Prospective and External Validation of Prognostic Machine Learning Models for Short- and Long-Term Mortality Among Acutely Admitted Patients Based on Blood Tests.

Abstract

The application of machine learning (ML) models in emergency departments (EDs) to predict short- and long-term mortality encounters challenges, particularly in balancing simplicity with performance. This study addresses this gap by developing models that uses a minimal set of biomarkers, derived from a single blood sample at admission, to predict both short-term and long-term mortality. Our approach utilizes biomarkers representing vital organs and the immune system, offering a comprehensive view of both acute and chronic disease states. Moreover, by integrating explainable machine learning methods, we ensured that clinicians can easily interpret the model's outputs. Our Analysis included 65,484 admissions from three cohorts at two large Danish university hospitals, demonstrating the models' efficacy with high accuracy, with AUC values between 0·87 and 0·93. These results underscore that a single assessment of routine clinical biochemistry upon admission can serve as a powerful tool for both short-term and long-term mortality prediction in ED admissions.