Predicting special care during the COVID-19 pandemic: a machine learning approach

Area of Science:

• Health Informatics
• Machine Learning in Medicine
• Epidemiology

Background:

• COVID-19 significantly strains global health systems, particularly in Brazil.
• Accurate prediction of patient care needs is crucial for effective hospital resource allocation.

Purpose of the Study:

• To develop a statistical and machine learning method for predicting COVID-19 patient hospitalization and length of stay.
• To create a decision support system for healthcare providers.

Main Methods:

• Utilized blood laboratory exam data for patient risk stratification.
• Employed a two-step procedure with Bayesian Optimization for model selection.
• Evaluated multiple candidate models to identify optimal predictive algorithms.

Main Results:

• Achieved an Area Under the ROC Curve of 0.94 for predicting the need for special care.
• Obtained a Root Mean Squared Error of 1.87 for predicting length of stay, a 77% improvement over the baseline.
• The developed models demonstrate readiness for deployment as a decision system.

Conclusions:

• The proposed method effectively predicts COVID-19 patient outcomes using routine blood tests.
• This approach offers a significant advancement in healthcare resource planning and patient management.
• The analytical framework is adaptable for predicting outcomes in other diseases.