Machine Learning-Based Scoring System for Early Prognosis Evaluation of Patients With Coronavirus Disease 2019 : Infectious Diseases & Immunity

Secondary Logo

Journal Logo

Original Article

Machine Learning-Based Scoring System for Early Prognosis Evaluation of Patients With Coronavirus Disease 2019

Zhang, Hao-Min1; Shi, Lei2,3; Chen, Hao-Ran4; Zhang, Jun-Dong1,5; Liu, Ge-Liang4; Wang, Zi-Ning1,5; Zhi, Peng1,4; Wang, Run-Sheng6; Li, Zhuo-Yang4; Chen, Xi-Meng1; Wang, Fu-Sheng2,3; Lu, Xue-Chun1,3,4

Editor(s): Wang, Haijuan

Author Information
Infectious Diseases & Immunity ():10.1097/ID9.0000000000000077, December 29, 2022. | DOI: 10.1097/ID9.0000000000000077

Abstract

Introduction

The coronavirus disease 2019 (COVID-19) is wreaking havoc worldwide. Currently, there is no specific therapeutic drug for this disease,[1] which is particularly concerning in the short term. Globally, as of August 26, 2022, there were 596,873,121 confirmed cases of COVID-19, including 6,459,684 deaths reported to the World Health Organization.[2] These numbers are still increasing. It is difficult to quickly reverse this disaster. Most COVID-19 cases are mild, but a proportion of them are critical, with rapid disease change, poor prognosis, and high mortality (39.2 of 100,000).[3,4] Combined with established clinical treatment in China, early identification and early and active intervention measures are the keys to effectively reducing the clinical stage and obtaining a good prognosis in critically ill patients with COVID-19. This cross-sectional study reports a univariate and multivariate Cox risk ratio model and a machine learning random forest model used to screen the key factors affecting the prognosis of patients with COVID-19. This study considered the basic data and clinical outcomes of 3,974 patients with COVID-19 from the Wuhan Huo-Shen-Shan Hospital and Maternal and Child Hospital, Hubei Province, China. A COVID-19 death risk assessment scoring system was established to provide a reference for the early identification of patients with poor COVID-19 prognosis.

Methods

Ethical approval

The studies involving human participants were reviewed and approved by the ethics committee of the Fifth Medical Center of PLA General Hospital, former PLA no. 302 Hospital (2020013D).

Patients and study design

Medical records were continuously collected from patients with COVID-19 diagnosed in the laboratories of the Huo-Shen-Shan Hospital and Maternal and Child Hospital of Hubei Province. Records were selected for treatments completed between February 4 and April 16, 2020, the deadline for this study. All enrolled patients were diagnosed according to the Guidance for Coronavirus Disease 2019 (Seventh Edition).[5] Laboratory confirmation was defined as a positive result from high-throughput sequencing or reverse transcription polymerase chain reaction of nasal and pharyngeal swab samples.

The collected electronic medical records were organized, and data on admission and discharge times, and basic characteristics such as age, sex, previous serology and imaging tests, history, and complications were extracted. Severity at admission (classified as mild, severe, or critical) and end point at discharge (cure or death) were recorded using the Guidance for Coronavirus Disease 2019 (Seventh Edition). All medical records were reviewed and extracted by a specialist, and the data were cross-checked. If the core data were lost, the attending physician was contacted. If the data were not retrieved, the patient was removed from the study.

To ensure the reliability of the study results, the 3,974 patients were simply randomized into two groups. The training data set, which covered 2,649 patients (two thirds of the patients), was used to select key prognostic factors and establish a scoring system. The remaining 1,325 patients (one third of the patients) were used as test data to verify the effectiveness of the scoring system. The pipeline used in this study is shown in Figure 1.

F1
Figure 1:
Flow chart of enrollment of patients for this study. *Patients with COVID-19 whose outcome was not properly recorded.

Research definitions

The clinical type and cure criteria were determined according to the Guidance for Coronavirus Disease 2019 (Seventh Edition).[5] The age groups were as follows: juvenile, 7–17 years; youth, 18–40 years; middle-aged, 41–65 years; and senior, >65 years.

