperinsulinemia, insulin resistance, hyperglycemia, chronic inflammation, and exogenous insulin treatment in diabetic patients [1]. Furthermore, diabetes is also associated with gallstone diseases [5, 6]; one of the major risk factors for GBC [7]. Diabetes is associated with nonalcoholic fatty liver disease, which might be another potentially link between diabetes and HCC [8].

However, it is not completely understood whether the association between diabetes and these cancers is mediated by obesity. Particularly, findings on BTC were inconsistent after controlling for obesity [2]. Furthermore, most of the studies accounted for general obesity, but not for abdominal obesity, which we have previously shown to be an independent risk factor for GBC and HCC [9]. Also, most previous efforts have focused on diabetes status only, and little is known about to what extent diabetes duration and treatment may influence the risk [1]. Indeed, previous studies provided inconsistent results on the association between diabetes duration and HCC [10–17]. For BTC, the few studies that have examined these factors did not find any associations [10, 12].

The main aim of this study was to investigate the associations between diabetes status, diabetes duration, age at diabetes diagnosis and insulin treatment with BTC (including GBC), and HCC. Particular emphasis was put on controlling for general and abdominal obesity.

methods

study design

The EPIC study is an ongoing prospective cohort study investigating associations between lifestyle factors and cancer incidence. Details on the EPIC study have been reported elsewhere [18]. In brief, between 1992 and 2000, ∼520 000 participants (25–70 years) were recruited in 10 European countries. Participants who gave informed consent were asked to complete questionnaires about their diet, lifestyle, and medical history. Blood samples were collected, and in most centers, anthropometric measurements were taken by trained staff. Ethical approval was obtained from the IARC Ethics Review Committee and EPIC centers.

The present analysis was based on 363 426 eligible participants, after exclusion of participants with any prevalent cancer at baseline (n = 23 818), incomplete follow-up data (n = 4385), metastasis or ineligible histology code (n = 78), unknown diabetes status at baseline (n = 19 961), missing information on diabetes treatment (n = 3466), and missing information on date of diagnosis of diabetes or an implausible date after baseline (n = 918). Furthermore, we excluded participants with missing information on weight, height, and waist or hip circumference, which excluded the cohort of Norway (n = 28 624), 47 488 participants from the French cohort and 24 389 participants from Sweden, and with missing information on alcohol consumption at baseline (n = 4777).

assessment of diabetes, anthropometric data, and lifestyle factors

Information on self-reported diabetes was inquired by using standardized questionnaires at baseline. Participants were asked if they had ever been diagnosed with diabetes, and information on age at diabetes diagnosis and use of insulin treatment was obtained. Diabetes duration at baseline was calculated by subtracting self-reported age at diabetes diagnosis from the age at baseline examination.

Weight, height, and waist circumference were measured at baseline as described elsewhere [19]. Body mass index (BMI; weight/height²) and waist-to-height ratio (WHtR; waist circumference/height), a measure of abdominal obesity [20], were calculated. Information on past alcohol consumption (at ages 20/30/40/50) was available for 311 706 participants. Information on lifestyle-related and socioeconomic characteristics (e.g. smoking status and level of education) was obtained by using standardized questionnaires [18].

assessment of end points

Assessment of vital status has been reported in detail elsewhere [9]. Briefly, incident BTC and HCC cases were determined through record linkage with regional cancer registries or by a combination of methods, including health insurance records, contacts to cancer, and pathology registries together with an active follow-up. Time of follow-up began with baseline and ended with diagnosis of first primary cancer, death, emigration, or end of follow-up. Cancer incidence data were coded following ICD-10. HCC was defined as tumors in the liver (C22.0); and BTC as gallbladder cancer (GBC) (C23), and non-GBC, including tumors of the ampulla of Vater (C24.1) and biliary tract (C24.0/C24.8/C24.9). Intrahepatic bile duct (IBD)cancer was defined as tumor in the IBDs (C22.1) and cholangiocarcinoma was defined as IBD and EBD with morphology code ‘8160/3’.

nested case–control subset

In a nested case–control subset, information on HBV/HCV infection status was available for 113 HCC cases and 221 matched controls. The design and methods have been described previously [21] and are shown in the supplementary Data, available at Annals of Oncology online.

