Purpose: In AARTS, patients were randomized to MDOT or SOC, but at 3 months, patients with poor outcomes with SOC crossed over to MDOT. Marginal structural modeling has been used to minimize bias from time dependent co-factors. To estimate the benefit of MDOT in the absence of crossover from SOC to MDOT at 3 months, we developed marginal structural models and incorporated them into a Monte Carlo simulation.

Methods: The Monte Carlo simulation compared 6 months of MDOT with SOC. Long-term survival was based on a regression-based Weibull model. . Marginal structural modeling was used to estimate viral load, CD4 count and costs from the trial in the absence of SOC crossover from inverse-probability of treatment weights (IPTW) of baseline, 3-month and 6-month demographic and clinical characteristics. IPTW adjust for the probability of allocation of SOC patients to MDOT and that of censoring. Model covariates included age, gender and time-dependent measures of HIV resistance, CD4 counts, viral load and MDOT allocation at baseline and 3 months based on clinical relevance and statistical significance.

Results: In the base case analysis, the MDOT intervention was dominant over the SOC. In all one-way and multi-way analyses, six months of MDOT was dominant over SOC. The lifetime costs in US$ for the MDOT and SOC interventions were 50,654 respectively. The effectiveness in QALYs for the MDOT and SOC interventions was 13.17 and 12.51 respectively.

Conclusion: Six months MDOT was dominant over SOC, suggesting that MDOT for six months may have more long-term benefits than previously found for three months. The use of marginal structural modeling in regression-based Markov modeling in CEA is a practical and novel approach.