key: cord-350338-lcsa06gm
authors: Wang, Kun; Zuo, Peiyuan; Liu, Yuwei; Zhang, Meng; Zhao, Xiaofang; Xie, Songpu; Zhang, Hao; Chen, Xinglin; Liu, Chengyun
title: Clinical and laboratory predictors of in-hospital mortality in patients with COVID-19: a cohort study in Wuhan, China
date: 2020-05-03
journal: Clin Infect Dis
DOI: 10.1093/cid/ciaa538
sha: 
doc_id: 350338
cord_uid: lcsa06gm

BACKGROUND: This study aimed to develop mortality-prediction models for patients with Coronavirus disease 2019 (COVID-19). METHODS: The training cohort were consecutive patients with COVID-19 in the First People’s Hospital of Jiangxia District in Wuhan from January 7, 2020 to February 11, 2020. We selected baseline clinical and laboratory data through the stepwise Akaike information criterion and ensemble XGBoost model to build mortality-prediction models. We then validated these models by randomly collecting COVID-19 patients in the Infection department of Union Hospital in Wuhan from January 1, 2020, to February 20, 2020. RESULTS: 296 patients with COVID-19 were enrolled in the training cohort, 19 of whom died during hospitalization and 277 were discharged from the hospital. The clinical model developed with age, history of hypertension and coronary heart disease showed AUC of 0.88 (95% CI, 0.80-0.95); threshold, -2.6551; sensitivity, 92.31%; specificity, 77.44% and negative predictive value (NPV), 99.34%. The laboratory model developed with age, high-sensitivity C-reactive protein (hsCRP), peripheral capillary oxygen saturation (SpO2), neutrophil and lymphocyte count, D-dimer, aspartate aminotransferase (AST) and glomerular filtration rate (GFR) had a significantly stronger discriminatory power than the clinical model (p=0.0157), with AUC of 0.98 (95% CI, 0.92-0.99); threshold, -2.998; sensitivity, 100.00%; specificity, 92.82% and NPV, 100.00%. In the subsequent validation cohort (N=44), the AUCs (95% CI) were 0.83 (0.68, 0.93) and 0.88 (0.75, 0.96) for clinical model and laboratory model, respectively. CONCLUSIONS: We developed two predictive models for the in-hospital mortality of patients with COVID-19 in Wuhan and validated in patients from another center.

We developed a clinical model and laboratory model for predicting the in-hospital mortality of COVID-19 patients, the AUCs (95% CI) were 0.88 (0.80, 0.95) and 0.98 (0.92, 0.99) in training cohort, and 0.83 (0.68, 0.93) and 0.88 (0.77, 0.95) in validation cohort, respectively.

Several cases of "unknown viral pneumonia" have been reported in Wuhan, Hubei 2 Province, China since December 2019. The causative agent was revealed as a novel 3 coronavirus named as severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) by the 4

International Committee on Taxonomy of Viruses. The disease caused by SARS-CoV-2 was 5 named coronavirus disease 2019 (COVID-19) by the World Health Organization (WHO). 1 6 This infectious disease has rapidly spread from Wuhan to other Chinese regions 2 . Since mid-7 march 2020, cases have been detected in most countries worldwide and community spread is Mild acute respiratory infection symptoms, such as fever, dry cough, and fatigue, 13 commonly occur in the early stages of COVID-19, but some patients might rapidly develop 14 acute respiratory distress syndrome, acute respiratory failure, multiple organ failure, and other 15 fatal complications. 3, 4 No specific treatment has been fully developed for COVID-19; thus, 16 early identification of patients with poor prognosis may facilitate the provision of proper 17 supportive treatment in advance and reduce mortality. 

The participants in the training cohort were all the consecutive patients diagnosed with 28 COVID-19 in the First People's Hospital of Jiangxia District in Wuhan, a major hospital in 29 the Jiangxia District. We collected data on patients hospitalized from January 7, 2020, 17:58 30

to February 11, 2020, 22:01. A total of 296 patients with final outcome (i.e. discharged or 31 dead) were enrolled in this study before February 12, 2020, 14:00. We then randomly 32 collected patients with COVID-19 who had been hospitalized in the Infection department of 33

Union Hospital in Wuhan from January 1, 2020, to February 20, 2020 to form our validation 34 cohort. A flow diagram is showed in Figure 1 . 35

The data of these participants were used to construct two predictive models for in-36 hospital mortality. The study protocol was approved by the Medical Ethics Committee of the 37 First People's Hospital of Jiangxia District and Union Hospital, and was complied with the 38 Declaration of Helsinki. We verbally informed the patients that their data would be used 39 anonymously for medical studies and obtained their permission. Written informed consent 40 was not gathered, because the data were anonymous and the study was observational. 41

