medRxiv Oncology· Fu, C., Yu, J., Shi, R., Waters, M., Pavlak, M., Arfe, A., Shah, S. P., Battle, A., kehl, K. L., Bot, B. M., US Cancer AI Alliance,, Schultz, N., Morris, Q. D., Jee, J. ·· 20 hours agoPickScore64
Longitudinal AI Models Predict Cancer Adverse Events
In brief
SPARC, a transformer-based model, predicts cancer-related adverse event risk from routine longitudinal health data and achieved state-of-the-art performance in development, prospective, and external multi-center cohorts. The model was pre-trained and fine-tuned on 101,920,379 measurements from 224,047 patients with cancer; the prospective validation cohort included 23,126 patients, public external datasets 26,012 patients, and external multi-hospital cohorts 562,620 and 206,840 patients.
Why it matters
Validation across a same-center prospective cohort, public external data, and external multi-hospital cohorts tested how pooled multi-institution fine-tuning affects generalizability and performance.
Source: medRxiv Oncology · medrxiv.org