OncoCast

Overview

OncoCast is an elastic-net Cox regression model combined with a random-forest residual correction, designed for survival analysis in left-truncated observational oncology cohorts. It is used to predict overall survival from multi-modal clinical, genomic, and NLP-derived features while handling time-on-study biases inherent to real-world data.

Used by

• PMID:39506116 — OncoCast (elastic-net Cox plus random-forest residual correction, 500 trees, 5 terminal nodes, 50 runs) applied within the MSK-CHORD clinicogenomic dataset (24,950 patients) alongside random survival forest models for left-truncated OS prediction across five cancer types (NSCLC, BRCA, COADREAD, PRAD, PAAD); combined-modality models outperformed single-modality in every cancer type PMID:39506116.

Notes

• Corpus-grown slug; not present in canonical ontology. Note: a separate page exists for oncocast-mpm, which is a specialized variant.
• Designed specifically for left-truncation bias correction in real-world datasets where cohort entry time does not equal start of follow-up.

Sources

• PMID:39506116

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