Random survival forest

Overview

Random survival forest (RSF) is an ensemble machine-learning method that extends the random forest algorithm to right-censored time-to-event data. It can capture non-linear relationships and interactions between features without requiring proportional hazards assumptions PMID:39147831.

Used by

• PMID:39147831 — a random survival forest (LB+ model) incorporating liquid biopsy variables (ctDNA detection, cfDNA concentration, specific gene alterations) plus cancer type and chemotherapy receipt was trained on 4,141 patients and validated prospectively on 1,426 patients; achieved c-indices of 0.74 (discovery), 0.73 (validation), and 0.67 (generalizability for NSCLC ctDx Lung cohort), substantially outperforming Khorana score (0.57, 0.61, 0.54) and RAM (0.62) for cancer-associated VTE prediction PMID:39147831.
• PMID:39506116 — RSF (n_trees=1,000, min_splits=10, min_samples_per_leaf=15) was the primary OS prediction model in the MSK-CHORD clinicogenomic dataset (24,950 patients); multi-modal RSF combining NLP-derived features (tumor sites, prior treatment), genomics, structured data, and imaging outperformed any single-modality model in all five cancer types; c-indices ranged from 0.58 (stage IV PAAD) to 0.83 (stage I–III BRCA); NLP-derived tumor-site features were the single most prognostic modality in stage-IV patients PMID:39506116.

Notes

• RSF outperformed traditional clinical risk scores for VTE prediction by capturing interactions between ctDNA variables and clinical features PMID:39147831.
• Corpus-grown slug; not present in canonical ontology.

Sources

• PMID:39147831
• PMID:39506116

This page was processed by crosslinker on 2026-04-30.