OBJECTIVES: In cost-effectiveness literature, several Markov models with different health states and assumptions are used to estimate cost-effectiveness for treatment of advanced breast cancer (ABC). This study aims to (1) derive the relationships among the Markov models for ABC, and (2) examine the impact of using different Markov models on cost-effectiveness results for treatment of HER2-positive advanced breast cancer. METHODS: There are four Markov models for ABC: (i) model 1 includes 4 health states (stable, response, progression, and dead) with a possibility of death can occur in all health states; (ii) model 2 also consists of 4 health states but the chance of dying can only occur at disease-progression state; (iii) model 3 is the most common model and has 3 health states (stable, progression, and dead) with a possibility of death can occur in all health states; and (iv) model 4 also has 3 health states but the chance of death can only occur at disease-progression state. Relationships of transition probabilities among the Markov models were derived. A simulation method was used to generate 10,000 samples with transition probabilities, estimated costs and utilities for each health state based on a previously published cost-effectiveness study of lapatinib in treatment of HER2-positive advanced breast cancer. RESULTS: Markov models 2 and 4 yielded similar and lowest incremental cost-effectiveness ratios (ICER), while the commonly used Markov model (model 3) produced the highest ICER. All four Markov models produced ICERs > $150,000 per additional quality-adjusted life year gained for the combination therapy with lapatinib as compared to monotherapy with capecitabine. CONCLUSIONS: This study suggests that modeling advanced breast cancer with different health-state Markov models may produce different cost-effectiveness results. Combination therapy with lapatinib in treatment of HER2-positive advanced breast cancer is not a cost-effective treatment strategy regardless which Markov models used.