Determination of factors relevant to the outcome

This study analyzed the clinical outcomes, sociological characteristics, and clinical characteristics of the patients to determine the factors influencing their clinical outcomes.

Screening of key prognostic factors

A univariate Cox regression analysis was used to screen for factors affecting survival. Screening indicators were based on the following criteria: β (regression coefficient) > 0 and hazard ratio (HR) > 1 were risk factors, and β < 0 and HR < 1 were protective factors.

Machine learning subsequently established a random survival forest model for the prognostic factors that were initially screened by the univariate Cox analysis, to further screen them. The random forest model was applied using the randomForestSRC (2.9.3) software in the R language.[6] The model chose to build 4,000 trees based on error stability. The number of test set trees is selected to provide a stable error rate for the test set.

Finally, a multivariate Cox regression analysis of the important factors identified by the random forest model was performed to determine the synergistic effect of these factors on the prognosis of patients with COVID-19. The analysis used a threshold of P < 0.05, which was applied using the RMS (6.0) software package in the R language.[7]

The previously mentioned results were summarized and stored until the establishment of a scoring system for the early prognosis assessment of patients with COVID-19.

Establishment of an early prognosis evaluation score system for patients with COVID-19

An early prognosis evaluation integral system for COVID-19 patient was established using nomogram modeling, which established an integral rating system for important factors determined by the random survival forest model. The model was based on the RMS software package in R language. The median of the deceased patients’ points was calculated as the cutoff value to distinguish low risk from high risk. The C index was subsequently calculated, and calibration curves were established to evaluate the reliability of the scoring system. Using the DynNom (5.0) software package in the R language, we built a dynamic and interactive webpage based on the static nomogram to facilitate subsequent use by clinicians.[8]

Evaluation of the efficiency of an integral system

According to the proposed scoring system, 1,325 patients in the test set were scored, and the risk ranges were determined. Survival images were drawn to assess the survival distribution. The validity of the system was verified by calculating the C-index and plotting a correction curve.

Statistical analysis

Normally distributed continuous variables were presented as mean ± standard deviation and nonnormally distributed continuous variables as median (interquartile 1–interquartile 3). Categorical variables were summarized as number (percentage). Continuous variables were compared by the Student t test or the Mann-Whitney U test, and categorical variables by the χ2 test or Fisher exact test. P < 0.05 was considered statistically significant.

Results

Determination of factors relevant to the outcome

As of April 16, 2020, we collected data from 4,522 patients, of whom 548 were excluded because of incomplete data. The remaining 3,974 patients were included in the analysis. Supplementary Table S1, https://links.lww.com/IDI/A19, lists the basic information and assessment status of patients upon admission. The median age was 60 (49–68) years, with a range of 6–100 years. There were slightly more women (2,095 patients, 52.7%) than men (1,879 patients, 47.3%). Most cases (3463, 87.1%) were mild, 440 (11.1%) were severe, and 71 (1.8%) were critical. The median age in each group was 59 (48–67), 65 (53–73), and 72 (65–81) years, respectively. The age group distribution was as follows: 16 juvenile (0.4%), 534 young (13.4%), 2,121 middle-aged (53.4%), and 1,303 senior (32.8%) patients. The severity of COVID-19 was assessed in 1,994 patients with underlying disease (50.2%), of whom 1,669 (83.7%) were classified as mild, 268 (13.4%) as severe, and 57 (2.9%) as critical. The most common underlying diseases were hypertension (1,131 patients, 28.5%), diabetes (513 patients, 12.9%), and coronary heart disease (220 patients, 5.5%).

Table 1 lists the patients’ outcomes. Overall, 3,876 patients (97.5%) were cured and 98 died (2.5%). Among the cured patients, 16 (0.4%) were juvenile, 533 (13.8%) were young, 2,097 (54.1%) were middle-aged, and 1,230 (31.7%) were older adults. The median hospital stay for all patients, cured patients, and patients who died was 12 (8–18) days, 12 (8–18) days, and 9 (4–17) days, respectively. There was a significant difference between cured patients and those who died.

