Explainable artificial intelligence to assess hospital financial performance during the COVID-19 pandemic through operating margin metric.

Abstract

Purpose: This study addresses critical gaps in understanding hospital financial resilience during the COVID-19 pandemic by developing an AI framework that enhances traditional analyses through capturing nonlinear interactions among clinical, financial and operational drivers of operating margins.

Design/methodology/approach: Using 21 clinical, financial and operational parameters from US hospital financial reports, we evaluated 12 machine learning, ensemble and deep learning regression models to predict hospital operating margins. SHapley Additive exPlanations (SHAP) quantified feature importance, while Spearman correlation validated pairwise relationships. Interactive SHAP dependence plots analyzed hospital size-dependent dynamics.

Findings: The Gradient Boosting model outperformed other regressors (MSE = 1,253, R2 = 0.79). SHAP analysis identified eight key determinants of operating margins during the COVID-19 era: supply expenses, licensed beds, patient days, charity care, Medicaid/Medicare revenues, employee benefits and the case mix index (CMI). Interactive SHAP analysis revealed hospital size-dependent financial dynamics - while supply and employee benefit expenses had a less pronounced impact on financial stability in larger hospitals, higher patient days and Medicaid/Medicare revenues played a more critical role. The impact of CMI in mid-sized hospitals remained uncertain. Smaller hospitals exhibited greater financial vulnerability, necessitating targeted cost management strategies.

Originality/value: This study is among the first to integrate explainable AI with SHAP for hospital financial analysis during COVID-19, offering a blueprint for AI-powered, equity-focused stewardship. It can help leaders to balance cost containment with care quality while exposing systemic disparities in reimbursement structures.