Pancreatic cancer risk and obesity/BMI
Cachexia is a known characteristic of pancreatic cancer with estimates as high as 80% of patients cachexic at diagnosis [8
]. Therefore to determine whether BMI is a risk factor for rather than a result of pancreatic cancer, recent epidemiological case-control and cohort studies of pancreatic cancer frequently have used self-reported data for usual adult weight and height, or weight and height at specific ages to assess obesity and associations with increasing BMI[9
]. Prospective cohort studies also used weight and height at baseline (self-reported or measured) to compute BMI and frequently excluded cases who were diagnosed shortly after study enrolment, typically ≤2 years, to ensure that weight was not a result of the disease. Few studies have collected data for other measures of adiposity such as waist circumference and waist to hip ratio (WHR).
Overall, results have suggested a positive association between obesity/high BMI and pancreatic cancer risk that has been confirmed in three recent large pooled analyses [20
] and in two of three meta-analyses [23
]. These analyses provide the basis for the discussion presented here as they encompass a range of well-designed independent observational epidemiological studies that investigated BMI and pancreatic cancer risk. As summarized below, risk estimates from these pooled and meta-analyses usually were reported as associations with a 5 kg/m2
increase in BMI and with BMI categorized using WHO cutpoints (overweight: 25≤BMI <30; obese 30≤BMI<35, extreme/severe obesity: 35≤BMI.). Effects were relatively consistent across studies with approximate 10% or greater increases in risk for a 5 kg/m2
unit increase in BMI, or a 20%-50% increased risk among obese relative to normal BMI participants. Several studies that assessed measures of adiposity in addition to BMI also were included in these pooled and meta-analyses with results supporting positive associations with WHR [20
]. Effect modification and confounding associated with known pancreatic cancer risk factors (e.g. smoking, diabetes), geographic location, sex, and for study design/methodology (self-reported vs. measured anthropometrics) were assessed when possible and provided further insight into the BMI and pancreatic cancer relationship that was not possible in many independent studies.
The first of the published meta-analyses[23
] included 14 studies, 6 of which were case-control studies thereby allowing for a formal evaluation of effect estimates by study design. The summary estimate for case-control studies was slightly attenuated compared with cohort studies (RR=1.02, RR=1.03 respectively), despite that BMI was computed based on pre-diagnostic weight values [usual adult weight (1 study), weight 2 years before interview (4 studies), unknown (1 study)][23
]. Given that most of these case-control studies assessed BMI based on anthropometric characteristics 2 years before enrollment, a time frame that many prospective cohort studies used to exclude newly diagnosed cancer cases from their analyses, results suggest that use of self-reported data, recall or selection bias in case-control studies may have contributed to the observed differences. Further evaluations of effect by type of study design were not conducted in the five subsequently published pooled/meta-analyses as only prospective cohort studies were included in these analyses (with one exception; one case-control study was included in the pooled PanScan analyses after sensitivity analyses showed no undue influence on study estimates[20
Another fundamental study design characteristic that also was investigated for its effect on the association between BMI and pancreatic cancer risk was the type of anthropometric factor assessment, self-reported or measured. Analysis by type of assessment is of interest because although measured and self-reported anthropometric data are well-correlated, reporting biases exist by sex, age and weight which can affect measures of association, such as estimates of mortality risk[26
]. BMI from self-reported compared with measured factors was evaluated best by the pooled and meta-analyses that included prospective cohort studies with weight and height data collected at baseline prior to diagnosis. Summary estimates for BMI based on self-reported data were either somewhat attenuated[24
] or similar[20
] to summary risk estimates from studies that measured anthropometric factors. Importantly, in studies that noted a variation in estimates the overall inference that obesity/high BMI is associated with increased pancreatic cancer risk was not altered.
Other key factors that were evaluated in most studies as potential confounders or effect modifiers of the association between BMI and pancreatic cancer risk included sex and age, the established risk factors smoking and diabetes, and geographic region although few studies formally tested statistical interactions. All studies assessed differences by sex with any observed differences in estimates statistically non-significant and no one sex having consistently higher risk estimates. Interestingly, in the PanScan[20
] study the greatest differential in risks by sex were observed for those severely obese (BMI≥35) as estimated in models that excluded smokers, diabetics and cases diagnosed within two years of follow-up and included BMI categorized using WHO cutpoints (women: RR=1.98, men: RR=0.90). Age was similarly included in all models in all studies and was not found to be an effect modifier in the studies that stratified by age at study entry/diagnosis[21
Smoking and long-standing diabetes are among the few known risk factors for pancreatic cancer and as both also are associated with weight they conceivably can alter the association between BMI and pancreatic cancer risk. Nearly all studies considered smoking as a confounder whereas statistical interaction or effect modification due to smoking was determined in a subset of these pooled and meta-analyses. There was some evidence that risk estimates were slightly increased when smoking was adjusted for in models[23
] although most studies did not provide effect estimates from models without smoking included. In analyses stratified by smoking, results provided support for an increased risk of pancreatic cancer with increased BMI mainly among nonsmokers[20
] and never/former smokers [21
] (p for interaction= 0.08 to 0.12), but not current smokers. This intriguing result requires further study as smoking status reflects baseline behavior in these cohort studies and if smokers are thinner than the rest of the population then power may be low to detect a joint effect of current smoking and obesity.
