COMPARING THREE- AND FOUR-HEALTH-STATE SEMI-MARKOV MODELS DERIVED FROM A COMMON DISCRETE-EVENT SIMULATION IN EARLY-STAGE ONCOLOGY
Author(s)
Benjamin White, MSc1, Sandra Milev, MSc2;
1Red Nucleus, HEOR, Yardley, PA, USA, 2Red Nucleus, HEOR, Walnut Creek, PA, USA
1Red Nucleus, HEOR, Yardley, PA, USA, 2Red Nucleus, HEOR, Walnut Creek, PA, USA
OBJECTIVES: Cohort-based semi-Markov models in oncology often differ in structural complexity, particularly in how post-progression disease states are represented. This study compared outcomes from three-health-state (3HS) and four-health-state (4HS) semi-Markov models derived from the same underlying patient-level data, using a discrete-event simulation (DES) as a common ground-truth generator.
METHODS: A lifetime DES generated patient-level event histories across event-free (EF), locoregional recurrence (LR), distant metastasis (DM), and death for two treatments. These data were used to parameterize a 3HS semi-Markov model (EF, progressed disease [PD], death) and a 4HS model (EF, LR, DM, death). One-off treatment costs were applied using DES-derived mean time on treatment. Outcomes included life-years, state occupancy, costs, and incremental effects.
RESULTS: In the base case analysis, the models showed close agreement. The model estimates of lifetime incremental LYs, QALYs, and costs were 0.99, 0.88, and $25,316 for the 4HS model and 0.99, 0.88, and $25,304 for the 3HS model. These compared closely with DES estimates of 1.00, 0.88, and $26,433. One-way sensitivity analysis revealed deviations between the 3HS and 4HS models of less than 5% across all parameters. Scenario analysis tested progression patterns in which the split between LR and DM following EF varied substantially over time (e.g., early-slow and late-fast EF to LR with the opposite for EF to DM). These scenarios showed larger differences between the two models, with the 3HS model tracking DES outputs more closely. Additional scenarios showed limited impact on outcomes, including a functional cure component in EF and use of median rather than mean time on treatment.
CONCLUSIONS: When derived from a common data source, 3HS and 4HS semi-Markov models rarely yielded meaningfully different results due structural choice. However, differences may emerge under alternative progression schemes (particular to the underlying disease), underscoring the importance of evaluating structural uncertainty when selecting model structure.
METHODS: A lifetime DES generated patient-level event histories across event-free (EF), locoregional recurrence (LR), distant metastasis (DM), and death for two treatments. These data were used to parameterize a 3HS semi-Markov model (EF, progressed disease [PD], death) and a 4HS model (EF, LR, DM, death). One-off treatment costs were applied using DES-derived mean time on treatment. Outcomes included life-years, state occupancy, costs, and incremental effects.
RESULTS: In the base case analysis, the models showed close agreement. The model estimates of lifetime incremental LYs, QALYs, and costs were 0.99, 0.88, and $25,316 for the 4HS model and 0.99, 0.88, and $25,304 for the 3HS model. These compared closely with DES estimates of 1.00, 0.88, and $26,433. One-way sensitivity analysis revealed deviations between the 3HS and 4HS models of less than 5% across all parameters. Scenario analysis tested progression patterns in which the split between LR and DM following EF varied substantially over time (e.g., early-slow and late-fast EF to LR with the opposite for EF to DM). These scenarios showed larger differences between the two models, with the 3HS model tracking DES outputs more closely. Additional scenarios showed limited impact on outcomes, including a functional cure component in EF and use of median rather than mean time on treatment.
CONCLUSIONS: When derived from a common data source, 3HS and 4HS semi-Markov models rarely yielded meaningfully different results due structural choice. However, differences may emerge under alternative progression schemes (particular to the underlying disease), underscoring the importance of evaluating structural uncertainty when selecting model structure.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
P40
Topic
Study Approaches
Topic Subcategory
Decision Modeling & Simulation
Disease
SDC: Oncology