Study design choices for evaluating the comparative safety of diabetes medications: An evaluation of pioglitazone use and risk of bladder cancer in older US adults with type‐2 diabetes

摘要

Abstract Aim The aim of the study was to empirically demonstrate the effect of varying study designs when evaluating the safety of pioglitazone in treating bladder cancer. Methods We identified Medicare beneficiaries above 65?years of age with diabetes between 2008 and 2015 and with classified exposure (at least two claims within 180?days) to glucose‐lowering drugs (GLD), pioglitazone or another drug. The effects of varying the following study design parameters on bladder cancer risk were assessed: use of a new vs existing drug, choice of referent (all non‐users and users of GLDs, non‐insulin GLDs and DPP‐4s) and whether or not censoring accounted for treatment change. We used the Cox proportional hazards model to obtain adjusted HRs and 95% CIs. Results We included 1,510,212 patients classified as pioglitazone users (N = 135,188) or non‐users (N = 1,375,024). Users had more diabetic complications than non‐users, but fewer than insulin users. The HR ranged from 1.10 (1.01‐1.20) to 1.13 (0.99‐1.29) when censoring ignored treatment change, suggesting a weak association or none between pioglitazone and bladder cancer, probably under‐estimating risk. However, the HR was 1.20 (1.01‐1.42) when cohorts were restricted to new users, censored upon treatment change, and when DPP‐4 was used as the referent, suggesting an increased risk of bladder cancer associated with pioglitazone. Conclusions The continued demand for new GLDs indicates the need for more robust observational methods to improve the value of generating real‐world evidence in equipping clinicians to make informed prescribing decisions. Although there is no one‐size‐fits‐all approach, we recommend active comparator new user study designs that compare therapeutically equivalent drugs and account for treatment changes during follow‐up to present the least biased comparative safety estimates.