Pancreatic cancer, the sixth leading cause of cancer-related death in the European Union1 and fourth in the United States,2 is one of the most aggressive malignancies. There is currently no effective means to screen for this cancer, and it is usually diagnosed at an advanced incurable stage. The 5-year survival rate is less than 5%.3 Consequently, identification of modifiable risk factors for pancreatic cancer is of importance as it may lead to prevention opportunities. Cigarette smoking is one of the few accepted modifiable risk factors for pancreatic cancer, but smoking may explain less than 30% of the cases.4, 5
Epidemiologic studies have associated overweight and obesity with elevated risk of many cancer types, including endometrial cancer, breast cancer (in postmenopausal women), renal cell carcinoma, colon cancer (especially in men) and adenocarcinoma of the esophagus.6 However, uncertainty remains about the relation between obesity and the risk of pancreatic cancer. In 2003, Berrington de Gonzalez et al.7 conducted a systematic review of case-control and prospective studies of body mass index (BMI), as a measure of overall obesity, and risk of pancreatic cancer. In their meta-analysis of 6 case-control and 8 prospective studies, BMI was weakly positively associated with pancreatic cancer risk (2% increase in risk per 1 kg/m2 increase in BMI), but there was evidence of heterogeneity among studies. A number of new prospective studies have been published since that meta-analysis. Therefore, we performed an updated systematic review with meta-analysis to summarize the available evidence from prospective studies on the association between BMI and pancreatic cancer risk, and to explore potential sources of heterogeneity among studies.
Material and methods
We searched the electronic database MEDLINE (from 1966 to November 2006) using the keywords or Medical Subject Headings “body mass index,” “BMI,” or “obesity” combined with “pancreatic cancer” or “pancreatic neoplasm.” Moreover, we searched for any additional studies in the reference lists of the identified articles. No language restrictions were imposed.
Studies were included if: (i) they presented original data from prospective studies, (ii) the exposure of interest was BMI (body weight in kg divided by the square of height in meters), (iii) the outcome was pancreatic cancer incidence or mortality and (iv) they reported relative risks (RRs) with corresponding 95% confidence intervals (CIs) for at least 3 categories of BMI or a RR per unit increase in BMI.
We identified 20 publications with data that were potentially eligible for inclusion in the meta-analysis. One study was excluded because BMI was classified in 2 categories only.8 The remaining 19 reports,9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27 consisting of 21 independent prospective studies (2 publications had 2 separate cohorts12, 23), were included in the meta-analysis.
We extracted from each publication the following information: the first author's last name, publication year, country, in which the study was performed, sample size, age at baseline, exposure assessment (self-reported vs. measured anthropometry), years of follow-up, covariates for adjustment in the analysis and the RRs and their 95% CIs for every category of BMI or per unit increase in BMI. From each study, we extracted the RRs that reflected the greatest degree of control for potential confounders.
We performed meta-analysis of the dose-response relationship between BMI and pancreatic cancer risk. Only four publications9, 12, 23, 26 had provided a RR per unit increase in BMI. For the other studies, we estimated the RR per 5 kg/m2 increase in BMI by regressing the natural log RRs according to BMI categories on the midpoint of each category. This was done using the method described by Greenland and coworkers,28, 29 which takes into account that the level-specific RRs are correlated. This method requires that the number of cases and person-time or noncases for each category are known. When this information was not available,13, 16, 25, 27 we estimated the dose-response slopes using variance-weighted least squares regression analysis. For open-ended categories (e.g., ≥30 kg/m2), we estimated the median values using data from the Swedish Mammography Cohort and the Cohort of Swedish Men,23 or obtained the values from the authors (for studies in Asia17, 21). Summary estimates of RR were obtained from random-effects models30 applied to the study-specific linear trends.
Statistical heterogeneity among studies was evaluated using the Q and I2 statistics.31 As statistical tests for heterogeneity have low power, heterogeneity was considered present for p-values <0.1. I2 is the proportion of total variation contributed by between-study variation.31 To investigate the influence of individual studies on the summary RR, a sensitivity analysis was conducted by omitting one study at a time and calculating the resulting summary RR. We performed subgroup analyses to examine potential sources of heterogeneity according to sex, geographic region, duration of follow-up, publication year, assessment of weight and height (self-reported vs. measured) and adjustment for diabetes. Publication bias was evaluated with the Egger's regression asymmetry test.32 All analyses were conducted using Stata, version 9.0 (StataCorp, College Station, TX).
Of the 21 prospective studies included in this meta-analysis, 10 were from the United States, 9 from Europe and 2 from Asia (Table I). One study was a nested case-control study within a prospective cohort9; the remaining were prospective cohort studies. All studies controlled for cigarette smoking and 13 studies also adjusted for diabetes (Table I).
Table I. Characteristics of Prospective Studies of Body Mass Index and Pancreatic Cancer1
Abbreviations: ATBC, α-Tocopherol β-Carotene Cancer Prevention Study; CHA, Chicago Heart Association Detection Project in Industry; CPS-II, American Cancer Society Cancer Prevention Study II; EPIC, European Prospective Investigation into Cancer and Nutrition; KNHIC, Korea National Health Insurance Corporation; NA, information not available; VHM&PP, Vorarlberg Health Monitoring and Promotion Program; M, male; F, female.
