Performance of advanced machine learning algorithms overlogistic regression in predicting hospital readmissions: A...

Area of Science:

• Health Informatics
• Medical Artificial Intelligence
• Predictive Analytics in Healthcare

Background:

• Hospital readmissions pose a significant challenge in healthcare systems.
• Accurate prediction of 30-day readmissions is crucial for patient care and resource management.
• Machine learning (ML) algorithms are increasingly explored for their potential in predicting patient outcomes.

Purpose of the Study:

• To conduct a meta-analysis evaluating the performance of logistic regression (LR) and various machine learning (ML) models.
• To compare the predictive accuracy of ML models versus LR for 30-day hospital readmissions in US patients.
• To identify which ML techniques offer the best performance for readmission prediction.

Main Methods:

• A systematic search of electronic databases (Medline, PubMed, Embase) was conducted for studies published between January 2015 and December 2019.
• Nine studies meeting selection criteria were included, with quality assessed using the Quality in Prognosis Studies (QUIPS) tool.
• Model performance was evaluated using the Area Under the Curve (AUC), with a random-effects meta-analysis performed using STATA 16.

Main Results:

• Machine learning models demonstrated a statistically significant improvement in predicting 30-day all-cause hospital readmissions compared to logistic regression (Mean Difference in AUC: 0.03).
• Subgroup analyses revealed that deep-learning methods (MD: 0.06) and neural networks (MD: 0.03) outperformed logistic regression.
• ML models also showed superior performance in predicting heart failure-related readmissions (MD in AUC: 0.04).

Conclusions:

• Machine learning offers enhanced predictive capabilities for 30-day hospital readmissions compared to traditional logistic regression.
• Deep learning methods represent the most effective ML approach for predicting hospital readmissions.
• These findings support the integration of advanced ML techniques into clinical practice for proactive patient management.