An interpretable machine learning model based on a quick pre-screening system enables accurate deterioration risk...

Area of Science:

• Computational biology and bioinformatics
• Clinical informatics
• Epidemiology

Background:

• Predicting deterioration in COVID-19 patients is crucial for timely intervention.
• Existing models may lack interpretability or require extensive data.
• Identifying key risk factors can improve clinical decision-making.

Purpose of the Study:

• To develop a high-performing, interpretable model for predicting COVID-19 patient deterioration.
• To identify significant risk factors and their warning ranges.
• To create a streamlined model and online tool for rapid pre-screening.

Main Methods:

• Developed a predictive model using a cohort of 3028 COVID-19 patients.
• Identified 15 high-risk factors for deterioration, including laboratory values and clinical measurements.
• Created a streamlined model using four key indicators (prothrombin time, heart rate, BMI, HCT) and comorbidities.

Main Results:

• The comprehensive model achieved an AUC of 0.8517.
• Fifteen significant risk factors for deterioration were identified.
• The streamlined model demonstrated good predictive performance with an AUC of 0.7941.
• An online pre-screening website was developed.

Conclusions:

• An interpretable model effectively predicts COVID-19 patient deterioration risk.
• Key indicators like PT, heart rate, BMI, and HCT, along with comorbidities, are vital for rapid assessment.
• The developed tools can aid in early identification and management of high-risk COVID-19 patients.