Characteristic of 523 COVID-19 in Henan Province and a Death Prediction Model

Front Public Health. 2020 Sep 8;8:475. doi: 10.3389/fpubh.2020.00475. eCollection 2020.

Abstract

Certain high-risk factors related to the death of COVID-19 have been reported, however, there were few studies on a death prediction model. This study was conducted to delineate the clinical characteristics of patients with coronavirus disease 2019 (covid-19) of different degree and establish a death prediction model. In this multi-centered, retrospective, observational study, we enrolled 523 COVID-19 cases discharged before February 20, 2020 in Henan Province, China, compared clinical data, screened for high-risk fatal factors, built a death prediction model and validated the model in 429 mild cases, six fatal cases discharged after February 16, 2020 from Henan and 14 cases from Wuhan. Out of the 523 cases, 429 were mild, 78 severe survivors, 16 non-survivors. The non-survivors with median age 71 were older and had more comorbidities than the mild and severe survivors. Non-survivors had a relatively delay in hospitalization, with higher white blood cell count, neutrophil percentage, D-dimer, LDH, BNP, and PCT levels and lower proportion of eosinophils, lymphocytes and albumin. Discriminative models were constructed by using random forest with 16 non-survivors and 78 severe survivors. Age was the leading risk factors for poor prognosis, with AUC of 0.907 (95% CI 0.831-0.983). Mixed model constructed with combination of age, demographics, symptoms, and laboratory findings at admission had better performance (p = 0.021) with a generalized AUC of 0.9852 (95% CI 0.961-1). We chose 0.441 as death prediction threshold (with 0.85 sensitivity and 0.987 specificity) and validated the model in 429 mild cases, six fatal cases discharged after February 16, 2020 from Henan and 14 cases from Wuhan successfully. Mixed model can accurately predict clinical outcomes of COVID-19 patients.

Keywords: death prediction model; epidemiology investigation; novel coronavirus pneumonia; random forest; risk factors.