OBJECTIVES : To compare different indirect comparison methods of proportional hazard model used in Markov model.

METHODS : A 3-state Markov model was constructed to estimate the cost-effectiveness of fruquintinib vs. regorafenib treated for metastatic colorectal cancer. Costs and utilities were obtained from published data. Weibull-distribution functions were applied to calculate transition probabilities using the data from clinical trials FRESCO (fruquintinib vs. placebo) and CONCUR (regorafenib vs. placebo) with 4 indirect comparison methods. In method 1, survival data of fruquintinib and regorafenib was compared directly. In method 2, survival data from FRESCO was used to calculate transition probabilities of fruquintinib and placebo, and transition probabilities of regorafenib were calculated by transition probabilities of placebo with hazard ratio (HR) of regorafenib vs. placebo. In method 3, transition probabilities of regorafenib were calculated by transition probabilities of fruquintinib with HR of fruquintinib vs regorafenib, which was gained from a network-meta analysis. In method 4, survival data of placebo from FRESCO and CONCUR was pooled, and transition probabilities of fruquintinib and regorafenib were calculated by pooled data of placebo with HRs from FRESCO and CONCUR.

RESULTS : In method 1, fruquintinib provided more QALYs of 0.132 with less cost than regorafenib. In method 2, regorafenib provided more QALYs of 0.057 with additional costs of CNY 73,531.85, the ICER was CNY 1,293,590.32/QALY. In method 3, regorafenib provided more QALYs of 0.072 with additional costs of CNY 75,724.16, the ICER was CNY 1,046,151.61/QALY. In method 4, regorafenib provided more QALYs of 0.062 with additional costs of CNY 57,799.96, the ICER was CNY 934,333.19/QALY. Sensitivity analyses demonstrated that all results were robust.

CONCLUSIONS : Except method 1, the results of the other 3 methods were similar. Indirect treatment comparisons of survival data in Markov model were necessary and indirect comparison methods should be chosen according to the actual conditions.