ASSESSING TIME-VARYING TREATMENT EFFECTS USING FRACTIONAL POLYNOMIAL NETWORK META-ANALYSIS OF SURVIVAL OUTCOMES
Author(s)
Rittwik Das, M.Sc.1, Udayraj Dnyaneshwar Patil, M. Pharm2, Biboshwan De, M.Sc.1, Aachal Shinde, M.Tech1, Varish Bhatia, PharmD1, Varun Ektare, MPH3, Manikanta Dasari, M.Tech (Pharma)1.
1Indence Research Private Limited, Kolkata, India, 2National Institute of Pharmaceutical Education and Research, Mohali, India, 3Indence Research Private Limited, Thane West, India.
1Indence Research Private Limited, Kolkata, India, 2National Institute of Pharmaceutical Education and Research, Mohali, India, 3Indence Research Private Limited, Thane West, India.
OBJECTIVES: Oncology network meta-analyses (NMAs) of survival summarize treatment effects using a single hazard ratio (HR), implicitly assuming proportional hazards (PH). When hazards vary over time, this may obscure time-specific changes in relative benefit and affect indirect comparisons. Using metastatic hormone-sensitive prostate cancer (mHSPC) as a case study, we applied fractional polynomial network meta-analysis (FP-NMA) to the same evidence base as a previous constant HR-based synthesis.
METHODS: An evidence network of seven mHSPC trials including 6,444 patients was constructed. Individual patient data (IPD) were reconstructed from published Kaplan-Meier curves. The PH assumption was tested for each trial. First-order fractional polynomial models were fitted to allow time-varying treatment effects and compared with standard parametric distributions using Akaike Information Criterion. Treatments were compared using time-specific HRs and model-predicted survival probabilities at 12, 24, 36, 48, and 60 months.
RESULTS: Formal diagnostics indicated PH violation in one trial. Log-normal and log-logistic models best fit 9 of 14 reconstructed curves. Among FP specifications, the log-time hazard model showed the most consistent fit and was used for synthesis. FP-NMA produced time-dependent, numerically attenuated estimates for darolutamide plus androgen deprivation therapy (ADT) plus docetaxel versus key comparators. Versus ADT alone, the HR changed from 0.90 at 12 months to 1.04 at 60 months, compared with 0.43 in the prior synthesis. Similar attenuation was observed versus abiraterone plus ADT (0.81 to 1.09; prior HR 0.65) and abiraterone acetate plus docetaxel plus ADT (0.69 to 1.17; prior HR 0.66). Uncertainty intervals crossed or approached unity.
CONCLUSIONS: FP-NMA provides a structured framework for assessing the robustness of survival evidence synthesis when PH assumptions are uncertain. In mHSPC, relaxing PH showed relative treatment effects may attenuate and vary by time horizon, suggesting constant HRs can obscure time-specific patterns. Interpretation is limited by reconstructed IPD and lack of population adjustment.
METHODS: An evidence network of seven mHSPC trials including 6,444 patients was constructed. Individual patient data (IPD) were reconstructed from published Kaplan-Meier curves. The PH assumption was tested for each trial. First-order fractional polynomial models were fitted to allow time-varying treatment effects and compared with standard parametric distributions using Akaike Information Criterion. Treatments were compared using time-specific HRs and model-predicted survival probabilities at 12, 24, 36, 48, and 60 months.
RESULTS: Formal diagnostics indicated PH violation in one trial. Log-normal and log-logistic models best fit 9 of 14 reconstructed curves. Among FP specifications, the log-time hazard model showed the most consistent fit and was used for synthesis. FP-NMA produced time-dependent, numerically attenuated estimates for darolutamide plus androgen deprivation therapy (ADT) plus docetaxel versus key comparators. Versus ADT alone, the HR changed from 0.90 at 12 months to 1.04 at 60 months, compared with 0.43 in the prior synthesis. Similar attenuation was observed versus abiraterone plus ADT (0.81 to 1.09; prior HR 0.65) and abiraterone acetate plus docetaxel plus ADT (0.69 to 1.17; prior HR 0.66). Uncertainty intervals crossed or approached unity.
CONCLUSIONS: FP-NMA provides a structured framework for assessing the robustness of survival evidence synthesis when PH assumptions are uncertain. In mHSPC, relaxing PH showed relative treatment effects may attenuate and vary by time horizon, suggesting constant HRs can obscure time-specific patterns. Interpretation is limited by reconstructed IPD and lack of population adjustment.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
SA84
Topic
Clinical Outcomes, Study Approaches
Topic Subcategory
Meta-Analysis & Indirect Comparisons
Disease
No Additional Disease & Conditions/Specialized Treatment Areas