Truace
Subscribe

predictive performance of random survival forest versus Cox model for patient-specific survival in RCT time-to-event data

Source article: Comparison of the Cox Proportional Hazards Model and Random Survival Forest Algorithm for Predicting Patient-Specific Survival Probabilities in Clinical Trial Data

The Cox proportional hazards model is often used to analyze data from randomized controlled trials (RCTs) with time-to-event outcomes. Random survival forest (RSF) is a machine-learning algorithm known for its high predictive performance. We conduct a comprehensive neutral comparison study to compare the performance of Cox regression and RSF in various simulation scenarios based on two reference datasets from RCTs. The motivation is to identify settings in which one method is preferable over the other when compa…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1232· cite & verify
Comparison of the Cox Proportional Hazards Model and Random Survival Forest Algorithm for Predicting Patient-Specific Survival Probabilities in Clinical Trial Data
Jackknifing the Kaplan-Meier survival estimator for censored data : simulation results and asymptotic analysis by Gaver, Donald Paul.Miller, R. G.. Public domain
Trace impact readingPositive state
P 66The P score combines the specificity and measured human impact of the grounded problem claim with the strength of this Trace’s cited sources.

Both sides are scored from claims and sources, not community votes.

G 74The G score combines the specificity and measured human impact of the grounded gain claim with the strength of this Trace’s cited sources.

In brief

By 2026-10-01, authors reported a neutral comparison of Cox proportional hazards regression and random survival forest for predicting patient-specific survival probabilities, using varied simulation scenarios based on two RCT reference datasets and TRIPOD-aligned metrics. They observed that C-index-only conclusions did not generalize, that overall performance measures appeared more reasonable, and that splitting rule choice mattered for RSF.

The comparison matters for clinical trial analysis because model choice and RSF splitting rule affect reliability of individualized survival predictions, especially with treatment-covariate interactions and nonproportional hazards. Uncertainty remains about generalizability beyond the two reference datasets and simulated conditions, and about which overall performance measures and alternative splitting rules are optimal in practice.

Main points

  1. Neutral comparison study simulated various scenarios based on two reference datasets from RCTs with time-to-event outcomes.
  2. Evaluation followed TRIPOD recommendations and examined multiple aspects of predictive performance beyond C index.
  3. Standard log-rank splitting rule for RSF was outperformed by alternative splitting rules in nonproportional hazards settings.
  4. Cox proportional hazards model performance was affected by violation of proportional hazards assumption.

The gain

In simulated RCT time-to-event data, random survival forest retained predictive performance better when treatment-covariate interactions were present than when they were absent.

The problem

In the same RCT simulations, the standard log-rank splitting rule for random survival forest was outperformed by alternative rules in nonproportional hazards settings, limiting its predictive advantage.

The rundown

The study used simulation scenarios anchored to two RCT reference datasets and assessed performance using TRIPOD-recommended measures, noting that overall performance measures may give more reasonable results than C index alone.

Results differentiated contexts: RSF was less degraded by treatment-covariate interactions, while Cox-PH degraded when proportional hazards was violated and RSF's default log-rank splitting lost to alternatives in those nonproportional settings.

What this doesn’t fix

Findings are based on simulation scenarios derived from only two RCT reference datasets, and conclusions based solely on C index may not generalize to other performance aspects.

Sources

  1. Peer-reviewedBiometrical Journal2026-10-01

The debate