Objectives: Several preclinical and observational studies have shown that anti-diabetic medications (ADMs) may modify the risk of pancreatic cancer (PaC). We performed a systematic review and meta-analysis evaluating the effect of metformin, sulfonylureas (SUs), thiazolidinediones (TZDs), and insulin on the risk of PaC in patients with diabetes mellitus (DM).
Methods: We conducted a systematic search of Medline, EMBASE, and Web of Science, up to June 2012, and published abstracts from major gastroenterology and oncology meetings from 2003 to 2012. Studies were included if they (1) evaluated and clearly defined exposure to metformin, SU, TZDs, and/or insulin, (2) reported PaC outcomes in patients with DM and (3) reported relative risks or odds ratio (OR) or provided data for their estimation. Summary OR estimates with 95% confidence intervals (CIs) were estimated using the random-effects model.
Results: Eleven studies (6 cohort, 3 case-control, and 2 randomized controlled trials (RCTs)) reported 1770 cases of PaC in 730,664 patients with DM. Meta-analysis of observational studies showed no significant association between metformin (n=9 studies; adjusted OR 0.76, 95% CI 0.57-1.03, P=0.073), insulin (n=7 studies; adjusted OR 1.59, 95% CI 0.85-2.96, P=0.144), or TZD (n=4 studies; adjusted OR 1.02, 95% CI 0.81-1.30, P=0.844) use and risk of developing PaC. Use of SU was associated with a 70% increase in the odds of PaC (n=8 studies; adjusted OR 1.70, 95% CI 1.27-2.28, P<0.001). There was considerable inherent heterogeneity between studies not explained by study design, setting, or comparator drug and is likely related to confounding by indication and reverse causality. The pooled analyses of the two RCTs were underpowered and provided non-significant results with wide, non-informative 95% CIs.
Conclusions: Although SU seems to be associated with increased risk of PaC, meta-analysis of existing studies does not support a protective or harmful association between ADM use and risk of PaC in patients with DM. There was considerable heterogeneity across studies, and future, well-designed, prospective studies would be required to understand this association better.