OBJECTIVES: Population adjustment methods are becoming increasingly popular in health technology assessments, particularly with single-arm trials. This research compares the performance of these methods in anchored and unanchored settings.
METHODS: Individual patient-level data were obtained from the COU-AA-301 trial (abiraterone versus placebo) and compared with summary data from the AFFIRM trial (enzalutamide versus placebo) in patients with metastatic castration resistant prostate cancer. Prognostic factors, and treatment-effect modifiers were investigated in the COU-AA-301 trial. The models fitted were unadjusted indirect comparison (anchored), naïve comparison (unanchored), matching-adjusted indirect comparison (MAIC), simulated treatment comparison (STC), and multiple imputation marginalization (MIM). Anchored and unanchored methods (which disregarded the placebo arms) were fitted.
RESULTS: There was little evidence of treatment effect modifiers in the COU-AA-301 trial. Out of 17 prognostic factors identified through desktop research, 14 were reported by both trials and were included in the population adjustments; the remaining 3 factors were reported to have relatively low importance. For the unadjusted anchored indirect comparison, the hazard ratio for abiraterone versus enzalutamide was 1.056 (95% confidence intervals (ci): 0.827, 1.347), MAIC: 1.089 (95% ci: 0.767, 1.546), STC: 0.911 (95% ci: 0.650, 1.277), and MIM: 1.059 (95% ci: 0.831, 1.349). For the unanchored indirect comparison, the hazard ratio for abiraterone versus enzalutamide from a naïve comparison was 1.371 (95% ci: 1.171, 1.606), MAIC: 1.557 (95% ci: 1.284, 1.888), STC: 1.457 (95% ci: 1.248, 1.701), and MIM: 1.508 (95% ci: 1.288, 1.764).
CONCLUSION: Assuming there were no treatment effect modifiers present, the performance of MIM was favourable in the anchored setting with hazard ratios and confidence intervals matching those from the unadjusted anchored comparison. In the unanchored setting, neither MAIC, STC, nor MIM were able to improve upon the naïve comparison, with all hazard ratios significantly different to those from the point estimates in the anchored setting.
