How Much Can a Crossover Blur the Bottomline? A Montecarlo Evaluation of Five Adjustment Methods in Oncology Cost-Effectiveness Analyses
Author(s)
Fenghao SHI, II, MSc1, SHENG HAN2.
1International Research Center for Medicinal Administration, Peking University, Beijing, China, 2International Research Center of Medicinal Administration. Peking Universit, Beijing, China.
1International Research Center for Medicinal Administration, Peking University, Beijing, China, 2International Research Center of Medicinal Administration. Peking Universit, Beijing, China.
OBJECTIVES: To quantify how treatment-switching adjustment methods alter cost-effectiveness results under Chinese reimbursement conditions.
METHODS: Phase-III oncology trials (N = 600; 36-month follow-up) were simulated across 16 factorial scenarios varying effect size (HR 0.65/0.80), crossover rate (40 %/75 %), time-dependent confounding (yes/no) and sample size (600/1 200). “True” incremental outcomes without crossover were 0.40 QALYs and ¥120 000. Five approaches—no adjustment (ITT), simple two-stage (TSE-simp), G-estimated two-stage (TSE-gest), rank-preserving structural failure-time model (RPSFTM) and inverse-probability-of-censoring weights (IPCW)—were applied to 1 000 replicates per scenario. Adjusted survival curves fed a lifetime partitioned-survival model (1-month cycle, 5 % discount). Costs used 2024 negotiated national prices; utilities were 0.83 (PFS) and 0.71 (PD). Outcomes were mean ICERs (¥/QALY), bias versus the no-crossover truth, cost-effectiveness probability at ¥90 000/QALY, and expected value of perfect information (EVPI) attributed to method choice.
RESULTS: With 75 % crossover and confounding, ITT overstated benefit by 30 % (ΔQALY = 0.52, ICER = ¥36 800) relative to the truth (0.40 QALYs, ¥47 500). Bias dropped to +12 % (TSE-simp), ±0 % (IPCW), -5 % (TSE-gest) and -8 % (RPSFTM). Across all scenarios, mean absolute QALY bias ranked: RPSFTM 2 % < TSE-gest 4 % < IPCW 9 % < TSE-simp 11 % < ITT 24 %. Decision reversals versus the truth occurred in 2 % (RPSFTM) to 16 % (ITT) of simulations. Method uncertainty contributed 9-41 % of total EVPI.
CONCLUSIONS: Under Chinese prices and thresholds, neglecting crossover can markedly skew ICERs. RPSFTM and TSE-gest most closely recovered true value, whereas TSE-simp and IPCW were more assumption-sensitive. HTA submissions should present multiple adjustments and propagate method uncertainty in probabilistic analyses.
METHODS: Phase-III oncology trials (N = 600; 36-month follow-up) were simulated across 16 factorial scenarios varying effect size (HR 0.65/0.80), crossover rate (40 %/75 %), time-dependent confounding (yes/no) and sample size (600/1 200). “True” incremental outcomes without crossover were 0.40 QALYs and ¥120 000. Five approaches—no adjustment (ITT), simple two-stage (TSE-simp), G-estimated two-stage (TSE-gest), rank-preserving structural failure-time model (RPSFTM) and inverse-probability-of-censoring weights (IPCW)—were applied to 1 000 replicates per scenario. Adjusted survival curves fed a lifetime partitioned-survival model (1-month cycle, 5 % discount). Costs used 2024 negotiated national prices; utilities were 0.83 (PFS) and 0.71 (PD). Outcomes were mean ICERs (¥/QALY), bias versus the no-crossover truth, cost-effectiveness probability at ¥90 000/QALY, and expected value of perfect information (EVPI) attributed to method choice.
RESULTS: With 75 % crossover and confounding, ITT overstated benefit by 30 % (ΔQALY = 0.52, ICER = ¥36 800) relative to the truth (0.40 QALYs, ¥47 500). Bias dropped to +12 % (TSE-simp), ±0 % (IPCW), -5 % (TSE-gest) and -8 % (RPSFTM). Across all scenarios, mean absolute QALY bias ranked: RPSFTM 2 % < TSE-gest 4 % < IPCW 9 % < TSE-simp 11 % < ITT 24 %. Decision reversals versus the truth occurred in 2 % (RPSFTM) to 16 % (ITT) of simulations. Method uncertainty contributed 9-41 % of total EVPI.
CONCLUSIONS: Under Chinese prices and thresholds, neglecting crossover can markedly skew ICERs. RPSFTM and TSE-gest most closely recovered true value, whereas TSE-simp and IPCW were more assumption-sensitive. HTA submissions should present multiple adjustments and propagate method uncertainty in probabilistic analyses.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR123
Topic
Economic Evaluation, Health Technology Assessment, Methodological & Statistical Research
Disease
Oncology