Demographic and Clinical Features in Training and Validation Cohorts
Overall 492 patients were recruited in this study. The demographic characteristics and clinical features of patients from the training cohort (n=237; TJH), validation cohort 1 (n=120; RHWU+WNH), and validation cohort 2 (n=135; WPH) cohorts were listed in the Table 1. The mortality rates in the three cohorts were 44.3%, 25.8% and 33.3% for the training cohort, validation cohort 1 and validation cohort 2, respectively. A total of 105 events occurred in the training cohort, 31 in the validation cohort 1, and 45 in the validation cohort 2. The median survival time were comparable among these three cohorts (15.0 days for the training cohort, 17.0 days for the validation cohort 1 and 14.0 days for the validation cohort 2. The patients in the training cohort with median age of 62 (IQR 50 - 70) was older than the validation cohort 1 (median age 46, IQR 37 - 66), but similar to the validation cohort 2 (median age 63, IQR 52 - 70.5). There was no significant difference in sex distribution among the three cohors. Most of patients were non-smokers (92% [training], 97.5% [validation 1], and 83% [validation 2]). The number of patients associated with comorbidities varied somewhat between three cohorts (52.7% [training] vs. 37.5% [validation 1] vs. 73.3% [validation 2]; Table 1). The number of severe cases requiring intensive care unit (ICU) admission varied among the three cohorts (8.9% [training], 1.7% [validation 1], and 20% [validation 2]). Lymphopenia occurred in the majority of patients in three cohorts (69.2% [training], 60.8% [validation 1], and 75.6% [validation 2], Table 1). Leucocytosis was observed 24.5% in the training cohort, 21.7% in the validation cohort 1, and 20.7% in the validation cohort 2.
Neutrophilia was observed in 34.6% in the training cohort, 3.1% in the validation cohort 1, and 31.1% in the validation cohort 2.
The median time of symptom onset to hospital admission was longer in the training cohort (median 10.0 days, IQR 7 - 14 days), than the validation cohort 1 (median 7.0 days, IQR 4 – 10 days), but the same as the validation cohort 2 (median 10.0 days, IQR 7 - 13 days) with fever as the most common symptom on admission. The duration of hospitalization and treatment were 15.0 days (IQR 7 – 24 days) for the training cohort, 17.0 days (IQR 11.8 - 25.0 days) for the validation cohort 1, and 14.0 days (IQR 10.0 - 19.0 days) for the validation cohort 2. The majority of patients treated with antibiotics (86.5%, 92.5, and 85.9% for the trainig cohort, validation cohort 1 and validation cohort 2, respectively and antivirals (lopinavir/ritonavir) (99.6%, 95.8% and 97.0% for the training cohort, validation cohort 1 and validation cohort 2, respectively).
Potential Risk Factors Associated with Vital Status for COVID-19
In an univariate analysis, advanced age, increased body temperature, and presence of underlying diseases were associated with higher mortality rate in patients with high COVID-19 infection (Table 2). Tachypnoea and hypertension, as well as treatment with antibiotics, corticosteroids or intravenous immunoglobulin were also associated with increased mortality (Table 2). Several laboratory parameters including serum bilirubin, urea, D-dimer, potassium level, prothrombin time (s), lactate dehydrogenase, aspartate transaminase (AST), and urea were also found to be associated with death. In addition, patients with lymphopenia, leucocytosis, or neutrophilia also had an increased risk of death (Table 2).
Of note, the ratio of neutrophils and lymphocytes, lymphocyte ratio, and neutrophile ratio were also significant risk factors for mortality.
