Structural Uncertainty of Markov Models for Advanced Breast Cancer

Abstract

Objective. To examine the impact of structural uncertainty of Markov models in modeling cost-effectiveness for the treatment of advanced breast cancer (ABC). Methods. Four common Markov models for ABC were identified and examined. Markov models 1 and 2 have 4 health states (stable-disease, responding-to-therapy, disease-progressing, and death), and Markov models 3 and 4 only have 3 health states (stable-disease, disease-progressing, and death). In models 1 and 3, the possibility of death can occur in any health state, while in models 2 and 4, the chance of dying can only occur in the disease-progressing health state. A simulation was conducted to examine the impact of using different model structures on cost-effectiveness results in the context of a combination therapy of lapatinib and capecitabine for the treatment of HER2-positive ABC. Model averaging with an assumption of equal weights in all 4 models was used to account for structural uncertainty. Results. Markov model 3 yielded the lowest incremental cost-effectiveness ratio (ICER) of $303,909 per quality-adjusted life year (QALY), while Markov model 1 produced the highest ICER ($495,800/QALY). At a willingness-to-pay threshold of $150,000/QALY, the probabilities that the combination therapy is considered to be cost-effective for Markov models 1, 2, 3, and 4 were 14.5%, 14.1%, 21.6%, and 17.0%, respectively. When using model averaging to synthesize different model structures, the resulting ICER was $389,270/QALY. Conclusions. Our study shows that modeling ABC with different Markov model structures yielded a wide range of cost-effectiveness results, suggesting the need to investigate structural uncertainty in health economic evaluation. When applied in the context of HER2-positive ABC treatment, the combination therapy with lapatinib is not cost-effective, regardless of which model was used and whether uncertainties were accounted for.