Using propensity score methods to analyse individual patient-level cost-effectiveness data from observational studies

The methodology relating to the statistical analysis of individual patient-level cost-effectiveness data collected alongside randomised controlled trials (RCTs) has evolved dramatically in the last ten years. This body of techniques has been developed and applied mainly in the context of the randomised clinical trial design. There are, however, many situations in which a trial is neither the most suitable nor the most efficient vehicle for the evaluation. This paper provides a tutorial-like discussion of the ways in which propensity score methods could be used to assist in the analysis of observational individual patient-level cost-effectiveness data. As a motivating example, we assessed the cost-effectiveness of CABG versus PTCA – one year post procedure - in a cohort of individuals who received the intervention within 365 days of their index admission for AMI. The data used for this paper were obtained from the Ontario Myocardial Infarction Database (OMID), linking these with data from the Canadian Institute for Health Information (CIHI), the Ontario Health Insurance Plan (OHIP), the Ontario Drug Benefit (ODB) program, and Ontario Registered Persons Database (RPDB). We discuss three ways in which propensity score can be used to control for confounding in the estimation of average cost-effectiveness, and provide syntax codes for both propensity score matching and cost-effectiveness modelling.