Key result
Random survival forests model shows no prognostic advantage over Cox nomogram for adverse events.
Why the study?
The study aimed to compare the prognostic value of machine learning and traditional Cox regression methods for predicting adverse clinical events in patients with ACS.
Does a machine learning model or a simplified Cox regression nomogram improve the prediction of adverse clinical events in patients with acute coronary syndrome compared to traditional GRACE risk scores?
Cohort (n=3,272)
Yes
Does a machine learning model or a simplified Cox regression nomogram improve the prediction of adverse clinical events in patients with acute coronary syndrome compared to traditional GRACE risk scores?
Effect estimate: C-index 0.836 (95% CI 0.813-0.866)
Absolute Event Rate: 0.836% vs 0.854%
p-value: p=0.427
RSF shows no superiority over Cox nomogram for ACS event prediction; leaves open whether ML improves risk stratification in larger cohorts.
We aimed to compare the prognostic value of Machine learning (ML) and traditional Cox regression method for predicting adverse clinical events in patients with acute coronary syndrome (ACS). In a prospective observational cohort of 2730 patients with ACS, we analyzed several candidate variables for median follow-up of 8.1 years from the Observation of Cardiovascular Events of ACS patients (OCEA) study. External validation was performed in 542 patients with ACS for median follow-up of 2.8 years from the management of cardiovascular risk and health (MCRH). The random survival forests (RSF)-based ML model including clinical traditional risk factors and plasma biomarkers were evaluated. A multivariable Cox regression model was subsequently used to develop a nomogram. Adverse clinical events refer to cardiovascular events and non-cardiovascular death. Cardiovascular events were defined as a combined endpoint of MI, ischemic stroke, cardiovascular death and heart failure for hospital. RSF based model exhibited the highest prognostic ability, achieving a C-index of 0.960 with 95% CI (0.952–0.970) in the derivation cohort compared with nomogram model [C-index 0.806, 95% CI:0.790–0.822) and GRACE 2.0 (C-index 0.730, 95% CI:0.711–0.749) (both p < 0.001). In the validation cohort with a median follow-up of 2.8 years, RSF model achieved C-index of 0.836 (95% CI: 0.813–0.866), without superiority compared with Cox regression model with C-index of 0.854 (95%CI: 0.785–0.922), both of which outperformed GRACE 2.0 risk score in derivation and validation cohorts ( p < 0.001). Also, our Cox model achieved significantly higher C-index than GRACE 3.0 across both derivation (0.806 vs 0.710) and external validation cohorts (0.854 vs 0.701; all p < 0.001). RSF model and Cox regression model outperformed GRACE 2.0/GRACE 3.0 in the discrimination of adverse clinical events in ACS, with former two carrying no significant difference. RSF showed high apparent performance in the derivation cohort, but the simpler seven-variable Cox nomogram demonstrated comparable external validation performance and may be more practical for clinical use.
No takes yet. Share an insight, caveat, or question.
Xue et al. (2026) conducted a cohort in Acute coronary syndrome (n=3,272). Random survival forests (RSF) machine learning model vs. Cox regression nomogram was evaluated on Discrimination of adverse clinical events (cardiovascular events and non-cardiovascular death) in the validation cohort (C-index 0.836, 95% CI 0.813-0.866, p=0.427). The random survival forests machine learning model achieved a C-index of 0.836 for predicting adverse clinical events in the validation cohort, showing no significant superiority over the Cox nomogram.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: