Bayesian Evidence Synthesis for Indirect Treatment Comparisons of Extrapolated Survival Outcomes Leveraging Historical Trial Data
Medical Decision Making · 04.11.2026
DOI: 10.1177/0272989x261493545 ↗Kurz gefasst
Fachartikel in der Zeitschrift Medical Decision Making, online erschienen am 04.11.2026. Verfasst von Daniel J. Sharpe und 3 weiteren Autor:innen. Die Arbeit nennt 92 Literaturangaben.
Abstract (Originalsprache)
Objectives
Flexible survival models can accommodate time-varying treatment effects in network meta-analyses (NMAs) but may yield biased estimates from naïve extrapolations when trial data have varying maturity. To enable more reliable comparison of long-term survival outcomes when there are differences in follow-up duration between studies, we proposed employing parametric spline NMA models leveraging historical trial data in a Bayesian multiparameter evidence synthesis (B-MPES) framework.
Methods
B-MPES NMA models were applied to reconstructed interim overall survival data from 4 randomized phase 3 trials of first-line immunotherapy plus tyrosine kinase inhibitor (IO+TKI) versus sunitinib in metastatic renal cell carcinoma, for which median follow-up duration ranged between 18.1 and 30.6 mo. Trial observations were supplemented with conditional survival estimates from a historical trial of dual IO therapy versus sunitinib with a median follow-up of 55 mo. An alternative B-MPES model was informed by older historical trial data for sunitinib and a hazard ratio condition. Treatment effects were quantified by 5-y restricted mean survival time. B-MPES model performance was compared with a naïve proportional hazards (PH) spline model.
Results
Leveraging historical trial data via B-MPES stabilized non-PH models that captured the diminishing treatment effects, whereas naive PH models were biased in favor of the IO+TKI therapy for which data were least mature, and naïve non-PH models yielded spurious and highly uncertain predictions. Relative effect estimates and treatment rankings were only moderately sensitive to the choice of historical trial data, while absolute survival probabilities exhibited high sensitivity.
Conclusions
B-MPES survival models informed by suitable external data provide an approach to estimate long-term relative effectiveness in a network of treatments when there are nonproportional hazards and data from some studies remain immature.
Zitieren
APA
Sharpe, D. J., Tate, A. E., De, T., & Vanderpuye-Orgle, J. (2026). Bayesian Evidence Synthesis for Indirect Treatment Comparisons of Extrapolated Survival Outcomes Leveraging Historical Trial Data. Medical Decision Making. https://doi.org/10.1177/0272989x261493545
Vancouver
Sharpe DJ, Tate AE, De T, Vanderpuye-Orgle J. Bayesian Evidence Synthesis for Indirect Treatment Comparisons of Extrapolated Survival Outcomes Leveraging Historical Trial Data. Medical Decision Making. 2026. doi:10.1177/0272989x261493545
BibTeX
@article{Sharpe2026,
author = {Sharpe, Daniel J. and Tate, Ashley E. and De, Tuli and Vanderpuye-Orgle, Jackie},
title = {Bayesian Evidence Synthesis for Indirect Treatment Comparisons of Extrapolated Survival Outcomes Leveraging Historical Trial Data},
journal = {Medical Decision Making},
year = {2026},
doi = {10.1177/0272989x261493545}
}
EUVIMED