statistical analyses

We calculated relative risks (RR) and 95%confidence intervals (95% CI) as hazard rate ratios using Cox proportional hazard models to investigate the associations between diabetes, diabetes duration (by median: <6/≥6 years; and continuous), age at diabetes diagnosis (by median: <50/≥50 years), insulin treatment (no/yes), and risk of BTC or HCC, where participants without diabetes were used as the reference category. For subsites of BTC, GBC and non-GBC, only analyses on diabetes status were conducted, because of small numbers. In all models, age was used as the underlying time variable, with entry time defined as age at recruitment and exit time as the participant's age at cancer diagnosis or end of follow-up. All models were stratified by age at recruitment in 1-year intervals, sex and study center. In multivariable analyses, we adjusted for the following potential confounders (model 1): education, smoking status, and alcohol consumption at baseline (compare Table 2 and 3 footnotes). Because of small numbers, we analyzed men and women combined.

Table 2.

Relative risks (RR) and 95% confidence intervals (95% CI) of biliary tract cancer (BTC), including gallbladder cancer (GBC) and non-gallbladder cancer (non-GBC) according to diabetes

	No. of person-years	No. of cases	Model 1a	Model 2b
			RR (95% CI)	RR (95% CI)
Total BTC		204
Diabetes status
No	3 103 576	190	1	1
Yes	70 488	14	1.80 (1.02–3.18)	1.77 (1.00–3.13)
Diabetes duration
No diabetes	3 103 576	190	1	1
<6 years	36 198	12	3.10 (1.68–5.71)	3.06 (1.66–5.67)
≥6 years	34 290	2	0.52 (0.13–2.10)	0.50 (0.12–2.06)
Age at diabetes
No diabetes	3 103 576	190	1	1
<50 years	32 552	5	1.76 (0.71–4.33)	1.75 (0.71–4.31)
≥50 year	37 935	9	1.83 (0.91–3.70)	1.79 (0.88–3.62)
Insulin treatment
No diabetes	3 103 576	190	1	1
No insulin	52 946	12	2.16 (1.17–4.00)	2.11 (1.14–3.91)
Insulin use	17 542	2	0.92 (0.22–3.77)	0.92 (0.23–3.79)
GBC		75
Diabetes status
No	3 102 808	68	1	1
Yes	70 455	7	3.11 (1.35–7.13)	2.72 (1.17–6.31)
Non-GBC		129
Diabetes status
No	3 103 101	122	1	1
Yes	70 449	7	1.26 (0.58–2.77)	1.28 (0.58–2.82)

^aModel 1 was calculated with the Cox proportional hazards regression using age at underlying time variable and stratified by sex, center and age at recruitment and were adjusted for education (none or unknown, primary school, technical or professional or secondary school, university degree), smoking status (never, former <10 and ≥10 years, current <15, 15–24, and ≥25 cigarettes/day, other than cigarettes, unknown), and baseline alcohol consumption [g/day; continuous and status of abstainer (yes, no)].

^bModel 2 were based on model 1 with additional adjustment for BMI (continuous) and waist-to-height ratio (continuous).

Table 3.

Relative risks (RR) and 95% confidence intervals (95% CI) of hepatocellular carcinoma (HCC) according to diabetes

	No. of person-years	No. of cases	Model 1a	Model 2b
			RR (95% CI)	RR (95% CI)
Total		176
Diabetes status
No	3 103 256	153	1	1
Yes	70 560	23	2.57 (1.61–4.10)	2.17 (1.36–3.47)
Diabetes duration
No diabetes	3 103 256	153	1	1
<6 years	36 206	11	2.80 (1.48–5.30)	2.25 (1.19–4.28)
≥6 years	34 354	12	2.38 (1.28–4.42)	2.09 (1.12–3.90)
Age at diabetes
No diabetes	3 103 256	153	1	1
<50 years	32 577	11	3.44 (1.82–6.52)	3.08 (1.62–5.83)
≥50 year	37 983	12	2.06 (1.11–3.83)	1.68 (0.90–3.13)
Insulin treatment
No diabetes	3 103 256	153	1	1
No insulin	52 930	8	1.18 (0.57–2.47)	1.02 (0.49–2.12)
Insulin use	17 630	15	6.19 (3.50–10.98)	5.25 (2.93–9.44)

^aModel 1 was calculated with the Cox proportional hazards regression using age at underlying time variable and stratified by sex, center and age at recruitment and were adjusted for education (none or unknown, primary school, technical or professional or secondary school, university degree), smoking status (never, former <10 and ≥10 years, current <15, 15–24, and ≥25 cigarettes/day, other than cigarettes, unknown), and baseline alcohol consumption [g/day; continuous and status of abstainer (yes, no)].

