Clinical characteristics and risk factors of fatal patients with COVID-19: a retrospective cohort study in Wuhan, China

This retrospective cohort study was performed in patients with confirmed COVID-19 from Tongji Hospital, Tongji Medical College, HUST, a designated hospital for severely or critically ill COVID-19 patients. All fatal patients with COVID-19 (N = 281) were enrolled from January 13 to March 31, 2020. Subsequently, 712 survivors were statistically matched by propensity score matching [12] at an approximate ratio of 3:1 based on age, sex and comorbidities. This study was approved by the Ethics Committee of Tongji Hospital, Tongji Medical College, HUST and granted a waiver of informed consent from study participants.

Epidemiological, clinical, radiological, laboratory, clinical treatments, and clinical outcomes data of all patients with laboratory-confirmed SARS-CoV-2 were obtained with data collection forms from electronic medical records of Tongji Hospital. The researches on other topics [9, 13] may obtain some information of non-survivors in Tongji hospital, however, our study collected more information about cases and detailed data. The admission and in-hospital data of these patients were collected, reviewed and verified by a trained team of physicians. Any missing or uncertain records were collected and clarified through communication with involved health-care providers and their families. The detailed and standardized information of demographic data, underlying comorbidities, initial symptoms, vital signs, and chest computed tomographic (CT) were recorded or diagnosed at hospital admission. The complications, treatments, clinical outcomes (survivors and non-survivors) and hospital length of stay were monitored to March 31, 2020, the final data of follow-up. Laboratory examinations including blood routine, immune cells subsets, inflammatory cytokines and biomarkers, blood gas assay, cardiac function test, renal function test, liver function test, pancreatic function test, coagulation test, and metabolism test were detected on the first diagnosed date. The time of follow-up was defined as the duration from admission to outcomes (survivor/non-survivor) of patients. There were no cases lost to follow-up in this study attributed to standardized government managements and close tracking for COVID-19 pandemic.

The detailed definitions of complications including acute respiratory distress syndrome (ARDS), acute pancreatic injury, acute liver damage, acute cardiac injury, disseminated intravascular coagulation (DIC), SOFA, qSOFA, APACHE II and SIRS scores were listed in the Additional file 1: Methods.

Continuous variables were presented as median and interquartile range (IQR) or mean and standard deviation (SD). Categorical variables were expressed as number (%). For continuous variables, Student’s t-test was used for normal distributed data, and Mann–Whitney U non-parameter test was used for non-normal distributed data. The Pearson’s χ² test or Fisher’s exact test were applied for categorical variables. Time to events (recovered or death) was defined as the time from hospital admission to events. The univariate Cox regression models were used to determine hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between individual factors and death risk of COVID-19. The multivariate Cox proportional hazards model analysis was used to establish different comprehensive death risk score models including four score systems, biochemical biomarkers, complications, and a combined system integrating all scores. The significant variables from the univariable analysis of the death risk were considered as the candidates. The stepwise regression was used for the selection of the prediction for the model. Each comprehensive risk score model of candidate variables was established according to the following formula: Risk score (RS) = ∑β variables × Expvariables. The β variables represented the estimated contribution coefficient of independent candidate variables in the multivariate Cox regression analysis and Expvariables denoted the level of independent candidate variables. According to this formula, the risk scores of each patient of different death risk score prediction models were computed. The combined system was performed based on the integrating risk scores of different comprehensive models. Then we calculated sensitivity, the specificity of different death risk score prediction models with survival, timeROC packages for all analyses in R version 3.50. Time-dependent receiver operating characteristic (ROC) curves and areas under the curves (AUCs) were used to assess the prognostic accuracy of each model [14]. A two-sided P value < 0.05 was considered statistically significant.

281 (28.30%) non-survivors and 712 (71.70%) survivors with COVID-19 were enrolled in this study. Considering that the age, sex, and comorbidities had been reported to be the common death risk factors of COVID-19, enrolled survivors and non-survivors were statistically matched based on these risk factors. As summarized in Table 1 and Additional file 1: Table S1, the most common symptoms of all patients were fever and cough, non-survivors were more likely to have dyspnea (50.39% vs 42.23%, P = 0.026) than survivors. Besides, non-survivors tended to have more abnormal vital signs on admission, such as higher temperature, heart rate and respiratory rate.

The abnormalities in chest CT images among these patients were also observed (Table 1 and Fig. 1). The patchy shadows presence was the most typical manifestation of both groups, while pleural thickening (41.38% vs 26.90%; P = 0.018) was more frequently observed in non-survivors, compared with survivors. Consistently, blood gas analysis revealed the degree of hypoxia was more evident in non-survivors, which presented lower concentration of PaO₂, SaO₂, PaCO₂, CtCO₂ and bicarbonate than survivors. These findings suggest more serious impairment of lung function in non-survivors.

