Predicting COVID-19 disease progression and patient outcomes based on temporal deep learning

Area of Science:

• Artificial Intelligence
• Biomedical Informatics
• Computational Biology

Background:

• The COVID-19 pandemic necessitates early identification of disease progression for effective patient management.
• Existing research lacks clear definitions for COVID-19 stages and characterization of disease progression.
• Accurate prediction of patient outcomes is crucial for resource allocation and targeted treatment.

Purpose of the Study:

• To develop a temporal deep learning model for predicting COVID-19 patient outcomes.
• To identify and characterize distinct stages of COVID-19 disease progression.
• To assist clinicians in better assessing and treating COVID-19 patients.

Main Methods:

• A time-aware long short-term memory (T-LSTM) neural network was employed.
• An online open dataset of 485 COVID-19 patients from Wuhan was utilized for model training.
• The model considered patient biomarkers and irregular time intervals for dynamic relation grasping.

Main Results:

• The model achieved over 90% accuracy in outcome prediction by 12 days, with high accuracy at earlier time points (3-9 days).
• Four distinct stages of COVID-19 progression were identified, each associated with different patient statuses and mortality risks.
• Forty biomarkers were ranked, with reference values provided for each stage, and three major complications (myocardial, liver, and renal injury) were identified.

Conclusions:

• The developed T-LSTM model accurately predicts COVID-19 patient outcomes and identifies disease progression stages.
• The identified four-stage progression model and associated biomarkers can aid clinicians in patient assessment and treatment.
• This research contributes to a better understanding of COVID-19 dynamics and facilitates more effective medical resource utilization.