138
Using Nested Markov Models to Model the Budget Impact of Treatment Sequences in Oncology: A Worked Example in Hepatocellular Carcinoma
Markov models are widely used to evaluate the cost-utility of oncology treatments. Three-state Markov model structures are commonly adopted to capture transitions between stable disease, disease progression, and death. Typically, transition probabilities between states are calculated using parametric models based on Kaplan-Meier (KM) data. Despite the substantial economic implications of treatment choices after failure of first-line therapy, much less sophisticated methodologies are typically employed in modeling the budget impact of transitions between subsequent treatment lines.
We developed a nested Markov model to simultaneously model health state and treatment line transitions. To illustrate the approach, we used KM data from the SARAH randomized controlled trial (RCT) comparing selective internal radiation therapy (SIRT) using SIR-Spheres® Y-90 resin microspheres with sorafenib in patients with hepatocellular carcinoma (HCC). The objective was to simultaneously capture disease progression and transition to subsequent treatments including regorafenib, lenvatinib and best supportive care (BSC).
Methods
A budget impact model was developed in Microsoft Excel using a partitioned survival, area-under-the-curve cohort simulation approach with three Markov states at the top level: progression-free survival (PFS), post-progression survival (PPS) and death. KM data for PFS and overall survival were obtained from the SARAH RCT and used to fit parametric curves, which informed the transition probabilities. A nested Markov model was then implemented within the PPS state to model transitions from sorafenib and SIRT to regorafenib, lenvatinib, and BSC. A 3-year time horizon was adopted with a 28-day cycle length.
Data from the per-protocol population of SARAH were used, including only patients receiving the treatments to which they were randomized. Utilization of subsequent treatments was informed by expert opinion, while median time on treatment was informed by published data.
Results
Using a nested Markov approach yielded modeled projections of both time in the PFS and PPS states, but also the average time on each treatment and average time at the point of initiating subsequent treatment lines. In the example analysis, in patients starting treatment with sorafenib, regorafenib use was highest after 24 weeks, while BSC was highest after 48 weeks. The absolute modeled peak use of regorafenib was 35% lower in the SIRT arm than in the sorafenib arm. Systemic therapy costs were projected to be EUR 37.6m over 3 years without SIRT, relative to EUR 24.9m with SIRT in a population of 1,533 SIRT-eligible patients with HCC in France. Although SIRT was also associated with implantable device, hospital and intervention room costs, the reduction in systemic therapy acquisition costs was projected to offset 93.6% of these additional costs, and SIRT was cost saving overall when also capturing adverse event and drug administration costs.
Conclusions
Modeling treatment lines after failure of first-line therapy can have a substantial impact on the findings of budget impact analyses in oncology. The present study employed a novel nested Markov modeling technique to estimate the budgetary implications of both disease progression and subsequent treatment lines, with the latter contributing an overwhelming majority of the cost offsets in an analysis comparing SIRT with sorafenib in patients with HCC.
