Prognosis patients with COVID-19 using deep learning

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Epidemiology

Background:

• COVID-19 pandemic highlights the need for effective mortality risk assessment.
• Current methods may be costly or inaccessible for some hospitals.
• Accurate prediction of mortality risk in COVID-19 patients is crucial for hospital management.

Purpose of the Study:

• To develop a COVID-19 mortality risk calculator using a deep learning model.
• To address challenges of data imbalance and missing biomarker data.
• To provide an accessible tool for assessing mortality risk in critically ill patients.

Main Methods:

• A deep learning (DL) model was developed using patient data from HM Hospitals Madrid.
• Pre-processing strategies included handling unbalanced classes and feature selection.
• Synthetic Minority TEchnique (SMOTE) and K-nearest neighbour imputation were used for data evaluation.

Main Results:

• The model achieved an Area Under the Curve (AUC) of 0.93 and an accuracy of 0.95.
• High recall (1.00) and precision (0.91) were reported.
• The deep learning model demonstrated superior performance, even with over-sampling techniques.

Conclusions:

• The proposed deep learning model is effective for COVID-19 mortality risk prediction.
• The tool can assist hospitals in managing critically ill patients and allocating resources.
• The method offers a valuable approach for assessing prognosis in COVID-19 patients.