Previous medical history, age, cough and fever (the oral temperature>37.5 ℃, the 43 axillary temperature>37℃, or the body temperature fluctuates more than 1℃ in a day) for 44 every subject were obtained by trained nurses. The laboratory data of the first examination 45 after admission of every subject were also collected. 46

All blood and urinary samples were processed within two hours of collection. Routine 47 106 range (IQR), 8.6-15.5] days, respectively. The mean and median hospital stay of the 107 survivors were 6.2 ± 5.0 and 4.9 (IQR, 2.6-10.5) days, respectively. The mean and median 108 time interval between symptom onset and admission of the non-survivors were 5.2 ± 3.7 and 109 5.0 (IQR, 3.0-7.0) days, respectively. And for the survivors were 6.8 ± 4.0 and 5.5 (IQR, 3.0-110 9.2) days, respectively. 111

Baseline clinical and laboratory characteristics of study population by training and 112 validation cohort are shown in Table 1 . We observed significant differences between the two 113 cohorts in age, outcome, symptoms, and clinical indicators. The patients in validation cohort 114 were remarkably older, with higher rates of diabetes and hypertension, lower SpO2, and 115 worse markers of inflammation, clotting status, and liver and kidney function. 116

The comparison between the survivors and the non-survivors were shown in Table 2 . The 117 mean age of the non-survivor group was remarkably higher than that of the survivor group in 118 both cohorts. Medical history showed that the non-survivor group had a higher proportion of 119 basic disease. No substantial difference was observed in the sex composition and habits of 120 smoking and drinking between survivors and non-survivors. In the training cohort, non-121 survivors had remarkably lower SpO2 than survivors. Inflammatory cells, namely, WBC and 122 neutrophil, were considerably higher whereas lymphocyte was remarkably lower in the non-123 survivor group than in the survivor group. Meanwhile, hsCRP, a marker of inflammation, was 124 also substantially elevated in the non-survivor group. In terms of blood coagulation indexes, 125 the non-survivor group had higher D-dimer and thrombin time and lower activated partial 126 thromboplastin time than the survivor group. Cr, BUN, ALT, AST, LDH, and blood ammonia 127 were remarkably higher whereas GFR and serum ALB were significantly lower in the non-128 survivor group. 129

In the model-development phase, the clinical model developed according to age, history 130 of hypertension and coronary heart disease showed good discriminatory power with AUC of Table 4 ) 142

The ROC of the two models in training and validation cohort were plotted in Figure 2 . exhibited relatively good discriminatory power the and the external verification was also 151 satisfactory. We believe that this is the first study to establish models for predicting the 152 mortality of patients with COVID- 19 . 153

The clinical model based on age, history of hypertension, and coronary heart disease had 154 achieved good predictive power. Elderly people are at higher risks for chronic diseases and 155 more susceptible to infection. Age might be the risk factor for worse outcomes in patients 156 with COVID-19 partially because age-related immune dysfunctions result from low-grade 157 chronic inflammation according to our speculation. 5, 11 In addition, elderly patients may 158 possess other risk factors, such as comorbidities and sarcopenia. Hypertension is one of the 159 most common diseases in the elderly. History of hypertension is an important risk indicator in 160 the MuLBSTA score, which is a viral pneumonia death warning model developed by Chinese 161 scholars. 12 Our results are consistent with the above research. In addition, angiotensin- with CHD history and infected with SARS-CoV-2 has to work harder to ensure that sufficient 168 blood oxygen is provided throughout the body. The problem of increased heart burden will 169 become more prominent. Reasonable precautions must be taken to prevent these patients from 170 the viral infection. 171

XGBoost showed that hsCRP was the most important predictor for the mortality of patients 172 with COVID-19, followed by age, SpO2, AST, neutrophil count, D-dimer, GFR and 173 lymphocyte count. This finding is consistent with our clinical observation. 174

A low SpO2 level suggests that the patients might have a serious illness at the time of 175 admission. We found that most of the patients with COVID-19 had mild acute respiratory 176 infection symptoms initially; however, the conditions of some patients would rapidly 177 exacerbate and result in multiple organ failure or even death. We suspected this exacerbation 178 was primarily due to the "cytokine storm" and consequent immunologic abnormality. 179

Cytokine storm is an important cause of death in severe acute respiratory syndrome (SARS), 180

