CM3 Multivariable Methods for Measuring Treatment Costs in Randomized Trials

As treatment costs are increasingly determined from individual level cost data, the number of proposed multivariable methods for use in this analysis has multiplied. These methods involve estimation of multivariable cost functions that yield predictions at the individual level, coditional on interventions, patient characteristics, and other factors. What are these methods, how are they used properly, and what are the circumstances when one method is preferred over others? The purpose of this workshop will be to develop skills in conducting multivariable analysis of cost data from randomized trials. We will instruct participants in the use of ordinary least squares regression techniques and survival analysis techniques. We will discuss how non-normal cost data and censored cost data are properly and improperly handled in these methods. Participants will learn when it is appropriate to use log transformation of costs in their analysis and how to estimate unbiased treatment costs using smearing techniques. Participants will also learn how to apply the Cox proportional hazard model to analysis of costs. How does one determine which model is best given the circumstances? We will develop concepts important for evaluating the superior model: predictive validity and adherence to assumptions for unbiased estimators. We will present results from a simulation designed to evaluate how well the various methods perform under different circumstances. Those who want to learn the techniques of multivariable cost analysis and develop criteria for choosing the best technique will benefit from this workshop. Participants who would benefit include analysts of cost data and those who want to increase their understanding of the literature of economic evaluation.