Predicting SARS-CoV-2 infection duration at hospital admission:a deep learning solution

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Infectious Disease Modeling

Background:

• The global increase in COVID-19 cases and deaths necessitates improved patient management strategies.
• Accurate prediction of infection duration is crucial for optimizing hospital resource allocation and patient care pathways.

Purpose of the Study:

• To develop and validate a deep learning model for predicting the duration of COVID-19 infection.
• To utilize readily available patient data at hospital admission for predictive modeling.

Main Methods:

• An observational study involving 222 COVID-19 patients.
• A deep learning approach using a sequential convolutional neural network (CNN) ensemble model.
• Feature selection based on principal component analysis (PCA) and k-fold cross-validation.

Main Results:

• The developed CNN-based ensemble model achieved a median absolute error of 2.7 days (IQR = 3.0 days) in predicting infection duration.
• Prediction accuracy was consistent across the range of infection durations.
• The model effectively integrated 55 features from patient admission data.

Conclusions:

• The proposed deep learning model offers a reliable tool for forecasting COVID-19 patient hospitalization duration.
• This predictive capability can significantly aid in managing hospital burdens and logistical complexities during pandemic waves.
• The model's reliance on early admission data makes it a practical solution for preemptive patient pathway planning.