Previous studies reported that cytokine storm and lymphopenia were common features in severe COVID-19 patients [2, 15]. We observed that the serum levels of inflammatory cytokines including IL6 (61.35 vs 6.85 pg/mL; P < 0.0001), IL10 (10.30 vs 5.00 pg/mL; P < 0.0001), IL8 (28.40 vs 11.60 pg/mL; P < 0.0001), TNF-α (11.45 vs 8.10 pg/mL; P < 0.0001), IL-1β (8.15 vs 6.49 pg/mL; P < 0.0001) and IL2R (1148.00 vs 599.50 U/mL; P < 0.0001) were significantly elevated in non-survivors compared with survivors. Moreover, the infection-related biomarkers including ferritin, CRP and procalcitonin also exhibited higher levels in non-survivors. Conversely, the baseline counts of lymphocytes (0.63 vs 1.09 × 10⁹/L; P < 0.0001), CD3⁺CD19⁻ T cells (276.50 vs 905.00/μL, P < 0.0001), CD3⁺CD8⁺ T cells (62.00 vs 280.00/μL; P < 0.0001), B cells (73.50 vs 156.00/μL, P < 0.0001), NK cells (36.50 vs 209.00/μL, P < 0.0001) as well as the total number of T cells, B cells and NK cells (406.00 vs 1337.00/μL; P < 0.0001) were drastically decreased in non-survivors, compared with survivors. Collectively, these findings demonstrate that aggravated inflammatory responses and severe lymphopenia might be correlated with the poor clinical outcome of COVID-19 patients.

Multiple-organ damage was more pronounced in non-survivors. We observed higher levels of ALT, TBIL, LDH, homocysteine, NT-proBNP, hs-cTnI, CK-MB and lower level of ALB/GLO in non-survivors. Besides, non-survivors had a falling count of eosinophils while elevated leukocytes and neutrophils, compared with survivors. Of note, in non-survivors, coagulation-related biomarkers of platelets counts were also substantially decreased, followed by the increased D-dimer and prolonged PT and APTT.

Additionally, in term of several scores evaluating disease severity, the fatal cases had higher SOFA (4.00 vs 2.00, P < 0.0001), qSOFA (1.00 vs 0.00, P < 0.0001), APACHE II (17.00 vs 10.00, P < 0.0001) and SIRS scores (2.00 vs 1.00, P < 0.0001), compared to recovered patients (Table 1).

SARS-COV-2 infection can cause both pulmonary and multi-system inflammation, leading critical complications (Table 2). The frequency of acute cardiac injury (79.72% vs 11.80%, P < 0.0001), heart failure (71.53% vs 6.32%, P < 0.0001), acute kidney injury (48.40% vs 4.35%, P < 0.0001), acute liver injury (24.20% vs 2.39%, P < 0.0001), acute pancreatic injury (4.27% vs 1.54%, P = 0.010) especially acute respiratory distress syndrome (ARDS, 96.80% vs 53.79%, P < 0.0001) and disseminated intravascular coagulation (DIC, 19.22% vs 0.56%, P < 0.0001) were more higher in non-survivors than survivors. These critical complications could be the main cause of death, and the underlying mechanisms warrant further investigations.

Almost all non-survivors received antibiotic treatment (93.95% vs 88.48%, P = 0.0094), more than the number of recovered patients (Table 2). Fatal patients received more glucocorticoid therapy, intravenous immunoglobulin therapy or transfusion than recovered cases, since the cytokine storm and DIC were more often observed in non-survivors. Similarly, due to the higher proportions of patients developed acute kidney injury or ARDS in non-survivors, deceased patients received more mechanical ventilation, continuous renal replacement therapy, high flow nasal cannula and extracorporeal membrane oxygenation than recovered patients. These factors might result in more frequent ICU admission (52.69% vs 0.56%, P < 0.0001) and shorter time of hospital stay (22[IQR 15–28] vs 8 [IQR 4–14], P < 0.0001) in non-survivors than survivors. The clinical interventions were more intensive in non-survivors due to the more severe illness in fatal cases. Of note, recovered patients were undergoing more, antiviral therapy (91.15% vs 80.07%, P < 0.0001) as compared to decreased cases.

