key: cord-0733316-jkmtpin4 authors: Chen, Chuming; Wang, Haihui; Liang, Zhichao; Peng, Ling; Zhao, Fang; Yang, Liuqing; Cao, Mengli; Wu, Weibo; Jiang, Xiao; Zhang, Peiyan; Li, Yinfeng; Chen, Li; Feng, Shiyan; Li, Jianming; Meng, Lingxiang; Wu, Huishan; Wang, Fuxiang; Liu, Quanying; Liu, Yingxia title: Predicting Illness Severity and Short-Term Outcomes of COVID-19: A Retrospective Cohort Study in China date: 2020-05-21 journal: The Innovation DOI: 10.1016/j.xinn.2020.04.007 sha: 18aa00ecfa05a716590998599ff9ffa9a404d622 doc_id: 733316 cord_uid: jkmtpin4 Among 417 COVID-19 patients in Shenzhen, demographic characteristics, clinical manifestations and baseline laboratory tests showed significant differences between mild-moderate cohort and severe-critical cohort.Based on these differences, a convenient mathematical model was established to predict the illness severity of COVID-19. The model includes four parameters: age, BMI, CD4(+) lymphocytes and IL-6 levels. The AUC of the model is 0.911.The high risk factors for developing to severe COVID-19 are: age ≥ 55 years, BMI > 27 kg / m(2), IL-6 ≥ 20 pg / ml, CD4(+) T cell ≤ 400 count / μ L.Among 249 discharged COVID-19 patients, those who recovered after 20 days had a lower count of platelet, a higher level of estimated glomerular filtration rate, and higher level of interleukin-6 and myoglobin than those who recovered within 20 days. COVID-19, caused by SARS-CoV-2, is a highly contagious disease. 1 By April 8, 2020, more than 1,350,000 patients were diagnosed with COVID-19 globally, with more than 79,000 deaths worldwide attributable to the disease. 2 Recent clinical data reported that mild and critical patients manifested different symptoms. Most of the mild patients with COVID-19 had symptoms such as fever, cough, and mild pneumonia, whereas the critical cases presented dyspnea, respiratory failure, sepsis, organ dysfunction, and even eventual death. [3] [4] [5] Therefore, we hypothesize that the clinical and laboratory characteristics may be associated with patients' severity and short-term outcomes. A large cohort study indicated that the total mortality rate of COVID-19 was 2.3%; however, the rate of severe cases was 8.1% while that of critical cases reached 14%. 6 In particular, patients with complications, such as septic shock and organ dysfunction, had the highest mortality rate. 4, 5 The median length of severe patients' aggravation period was 10 days. 5 Without specific antiviral agents, an early intervention and supportive treatments are essential for improving clinical outcomes and reducing mortality in severe patients. The ongoing outbreak of COVID-19 has been overloading medical systems worldwide. An accurate prediction of disease progression is conducive to informative clinical decision making and optimized allocations of hospital resources. Several modeling studies on COVID-19 have been reported to predict the spreading profile using the number of infected patients and transportation, 7, 8 as well as to predict the illness severity and mortality based on the clinical features. 9 In our previous clinical practice and research, we found that the individual immune status plays an important role in the progress and prognosis of the disease. 10, 11 In this study we aimed to build a model combining clinical characteristics and laboratory findings, including the immune status, to predict the progression and outcomes of COVID-19. Our study included all 417 confirmed COVID-19 cases in Shenzhen by February 29, 2020, who were all admitted to the Third People's Hospital of Shenzhen. The patients were divided into two groups, comprising 325 (77.9%) mildly to moderately ill patients and 92 (22.1%) severely to critically ill patients. The sample consisted of 198 males (47.5%) and 219 females (52.5%), but the sex ratio was skewed toward males in the severe-critical group (64.1%). No patient had an exposure history (shopping) in the Huanan Seafood Market in Wuhan, although 309 (74.1%) had a traveling history in Hubei province. In addition, 279 (66.9%) patients had a fever before admission and 148 (25.5%) presented with cough. Other symptoms that were not common included expectoration (3.1%), headache (3.4%), myalgia (3.2%), chill (0.7%), nausea or vomiting (0.2%), and diarrhea (2.6%). Most symptom profiles were comparable between the mild-moderate group (n = 325) and the severe-critical group (n = 92) while fever and cough occurred in a significantly higher proportion among severely to critically ill patients (p < 0.001 for fever and p = 0.015 for cough). Hypertension (23 [5.5%]), chronic liver disease (23 [5.5%]), and diabetes (15 [10.1%]) were the most common co-existing conditions. Moreover, 367 (88%) patients had pneumonia first time on admission (Table 1) . Levels of white blood cells, hemoglobin, triglyceride, low-density lipoprotein, high-density lipoprotein, and cholesterol did not differ between the two groups. These measures were recorded on the day of admission for all patients. They were then divided into groups according to illness severity. There were numerous differences in laboratory findings between the two groups, including the level of lymphocytes ( On the day of admission, those of the 249 discharged patients who recovered after 20 days had a lower count of platelets, a higher estimated glomerular filtration rate, and higher level of IL-6 and myoglobin than those who recovered within 20 days; among 249 discharged patients, the median level of platelets, estimated Figure 1 shows the sensitivity and specificity of our model. After model training, cut-off values of the four variables were selected