Bayesian propensity score analysis for pharmacoepidemiologic research

Data on prescription claims, health services provided, and hospital discharges are routinely collected in the Canadian health care system for the management of health services. This information is valuable for observational research assessing the effectiveness and safety of drug therapies, allowing more rapid and cost-effective investigations. In addition to yielding large sample sizes with long periods of follow up, they also better reflect the spectrum of medical practice in large populations compared to randomized clinical trials. However, because study participants are not randomly assigned to treatment and control groups through this methodology, findings may be biased if there is inadequate control of the variables. There is a need for new statistical tools to analyze healthcare administrative data. Lawrence McCandless is developing and investigating a new statistical method called a Bayesian propensity score analysis. Through computer simulations, mathematical techniques and models for drug prescribing patterns, the Bayesian analysis will improve the validity of observational investigations of the effectiveness and safety of drugs using large health care administrative databases.