Abstract
PURPOSE: Patients with advanced cancer may undergo multiple lines of treatment, switching therapies as their disease progresses. We developed a general microsimulation framework to study therapy sequence and applied it to metastatic prostate cancer. METHODS: We constructed a discrete-time state transition model to study 2 lines of therapy. Using digitized published survival curves (progression-free survival, time to progression, and overall survival [OS]), we inferred event types (progression or death) and estimated transition probabilities using cumulative incidence functions with competing risks. We incorporated within-patient dependence over time; first-line therapy response informed subsequent event probabilities. Parameters governing within-patient dependence calibrated the model-based results to a target clinical trial. We applied these methods to 2 therapy sequences for metastatic prostate cancer, wherein both docetaxel (DCT) and abiraterone acetate (AA) are appropriate for either first- or second-line treatment. We assessed costs and quality-adjusted life-years (5-y QALYs) for 2 treatment strategies: DCT → AA versus AA → DCT. RESULTS: Models assuming within-patient independence overestimated OS time, which corrected with the calibration approach. With generic pricing, AA → DCT dominated DCT → AA, (higher 5-y QALYs and lower costs), consistent for all values of calibration parameters (including no correction). Model calibration increased the difference in 5-y QALYs between treatment strategies (0.07 uncorrected v. 0.15 with base-case correction). Applying the correction decreased the estimated difference in cost (-$5,360 uncorrected v. -$3,066 corrected). Results were strongly affected by the cost of AA. Under a lifetime horizon, AA → DCT was no longer dominant but still cost-effective (incremental cost-effectiveness ratio: $19,463). CONCLUSIONS: We demonstrate a microsimulation approach to study the cost-effectiveness of therapy sequences for advanced prostate cancer, taking care to account for within-patient dependence. HIGHLIGHTS: We developed a discrete-time state transition model for studying therapy sequence in advanced cancers.Results are sensitive to dependence within patients.A calibration approach can introduce dependence across lines of therapy and closely match simulation outcomes to target trial outcomes.