Evaluation of Cost-Saving Machine Learning Methods for Patient Blood Management

Abstract

Our objective is the development of cost-saving methods for the patient blood management in Galeazzi Orthopedic Institute, a large Italian hospital. The methods have been developed in relation to the known costs of the hospital, both in terms of unused blood bags and drugs. Observational data about 4593 patients have been retrieved, with anagraphical and pre-operational clinical features. Model's performances have been compared to an existing baseline in terms of both accuracy measures (F1, recall, AUC) and saved costs per patient. The proposed methods recorded an enhancement of performances for the adopted measures, demonstrating a possible useful application of machine-learning-based methods for the patient blood management task.