Effective Use of Reconstructed Survival Data for Individual-Level Correlation Assessment in Oncology: A Case Study in Metastatic-Hormone Sensitive Prostate Cancer (mHSPC)
Author(s)
Paul Serafini, BA, Victoria Wan, MSc, Mir Sohail Fazeli, PhD, MD, Murat Kurt, BS, MS, PhD.
Evidinno Outcomes Research Inc, Vancouver, BC, Canada.
Evidinno Outcomes Research Inc, Vancouver, BC, Canada.
OBJECTIVES: Access to IPD is often a barrier to investigating individual-level correlation between surrogate endpoints and overall survival (OS), a key component of surrogate endpoint validation. We devised a novel framework for estimating individual-level correlation using reconstructed survival data and demonstrated its predictive performance in a case study from mHSPC assessing the association between radiographic progression-free survival (rPFS) and OS.
METHODS: Pseudo-patient level rPFS and OS data were reconstructed from Kaplan-Meier curves previously published in a surrogacy analysis [Halabi et al. (2024)]. Standard parametric and spline-based models were fitted to both endpoints to estimate mean rPFS and mean OS. Pre-progression death (PPD) probability and exponentially-distributed post-progression survival (PPS) were elicited by a moment-matching approach embedded in an enumeration algorithm utilizing extrapolated rPFS and OS curves, mean rPFS and mean OS simultaneously. Paired rPFS-OS data were generated in a microsimulation leveraging modeled rPFS, PPD probability and PPS, and used to estimate Kendall’s (tau), Spearman’s (rho) and Pearson’s (r) correlation coefficients between the endpoints. Sensitivity of the results were explored with respect to changes in PPD probability (+/−10%), hazard rate of PPS (+/−10%), and time horizon (+/−5 years) considered for survival extrapolations.
RESULTS: Predicted Kendall's tau was 0.71 (95% CI: 0.70-0.72) compared to Kendall’s tau 0.83 (95% CI: 0.82-0.84) obtained from actual IPD in Halabi et al. Predicted Pearson’s r and Spearman’s rho were 0.97 (95% CI: 0.96-0.97) and 0.86 (95% CI: 0.85-0.87), respectively. Across all experiments, predicted correlation measures were minimally sensitive to changing parameter settings (≤2.5% deviation from base estimates) with sufficiently narrow 95% CIs (width ≤2.9%)
CONCLUSIONS: This proposed approach can estimate both rank and product moment correlation coefficients without access to IPD and be generalized to other cancer types. Its predictions in the presented case study can be conservative due to independent sampling of rPFS and PPS.
METHODS: Pseudo-patient level rPFS and OS data were reconstructed from Kaplan-Meier curves previously published in a surrogacy analysis [Halabi et al. (2024)]. Standard parametric and spline-based models were fitted to both endpoints to estimate mean rPFS and mean OS. Pre-progression death (PPD) probability and exponentially-distributed post-progression survival (PPS) were elicited by a moment-matching approach embedded in an enumeration algorithm utilizing extrapolated rPFS and OS curves, mean rPFS and mean OS simultaneously. Paired rPFS-OS data were generated in a microsimulation leveraging modeled rPFS, PPD probability and PPS, and used to estimate Kendall’s (tau), Spearman’s (rho) and Pearson’s (r) correlation coefficients between the endpoints. Sensitivity of the results were explored with respect to changes in PPD probability (+/−10%), hazard rate of PPS (+/−10%), and time horizon (+/−5 years) considered for survival extrapolations.
RESULTS: Predicted Kendall's tau was 0.71 (95% CI: 0.70-0.72) compared to Kendall’s tau 0.83 (95% CI: 0.82-0.84) obtained from actual IPD in Halabi et al. Predicted Pearson’s r and Spearman’s rho were 0.97 (95% CI: 0.96-0.97) and 0.86 (95% CI: 0.85-0.87), respectively. Across all experiments, predicted correlation measures were minimally sensitive to changing parameter settings (≤2.5% deviation from base estimates) with sufficiently narrow 95% CIs (width ≤2.9%)
CONCLUSIONS: This proposed approach can estimate both rank and product moment correlation coefficients without access to IPD and be generalized to other cancer types. Its predictions in the presented case study can be conservative due to independent sampling of rPFS and PPS.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
CO84
Topic
Clinical Outcomes
Topic Subcategory
Relating Intermediate to Long-term Outcomes
Disease
Oncology