Diabetes and Pancreatic Cancer: Both Cause and Effect

By now, it is known that diabetes mellitus is a risk factor for pancreatic cancer. The special section on pancreatic cancer in the American Cancer Society’s Cancer Facts & Figures 2013 (1) states, “About 25% of patients with pancreatic cancer have diabetes mellitus at diagnosis, and roughly another 40% have pre-diabetes,” that “patients with long-term (five or more years) type II diabetes have a 50% increased risk of pancreatic cancer” and “pancreatic cancer can cause diabetes, and sometimes diabetes is an early sign of the tumor.” The direction of association goes both ways. Even though it takes about 10 to 15 years from pancreatic cancer initiation to diagnosis (2), most observable changes occur late in the disease course, though the trend in weight loss and cachexia is seen beginning perhaps 18 to 21 months before diagnosis (3), and we have shown that behaviors such as quitting cigarette smoking and starting proton pump inhibitor heartburn medication use and episodes of pancreatitis can be observed up to about two to three years before diagnosis as well (4). To date, most studies of the bi-directional temporal association between diabetes and pancreatic cancer have largely examined white and Asian populations. In this issue of the Journal, Setiawan and colleagues have extended this association to African and Hispanic Americans (5). In spite of the 60% higher lifetime risk of diabetes in these groups compared with white Americans (6), Setiawan and colleagues have shown that the same bi-directional association between diabetes and pancreatic cancer applies to them, with risk magnitudes comparable to those in other studies. It is a challenge to separate a causal involvement of diabetes mellitus (T2DM) in the origin of pancreatic cancer from diabetes as a result of pancreatic disease (T3cDM), in part because the risk relationship in the time before diagnosis is a continuum. Figure 1 shows the temporal pattern of risk according to years before diagnosis. In this figure, risk associations given in 13 reports (4,5,7–16) representing 37 to 41 individual studies (depending upon time point) are plotted, along with a random-effects meta-analysis value for each of the time intervals. Within two to three years of diagnosis, the average risk association is greater than twofold, whereas by five years before diagnosis, it settles down to near the 50% elevation noted by the American Cancer Society. Relative risk of pancreatic cancer by years in the past of diabetes diagnosis. Denoted as indicated in the upper-right of the figure, 37 to 41 studies of the risk association between diagnosis of diabetes mellitus and subsequent diagnosis of pancreatic cancer are plotted according to time interval between the diagnoses. In black (dashed line), the random-effects meta-analysis summary relative risk for each time interval is shown; the shaded boxes denote the 95% confidence intervals of this relative risk. Not all studies contributed data to each time interval. CI = confidence interval; T3cDM = diabetes as a result of pancreatic disease; T2DM = diabetes mellitus. It would seem that a new diagnosis of diabetes could be a warning sign that pancreatic cancer is underway, and thus that such individuals should be worked up for the disease. However, the magnitude of increased risk, about fivefold in the first year, does not help all that much with the biomarker specificity problem, because pancreatic cancer has a lifetime risk of about 1.5%, and thus the risk of its diagnosis over the next year or two among newly diagnosed diabetes patients is still less than 1%, requiring biomarker specificities greater than 99% in order to avoid large numbers of false positives for every true positive found (17). Further, only 5% of pancreatic cancer patients with diabetes have had the diabetes diagnosed within one year of their cancer diagnoses, and only 10% within two years (4); thus, recent diabetes diagnosis, by itself, does not establish with any efficiency a “high-risk group” for pancreatic cancer. Nevertheless, a diagnosis of diabetes involves an interaction with the health care system and an opportunity to add an additional blood test or two. Even though adequate population biomarkers for pancreatic cancer have not yet been established—including CA19-9 and CancerSEEK (18), both of which have issues—if T2DM could be physiologically distinguished from pancreatic cancer–related T3cDM, that might provide some degree of cancer risk prediction. If chronic pancreatitis is ruled out as a risk factor for newly diagnosed diabetes, then low circulating pancreatic polypeptide (PP) levels or lack of PP response to a mixed-food meal could suggest a pancreatic cancer etiology of the diabetes (19), which would likely be worth further screening with CA19-9. What should one make of a risk factor with a magnitude of association relative risk of about 1.5? Even though such a low magnitude does not provide much evidence for its causal nature, the large number of consistent studies showing this for long-term diabetes and risk of pancreatic cancer (Figure 1) across various study designs, with both medical records and individual questionnaire information and adjusted for various possible confounding factors, suggests that the relationship is real. Perhaps it is operating through some further action of the diabetic state. Over the last two decades, pancreatic cancer incidence in the United States has been steadily increasing by about 1% per year (20). This increase has occurred in spite of long-term trends in reduced cigarette smoking. We have suggested elsewhere that declining trends in the frequency of CagA-positive Helicobacter pylori colonization may be partly responsible (21), but the growing obesity epidemic with its parallel rise in frequency of T2DM could contribute to this trend as well. Even though the magnitude of risk of 1.5 for a five-year or longer history of T2DM is not large, the very large fraction of the population now subject to this condition makes it an important and potentially modifiable contributor to the burden of pancreatic cancer. Setiawan and colleagues (5) have shown that the involvement of diabetes in the risk of pancreatic cancer applies just as much to African and Hispanic Americans as to white Americans, and thus that the continuing trend of increasing risk of this disease will likely involve them as well. Affiliation of author: Department of Chronic Disease Epidemiology, Yale School of Public Health, New Haven, CT. Dr. Risch has no conflicts of interest with this work.