A Machine Learning Algorithm Predicts Duration of hospitalization in COVID-19 patients

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Epidemiology

Background:

• The COVID-19 pandemic significantly strained healthcare resources, especially hospital bed capacity.
• Variations in patient length of stay (LOS) complicated resource management during the pandemic.

Purpose of the Study:

• To develop machine learning models for predicting prolonged COVID-19 hospital length of stay (LOS).
• To assist healthcare systems in managing bed capacity and inform clinical decision-making regarding patient hospitalization duration.

Main Methods:

• Utilized electronic health record data from 966 COVID-19 patients.
• Developed three machine learning algorithms using 353 variables, trained on 80% of data and validated on 20%.
• Models were created on hospital days 1, 2, and 3, incorporating data available up to each respective day.

Main Results:

• Predictive accuracy improved sequentially, reaching 0.765 with an Area Under the Curve (AUC) of 0.819 by hospital day 3.
• Models demonstrated increasing capability to identify patients likely to have a prolonged LOS (>8 days).

Conclusions:

• Machine learning models using readily available EHR data can effectively predict prolonged COVID-19 hospitalizations.
• These predictive tools can support hospital operational planning and patient communication regarding expected LOS.