Overall, there was evidence of a small positive increase in the risk of pancreatic cancer per unit increase in body mass index. A unit increase in body mass index for a male patient of 1.78
in) represents weight gain of about 3
lbs). For a female patient of 1.64
m (5 feet 5 inches), a unit increase in body mass index represents weight gain of about 2.5
lbs). The observed per unit increase in relative risk translates into a 19% higher risk of pancreatic cancer for obese people (body mass index >30
) compared to those of normal body weight (22
). There was, however, some evidence of heterogeneity between the studies' results (P
=0.1). The summary relative risk estimates were slightly higher for studies that had adjusted for smoking and for case–control studies that had not used proxy respondents.
As this meta-analysis was based on published data, there is the possibility that publication bias could have affected the results. The relative risks are presented in chronological order of publication, and there was no visual evidence of publication bias (). Neither was there evidence that the smaller studies tended to have positive results more frequently. However, it is not possible to rule out the possibility that further data, which show no evidence of an association between body mass index and pancreatic cancer risk, exist but have not been published.
Two case–control studies that had published information on body mass index and the risk of pancreatic cancer could not be included in this analysis. A US case–control study of 149 cases of pancreatic cancer conducted in Utah (Lyon et al, 1993
), which was not included because the authors only published crude relative risks for the highest to lowest tertile and provided no information on the tertile boundaries, found a relative risk for the highest tertile of body mass index compared to the lowest of 0.83 (95% CI: 0.43–1.58) for men and 1.05 (95% CI: 0.53–2.08) for women. Although the study findings were null, the CI were wide and were therefore not inconsistent with our summary relative risk of 1.19 for obese people (body mass index >30
) compared to those of normal body weight (22
). A Chinese case–control study (Ji et al, 1996
) was not included because there is evidence that adiposity-associated health effects occur at lower levels of body mass index among Asian populations than among Western population (WHO, IASO and IOT, 2000
). It was not clear therefore that a relative risk per unit increase in body mass index would have the same meaning in an Asian as in a Western population. In this study, which included 483 cases of pancreatic cancer, there was evidence of an increasing risk of pancreatic cancer with increasing body mass index in men (P
=0.05) but not in women (P
=0.31). The relative risk for the top quartile of body mass index in male patients (>22.4 compared to <19.5
) was 1.38 (95% CI: 0.91–2.08) and for female patients, it was 1.46 (95% CI: 0.85–2.51) (>23.1
compared to <19.5
The studies that were included in this meta-analysis contained 91% of the cancer cases that were available and the results from the two studies that could not be included were not inconsistent with the findings from this meta-analysis. Two other record linkage cohort studies of obese individuals both found an elevated risk of pancreatic cancer compared to expected rates in the general population of 1.7 and 1.5 (95% CI: 1.1–1.9), respectively (Moller et al, 1994
; Wolk et al, 2001
). The study by Wolk et al
found that the risk decreased with age from 2.5 (95% CI: 1.5–4.0) for those aged <60 years, to 2.0 (1.3–2.8) for 60–69 year olds and 0.7 (0.4–1.2) for those aged 70+years.
Smoking and diabetes are both potential confounding factors for the relationship between obesity and pancreatic cancer. If there is an association between body mass index and smoking, however, it is more likely to be a negative one, as current smokers have been observed to weigh less than nonsmokers (Istvan et al, 1992
). Therefore, ignoring confounding by smoking could make the association between obesity and pancreatic cancer less positive. Only two of the studies that were included in this meta-analysis had not adjusted the risk estimates for smoking (Howe et al, 1990
; Zatonski et al, 1991
). When these studies and the cohort study of smokers (Stolzenberg-Solomon et al, 2002
) were excluded, the summary relative risk per unit increase in body mass index was slightly higher (1.03). In the cohort study by Calle et al
, the risk associated with obesity was investigated in men and women who had never smoked. In this subgroup, the relative risk of pancreatic cancer mortality for a body mass index of 35–39.9
was 2.61 compared to those with a normal body mass index (18.5–24.9
). In the analysis with adjustment for smoking, the relative risks were somewhat lower at 1.49 for men and 1.41 for women.
Long-standing diabetes has also been established as a risk factor for pancreatic cancer with duration of diabetes of 5 years or more being associated with a two-fold increased risk of pancreatic cancer (Everhart and Wright, 1995
). Hence, a history of diabetes could positively confound the relationship between the risk of pancreatic cancer and body mass index. However, in the six studies in this meta-analysis that had adjusted for a history of diabetes, the risks associated with a unit increase in body mass index were actually marginally higher than in the studies that had not adjusted for this risk factor (1.03 compared to 1.02). In their case–control study, Silverman et al
also published the risks associated with obesity cross-classified by a history of diabetes. Although there was evidence of an increase in risk for both those with and without the disease, within each level of body mass index diabetics had a higher risk of pancreatic cancer than nondiabetics (Silverman et al, 1999
). Future studies need to examine the relationship between obesity and pancreatic cancer in more detail in those who have never smoked and in those without a history of diabetes.
All of the studies except for Gapstur et al
, Friedman et al
and Stolzenberg-Solomon et al
relied upon self-reported height and weight and it is possible that weight may have been under-reported, especially by overweight or obese individuals (Spencer et al, 2002
). Such under-reporting could result in overestimation of the dose–response relationship. The summary relative risk estimate for the studies that had measured anthropometry was marginally lower than those that relied upon self-reporting (1.02 vs
1.03). In case–control studies, under-reporting of weight could be a potential bias if it occurred unequally among cases and controls. Case–control studies could also be biased if the individuals in the control group were more ‘health conscious’ and thus less likely to be overweight than the cases. However, the summary relative risks in the case–control studies (1.02) were actually slightly lower than those for the cohort studies (1.03).
Obesity may be related to an increased risk of several other cancers including those of the endometrium, colorectum, oesophagus, kidney and postmenopausal breast cancer (IARC, 2002
). Some of the mechanisms that have been suggested to explain these relationships may also be relevant for pancreatic cancer, including the hypothesis that insulin resistance and abnormal glucose metabolism may be a factor in pancreatic cancer development (Gapstur et al, 2000
). The association between diabetes and pancreatic cancer risk (Everhart and Wright, 1995
) supports this hypothesis, and further support for this is given by findings that physical activity may be associated with a decreased risk of pancreatic cancer (Hanley et al, 2001
; Michaud et al, 2001
This meta-analysis of the available observational data provides evidence that the risk of pancreatic cancer may increase slightly with increasing body mass index, and that obese individuals may have a risk that is 19% higher than those with a normal body mass index. However, the small magnitude of the summary relative risk means that the possibility of confounding cannot be excluded.