Leveraging Medicare Linkages to Identify New Associations: Prescription Drugs and Digestive Cancer Risk

Sloan-Kettering Inst Can Research, New York NY

Investigators

Abstract

PROJECT SUMMARY Nearly 90% of US adults ages 65 and older use prescription drugs; however, surprisingly little is known about the long-term health effects of these drugs beyond the conditions for which they are indicated. Such unforeseen risks and benefits, particularly for diseases with a long latency period, such as cancer, are not well suited to detection in clinical trials, given that such trials are inadequate in size and duration of follow-up to detect drug-cancer associations. Multiple lines of evidence suggest that use of common prescription drugs may influence the development of various digestive cancers. However, prior studies have largely focused on broad, common classes of drugs, rather than individual drugs. The availability of Medicare linkages—featuring an extremely large diverse population of older adults, detailed prescription drug data, and follow-up for cancer outcomes—enables the comprehensive scan of associations between individual drugs and cancer risk. Given the high prevalence of prescription drug use among older adults and increased risk of cancer in this group, it is especially critical to understand the long-term effects of prescription drugs in aging populations. We will leverage data from the National Cancer Institute Surveillance, Epidemiology, and End Results Program (SEER)-Medicare linked database to discover and follow-up novel associations between prescription drugs and risk of digestive cancer, both overall and specific to the five most common digestive sites (colorectum, pancreas, stomach, oropharynx, and liver). First, we will conduct a Drug-Wide Association Study (‘DrugWAS’) to identify associations between prescription drugs and risk of digestive cancers, using bootstrapping to correct for multiple testing. Second, we will assess the influence of potential confounding by indication by evaluating whether the associations hold when compared directly to drugs taken for the same indication. For associations that remain after applying the steps above, we will perform propensity score calibration using additional data from the Medicare Current Beneficiary Survey (MCBS); this will allow us to estimate associations after accounting for additional sources of confounding. Our proposed hybrid approach—integrating a hypothesis-free framework with careful adjustment for multiple testing and confounding—will identify novel drug signals in older adults and further refine these signals to those that withstand correction for multiple testing and adjustment for confounding. This exploratory study will generate hypotheses of novel drug-cancer associations as well as provide impetus for follow-up in resource- intensive settings. Moreover, we will provide proof of principle for a multi-phased approach that leverages large publicly available datasets created through decades of NIH investment to identify novel and meaningful cancer associations; this approach can be applied in long-term surveillance efforts of drug effects on health.