^bModel 2 was based on model 1 with additional adjustment for BMI (continuous) and waist-to-height ratio (continuous).

To investigate if diabetes was associated with BTC—GBC and non-GBC, separately—and HCC independently of general and abdominal obesity, we adjusted for BMI (continuous) and WHtR (continuous) in an additional model (model 2). To account for collinearity between anthropometric measures, we calculated residuals from the regression of WHtR on BMI to control for confounding on both factors. However, the results did not change and were not shown. Furthermore, to elucidate the interplay between diabetes and obesity, we investigated the associations between diabetes status and risk of BTC and HCC stratified for obesity (BMI ≥30 kg/m²).

We conducted several sensitivity analyses to examine the consistency of our results, presented in the supplementary data, available at Annals of Oncology online.

All analyses were carried out using SAS version 9.2 (SAS Institute, Cary, NC).

results

Among 363 426 participants during a mean follow-up time of 8.5 years, 204 incident cases of BTC (including 75 GBC and 129 non-GBC cases), and 176 incident cases of HCC were identified. In addition, 58 incident cases of IBD and 50 of cholangiocarcinoma were diagnosed; but we did not conduct analyses on these both outcomes, because of low sample size (two cases with IBD and one with cholangiocarcinoma reported to have diabetes).

Baseline characteristics of participants with and without diabetes are presented in Table 1.

Table 1.

Baseline characteristics among participants without and with diabetes in the European Prospective Investigation into Cancer and Nutrition (n = 363 426)

	Without diabetes	With diabetes
N	354 838	8588
Men, %	35.0	47.5
Age, years (mean, SD)	51.4 (10.5)	57.7 (7.9)
Acohol intake, g/day (median, IQR)	6.8 (1.1–17.9)	2.8 (0.0–16.0)
Smoking, %a
Never	47.3	48.0
Former
<10 years	10.2	11.4
≥10 years	16.5	17.4
Current
<15 cigarettes/day	10.1	7.6
15–24 cigarettes/day	8.3	7.5
≥25 cigarettes/day	3.2	4.1
Other than cigarettes	3.1	3.3
Education, %
No school degree/unknown	8.9	20.5
Primary school	27.6	41.4
Technical/professional/secondary school	39.6	25.6
University degree	23.9	12.5
BMI, kg/m2 (mean, SD)	25.8 (4.3)	29.2 (4.9)
Waist-to-height ratio (mean, SD)	0.51 (0.07)	0.59 (0.08)
Gallstones diseases, % b	5.0	12.5
Age at diabetes, years (median, IQR)		50.0 (43.0–57.0)
Diabetes duration, years (median, IQR)		6.0 (2.6–12.3)
Insulin treatment, %		25.1

P for differences between participants without and with diabetes was P < 0.0001 for all characteristics.

^aNumbers do not add up to 100% because of unknown values.

^bMissing values for gallstones (81 497).

Diabetes status was associated with an almost twofold higher risk of BTC, even after controlling for BMI and WHtR (Table 2). Participants with shorter disease duration were at higher risk of BTC compared with participants without diabetes, whereas only few participants with BTC had been diagnosed with diabetes longer than 6 years (only two cases). When we investigated diabetes duration as a continuous measure, we did not observe an association between diabetes duration and BTC risk [multivariable RR (95% CI) 0.98 (0.91–1.05)]. Furthermore, participants with diabetes, not treated with insulin were at higher risk, compared with participants without diabetes, whereas no association could be observed for participants using insulin (also based on two cases). When we analyzed subsites of BTC, diabetes status was associated with risk of GBC (Table 2). All cases of GBC with diabetes were in the short diabetes duration category (<6 years). For non-GBC, risk of diabetes status was slightly elevated, but not significant (Table 2).