The role of diabetes as a potential confounder is complicated as diabetes may be in the causal pathway between BMI and pancreatic cancer. Despite this possibility many studies adjusted for diabetes history in their models but similar to smoking characteristics, few studies evaluated effect modification/statistical interaction with diabetes. In a comparison of models that did and did not include adjustment for diabetes, adjustment for diabetes resulted in higher risk estimates in several studies[22
] (highest risk was found in non-diabetics although statistical interaction was non-significant),[22
] attenuated risk in one large pooled analysis[20
] and no substantial difference in risk in another[21
]. Given the analytic concerns that diabetes is not independently associated with both obesity and pancreatic cancer, and the mixed results that have been reported in some large studies, further carefully designed research that can clarify the interplay among obesity/high BMI, diabetes and pancreatic cancer risk are needed.
Although most studies did not consider the effect of geographic region in their analyses, pooled and meta-analyses that included studies from different continents provided an opportunity to investigate whether risk of pancreatic cancer associated with BMI differed across continents. Study data confirmed obesity statistics across populations with the highest median or average BMIs observed among participants from North American studies, followed by Europeans and Asians[21
]. In the studies that stratified analyses by region, BMI was modeled as a continuous variable and results showed positive associations between BMI and pancreatic risk in North American and European populations, with the highest RRs observed among North American groups[23
] and no association in Asian study populations[24
]. The lack of an association in the Asian study populations was based on two studies and is inconsistent with data suggesting that the negative effects of obesity/high BMI on disease occur at lower BMI values in Asian populations. Although compelling, further pooled and meta-analyses that include a larger number of Asian studies and that also can compare the effect estimates to those obtained for assimilated Asians in non-Asian countries will help to clarify this association.
Two recently published prospective cohort study pooling projects [20
] are noteworthy as analyses included evaluation of other important characteristics of obesity/high BMI that have not been investigated in earlier pooled analyses; measures of central adiposity, BMI as a young adult, change in BMI overtime, and the effect of physical activity. Both studies assessed central adiposity and in analyses of men and women combined[21
] or separately,[20
] results showed that hip circumference[21
] and waist circumference[20
] were not associated with pancreatic cancer risk. In contrast, WHR was associated with a statistically significant 20-30% increased risk for those with the highest ratios[20
]. Further analyses of WHR by sex were less consistent with data from one study suggesting similar risk by sex[21
] and the other showing a stronger effect among women (4th
quartile: RR=1.6 (1.03-2.50)) although risk also was elevated among men (RR=1.50 (0.77-2.93))[20
]. Central adiposity is a marker of visceral body fat that is metabolically active, secreting adipokines that have a downstream effect on insulin resistance, a risk factor for pancreatic cancer. Storage of body fat is hormonally driven and among women the tendency for centrally stored fat is more pronounced after menopause. Because BMI can be misleading measure of ‘fatness’ for those who are very fit or for those who are very thin, central adiposity has been shown to be a better marker of disease risk for some conditions, e.g. cardiovascular disease and metabolic syndrome. Given the accumulating evidence and biologic relevance of central adiposity in disease development, continued study of central adiposity that also includes waist-to-height ratio, evaluation of defined risk cutpoints, and consideration of pre and post-menopausal status among women are needed to improve our understanding of body fat in risk of pancreatic cancer.
Results for other obesity-related factors that were pooled for analyses in these two recent studies suggested that obesity was associated with earlier onset and that timing of obesity may be important[20
]. Obese participants were found to be diagnosed on average 1 year earlier than those of normal-weight [20
] which is consistent with results reported from a recent independent study [16
]. Participants who reported being overweight/obese as young adults (ages 18-21 years) had an approximate 1.2-fold increased risk of pancreatic cancer compared with those with normal BMI[21
]. Further, those whose BMI had increased >10 kg/m2
over their adult life had a 1.4-fold increased risk of pancreatic cancer compared with those whose BMI had changed no more than 2 kg/m2
]. Finally, although physical activity is a critical component of energy balance, it was evaluated in only one pooled analysis with results suggesting a statistically significant interaction between physical activity and BMI (p for interaction =0.02)[21
]. Those who were obese and highly active were at the greatest risk of pancreatic cancer (RR=1.29).[21
]. These results were based on subsets of studies within these pooled analyses and although they contribute new data that may help refine our study of obesity and pancreatic cancer, additional research is required to clarify these associations.
