Logistic regression combined with ROC curve model to predict risk of critically ill-patients with COVID-19

ABSTRACT

Objective: To build a model to predict critically ill-patients with coronavirus disease 2019 (COVID-19), and provide a new idea for the rapid identification of clinical progression in the early stage of critically ill-patients. Methods: A retrospective analysis of the general data of 152 general patients and 323 critically ill-patients diagnosed with COVID-19 from Jan 17th, 2020 to Feb 25th, 2020 in Wuhan Third Hospital was carried out;At the same time, the differences in fever, blood routine, liver and kidney function, coagulation function, C-reactive protein (CRP), and nucleic acid reagent testing results from the day of admission were statistically analyzed. Factors with statistical significance were included in a multivariate logistic regression analysis to obtain independent relevant factors that affect the critical ill-patients with COVID-19. Then a prediction model was built based on these factors and its accuracy was evaluated by the receiver operating characteristic (ROC) curve. Results: The sensitivities of age, fever, neutrophil ratio, lymphocyte ratio, serum creatinine (Scr) and combined diagnosis were 0.664, 0.671, 0.607, 0.669, 0.302 and 0.710, respectively;The specificities were 0.669, 0.585, 0.795, 0.685, 0.895 and 0.802, respectively;The area under the curve (AUC) were 0.725, 0.628, 0.721, 0.681, 0.590 and 0.795, respectively;The AUC of combined diagnosis was higher than that of single diagnosis (P < 0.05). Conclusion: The logistic regression and combined with ROC curve model based on multi-factors, including age, fever status, neutrophil ratio, lymphocyte ratio, and Scr, can play a good role in predicting the occurrence of critically ill-patients with COVID-19, which is worthy of further promotion and application.