Presentation (CTI)

OBJECTIVES: Individual patient-level data (IPD) from pivotal randomized controlled trials remain inaccessible for most health technology assessments, constraining reproducibility of model-based economic evaluations. This study evaluated the technical validity of the Guyot iterative reconstruction algorithm as a reproducible framework for cost-utility analysis, benchmarked against published trial-reported survival estimates.
METHODS: A partitioned survival model was developed from a healthcare payer perspective over a lifetime horizon using publicly available Kaplan-Meier curves from a published phase III RCT. Curves were digitized and reconstructed via the Guyot algorithm to generate pseudo-IPD for progression-free and overall survival. Parametric distributions (Weibull, log-normal, log-logistic, Gompertz) were selected using Akaike and Bayesian Information Criteria alongside clinical plausibility assessment. Validation compared reconstructed parametric predictions against published trial-reported survival probabilities at pre-specified landmark timepoints. Health state utilities and direct medical costs were sourced from published literature and national reimbursement schedules (2024 GBP), discounted at 3.5% annually per NICE reference case. Deterministic and probabilistic sensitivity analyses (PSA; 10,000 iterations) and cost-effectiveness acceptability curves (CEACs) characterized decision uncertainty.
RESULTS: Reconstruction validity was confirmed: parametric model predictions deviated from published trial-reported survival probabilities by a mean absolute error of 1.4% across all landmark timepoints. Survival extrapolation assumptions and drug acquisition costs were the primary uncertainty drivers in DSA. The base-case ICER was £22,800/QALY (incremental QALYs: 0.79; incremental cost: £18,000), falling within the NICE £20,000-£30,000/QALY threshold range. CEACs demonstrated 68% cost-effectiveness probability at £20,000/QALY, increasing to 86% at £30,000/QALY.
CONCLUSIONS: The Guyot reconstruction algorithm produces technically valid pseudo-IPD with under 1.5% deviation from published trial survival estimates, supporting application in HTA dossier development and early value assessment. This framework offers a transparent, auditable, and reproducible pathway for model-based economic evaluation aligned with ISPOR good research practices and NICE reference case requirements.