Development and validation of prognosis model of mortality risk in patients with COVID-19

Area of Science:

• Infectious Diseases
• Medical Informatics
• Pulmonology

Background:

• Coronavirus disease 2019 (COVID-19) poses a significant global health threat.
• Accurate prognostication of mortality risk in COVID-19 patients is crucial for clinical management.
• Existing predictive models may require refinement for optimal patient stratification.

Purpose of the Study:

• To identify key clinical features for predicting mortality risk in COVID-19 patients.
• To compare the performance of machine learning and logistic regression models against established scoring systems.
• To develop a robust model for early identification of high-risk individuals.

Main Methods:

• Retrospective analysis of inpatient data from Wuhan, China (January-March 2020).
• Collection of demographic, clinical, comorbidity, vital sign, CT scan, and laboratory data.
• Application of Random Forest, XGboost, and multivariate logistic regression for feature selection and model development.

Main Results:

• Lactate dehydrogenase (LDH), C-reactive protein (CRP), and age were identified as significant predictors of mortality.
• The developed multivariate logistic regression model achieved a high in-sample AUROC of 0.9521.
• The model demonstrated superior performance compared to CURB-65 and machine learning models in both in-sample and out-of-sample testing.

Conclusions:

• LDH, CRP, and age are reliable indicators for identifying severe COVID-19 cases upon hospital admission.
• A logistic regression model incorporating these features offers a more accurate prognostic tool than CURB-65 or machine learning approaches.
• These findings can aid clinicians in timely risk stratification and resource allocation for COVID-19 patients.