based on statistical results and clinical experience. The median probability of progressing to Values are number (%) except where indicated otherwise. p values indicate differences between onset to discharged date %20 days and >20 days. p < 0.05 was considered statistically significant. The Innovation severely to critically ill was significantly higher in patients with a BMI greater than 27 than in those with a BMI of or below 27 (53.18% versus 7.76%). This trend remained for the parameters of CD4 + lymphocytes, IL-6, and age. Patients with a CD4 + T lymphocyte level of %400/mL showed a lower median probability of progressing to severe-critical status (62.52%) than those with a level >400/mL. Patients with level of IL-6 R20 pg/mL were more likely to develop into a severely to critically ill case (median probability 70.01%) than those whose level was lower than 20 pg/mL (3.80%). Patients aged R55 years had a higher probability of developing into a severe-critical case (41.15%) than those younger than 55 (1.73%). The logistic regression model was then retrained with two separate ranges under these four variables. For model validation, we applied a repeated k-fold cross-validation method of 10-fold with 10 repeats. Values of R 2 , root mean squared error (RMSE), and mean absolute error (MAE) were used for model performance evaluation. Results showed R 2 = 0.596, MAE = 0.21, and RMSE = 0.31. In the end, we predicted the probability for each of 16 (2 4 = 16) types of initial test results, shown in Figure 2 . These results represent potential indicators for early initial auxiliary prediction. All laboratory values are in the form of medians(interquartile range). p values indicate differences between mild-moderate and severe-critical patients. p < 0.05 was considered statistically significant. WBC, white blood cells; LYM, lymphocytes; NEU, neutrophils; HGB, hemoglobin; PLT, platelets; CD4 + , CD4 + T lymphocyte; CD8 + , CD8 + T lymphocyte; D-DIC, D-dimer; ALT, alanine aminotransferase; AST, aspartate aminotransferase; K, potassium; Na, sodium; Cr, chromium; eGFR, estimated glomerular filtration rate; CK, creatine kinase; MYO, myoglobin; BNP, brain natriuretic peptide; TnI, troponin I; LDH, lactate dehydrogenase; TG, triglyceride; LDL, low-density lipoprotein; HDL, high-density lipoprotein; CHOL, cholesterol; PCT, procalcitonin; CRP, C-reactive protein; IL-6, interleukin-6; ESR, erythrocyte sedimentation rate. The Innovation Discussion COVID-19 has been spreading worldwide, 2 putting a huge amount of pressure on medical systems in a short time. Accurate prediction of disease progression for COVID-19 patients is therefore urgent to facilitate appropriate clinical decision making and optimized medical resource allocations. Risk stratification management will help to alleviate burdens of insufficient medical resources and further reduce mortality. 12 Our current research showed that there were significant differences between the mild-moderate group and the severe-critical group concerning several epidemiological and clinical features (age, gender, and fever). Results were consistent with previous reports that age, gender, and underlying diseases might be associated with disease severity. 4, 9 In terms of laboratory data, 18 out of 27 tested baseline parameters, including levels of blood cells, indicators of myocardial zymogram, liver and renal function, indicators of coagulation function, and levels of infection-related biomarkers were significantly different between the two groups. Hence, we recommended four parameters-age, BMI, and levels of CD4 + T lymphocytes and IL-6-as predictors in future multivar-iate regression analysis. The elderly (R55 years), a high level of BMI (>27 kg/m 2 ), a low level of CD4 + T lymphocytes (%400/mL), and a high level of IL-6 (R20 pg/mL) are markers associated with the progression to severe and critical illness among COVID-19 patients. In a large cohort study of COVID-19 in China, the median age of severe cases was 52 years while that of mild cases was 45 years. 4 Consistent with our findings, an advanced age was associated with immunity decline characterized by decreases in both humoral and cellular responses according to previous research. 13 Although underlying reasons remained unknown, we found a correlation between BMI and disease severity. This finding was in line with results showing BMI as an independent risk factor that could predict the incidence of pneumonia among patients with chronic obstructive pulmonary disease. Lymphocytopenia was present in 83.2% of the hospitalized patients. Severe patients showed more prominent laboratory abnormalities, including lymphocytopenia and leukopenia, than mild ones. 6 Among the severe-critical cohort, we also observed lower levels of lymphocytes and CD4 + T lymphocytes. Those with a level of CD4 + T lymphocytes at or below 400/mL suffered a higher risk of progression to severe cases. The cytokine storm caused by a vigorous immune response against the virus was a significant predicting factor for the severity of viral pneumonia. Plasma levels of IL-2, IL-7, IL-10, granulocyte colony-stimulating factor, interferon-g-inducible protein-10, monocyte chemotattractant protein-1, macrophage inflammatory protein-1a, and tumor necrosis factor a in COVID-19 were higher in intensive care unit (ICU) cases than non-ICU cases. 4 Being one of the important cytokines, IL-6 could complement the function of CD8 + T lymphocytes and might aggravate the clinical course. 