Clinical Factors And Discriminative Models Of COVID-19 Severity

Objective: This study explores factors related to COVID-19 severity,and constructs discriminative models of severe COVID-19.The purpose is to provide scientific basis for auxiliary screening of severe COVID-19,and reduce the occurrence of adverse outcomes.Methods: Medical records of 271 patients hospitalized with COVID-19 from a hospital in Wuhan from February to March 2020 were collected.All study subjects fulfilled the inclusion and exclusion criteria.Based on the COVID-19 severity,patients were grouped into the severe group and the mild group.Firstly,a descriptive analysis was conducted on the basic features of the two groups,and EFA was used to extract severe COVID-19 related factors with specific clinical significance from 20 indicators.Then,the discriminative model of COVID-19 severity was constructed based on the logistic regression,SVM,and BP neural network.Finally,the sensitivity,specificity,accuracy and ROC curve were used to evaluate the discriminant effects of the above three models,and 10-fold crossvalidation was used to evaluate the stability of the model.Results:1.A total of 271 COVID-19 patients were included,among the 97 severe patients,the median age was 68 years,and 63.9% were males.Age and sex were statistically significant between the mild group and the severe group(P<0.05).Hypertension and diabetes were common comorbidities in COVID-19 patients,accounting for 41.7% and 25.8%,respectively.Hypertension was highly prevalent among the severe group compared with the mild group(52.6% vs 35.6%,P=0.007).2.Fever,cough and fatigue were the most common clinical symptoms in COVID-19 patients,accounting for 86.0%,62.4% and 38.0%,respectively.There were no significant differences in clinical symptoms between the two groups(P>0.05).3.Seven independent factors were retained and named in the context of with their clinical roles.The first factor was identified as the inflammatory factor,followed by the immune factor,the liver function factor,the renal function factor,the coagulation factor,the hypertension factor,and the gender factor.4.The logistic regression model that discriminates COVID-19 severity showed a good discriminant ability.The results showed that the AUC value was 0.896,the sensitivity was 0.856,the specificity was 0.793 and the accuracy was 80.8%.The AUC was 0.870 and the accuracy was81.5% after 10-fold cross-validation.5.The SVM that discriminates COVID-19 severity showed a good discriminant ability.The results showed that the AUC value was 0.902,the sensitivity was 0.835,the specificity was 0.856 and the accuracy was84.1%.The AUC was 0.850 and the accuracy was 74.2% after 10-fold cross-validation.6.The BP neural network that discriminates COVID-19 severity showed a good discriminant ability.The results showed that the AUC value was 0.944,the sensitivity was 0.845,the specificity was 0.971 and the accuracy was 92.3%.The AUC was 0.870 and the accuracy was 80.4%after 10-fold cross-validation.Conclusions:1.The inflammatory factor,the liver function factor,the renal function factor,the hypertension factor,the immune factor,and the coagulation factor may be influential factors for severe COVID-19.2.The three discriminative models based on EFA can achieved effective results for severe COVID-19,and their AUC values and accuracy values reflected the validity of the discriminative model.However,the stability of SVM was relatively poor.