OBJECTIVES : Potentially curative therapies pose unique challenges for cost-effectiveness analysis, including estimating long-term survival from Kaplan-Meier curves, as clinical interest in these therapies often outpaces the maturity of survival data. Previous work has shown standard parametric modeling may underestimate long-term survival when flattening of the curve occurs. Mixture cure models have been proposed as one solution, but require disaggregated data. Our objectives were to develop an approach to account for flattening observed in aggregated Kaplan-Meier data for CAR-T therapies, and compare to survival and cost-effectiveness estimates using standard parametric approaches.

METHODS : We fit a parametric curve function on the portion of the Kaplan-Meier curve that had a negative slope. When the parametric curve intercepted the flattening of the survival curve (i.e. slope of zero), a knot was introduced. After the knot, mortality was restricted to general population rates. Cost-effectiveness estimates of tisagenlecleucel and axicabtagene ciloleucel (FDA approved CAR-T therapies) were calculated using this modified parametric modeling approach and compared to estimates generated using standard parametric modeling.

RESULTS : When modeling long-term survival using the modified parametric approach, tisagenlecleucel resulted in 10.34 life years [9.28 quality-adjusted life years (QALYs)] and axicabtagene ciloleucel resulted in 7.35 life years (5.87 QALYs). When compared to chemotherapy, the incremental cost-effectiveness ratio was approximately $57,000 per QALY gained for tisagenlecleucel and $140,000 per QALY gained for axicabtagene ciloleucel. Using standard parametric modeling, tisagenlecleucel resulted in 5.15 life years (4.49 QALYs) and axicabtagene ciloleucel resulted in 3.17 life years (2.19 QALYs). The incremental cost-effectiveness ratios increased to $98,000 per QALY gained for tisagenlecleucel and $267,000 per QALY gained for axicabtagene ciloleucel.

CONCLUSIONS : Without patient-level data, cost-effectiveness analyses for potentially curative therapies should use a modified parametric approach to estimate long-term survival. Standard parametric modeling could be a lower bound for survival scenario analysis to inform uncertainty in the duration of curve flattening.