14 Compared with other cytokines, IL-6 is more widely used and more meaningful clinically. Data for the four clinical test indexes proposed by our model are easy to obtain. Therefore, we recommend that all related characteristics should be tested for patients immediately after admission. Our model reflected The Innovation the real-world scenario in clinical practice. Receiveroperating characteristic (ROC) analysis suggested that this model had functional capacity in predicting severity among COVID-19 patients (AUC = 0.911). The score may be regarded as a predictor of prognosis to prompt appropriate early intervention. Hence, we present the probability graph of progressing to severe or critical illness, which could be of clinical reference for early diagnosis. Despite remarkable findings, the study has a few limitations. First, we did not measure broad-spectrum cytokines and viral loads, which may be related to disease progression and severity. Second, the retrospective single-center design led to missing data and unavoidable bias when identifying participants. A larger sample size would be in favor the generalizability of a prediction model. This model would need to be further verified in other cohorts and in clinical practice. Moreover, we did not compare our model performance with alternative machine-learning models, such as the random forest model and the extreme gradient boosting model. It will add value to compare efficacies of different models in the future. In summary, we proposed a model containing four routine baseline parameters. This allowed accurate prediction of severity among COVID-19 patients. The model accurately stratified these patients into relevant risk categories. It could further help to produce appropriate clinical decisions and optimized allocations of hospital resources when shifting attention to the server during the pandemic. Nasopharyngeal and oropharyngeal specimens for SARS-CoV-2 tests were collected with synthetic fiber swabs under the guidelines introduced by either the World Health Organization (WHO) or the Figure 2 The Probability under 16 Types of Initial Test Results Older age (R55 years), high BMI (>27 kg/m 2 ), low level of CD4 + T lymphocytes (%400/mL), and high level of IL-6 (R20 pg/mL) have the highest probability of progressing into severe-critical COVID-19. Index units: age (years), BMI (kg/m 2 ), IL-6 (pg/mL), CD4 + T lymphocytes (count/mL). The Innovation Chinese Center for Disease Control and Prevention (Chinese CDC). Stored at 2 C-8 C, these specimens were packaged with ice and ship- In the following analysis, we divided patients into two cohorts, namely a mild-moderate cohort and a severe-critical cohort according to the above classification criteria. The COVID-19 guidelines (seventh version) issued by the National Health Commission of the People' Republic of China required that all patients must meet certain criteria before being discharged. 16 Continuous variables were described using median and interquartile range (IQR) values and were compared using the Mann-Whitney U test. Categorical variables were described as frequency rates, percentages, and proportions and were compared using Chi-square test. When a p value of Chi-square test was around a (0.05), attention was paid to ensure that the approximation of the Chi-square test met with requirements. Under this circumstance, Fisher's test was performed instead when necessary. 17 Pairwise comparisons between groups were performed using Bonferroni's correction. All p values were from two-sided tests, and results were deemed statistically significant at p < 0.05. All statistical analysis was done using R (version 3.6.2). We selected statistically significant variables to develop a straightforward clinical predicting tool-a logistic regression model-used for early diagnosis. Variables with an insignificant p value were consid-ered as irrelevant and were eliminated during the training process. The area under the ROC curves in the model's sensitivity and specificity results and repeated k-fold cross-validation results were used to assess how well the model performed. Cut-off values of each variable were obtained from statistical results and clinical experience. We then retrained the clinical predicting model to establish an auxiliary diagnosis table showing the probability of developing into a severe-critical case. None. Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia Coronavirus disease 2019 (COVID-19) situation report-79 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 Clinical characteristics of coronavirus disease 2019 in China Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in Wuhan, China: a modelling study An updated estimation of the risk of transmission of the novel coronavirus (2019-nCov) Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study Treatment of 5 critically ill patients with COVID-19 with convalescent plasma Clinical and biochemical indexes from 2019-nCoV infected patients linked to viral loads and lung injury Potential association between COVID-19 mortality and health-care resource availability update World Health Organization. Clinical management of severe acute respiratory infection when Novel coronavirus (nCoV) infection is suspected: interim guidance Statistical Methods for Research Workers We thank the patients, nurses, and physicians who provided care for the patients, and the investigators at the Third People's Hospital of Shenzhen. This work is supported by grants from the Science and Technology Innovation Committee of Shenzhen Municipality (202002073000001), National Key Research and Development Program (2020YFC0841700), High-level University Fund (no. G02386301, G02386401), and Guangdong Natural Science Foundation Joint Fund (no. 2019A1515111038).