Tool demonstrated generalizability through high predictive performance in international clinical cohort
By Elana Gotkine HealthDay Reporter
WEDNESDAY, Jan. 15, 2025 (HealthDay News) — A machine learning model can make accurate survival predictions for Merkel cell carcinoma (MCC), according to a study published online Jan. 8 in npj Digital Medicine.
Noting there are no personalized prognostic tools in use in MCC, Tom W. Andrew, from Newcastle University in the United Kingdom, and colleagues employed explainability analysis to examine insights into mortality risk factors for MCC. Deep learning feature selection was combined with a modified XGBoost framework to develop a web-based prognostic tool for MCC.
The researchers found that the tool, termed DeepMerkel, can make accurate, personalized, time-dependent survival predictions from readily available clinical information. The tool demonstrated generalizability through high predictive performance in an international clinical cohort, and outperformed current prognostic staging systems.
“DeepMerkel can make time-dependent survival predictions providing personalized prognostication and clinical guidance in MCC,” the authors write. “This hybrid approach uses applied models which have been specifically adapted and integrated for disease-specific survival predictions and has demonstrated generalizability through high predictive performance in an international clinical cohort, outperforming current population-based prognostic staging systems.”
Copyright © 2025 HealthDay. All rights reserved.
