How to deal with cost differences at baseline

Background: To our knowledge, adjustment for baseline imbalances in costs has never been performed in trial-based cost-effectiveness analyses. Methods: We used data from a clinical trial performed in the Netherlands comparing two outpatient psycho therapies: schema-focused therapy (SFT) versus transference-focused psychotherapy (TFP). Costs were assessed with a cost interview. Outcome was the proportion of recovered patients measured with the Borderline Personality Disorder Severity Index (BPDSI-IV). We used three methods to adjust the costs for baseline differences: (i) mean difference adjustment, calculating total costs after baseline by adjusting the difference between groups with the difference of the mean baseline costs, (ii) delta adjustment, calculating the individual differences between patient baseline and the subsequent measurements (concerning incremental costs, this is the same as mean difference adjustment); and (iii) regression-based adjustment, adjusting total costs with a regression model, with total costs as the dependent variable and baseline costs as the independent variable. Results: Mean baseline costs were (sic)3339 for SFT and (sic)4238 for TFP, a mean difference of (sic)899. Total unadjusted follow-up costs were (sic)30822 for SFT and (sic)36812 for TFP. The fraction of recovered patients was 45% for SFT and 24% for TFP. Cost-effectiveness acceptability curves show that mean difference and delta adjustments are different from the regression-based methods. Conclusions: Although the routine starting point of an analysis should always be an unadjusted analysis of the cost effectiveness, a baseline difference between treatment groups should be adjusted for. This should be done by reported patient characteristics or, when these are not sufficiently present, by baseline costs as a substitute. This adjustment should be carried out most preferably with a regression-based method.