Precision Oncology

SPARC, a transformer-based foundation model trained on more than 100 million routinely collected longitudinal clinical measurements from 224,047 patients with cancer, predicts treatment-related adverse events and generalizes to external centers.

General-purpose large language models, used without any fine-tuning, match or complement specialized predictors on three tumor genomic interpretation tasks, with the largest gains on data from new assays and institutions.

Cancer-modulated allostatic load, a stress index computed from routine labs and vitals in 12,689 MSK cancer patients, rises with disease progression, tracks EGFR and KRAS mutation status in lung cancer, and is linked to fewer pre-diagnosis daily steps in both MSK and All of Us wearable cohorts.

Without any task-specific training, general-purpose large language models approach the accuracy of specialized supervised models at separating tumor mutations from clonal hematopoiesis variants, flagging oncogenic variants and predicting cancer type from tumor genomic profiles, and ensembling them with a supervised classifier improves generalization across institutions.

Benchmarking zero-shot large language models against supervised, tailor-made predictors on cancer-type, somatic-vs-CHIP and variant-oncogenicity classification across 100,000+ samples from 19 institutions, and combining them into LLM–specialist ensembles that raise cross-institution accuracy by up to 55%.

SPARC, a transformer-based foundation model pretrained on more than 100 million longitudinal laboratory and vital-sign measurements from 224,047 patients with cancer, used to predict treatment toxicities, adverse events and mortality.