Diabetes status was associated with a more than two-fold higher risk of HCC, even after adjustment for BMI and WHtR (Table 3). We did not find a difference in risk of HCC between short (<6 years) or long (≥6 years) diabetes duration. For diabetes duration as a continuous measure, the multivariable RR (95% CI) for HCC was 1.03 (0.97–1.09). In addition, younger age of diabetes diagnosis was associated with risk of HCC, also independent of BMI and WHtR. For HCC risk, only participants treated with insulin were at higher risk, whereas no association was observed for participants without insulin treatment.

For both, BTC and HCC, highest risk was observed for participants with diabetes and a BMI ≥30 kg/m² compared with the reference group (participants without diabetes and a BMI <30 kg/m²) (supplementary Figure S1, available at Annals of Oncology online).

In sensitivity analyses, findings, including results of the nested case–control subset, remained comparable (more details in the supplementary data, available at Annals of Oncology online).

discussion

Diabetes was associated with risk of BTC (particularly GBC), and HCC, independently of general and abdominal obesity. The findings indicated that participants with shorter diabetes duration seemed to be at higher risk of BTC, whereas participants treated with insulin were at higher risk of HCC, compared with participants without diabetes.

Our results are in accordance with findings of a meta-analysis that indicated a positive association between diabetes and BTC [summary RR (95% CI) 1.43 (1.18–1.72)] [2]. However, findings of the meta-analysis were not statistically significant anymore when the analysis was restricted to studies adjusting for BMI [summary RR (95% CI) 1.46 (0.88–2.42)]. The RR found in our study was somewhat stronger after adjustment for general and abdominal obesity [multivariable RR (95% CI) 1.77 (1.00–3.13)]. Also, consistent with our results, the meta-analysis found a positive association with GBC alone [summary RR (95% CI) 1.65 (1.22–2.23)]; however, our RR was slightly higher. To our knowledge, only one study adjusted for abdominal obesity in addition to adjustment for BMI, and provided similar results [5]. In concordance with this study, we did not observe a statistically significant association for other subsites of BTC, including EBD, tumors of the ampulla of Vater and other BTCs [5]; whereas other studies have shown different results [22, 23]. Furthermore, highest risks for BTC were observed for participants with diabetes and obesity, suggesting that participants with more metabolic conditions were at higher risk. To our knowledge, there is no other study investigating this interplay.

To our knowledge, little is known about the associations between diabetes duration, age at diabetes diagnosis, insulin treatment, and risk of BTC. Two cohort studies did not find statistically significant associations between diabetes duration, insulin treatment, and BTC [10, 12]. Our findings suggest that participants with shorter disease duration and without insulin treatment might be at increased BTC risk. However, this association could be driven by strong results found for GBC only, since all GBC cases were in the category with shorter disease duration and most GBC cases were not treated with insulin.

The underlying mechanisms linking diabetes and BTC, particularly GBC, are not clear. Findings of a case–control study suggested that gallstone diseases, the major risk factors for GBC [7], might be one of the most important mediators [5]. On the other hand, gallstone diseases might predict diabetes as demonstrated in one prospective study [6]. Therefore, a possible explanation for our findings might be the presence of pre-existing gallbladder diseases, which could have influenced diabetes status. However, when we restricted our analyses to participants free of self-reported gallstone diseases, the associations remained unaltered, pointing to other mechanisms underlying this association. Nonetheless, we could not fully account for the effect of gallstone disease as relevant information was missing for 22.4% of the participants and many cases may have been undetected.

In our study, diabetes status was associated with a more than twofold increased risk of HCC, which is in consistency with findings from two recent meta-analyses [3, 4]. Also in agreement with our findings, studies have shown that insulin treatment was associated with increased risk of HCC [10, 24, 25], and diabetes duration was not associated with HCC [12, 13, 15, 17], whereas other studies provided contrary results [10, 11, 16]. In our analyses limited to participants free of HBV/HCV, the risk estimates remained comparable to our main analyses, but results were more imprecise, possibly due to small sample size. Correspondingly, the meta-analyses reported that their findings were independent of HBV/HCV status [3, 4].

