Predicting the Length of Stay of Cardiac Patients Based on Pre-Operative Variables-Bayesian Models vs. Machine...

Area of Science:

• * Medical Informatics
• * Health Services Research
• * Statistical Modeling

Background:

• * Accurate prediction of patient length of stay (LoS) is vital for hospital operational efficiency and clinical preparedness.
• * Pre-operative clinical features are commonly used for LoS estimation.
• * Existing statistical and machine learning methods have limitations in handling data variability and skewness.

Purpose of the Study:

• * To evaluate and compare Bayesian (simple and hierarchical) and machine learning (ML) regression models for cardiac patient LoS prediction.
• * To assess the utility of hierarchical Bayesian regression in managing data variability and skewness without outlier removal.
• * To provide accurate LoS estimates for patients at Tabba Heart Institute (THI).

Main Methods:

• * Comparative analysis of simple Bayesian regression, hierarchical Bayesian regression, and ML regression models.
• * Application of models to predict LoS for cardiac patients admitted to THI (2015-2020).
• * Evaluation using Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE).

Main Results:

• * Hierarchical Bayesian regression yielded the lowest RMSE (1.49) and MAE (1.16).
• * Simple Bayesian regression showed higher errors (RMSE: 3.36, MAE: 2.05).
• * ML models had average errors comparable to simple Bayesian regression (RMSE: 3.36, MAE: 1.98).

Conclusions:

• * Hierarchical Bayesian regression demonstrates superior performance for cardiac patient LoS prediction.
• * This method effectively handles data variability and skewness, improving prediction accuracy.
• * Findings support the adoption of hierarchical Bayesian regression for enhanced hospital resource management.