Middle East respiratory syndrome coronavirus, and influenza A virus subtype H1N1 181 infection. [15] [16] [17] Cytokine storm also seems to be a remarkable mechanism in the present 182 outbreak of COVID-19 and contributed to the death of several patients, especially young 183 patients. A recent study showed that patients requiring ICU admission had higher 184 concentrations of granulocyte colony-stimulating factor, interferon-induced protein 10, 185 monocyte chemoattractant protein 1, macrophage inflammatory protein 1 alpha, and tumor 186 necrosis factor alpha than those who did not require ICU admission, suggesting that cytokine 187 storm is associated with disease severity. 4 A remarkable finding of our study was that the 188 increasing level of hsCRP and neutrophil counts had prominent power in predicting fatal 189 outcomes in patients with COVID-19. Neutrophil chemotaxis and transmigration are essential 190 components for host defense during infections, but excessive neutrophil infiltration 191 contributes to deleterious inflammatory processes, 18 which might deeply interact with 192 cytokine storm during virus invasion. 193

The substantially depressed total lymphocytes in the non-survivor group indicated that 194 SARS-CoV-2 might act on T lymphocytes, and high replication of the virus leads to the 195 depletion of T lymphocytes, which suppresses the body's immunity. 19 In addition, patients 196 with severe illness are more likely to be co-infected with bacteria because of depressed 197 immune function, which is another possible reason for the increased level of neutrophils and 198

hsCRP. Further studies are necessary to elucidate the cytokine storm and immunologic 199 abnormality in SARS-CoV-2 infection. 200

We found that coagulation indicators might play a role in identifying severe cases as 201 well. We observed that D-dimer was negatively associated with in-hospital mortality. high specificity and PPV were demonstrated in clinical models in the validation cohort, as 244 opposed to the training cohort. We hypothesized that the probable reason was that there were 245 more deaths in patients with a history of hypertensive or coronary artery disease in the 246 validation cohort. More external validation is needed to demonstrate the robustness of the 247 model, and we currently recommend that clinical models with limited information only be 248 used for preliminary screening of high-risk populations. 249

By comparing the training and validation populations in Table 1 , we had observed 250 significant differences between the two groups in age, symptoms, and examination index 251 

WK and LC conceived and designed the study. WK, ZP, CX analyzed the data, and wrote the first draft of the manuscript. WK, LY, ZM, ZX, XS and ZH recruited patients, gathered data and participated in manuscript revision. LC provided study oversight and participated in manuscript revision. All authors had access to study data and approved the decision to submit the manuscript.

We thank the patients and families who agreed to participate in this important study. We 

There was no funding directly relevant to this work.

Chengyun Liu received a grant from the National Natural Science Foundation of China (81974222) within the last 36 months, which was not directly related to this study. And all of the authors declare that they have no conflict of interest. ROC curves of in-hospital mortality from logistic regression models of patients with clinical data (red ) and laboratory data (black) using Bootstrap resampling (times = 500). ROC = receiver operator characteristic. AUC = area under the curve.

SpO2=peripheral capillary oxygen saturation. AST=aspartate aminotransferase. GFR= glomerular filtration rate. ALB=albumin. GLO=globulin. CK=creatine kinase.

-Data not collected in the validation cohort. AUC=area under the curve. AIC= Akaike information criterion.

*Bootstrap resampling (times = 500). 

Outbreak of COVID-19-an urgent need for good science to silence our fears?

Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in Wuhan, China: a modelling study

Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study

Clinical features of patients infected with 2019 novel coronavirus in Wuhan

Clinical Characteristics of 138 Hospitalized Patients With 2019 Novel Coronavirus-Infected Pneumonia in Wuhan, China

A familial cluster of pneumonia associated with the 2019 novel coronavirus indicating person-to-person transmission: a study of a family cluster

Machine learning for the prediction of volume responsiveness in patients with oliguric acute kidney injury in critical care

Boosted tree model reforms multimodal magnetic resonance imaging infarct prediction in acute stroke

Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD)

Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach

Innate immune responses in the ageing lung

Clinical Features Predicting Mortality Risk in Patients With Viral Pneumonia: The MuLBSTA Score

Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor binding

From gene to proteinexperimental and clinical studies of ACE2 in blood pressure control and arterial hypertension

Preparing for the next influenza pandemic

An interferon-γ-related cytokine storm in SARS patients

Corticosteroid Treatment Ameliorates Acute Lung Injury Induced by 2009 Swine Origin Influenza A (H1N1) Virus in Mice

The trafficking protein JFC1 regulates Rac1-GTP localization at the uropod controlling neutrophil chemotaxis and in vivo migration

Understanding the T cell immune response in SARS coronavirus infection

Biochemical Aspects of Coronavirus Replication and Virus-Host Interaction

Abnormal Coagulation parameters are associated with poor prognosis in patients with novel coronavirus pneumonia

Specific ACE2 Expression in Cholangiocytes May Cause Liver Damage After 2019-nCoV Infection

Clinical course and outcomes of critically ill patients with SARS-CoV-2 pneumonia in Wuhan, China: a single-centered, retrospective, observational study

Clinical characteristics of novel coronavirus cases in tertiary hospitals in Hubei Province

25 National health commission. Treatment scheme of COVID-19