Demographics and clinical characteristics
We enrolled 252 SARS-CoV-2 RNA positive COVID-19 patients admitted between January 19 and March 6. Fifty-two patients (20.6%) were critically ill (Table 1). Compared to non-critically ill patients, the critically ill patients were significantly older, with a median (IQR) age of 64 (52-73), compared to 45 (35.25-56) of non-critically ill patients. Although more of our patients were male, the gender ratio of the critically ill patients was similar to that of non-critically ill. On admission, the critically ill patients more often presented dyspnea (29 patients, 55.8%) and elevated respiratory rate (Respiratory rate >24 breaths per min, 19 patients, 36.5%), compared to 6 patients (3%) and 7 patients (3.6%), respectively, of the non-critically ill. On the other hand, critically ill patients less frequently presented fever on admission.
Critically ill patients more often had comorbidities than non-critically ill patients (Table 1). Twenty (38.5%) of the critically ill patients had hypertension, compared to 28 patients (14%) of non-critically ill. Similarly, more of the critically ill patients presented diabetes, cardiovascular disease and malignancy, compared to the non-critically ill patients. Correlation between chronic lung disease and critical illness was not observed, likely because of the small numbers of patients with chronic lung disease in both study groups.
Laboratory and radiographic findings
On admission, markers for coagulation function APTT, fibrinogen and d-dimer were consistently higher in critically ill patients, compared to non-critically ill patients (Table 2). Inflammatory biomarkers ESR, procalcitonin, CRP and NLR were markedly and consistently higher in critically ill patients compared to non-critically ill patients (Table 2). Note that procalcitonin is a marker for bacterial and fungal infection, but not for viral infection. Many of the markers for cell, tissue and organ damage including LDH, BUN, AST, ALT and total bilirubin were higher in the critically ill patients compared to non-critically ill patients (Table 2). Note that among all these markers, only CRP but no other inflammatory markers was out of the normal range. The underlying mechanism for this interesting phenomenon awaits further studies.
Chest CT results showed that 179 patients (71%) exhibited bilateral pulmonary infiltration, while 62 patients exhibited ground-glass opacity (24.6%) and 45 patients (17.9%) unilateral pulmonary infiltration (Table 2). Although not statistically significant, there were more patients with CT abnormality in the non-critically ill patients than in the critically ill patients on admission. This counter-intuitive observation is in line with a recent report that patients without fever at disease onset has an adverse prognosis[10].
Treatments and outcomes
Except for two critically ill patients, all other patients were given antiviral medicine (Table 3). More critically ill patients (47 patients, 90.4%) were given antibiotics, compared to non-critically ill patients (155 patients, 77.5%). More of the critically ill patients were treated with corticosteroids, immunoglobulin and albumin. In contrast, fewer critical patients were treated with a traditional Chinese medicine, the Lung Cleansing and Detoxifying Decoction, which is an extraction from a mixture of 21 herbal plants. Our data support a protective role of this therapy for COVID-19[11].
Similar portions of the critically and non-critically ill patients required supplementary oxygen, although all the patients requiring mechanical ventilation and ECMO (extracorporeal membrane oxygenation) were critically ill.
Twenty (38.5%) critically ill patients developed acute respiratory distress syndrome (ARDS), compared to one non-critically ill patient who developed ARDS (Table 3). All 43 patients admitted to ICU were critically ill, and 6 of them died. Median time from illness onset to dyspnea, ARDS, ICU admission, and death were 6, 9, 10 and 8.5 days, respectively.
Risk factors associated with critical illness
Based on the published work on SARS-CoV and SARS-CoV-2 infections, and the differential clinical presentations and outcomes between critically ill and non-critically ill patients in our study, we conducted univariable and multivariable logistic regression analysis to identify the potential risk factors for critical illness in COVID-19.
Higher proportions of older patients were critically ill, with an odds ratio (OR, (95%CI) of 2.03 (1.61-2.62) indicating an over 100% increase in the risk of developing critical illness for every additional 10 years in age (Table 4). Age older than 60 years was identified as a major risk factor for critical illness.
Among all the symptoms on admission, dyspnea was highly associated with the critical illness, with an adjusted (for age, gender and comorbidities) OR (95% CI) of 46.01 (15.36-169.48), p < 0.0001 (Table 4). “Respiratory rate >24 breaths per min” also highly associated with critical illness, with an adjusted OR of 5.84 (1.53-23.01), p < 0.001 (Table 4)
Univariable analysis indicated association of critical illness with comorbidities hypertension, diabetes, cardiovascular disease and malignancy (Table 4). However, statistical significance was not achieved after adjustment for the influences of age and gender.
Abnormal counts of white blood cell was significantly correlated with critical illness. After adjusting for confounding factors, white blood cell counts > 9.5 X 109/L was identified as a risk factor for critical illness, with an adjusted OR of 8.38 (2.82-26.64), p < 0.001. Neutrophilia was significantly correlated with critical illness. Neutrophil count > 6.3 X 109/L was identified as a risk factor for critical illness with an adjusted OR of 6.03 (2.31-16.18), p < 0.001 (Table 4). Lymphopenia was associated with the critical illness, and lymphocyte < 1.1 X 109/L was a risk factor with an adjusted OR of 2.41 (1.09-5.63), p<0.05 (Table 4). Accordingly, elevated NLR was highly correlated with the critical illness, NLR > 3.53 was a risk factor for critical illness with an adjusted OR of 3.78 (1.73-8.65), p = 0.001. Other infection related markers, ESR, procalcitonin, CRP and LDH, was not significantly correlated with critical illness, after adjustment for age, gender and comorbidities.
Higher levels of the markers for coagulation function, fibrinogen and d-dimer, were correlated with the critical illness. Fibrinogen > 4 g/L and d-dimer > 0.55µg/mL were risk factors for critical illness with adjusted ORs of 2.38 (1.07-5.33), p = 0.033, and 2.52 (1.14-5.63), p = 0.023, respectively (Table 4).
Higher levels of BUN, AST, ALT and total bilirubin were correlated with critical illness, with adjusted ORs of 9.20 (2.87-38.15), p < 0.001, 2.29 (1.05-5.03), p = 0.04; 3.67 (1.45-9.47), p < 0.01, 5.53 (2.12-14.92), p < 0.001, respectively (Table 4). Accordingly, higher SOFA score, an integrated reference for multi-organ failure, was highly correlated with critical illness, with an adjusted OR of 2.77 (1.95-4.16), p < 0.0001. SOFA scores 2 and greater was identified as risk factor for critical illness after adjustment for confounding factors.
The Area Under the Receiver-Operator Curve (AUROC) was measured to evaluate the ability of the above critical illness associated risk factors for the prediction of adverse outcome (Supplementary Table S2). The individual factors with the highest AUROC values were SOFA score (0.921), age (0.776), dyspnea (0.764) and leukocytosis (0.658) (Figure 2a). The AUROC for the combination of age and SOFA was 0.936 (Figure 2b). The AUROC for the combination of all 14 risk factors in Table 4 was 0.967 (Figure 2b).
Further, LASSO logistic regression analysis was conducted to select the best combination of predictors from 14 potential risk factors, and identified SOFA score, age, dyspnea, and white blood cell count and age as the most sensitive marker for the prediction of critical illness (Figure 2c, d). The AUROC of the combined 4 factors was 0.960 (Figure 2b, Supplementary Table S2).