Enhancement of Inpatient Mortality Prognostication With Machine Learning in a Prospective Global Cohort of Patients With Cirrhosis With External Validation

Abstract

Background & aims: Cirrhosis is a major global burden requiring frequent hospitalizations and has high inpatient mortality. Traditional prognostic tools focused on inpatient mortality are affected by global disparities, which impacts timely management. We aimed to deploy machine learning (ML) approaches to enhance inpatient mortality prognostication.

Methods: Using the prospective Chronic Liver Disease Evolution and Registry for Events and Decompensation (CLEARED) cohort, which enrolled inpatients with cirrhosis globally, we used admission-day data to predict inpatient mortality with ML approaches vs logistic regression. Internal validation (75/25 split) and subdivision using World Bank income status (low/low-middle-income countries, upper-middle income countries, and high-income countries) were performed. The ML model with the best area under the curve (AUC) was externally validated in a US veteran inpatient population with cirrhosis.

Results: The CLEARED cohort included 7239 inpatients with cirrhosis (64% were men; mean [SD] age, 56 [13] years; median Model for End-Stage Liver Disease-Sodium score = 25) from 115 centers globally; 22.5% were from low/low-middle-income countries, 41% were from upper-middle-income countries, and 34% were from high-income countries; 11.1% of patients (n = 808) died in the hospital. Random forest (RF) showed the best AUC (0.815) with high calibration, which was significantly better vs parametric logistic regression and LASSO models (AUC, 0.774; P < .001 and AUC, 0.787; P = .004, respectively). RF was the ML method with the highest AUC and remained better than logistic regression, regardless of country income level (high-income countries: AUC, 0.806; upper middle-income countries: AUC, 0.867; and low/low-middle-income countries: AUC, 0.768). External validation was performed in 28,670 veterans (96% were men, mean [SD] age, 67.8 [10.3] years, median Model for End-Stage Liver Disease-Sodium score = 15) with 4% (n = 1158) inpatient mortality. The AUC using the CLEARED-derived RF model was 0.859.

Conclusions: RF analysis trained on a global prospective cirrhosis cohort enhances mortality prediction over traditional methods, was consistent across country income levels, and was successfully validated externally in a US veteran population.

Keywords: CLEARED Cohort; Income Disparities; Random Forest; Veterans.