EXPLORATORY APPLICATION OF BIVARIATE NETWORK META-ANALYSIS TO PREDICT MISSING HAZARD RATIOS IN MCRPC USING SIMULATED CORRELATION DATA
Author(s)
Vikalp Maheshwari, B.Tech, MBA1, Abhiroop Chakravarty, BSc, MSc1, Urjashwal Vidhata, BSc, MSc1, Aakanksha Jaiswal, BSc, MSc2, Sudarshan Tewary, BSc, MSc2, Jackie Vanderpuye-Orgle, MSc, PhD3;
1Parexel International, Hyderabad, India, 2Parexel International, Bengaluru, India, 3Parexel International, Billerica, MA, USA
1Parexel International, Hyderabad, India, 2Parexel International, Bengaluru, India, 3Parexel International, Billerica, MA, USA
OBJECTIVES: Bivariate network meta-analysis (BvNMA) analyzes two correlated outcomes, leveraging their correlation structure to enhance precision, borrow strength across endpoints, and enable evidence synthesis when data are partially missing. This study applies BvNMA by using simulated data on metastatic castration-resistant prostate cancer (mCRPC) patients to simultaneously model overall survival (OS) and radiographic progression-free survival (rPFS) and predict hazard ratios despite partially missing data.
METHODS: A systematic literature review identified 19 published clinical trials that met the PICOS criteria for mCRPC. Five studies did not report HRs for OS and two did not report HRs for rPFS/PFS. For several studies, HRs were derived by digitizing Kaplan-Meier curves and generating pseudo-individual patient data using Guyot’s algorithm. The published data was supplemented by simulated patient-level data in estimating the correlation between OS and rPFS hazards via bootstrap resampling and Gaussian copula modeling. This correlation informed a BvNMA to predict missing HRs and reduce uncertainty. A univariate random effect NMA (RE NMA) was also performed for comparison. All results are considered exploratory for methodological purposes.
RESULTS: BvNMA successfully predicted HRs for comparisons where published data were unavailable and substantially narrowed credible intervals (CrI) compared to RE NMA. For example, in one comparison where OS HR was missing, RE NMA produced a wide CrI (e.g., 0.0127-94.86), whereas BvNMA estimated a more precise interval (e.g., 0.38-2.39). Similar improvements were observed for rPFS estimates. These findings demonstrate the methodological advantage of incorporating correlated outcomes in NMA.
CONCLUSIONS: BvNMA leveraging OS-rPFS correlation enhances robustness and completeness of comparative effectiveness estimates when data are missing. However, as this analysis uses simulated correlation data and predicted HRs, all results are exploratory and not intended for clinical or policy decision-making.
METHODS: A systematic literature review identified 19 published clinical trials that met the PICOS criteria for mCRPC. Five studies did not report HRs for OS and two did not report HRs for rPFS/PFS. For several studies, HRs were derived by digitizing Kaplan-Meier curves and generating pseudo-individual patient data using Guyot’s algorithm. The published data was supplemented by simulated patient-level data in estimating the correlation between OS and rPFS hazards via bootstrap resampling and Gaussian copula modeling. This correlation informed a BvNMA to predict missing HRs and reduce uncertainty. A univariate random effect NMA (RE NMA) was also performed for comparison. All results are considered exploratory for methodological purposes.
RESULTS: BvNMA successfully predicted HRs for comparisons where published data were unavailable and substantially narrowed credible intervals (CrI) compared to RE NMA. For example, in one comparison where OS HR was missing, RE NMA produced a wide CrI (e.g., 0.0127-94.86), whereas BvNMA estimated a more precise interval (e.g., 0.38-2.39). Similar improvements were observed for rPFS estimates. These findings demonstrate the methodological advantage of incorporating correlated outcomes in NMA.
CONCLUSIONS: BvNMA leveraging OS-rPFS correlation enhances robustness and completeness of comparative effectiveness estimates when data are missing. However, as this analysis uses simulated correlation data and predicted HRs, all results are exploratory and not intended for clinical or policy decision-making.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
CO64
Topic
Clinical Outcomes
Topic Subcategory
Comparative Effectiveness or Efficacy
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Oncology