Using machine learning methods to predict all-cause somatic hospitalizations in adults: A systematic review

Area of Science:

• Medical Informatics
• Health Services Research
• Computational Medicine

Background:

• Machine Learning (ML) is increasingly explored for predicting patient hospitalizations.
• Accurate prediction of hospital admissions and readmissions is crucial for healthcare resource management and patient outcomes.

Purpose of the Study:

• To systematically review and analyze the utilization of Machine Learning (ML) in predicting all-cause somatic hospital admissions and readmissions in adults.
• To identify common methodologies, features, and performance metrics used in ML-based hospitalization prediction models.

Main Methods:

• A comprehensive literature search was conducted across eight databases up to October 2023.
• Studies predicting adult hospital admissions/readmissions using ML were included, with data extraction using CHARMS and assessment via PROBAST and TRIPOD.

Main Results:

• 116 studies met inclusion criteria; 45 predicted admission, 70 predicted readmission. Boosting tree-based algorithms and Natural Language Processing (NLP) of clinical notes showed strong performance.
• Significant variability existed in datasets, algorithms, and validation methods. Reporting quality and methodological rigor were generally poor, with limited clinical implementation.
• Key inadequacies included lack of individual patient-level interpretation, missing code availability, insufficient external validation, and poor calibration.

Conclusions:

• Studies on ML for hospitalization prediction face substantial methodological and reporting quality challenges.
• Improving the quality and transparency of research is essential for the clinical acceptance and implementation of these predictive models.