OBJECTIVES: We previously presented evidence suggesting that clinical trial learning curves may affect clinical outcomes in patients in drug trials. In the current analysis, we demonstrate the potential effect of learning curves on economic outcomes (specifically, cost-effectiveness).
METHODS: The PROWESS trial, which evaluated drotrecogin alpha (DrotAA) for severe sepsis, was identified in our previous study and was chosen for further analysis based on several considerations: a published analysis suggested that a clinical trial learning curve may have had a substantial effect on outcomes in a subgroup of patients (APACHE II 25); and a published cost-effectiveness analysis (which did not account for the learning curve effect) was transparent and easily replicable. Furthermore, a health technology appraisal (HTA) of DrotAA conducted in the UK cited the cost-effectiveness analyses, which suggested that the incremental cost per quality-adjusted life year for patients with APACHE II scores 25 was US$400,000. Similarly, an Australian reimbursement decision excluded this patient subpopulation from coverage citing unacceptable cost-effectiveness. We replicated the cost-effectiveness analysis for DrotAA, and used it to model the cost-effectiveness of DrotAA in the subgroup of patients with APACHE II25, both with and without the patients enrolled earlier in the trial and thus potentially affected by the learning curve.
RESULTS: When patients who may have been affected by the trial learning curve were excluded from the analysis, cost-effectiveness of DrotAA improved significantly, from US$411,333 per LYG with all patients with APACHE II score 25 to US$46,395 per LYG when the first block of patients enrolled at each site was removed from the analysis.
CONCLUSIONS: Clinical trial learning curves potentially affect both clinical and economic outcomes, and impact reimbursement decisions. Consideration of learning curves may be important in HTAs and reimbursement decisions, particularly when evaluating trial data in which learning curves are more likely to be present.