Furthermore, we performed Cox analysis to identify the potential death risk factors of COVID-19 with adjustment of age, sex, comorbidities. As shown in Table 3 and Additional file 1: Table S2, the higher temperature, faster heart rate and respiratory rate and lower mean arterial pressure were associated with increased risk of death. In terms of laboratory parameters, elevated inflammatory cytokines and infection-related factors, such as TNF-α, IL-6, IL-10, IL-8, IL-1β, IL-2R, ferritin, hs-CRP and procalcitonin were significantly associated with higher death risk of COVID-19. Conversely, the immune cells subsets, such as lymphocytes (HR 0.220, 95% CI 0.161–0.302, P < 0.0001), B cells (HR 0.994, 95% CI 0.990–0.998, P = 0.002), NK cells (HR 0.985, 95% CI 0.981–0.990, P < 0.0001), CD3⁺CD19⁻ T cells (HR 0.996, 95% CI 0.995–0.997, P < 0.0001) and CD3⁺CD8⁺ T cells (HR 0.989, 95% CI 0.985–0.992, P < 0.0001), were significantly associated with the lower death risk of COVID-19.

Indicators that represented organ damages, such as elevated ALT, AST, LDH, creatinine, amylase, hs-cTnI, NT-proBNP, CK-MB, and decreased ALB/GLO aggrandized the risk for COVID-19 death. Furthermore, coagulation-related biomarkers, including declined platelet counts (HR 0.994, 95% CI 0.992–0.995, P < 0.0001), and increased PT, APTT, D-Dimer (HR 1.025–1.359, P < 0.0001), might be strong indicators for death risk of COVID-19. Moreover, apart from abnormal biochemical dynamics, metabolism indices and blood gas analysis such as decreased serum level of PaO₂, SaO₂, PaCO₂, and CtCO₂ could result in the degree of electrolyte disturbance or hypoxia, increasing the risk of death of COVID-19.

We also analyzed the risk of death for patients with complications, patients with ARDS (HR 16.216, 95% CI 8.341–31.527, P < 0.0001) had highest risk of death, followed by acute cardiac injury (HR 13.023, 95% CI 9.678–17.524, P < 0.0001), heart failure (HR 10.722, 95% CI 8.227–13.974, P < 0.0001), acute kidney injury (HR 6.063, 95% CI 4.759–7.724, P < 0.0001) and DIC (HR 5.819, 95% CI 4.281–7.910, P < 0.0001). Besides, the higher levels of SOFA, qSOFA, APACHE II and SIRS scores were significantly associated with increased death risk (HR 1.195–3.471, P < 0.0001). These findings provided evidence supporting that dynamic of inflammatory cytokines, immune cells subsets, blood gas, organ damage biomarkers, and especially complications should be closely monitored, in case of poor outcomes.

In light of the SOFA, qSOFA, APACHE II and SIRS scores have been reported as good diagnostic indicators for sepsis, septic shock and multi-organ failure [3, 4, 6‐8, 16], we then calculated the prediction accuracy of these four scores in assessing death risk of COVID-19. As presented in Fig. 2, these four scores had prominent prediction capacities evaluating COVID-19 death risk. The AUCs of SOFA, qSOFA, APACHE II and SIRS scores attained 0.697(95% CI 0.546–0.849), 0.610(95% CI 0.474–0.747), 0.826(95% CI 0.671–0.981) and 0.749(95% CI 0.629–0.869). Of note, discrimination of death risk models were better using APACHE II (AUC = 0.826, P = 0.022) and SIRS scores (AUC = 0.749, P = 0.013) than SOFA or qSOFA scores, which might partly be attributed to aggravated pro-inflammatory responses in non-survivors.

Besides, we also established another four prediction models based on inflammatory-related indices, immune cells subsets, organ damage biomarkers and complications, all of which were significantly associated with the COVID-19 death risk. Among death risk prediction models of each group alone, the predictive accuracy of the immune cells subsets group was the highest (AUC = 0.901, 95% CI 0.801–1.000). Similarly, multiple-organ damage biomarkers (AUC = 0.894, 95% CI 0.829–0.959), inflammatory-related indices (AUC = 0.757, 95% CI 0.665–0.850) and complications (AUC = 0.878, 95% CI 0.817–0.938) had better predictive effects in the discrimination of mortality, outperforming abovementioned SOFA, qSOFA scores (P < 0.05) (Fig. 2).

Finally, we integrated four score predictive systems, inflammatory-related indices, immune cells subsets, organ damage biomarkers and complications to construct a combined group. The combined score (AUC = 0.950, 95% CI 0.853–1.000) was significantly higher than that of each risk group alone (Fig. 2), suggesting the combined score system can comprehensively reflect the death risk of COVID-19.