Prediction of financial deficits of postoperative patients in the intensive care unit using machine learning.

Abstract

Background: Operational loss, defined as unanticipated financial deficits in intensive care unit (ICU) management, is challenging to predict yet critical for hospital sustainability. This study aimed to evaluate whether machine-learning models can predict financial loss events in postoperative ICU patients.

Methods: We conducted a retrospective analysis of postoperative patients admitted to the ICU at Tohoku University Hospital between April 2017 and March 2021. A total of 22 clinical and administrative variables collected within 24 h of ICU admission were used to develop machine-learning models. The outcome was defined as financial loss events, determined by a negative contribution margin below the break-even threshold of - 909 USD. The dataset was randomly split into training (70%) and test (30%) sets. Predictive performance was assessed using the area under the receiver operating characteristic curve (AUC) and accuracy.

Results: Among 6743 postoperative ICU patients, 425 (6.3%) experienced financial loss events. The random forest classifier demonstrated high predictive performance, with an AUC of 0.859 and accuracy of 0.785.

Conclusions: Machine-learning models may accurately predict financial loss events in postoperative ICU patients, potentially supporting efficient resource allocation and hospital financial planning.