Construction of a Prognostic Model for Vital Status and Survival in SARS-CoV-2
For the training cohort, a multivariate analysis was performed to analyze the association between vital status, survival time, and all the covariates listed in Table 1. Statistically significant predictors for vital status and survival time in a multivariable analysis were age (adjusted odds ratio (AOR): 1.1/years increase [95% CI 1.06 - 1.13]; Wald’s p<0.001), neutrophil-to-lymphocyte ratio (AOR: 1.14 [95% CI 1.08 - 1.2]; p<0.001), body temperature at admission (AOR: 1.53 / °C increase [95% CI 1.0 - 5.26]; p=0.005), aspartate transaminase (AST) (AOR: 2.47 [95% CI 1.16 - 5.26] for increase vs. normal; p=0.019), and total protein (AOR: 1.69 [95% CI 0.78 - 3.64] for decrease vs. normal ; p=0.018; Table 2). Based on the weights (coefficients) of these five significant covariates (Table 2), a model prognostic model was constructed and applied to predict the vital status of the training cohort. The results of this analysis yielded an AUC of 0.912 (95% CI 0.878 - 0.947; Fig. 1A). This indicated that the prognostic model was able to effectively dichotomize patients with SARS-CoV-2 pneumonia who subsequently discharged and those who later died. In the prediction of overall survival, the model reached a Harrell’s c-index of 0.758 (95% CI 0.723 - 0.793; Fig. 1C). The model was able to define a high-risk subgroup with a significantly increased likelihood of death due to SARS-CoV-2 pneumonia (hazard ratio [HR]: 24.22 [95% CI 10.57 - 55.5]) versus a low-risk subgroup. The predicted survival probabilities were compared with observed survival probabilities on the 7th, 14th, 21th, and 28th day after admission (Fig. 1B). The nomogram was constructed to assess impact of these factors (Supplement Figure 2). The predicted 30-days survival rates of the high- and low-risk subgroups in the training cohort were visualized in Fig. 1C (=>799 and <799). Here, 799 represented the cutoff in the model based on the average of minimum calculated scores among deceased patients.
Validation of the Model for Vital Status and Survival
In order to validate the prognostic value of the established prognostic model for SARS-CoV-2 pneumonia, external validation using 2 cohorts to test the predictive model was performed. The model reached an AUC of 0.928 [95% CI 0.884 - 0.971; validation cohort 1] and 0.883 [95% CI 0.815 - 0.952; validation cohort 2] to predict the vital status (Fig. 1A). For the prediction of survival of both validation cohorts, the model yielded C indices of 0.762 [95% CI 0.723 - 0.801; validation cohort 1] and 0.711 [95% CI 0.672 - 0.75; validation cohort 2] (Figure 2). By applying the same cutoff of model score, high-risk subgroups with lower survival rate were defined to clearly differentiated between the low-risk subgroups in both validation cohorts (HR: 11.53 [95% CI 4.01 - 33.15 for the validation cohort 1 and HR: 9.3 [95% CI 3.32 - 26.03] for the validation cohort 2) (Figure 2). Of note, the predicted 30-day survival rates in high- and low-risk subgroups in both validation cohorts were similar to the observed survival rates in the training cohort (Fig. 2), thereby confirming the strength of the model for the prognosis for SARS-CoV-2 pneumonia.
For the investigation of age-related impact on the prognostic model, these two validation cohorts were merged and then divided into three groups by age to form three subgroups: <50 year (cohort_5), 50~70 year (cohort_6), and >70 year (cohort_7), respectively. For the prediction of the vital status, the model yielded an AUC of 0.911 [95% CI 0.853 - 0.97; cohort_5], 0.809 [95% CI 0.713 - 0.904; cohort_6], and 0.825 [95% CI 0.719 - 0.931; cohort_7; Fig. 1A]. For the survival prediction, the model yielded C indices of 0.572 [95% CI 0.533 - 0.611; cohort_5], 0.721 [95% CI 0.682 - 0.76; cohort_6], and 0.706 [95% CI 0.667 - 0.745; cohort_7; Table 3]. Finally, to aid in the current clinical management of SARS- CoV-2, a web-based application (http://220.127.116.11:8013/SIMTaskMaster/COVID_Tool) was developed to enable broad testing and utilization of the developed prognostic model (Supplement Figure 3).