The predictive factors of US hospital bankruptcy - a multi-model comparison

TL;DR

This study compares different models to predict US hospital bankruptcies and introduces a new, effective model called BRKFSST.

Contribution

A new hospital-specific logistic regression model (BRKFSST) is developed and shown to outperform traditional bankruptcy prediction models.

Findings

The BRKFSST model achieved an AUC of 81.8%, balanced accuracy of 72.2%, and mean recall of 60.6%.

Key predictors identified include labor compensation ratio, adjusted patient days, and quality ratings.

The model outperformed traditional models like Altman's Z, Ohlson’s O-score, and Zmijewski’s model.

Abstract

In response to the growing number of hospital bankruptcies across the United States, this study sought to develop a predictive and interpretable model tailored specifically to the healthcare industry. Utilizing a longitudinal dataset of 3,091 short-term acute care hospitals from 2008 to 2021, we evaluated and compared traditional bankruptcy prediction models—Altman's Z'', Ohlson’s O-score, and Zmijewski’s model—against a newly developed hospital-specific logistic regression model (BRKFSST). We incorporated over 30 financial and hospital-level variables, including quality indicators, ownership type, and market characteristics. Unlike prior models, ours lagged all unknowable variables to ensure true out-of-sample prediction. The BRKFSST model achieved strong performance, with an Area Under the Curve (AUC) of 81.8%, balanced accuracy of 72.2%, and a mean recall of 60.6% across multiple test/train splits, outperforming all benchmark models. Importantly, the model retained interpretability, allowing for the identification of key predictors such as labor compensation ratio, adjusted patient days, and quality ratings. These findings provide actionable insights for hospital leaders and policymakers to identify at-risk institutions and implement early interventions to prevent financial collapse and preserve access to care.

Figures