Statistical analysis of cost–effectiveness data from randomized clinical trials

Since the mid-1990s, motivated by the availability of patient-level cost data in randomized clinical trials, there has been rapid development in the statistical methods for analyzing cost-effectiveness data. Initial efforts concentrated on inference about the incremental cost-effectiveness ratio (ICER), but due to difficulties associated with ratio statistics, interest has settled more recently on incremental net benefit (INB). Regardless of the approach, five parameters need to be estimated: the between-treatment arm differences in mean effectiveness and mean cost, and the corresponding variances and covariance. With the estimates of these parameters, the analyst can plot the cost-effectiveness acceptability curve and estimate the ICER and the INB, and calculate confidence limits for both. A review of these methods is given. The particular statistical procedure used for estimating the five parameters depends on whether or not censoring is present; whether or not covariates are adjusted for; whether or not random effects, such as country, are adjusted for; and the assumptions regarding the distribution for cost. A review of the statistical procedures, particular to each combination of these conditions, is given, where they exist.