Using machine-learning methods to predict in-hospital mortality through the Elixhauser index: A Medicare data...

Area of Science:

• Health Informatics
• Biostatistics
• Machine Learning in Healthcare

Background:

• Accurate in-hospital mortality prediction is crucial for patient prognosis and clinical resource allocation.
• Traditional logistic regression models have limitations in assessing comorbidity measures for mortality prediction.
• Machine learning methods are increasingly utilized for healthcare predictions.

Purpose of the Study:

• To compare the model performance of logistic regression, elastic net, and artificial neural network (ANN) for predicting in-hospital mortality.
• To evaluate these models using updated Present-on-Admission (POA) guidelines for Elixhauser's comorbidity measures.
• To assess the utility of machine learning in improving mortality prediction accuracy.

Main Methods:

• Retrospective analysis of 1,810,106 adult Medicare inpatient admissions.
• Utilized data from the Centers for Medicare and Medicaid Services data warehouse.
• Employed logistic regression, elastic net, and artificial neural network (ANN) models incorporating POA indicators.

Main Results:

• All models demonstrated good performance with C-statistics above 0.77.
• Elastic net produced a parsimonious model with similar predictive power to logistic regression but fewer comorbidities.
• Artificial neural network (ANN) achieved the highest C-statistic (0.800), outperforming logistic regression and elastic net.

Conclusions:

• Elastic net and ANN models can be successfully applied to predict in-hospital mortality under updated POA guidelines.
• Machine learning approaches offer enhanced predictive accuracy for in-hospital mortality compared to traditional methods.
• These findings support the use of advanced modeling techniques for better patient outcome prediction and healthcare management.