Clinical Profiling And Disease Severity Prediction Of COVID-19 By Multinomial Regression Analysis

Abstract

Background and Objectives: The biochemical evaluation of patients infected with SARS-CoV-2 is crucial for their on-site stratification, accurate assessment, and subsequent management. This cross-sectional study aims to evaluate the biochemical markers in patients of both genders with control (normal), ward (moderate), and ICU (severe) COVID-19. Furthermore, using the same dataset, disease severity prediction by multinomial regression analysis was performed to identify the major predictors of disease progression in COVID-19. Materials and Methods: Biochemical evaluations of all study subjects were performed using an automated chemistry analyzer. The statistical analysis was performed using the R programming language, while the multinomial regression analysis was performed using Python. Confusion matrix and ROC analysis were used to evaluate the predictive performance of this model. Cross-validation of the model was done to ensure that it generalizes well with new and unseen data. Results: Among ward patients, significantly increased levels of LDH (p = 0.0028) and CRP (p = 0.0138) were observed in females compared to males. In contrast, among ICU patients, CRP levels (p = 0.0309) rose more in males than in females. Moreover, according to the multinomial regression analysis, AST, IL-6, Direct Bilirubin, LDH, Neutrophils, ALT, and CRP were observed to be the major predictors of disease severity. The mean AUC obtained after 5-fold cross-validation was 0.90±0.06. Conclusion: The biochemical analysis revealed CRP as the most important predictor of disease severity in males, whereas in females, LDH emerged as the most important marker of disease severity. Accordingly, AST, IL-6, Direct bilirubin, Neutrophils, LDH, ALT, and CRP were observed to have the most impact on the predictability of the disease status, as evident by the multinomial regression analysis. The high mean AUC value obtained after cross-validation of the model indicates its excellent ability to determine the disease severity status in COVID-19 patients.