Symptom Prediction and Mortality Risk Calculation for COVID-19 Using Machine Learning.

Area of Science:

• Artificial Intelligence
• Public Health
• Medical Informatics

Background:

• Early prediction of COVID-19 patient outcomes is crucial for healthcare resource management and disease control.
• Existing prediction models lack public accessibility.
• Timely identification of high-risk patients can inform treatment and public health strategies.

Purpose of the Study:

• To develop and validate machine learning models for predicting COVID-19 symptoms and mortality.
• To create a publicly accessible tool for early risk assessment of COVID-19 patients.

Main Methods:

• Developed two machine learning models: a symptom prediction model (SPM) and a mortality prediction model (MPM).
• Utilized data from 23,749 hospitalized COVID-19 patients (February-September 2020), focusing on age, gender, and medical history.
• Created a free online interface (www.aicovid.net) for model accessibility.

Main Results:

• The SPM achieved ROC-AUCs of 0.53-0.78 for predicting 12 symptom groups, with consciousness disorders being most accurately predicted (74% sensitivity, 70% specificity).
• The MPM demonstrated a ROC-AUC of 0.79, predicting mortality with 75% sensitivity and 70% specificity.
• Approximately 90% of observed deaths were concentrated within the top 21% of predicted risk groups.

Conclusions:

• Machine learning models can effectively predict COVID-19 symptoms and mortality using easily accessible patient information.
• These models empower patients and clinicians to make informed decisions regarding hospitalization, self-isolation, and vaccine prioritization.
• The developed online tool facilitates rapid risk assessment, potentially improving healthcare outcomes and reducing disease prevalence.