HEALTH TECHNOLOGY ASSESSMENT RELEVANT VALIDATION OF PROGRESSION-FREE SURVIVAL AS A SURROGATE FOR OVERALL SURVIVAL IN FIRST-LINE METASTATIC TRIPLE-NEGATIVE BREAST CANCER: A BAYESIAN META-ANALYTIC FRAMEWORK
Author(s)
Anandaroop Dasgupta, PhD1, Ankita Kaushik, PhD1, Paul Serafini, BA2, Gaurang Nazar, MD, PhD2, Divya Pushkarna, B. Tech3.
1Gilead Sciences, Foster City, CA, USA, 2Evidinno Outcomes Research Inc, Vancouver, BC, Canada, 3Evidinno Outcomes Research Inc., Vancouver, BC, Canada.
1Gilead Sciences, Foster City, CA, USA, 2Evidinno Outcomes Research Inc, Vancouver, BC, Canada, 3Evidinno Outcomes Research Inc., Vancouver, BC, Canada.
OBJECTIVES: In advanced breast cancer, overall survival (OS) is often confounded by treatment crossover and subsequent therapies, reducing its reliability for health technology assessment (HTA), reimbursement, and cost-effectiveness modeling. This study validates progression-free survival (PFS) as a surrogate endpoint for OS, using methods that account for uncertainty and heterogeneity.
METHODS: A systematic literature review (November 2024) identified randomized and non-randomized studies in first-line metastatic triple-negative breast cancer (mTNBC). Study-level surrogacy was assessed using a Bayesian trivariate random-effects correlation meta-analysis, jointly modeling log transformed HROS and HRPFS, and PD-L1+ fraction. Posterior estimates were obtained via Markov Chain Monte Carlo. Surrogacy was evaluated using correlation estimates, a regression-based surrogacy function, and the surrogate threshold effect (STE), defined as the HRPFS at which the upper bound of the predicted HROS credible interval equals 1. Predictive validity was assessed via Bayesian leave-one-out cross-validation (LOOCV), targeting ≥95% coverage. Individual-level associations were estimated using reconstructed pseudo-patient-level data from Kaplan-Meier curves, with Pearson’s r and non-parametric metrics. Sensitivity analyses evaluated robustness to crossover, study design, and proportional hazards assumptions.
RESULTS: Fifty-two studies (56 comparisons) were included. A strong correlation between log transformed HROS and HRPFS was observed (r=0.87; 95% CrI: 0.71-0.94). The regression slope (~0.95) indicated near-proportional treatment effects. The STE was 0.81, suggesting HRPFS <0.81 predicts significant OS benefit. LOOCV demonstrated 100% predictive coverage, supporting model calibration. Individual-level correlations were moderate (r≈0.51-0.63) and consistent across analyses.
CONCLUSIONS: Bayesian meta-analysis supports PFS as a validated surrogate for OS in first-line mTNBC. Strong association, credible STE, and high predictive performance strengthen its applicability in HTA, enabling earlier assessment of clinical and economic value when OS data are immature or biased.
METHODS: A systematic literature review (November 2024) identified randomized and non-randomized studies in first-line metastatic triple-negative breast cancer (mTNBC). Study-level surrogacy was assessed using a Bayesian trivariate random-effects correlation meta-analysis, jointly modeling log transformed HROS and HRPFS, and PD-L1+ fraction. Posterior estimates were obtained via Markov Chain Monte Carlo. Surrogacy was evaluated using correlation estimates, a regression-based surrogacy function, and the surrogate threshold effect (STE), defined as the HRPFS at which the upper bound of the predicted HROS credible interval equals 1. Predictive validity was assessed via Bayesian leave-one-out cross-validation (LOOCV), targeting ≥95% coverage. Individual-level associations were estimated using reconstructed pseudo-patient-level data from Kaplan-Meier curves, with Pearson’s r and non-parametric metrics. Sensitivity analyses evaluated robustness to crossover, study design, and proportional hazards assumptions.
RESULTS: Fifty-two studies (56 comparisons) were included. A strong correlation between log transformed HROS and HRPFS was observed (r=0.87; 95% CrI: 0.71-0.94). The regression slope (~0.95) indicated near-proportional treatment effects. The STE was 0.81, suggesting HRPFS <0.81 predicts significant OS benefit. LOOCV demonstrated 100% predictive coverage, supporting model calibration. Individual-level correlations were moderate (r≈0.51-0.63) and consistent across analyses.
CONCLUSIONS: Bayesian meta-analysis supports PFS as a validated surrogate for OS in first-line mTNBC. Strong association, credible STE, and high predictive performance strengthen its applicability in HTA, enabling earlier assessment of clinical and economic value when OS data are immature or biased.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
HTA337
Topic
Clinical Outcomes, Health Technology Assessment, Methodological & Statistical Research
Topic Subcategory
Decision & Deliberative Processes
Disease
Oncology