Several mechanisms have been suggested to explain the association between diabetes and HCC, including hyperinsulinemia, insulin resistance, or exogenous insulin treatment, which results in elevated levels of insulin-like growth factor (IGF-I) [1]. Insulin and IGF-I may stimulate hepatic cell proliferation and inhibit apoptosis in the liver, which might be the leading factor for carcinogenesis [26]. In accordance, it has been shown that diabetes is related with a spectrum of obesity-driven liver diseases [27], which are associated with hepatocyte injury, inflammation, and fibrosis and are presumably risk factors for HCC [8]. In agreement, highest risk for HCC was detected for participants with diabetes and obesity, pointing out that individuals with more components of the metabolic syndrome were at higher risk, which could be confirmed in other studies [14, 28].

Our study has some limitations. First, analyses relied on a small number of cases, particularly within categories of diabetes duration, age at diabetes diagnosis and insulin treatment, limiting the interpretation of some of the results, and the number of stratified analyses that could be carried out. Secondly, diabetes status was self-reported, which could have led to misclassification. However, it has been shown that the accuracy of self-reported diabetes was high [29]. In addition, diabetes status was missing for ∼4% of participants, which might result in under- or overestimation of results, depending on diabetes prevalence among excluded participants. Thirdly, diabetes status was assessed at baseline, and we did not account for incident diabetes cases during follow-up. Fourthly, no information on other types of medication than insulin was available, including metformin, which could have affected our results in both directions, depending on medication use among participants with diabetes. Furthermore, duration and change in treatment over time was not available for our analyses. Finally, no data were available for the presence of cirrhosis and family history of HCC.

The strengths of our study are the prospective study design and the incorporation of several potential confounders, including general and abdominal obesity, and information on HBV/HCV in a subset of the cohort.

In conclusion, our findings demonstrate that diabetes is a risk factor for GBC—and possibly other BTC—and HCC, independent of general and abdominal obesity. Whether diabetes treatment or duration are directly associated with these cancers, or whether they are surrogates for severity of diabetes, remains to be answered by future research. Further studies should consider potential risk differences by subsites of BTC and investigate associations between diabetes duration and treatment.

funding

This work was supported by the Federal Ministry of Education and Research, the German Research Foundation, Excellence Cluster Inflammation at Interfaces (EXC306 and EXC306/2), grants from the German Research Foundation (DFG NO446/7-1) (Germany); and the French National Cancer Institute (L'Institut National du Cancer; INCA) (grant number 2009-139). The coordination of EPIC is financially supported by the European Commission (DG-SANCO); and the International Agency for Research on Cancer. The national cohorts are supported by Danish Cancer Society (Denmark); Ligue Contre le Cancer; Institut Gustave Roussy; Mutuelle Générale de l'Education Nationale; and Institut National de la Santé et de la Recherche Médicale (INSERM) (France); Deutsche Krebshilfe, Deutsches Krebsforschungszentrum; the Hellenic Health Foundation, the Stavros Niarchos Foundation and the Hellenic Ministry of Health and Social Solidarity (Greece); Italian Association for Research on Cancer (AIRC); National Research Council; and AIRE-ONLUS Ragusa, AVIS Ragusa, Sicilian Government (Italy); Dutch Ministry of Public Health, Welfare and Sports (VWS); Netherlands Cancer Registry (NKR); LK Research Funds; Dutch Prevention Funds; Dutch ZON (Zorg Onderzoek Nederland); World Cancer Research Fund (WCRF); and Statistics Netherlands (The Netherlands); European Research Council (ERC) (grant number ERC-2009-AdG 232997) and Nordforsk; and Nordic Center of Excellence Programme on Food, Nutrition and Health (Norway); Health Research Fund (FIS); Regional Governments of Andalucía, Asturias, Basque Country, Murcia (No. 6236) and Navarra; and ISCIII RETIC (RD06/0020) (Spain); Swedish Cancer Society; Swedish Scientific Council; and Regional Government of Skåne and Västerbotten (Sweden); Cancer Research UK; Medical Research Council; Stroke Association; British Heart Foundation; Department of Health; Food Standards Agency; and Wellcome Trust (UK).

disclosure

The authors have declared no conflicts of interest.