Mean or median duration of follow-up in parenthesis.
Number of controls (nested case-control study).
38% of cases (49 males and 44 females) in this study were also included in the CPS-II Mortality Cohort.
Figure 1 shows the estimated RRs of pancreatic cancer per 5 kg/m2 increase in BMI for each study. The 21 studies combined included 3,495,981 individuals and 8,062 pancreatic cancer cases. Overall, the estimated summary RR per 5 kg/m2 increase in BMI was 1.12 (95% CI, 1.06–1.17). There was no statistically significant heterogeneity among studies (Q = 27.07; p = 0.13; I2 = 26.1%). In a sensitivity analysis, in which one study at a time was omitted and the remaining analyzed, the summary RR ranged from 1.10 (95% CI, 1.05–1.16; when the study by Calle et al.16 was excluded) to 1.13 (95% CI, 1.08–1.19; when the study by Samanic et al.27 was excluded). There was no evidence of publication bias (Egger's test: p = 0.58).
In analysis stratified by sex, the estimated summary RR of pancreatic cancer per 5 kg/m2 increase in BMI was similar for men and women; there was heterogeneity among the RR estimates for men but not for women (Fig. 2). Continent-specific RR estimates showed that BMI was positively associated with pancreatic cancer risk in the United States and Europe, but not in the two studies in Asia (Table II). There was no statistically significant difference in summary RR estimates across strata of duration of follow-up or publication year (Table II). However, the summary RR estimate was statistically significantly higher among studies based on self-reported weight and height than among studies based on measured anthropometry (p = 0.003). Studies that adjusted for diabetes showed higher RR estimates than those that did not control for diabetes (p = 0.07).
Table II. Summary Relative Risk (RR) Estimates of Pancreatic Cancer for a 5 (kg/m2) Increment in Body Mass Index in Prospective Studies
The current meta-analysis of prospective studies supports a positive association between BMI and risk of pancreatic cancer in both men and women. Overall, a 5 kg/m2 increase in BMI was associated with a 12% increased risk of pancreatic cancer.
A positive association between BMI and pancreatic cancer risk is biologically plausible. Obesity, especially central obesity, has been related to glucose intolerance, insulin resistance, hyperinsulinemia and to the development of type 2 diabetes.33, 34 A number of epidemiologic studies have reported an increase in risk of pancreatic cancer associated with long-standing diabetes.35 Moreover, ample evidence from in vitro, animal and human studies indicates that abnormal glucose metabolism and insulin resistance may be implicated in the development of pancreatic cancer.11, 36, 37, 38 Three prospective studies have directly examined the associations of postload plasma glucose,11 fasting serum glucose,38, 39 insulin resistance39 and fasting serum insulin concentrations39 with pancreatic cancer risk. All studies showed an approximately 2-fold elevated risk of pancreatic cancer when the top and bottom categories were compared. In the 3 available prospective studies on waist circumference and/or waist-to-hip ratio, indicators of central obesity, in relation to risk of pancreatic cancer, a statistically significant23, 26 or a nonsignificant19 positive association was observed. Another cohort study found that men and women who reported “central” weight gain had a statistically significant increased risk of pancreatic cancer compared with those who reported “peripheral” weight gain.18
The possibility that the observed positive relation between BMI and pancreatic cancer risk was due to unmeasured or residual confounding should be considered. Although all studies controlled for smoking, residual confounding may exist. Individuals with high BMI generally were less likely to smoke than those with lower BMI.12, 18, 23 Hence, any residual confounding from smoking would most likely have attenuated rather than exaggerated any true relationship between BMI and pancreatic cancer.
About half of the studies in this meta-analysis relied on self-reported weight and height. Validation studies have shown that weight and height are assessed with high validity.12, 23 The correlations between body weight assessed by questionnaire and by direct measurements were 0.9 for Swedish men and women,23 0.97 for US men12 and 0.96 for US women.12 All studies assessed weight only once (or used only a baseline assessment in the analyses), and changes in weight may have weakened the observed association with BMI given the long follow-up period of many of the cohort studies.
In a meta-analysis of published data, it is possible that an observed association is the result of publication bias, because studies with null results tend not to be published. In the present meta-analysis, however, we found no indication of such bias. One prospective study that had examined the association between BMI and risk of pancreatic cancer8 could not be included in this meta-analysis because BMI had been classified in only 2 categories. In that cohort of about 63,000 Norwegian men and women, including 166 pancreatic cancer cases diagnosed during 12 years of follow-up, high versus low BMI (>25.2 vs. ≤25.2 kg/m2 in men; >24.7 vs. ≤24.7 kg/m2 in women) was not associated with pancreatic cancer risk.8 Three record-linkage cohort studies reported that individuals with a diagnosis of obesity had a significant increased risk of pancreatic cancer when compared with the general population.40, 41, 42
In summary, findings of this meta-analysis support a positive relationship between BMI and risk of pancreatic cancer. Future studies that evaluate measures of central obesity, such as waist circumference and waste-to-hip ratio, in relation to pancreatic cancer risk may provide further insight into the role of insulin and insulin resistance in the development of this fatal malignancy.