For trial-based economic evaluation where patient-specific cost data are not routinely available, cost prediction models are commonly used to estimate total cost for each patient. Typically, multiple regression techniques are used on data from diagnosis-matched, non-trial patients (where patient-level cost data are available) to model cost as a function of covariates that are observed on the trial subjects (e.g. length of hospital stay, procedures, etc.). The estimated beta coefficients provide a means of estimating the total cost for each patient in the trial. However, the variability of the beta coefficients due the measurement and sampling error is seldom included in the overall variance expression for mean costs by treatment group. In this paper we provide a method for estimating this variance and provide an example application
Copyright 2001 John Wiley & Sons, Ltd.