Demographics and baseline characteristics of moderate and severe patients
191 patients with COVID-19 were enrolled in this study. After excluding 24 patients who were not confirmed by SARS-CoV-2 RNA detection as of Feb 29, 2020, and 4 patients with incomplete core information in their electronic medical records, we included 163 patients in the final analysis. 87 (53.4%) of the 163 patients were clinically diagnosed as severe COVID-19, whereas 76 (46.6%) as moderate. The median age of the 163 patients was 65.0 years (range, 27-89 years), and nearly half were male. There were more male patients in severe cases, while opposite in moderate cases. Most patients in two groups had comorbidities, which was more pronounced in severe cases, with hypertension and diabetes commonly presented (Table 1). And severe patients were more likely to have hypertension. Among the 163 patients, the most common symptom was fever, followed by dry cough, expectoration, fatigue, dyspnea, diarrhea, and myalgia. Compared with moderate cases, less severe cases experienced expectoration. Only severe cases experienced dyspnea, with higher systolic blood pressure and increased respiratory rate on admission. The median time from illness onset to admission and hospital stay were similar between moderate and severe cases (13 vs 12; P > 0.05; 34 vs 36; P > 0.05; respectively).
Laboratory and radiographic findings
We observed extensive differences in laboratory findings between two groups (Table 2). 22 severe patients and only 3 moderate patients had leukocytosis (white blood cell [WBC] count ≥ 10×109/L) on admission. Increased neutrophils (≥ 6.3 × 109/L) were more frequent in severe patients (46.0% vs 6.6%; P < 0.001). More severe patients had lymphopenia (lymphocyte count < 0.8 × 109/L) and eosinopenia (eosinophil count < 0.02 × 109/L) than moderate patients. Severe patients had higher WBC and neutrophil counts, as well as lower lymphocyte and eosinophil counts.
Levels of aspartate aminotransferase (AST), total bilirubin, direct bilirubin, indirect bilirubin, creatinine, glucose, high-sensitivity cardiac troponin I (hs-cTnI), creatine kinase isoenzyme-MB (CK-MB), procalcitonin, D-dimer, prothrombin time (PT) and international normalized ratio (INR) were higher in severe patients than moderate patients. Serum albumin was significantly lower in severe patients than moderate patients, and hypoalbuminemia (albumin < 32 g/L) was more frequent in severe patients. In addition, levels of LDH, NT-proBNP, hsCRP and ferritin were remarkably higher in severe patients. Of patients with available data, a bunch of inflammatory cytokines were remarkably higher in severe patients than moderate patients, including interleukin-2 receptor (IL-2R), IL-6, IL-8, IL-10 and tumor necrosis factor-α (TNF-α). And increased IL-2R, IL-6 and TNF-α were detected in majority of severe patients.
On admission, abnormalities in thoracic computed tomography (CT) scans were observed in all patients. The typical CT features in severe patients included multifocal ground-glass opacities and subpleural consolidation, rapidly evolved to mass shadows of high density in bilateral lungs. Whereas representative CT features in moderate patients were unilateral or bilateral ground-glass opacification.
Temporal changes of blood cell counts during hospitalization
The dynamic profile of blood route was tracked after admission, from day 4 to day 25 at 3-day intervals (Fig. 1 and Supplementary Table 1). WBC and neutrophil counts were constantly higher in severe patients than moderate patients from admission to day 16 and 19, respectively, and decreased with recovery. In severe patients, WBC and neutrophil counts were highest on day 10 after admission. Eosinophil and lymphocyte counts were significantly lower in severe patients than moderate patients until day 13 and 19, respectively. In severe patients, lymphocyte and eosinophil counts were lowest on admission, gradually increased during hospitalization. Platelet counts were slightly lower in severe patients than moderate patients in the first week of hospitalization.
Treatments, complications, and clinical outcomes
Antiviral coverage (oseltamivir, arbidol or lopinavir/ritonavir) was similar between two groups. 85 severe patients and 57 moderate patients received empirical antibiotics (moxifloxacin or cephalosporin, etc). More severe patients were prescribed corticosteroids, immunoglobulin and thymosin. A large proportion of severe patients required high flow nasal cannula or non-invasive mechanical ventilation (Table 3). The most common complication in severe patients was respiratory failure (77.0%). Less frequent complications in severe patients included acute cardiac injury (17.2%), acute kidney injury (12.5%) and shock (9.2%), most of which were not observed in any moderate cases. As of March 15, 2020, 133 (81.6%) patients had fully recovered and been discharged, 30 (27.4%) were still in treatment.
Exploring independent factors for discrimination of disease severity
Univariate analysis identified 10 variables associated with disease severity, including hypertension, sex (male), neutrophil count, eosinophil count (< 0.04 × 109/L), lymphocyte count, LDH, albumin, creatinine, PT and D-dimer. A multiple logistic regression model was fitted using these variables. Backward elimination at this point removed 6 variables from the model. The final independent variables in the model were neutrophil count, eosinophil count (< 0.04 × 109/L), LDH and D-dimer with odds ratios of 1.34, 19.93, 1.02 and 1.15, respectively (Table 4).
Among 4 variables, LDH demonstrated the highest predictive value to discriminate COVID-19 severity with an AUC of 0.86 and a cut-off level of 310 in ROC curve analysis. Moreover, we used several eosinophil count cutoff values to discriminate severe and moderate cases. The result showed that the best eosinophil cell count cut-off value was 0.04 × 109/L, with AUC of 0.82, (95% CI, 0.77–0.88), and sensitivity of 83.3% and specificity of 68.4% (Table 5, Figure 2). Nagelkerke pseudo R2 for the model was 0.695, indicating that 4 variables accounted for 69.5% of the variance in COVID-19 severity. The Hosmer-Lemeshow test indicated satisfactory fit (P = 0.934). The performance of combined variables in modeling for discriminating disease severity was tested. Neutrophil + eosinophil (< 0.04 × 109/L) + LDH + D-dimer combinations demonstrated high predictive values (Table 5 and Fig. 2).