Based on data described above, obesity/high BMI has been consistently associated with increased risk of pancreatic cancer. However, as implied by results from these large pooled and meta-analyses, effects have been influenced by study design characteristics and are likely to vary by subgroups of at-risk patients. Therefore, although the wealth of evidence indicates that obesity increases risk of pancreatic cancer, the additional refinement of hypotheses and continued study of adiposity and obesity-related factors will increase our understanding of the underlying biological mechanisms relevant to pancreatogenesis and help to better define an at-risk population.
Pancreatic cancer survival and obesity/BMI
Studies that have assessed obesity/BMI associated with pancreatic cancer survival include those among patients who had undergone surgical resection[28
] as well as in observational case-control and cohort study populations [16
]. Reported results have been inconsistent with some studies reporting decreased survival associated with obesity/increased BMI [16
], others finding no association [29
] and one study finding longer survival duration [32
The studies of surgical resection patients typically used BMI at time of surgery (from medical records) to assess primary hypotheses related to specific surgical complications/outcomes in obese compared with non-obese patients, in addition to the more general question of whether obesity was related to survival in these patients. Two studies reported no association with obesity [29
] whereas another also found no association with BMI but a poor survival for those with high intra-abdominal fat as measured by CT [28
]. Poorer survival also was reported for extremely obese patients (BMI≥35) compared with all others in a study of consecutive patients seen for curative resections at a cancer referral center [30
]. The single study that reported improved survival was a relatively large series (795 cases) of surgical-resection patients where both overweight (HR=0.68) and obese patients (HR=0.72) had improved survival relative to normal BMI patients [32
In contrast to studies of surgical patients, the observational case-control and cohort studies tested whether obesity prior to diagnosis was associated with survival. In these studies BMI was computed based on usual adult weight, weight at specific ages or weight at baseline (prospective cohort studies) before diagnosis. Obesity/high BMI was associated with poor survival in three case-control or case series studies[16
Despite some suggestion that pre-diagnostic obesity/high BMI may be associated with poor pancreatic cancer survival, overall results are inconclusive. Additionally, the differences in study designs, study populations, measures and classifications of obesity, and slight differences in hypotheses being tested prevent generalization of study results. Because surgical-resection studies computed BMI based on weight at time of surgery and are likely to have included only those patients with localized disease who tend to have a better prognosis compared to other pancreatic cancer patients, results and inferences from these populations may not reflect associations in patients with more advanced inoperable disease. Also, underlying differences in populations of surgically resected patients that were related to varying medical practices between countries and types of treatment centers may be a concern when interpreting results. Finally, as the negative effects of increased BMI may occur at lower levels in Asian compared with non-Asian populations, inferences based on results from studies of largely Asian populations may not be applicable to other racial groups. Because of the low variability in survival duration, and confounding by indication and clinical prognostic factors, large pooled analyses of case-control and cohort studies are needed to have sufficient sample size and power to test the association between obesity/high BMI determined at different times, and pancreatic cancer survival. Case-control studies can be affected by survival bias but will have a larger number of patients and pre-diagnostic data specific to pancreatic cancer. In contrast, prospective cohort studies will have fewer cases and events, but by design are more likely to include a greater proportion of the sickest patients who tend to be under-represented in case-control studies, and can directly measure anthropometric factors and adiposity before diagnosis. Integration of complementary information from both case-control and cohort studies will provide a more comprehensive evaluation of survival-related associations.
Pancreatic cancer mortality and obesity/BMI
Results from studies of pancreatic cancer mortality associated with obesity/high BMI also have been inconsistent. Obesity/high BMI at baseline was not associated with pancreatic cancer mortality in three large prospective cohorts[35
]. In contrast, obesity/high BMI was associated with increased mortality in a series of patients after pancreatectomy[30
], in the Million Women Study (a large cohort of middle aged-women) [38
], in a large cohort of U.S. adults [39
] where later follow-up showed the magnitude of risk was greatest in nonsmoking men [40
], and in a prospective cohort of men in Chicago, U.S. area [41
]. An analysis of individual data from 30 cohort studies in the Asia Pacific showed an increase in pancreatic cancer mortality with weight change that was measured as a 2cm increase in waist circumference [42
]. Although several of these studies found that BMI and other known risk factors were positively associated with pancreatic cancer risk, there was insufficient evidence to conclude that obesity/high BMI was associated with pancreatic cancer mortality. Mortality results may have been affected by type of assessment of BMI as shown in a recent study[26
]. However, similar to the determination of pancreatic cancer risk and survival associated with BMI, the inconsistency in results from these studies also may be related to the potential for misclassification of obesity related to assessing BMI at baseline only and possible inadequate adjustment for potential confounders. Assessment of other obesity-related characteristics with pancreatic cancer mortality including measures of adiposity, use of measured anthropometrics, and more detailed data about timing of obesity may help to clarify and increase our understanding of a possible association between BMI and pancreatic cancer mortality.