Presentation (CTI)

OBJECTIVES: In health technology assessment (HTA) submissions, partitioned survival models are frequently implemented using Monte Carlo simulation. A key challenge is that incremental quality-adjusted life-years (QALY) often exhibit high sampling variance, requiring large simulation sizes to achieve convergence. This study evaluated the use of common random numbers (CRN) as a variance reduction technique in partitioned survival models, focusing on appropriate application for survival outcomes.
METHODS: We conducted a simulation analysis based on a cohort of >900 patients using a partitioned survival model. Outcomes were generated under two scenarios: (1) independent sampling between treatment arms (no CRN) and (2) aligned sampling using CRN, where the same random draws informed both arms. Variance of incremental QALYs, confidence intervals (CI), and convergence rates was compared. Justification for CRN use emphasized a patient-level perspective: both arms represented the same hypothetical cohort, allowing CRN to isolate treatment effects from random heterogeneity. CRN is not appropriate when interventions target independent populations where shared randomness would impose artificial correlation.
RESULTS: In the no CRN scenario, incremental QALYs yielded a 95% CI of 0.66, 3.46, variance of 0.53, and convergence at approximately 120 simulations. With CRN applied, the 95% CI narrowed to 1.34, 2.89, variance decreased to 0.15, and convergence occurred at around 10 simulations. This represented a variance reduction factor of 3.49, with convergence achieved in just 8% of the iterations required in the non-CRN case.
CONCLUSIONS: CRN markedly improved simulation efficiency in partitioned survival models by reducing variance and accelerating convergence. While results are specific to this example, the pattern of improved precision is generalizable to other survival-based HTA models when both arms represent the same patient cohort. Where comparators instead reflect independent populations, CRN is not justified. For HTA submissions, careful use of CRN enhances transparency and computational efficiency while preserving the integrity of survival-based outcomes.