Table 1 - Patients’ discharge status and outcome
Variables All patients (n = 3974) Cured patients (n = 3876) Dead patients (n = 98) P
Age, [y, M (Q1–Q3)] 60 (49–68) 59 (49–68) 73 (65–81) <0.001
Juvenile [n (%)] 16 (0.4) 16 (0.4) 0 (0) 0.999
Youth [n (%)] 534 (13.4) 533 (13.8) 1 (1.0) <0.001
Middle-aged [n (%)] 2121 (53.4) 2097 (54.1) 24 (24.5) <0.001
Senior [n (%)] 1303 (32.8) 1230 (31.7) 73 (74.5) <0.001
Sex [n (%)] 0.001
Male 1879 (47.3) 1817 (46.9) 62 (63.3)
Female 2095 (52.7) 2059 (53.1) 36 (36.7)
Hospital stay [d, M (Q1–Q3)] 12 (8–18) 12 (8–18) 9 (4–17) 0.049
Mild 12 (8–17) 12 (8–17) 5 (2–10) <0.001
Severe 13 (8–19) 14 (8–20) 6 (4–10) <0.001
Critical 14 (6–21) 18 (12–23) 13 (6–21) 0.613
Underlying disease [n (%)] 1994 (50.2) 1915 (49.4) 79 (80.6) <0.001
Hypertension 1131 (28.5) 1084 (28.0) 47 (48.0) <0.001
Diabetes 513 (12.9) 488 (12.6) 25 (25.5) 0.001
Coronary heart disease 220 (5.5) 203 (5.2) 17 (17.3) <0.001
Digestive system chronic diseases 264 (6.6) 258 (6.7) 6 (6.1) 0.999
Chronic respiratory diseases 163 (4.1) 152 (3.9) 11 (11.2) 0.002
Chronic diseases of the urinary system 164 (4.1) 146 (3.8) 18 (18.4) <0.001
Blood diseases 16 (0.4) 12 (0.3) 4 (4.1) <0.001
Cerebrovascular disease 133 (3.3) 118 (3.0) 15 (15.3) <0.001
Malignant tumor 84 (2.1) 75 (1.9) 9 (9.2) <0.001
Gout 92 (2.3) 91 (2.3) 1 (1.0) 0.727
Autoimmune disease 38 (1.0) 36 (0.9) 2 (2.0) 0.241
Complication [n (%)] 450 (11.3) 378 (9.8) 72 (73.5) <0.001
Hypoproteinemia 75 (1.9) 56 (1.4) 19 (19.4) <0.001
Electrolyte disturbance 58 (1.5) 46 (1.2) 12 (12.2) <0.001
Arrhythmia 26 (0.7) 23 (0.6) 3 (3.1) 0.025
Liver failure 83 (2.1) 75 (1.9) 8 (8.2) <0.001
Anemia 109 (2.7) 99 (2.6) 10 (10.2) <0.001
DIC 2 (0.1) 0 (0) 2 (2.0) <0.001
Myocardial injury 4 (0.1) 2 (0.1) 2 (2.0) 0.004
Pleural effusion 18 (0.5) 14 (0.4) 4 (4.1) <0.001
Digestive tract bleeding 7 (0.2) 1 (0) 6 (6.1) <0.001
Lung infection 9 (0.2) 8 (0.2) 1 (1.0) 0.201
Acidosis 3 (0.1) 0 (0) 3 (3.1) <0.001
Hyperlactemia 2 (0.1) 0 (0) 2 (2.0) <0.001
Ascites 1 (0) 0 (0) 1 (1.0) 0.025
MODS 16 (0.4) 0 (0) 16 (16.3) <0.001
ARDS 23 (0.6) 5 (0.1) 18 (18.4) <0.001
Renal insufficiency 46 (1.2) 46 (1.2) 0 (0) 0.629
Septic shock 15 (0.4) 15 (0.4) 0 (0) 0.999
Respiratory failure 67 (1.7) 19 (0.5) 48 (49.0) <0.001
Cardiac insufficiency 56 (1.4) 31 (0.8) 25 (25.5) <0.001
ARDS: Acute respiratory distress syndrome; DIC: Disseminated intravascular coagulation; M (Q1–Q3): median (interquartile 1–interquartile 3); MODS: Multiple-organ dysfunction syndrome.

Altogether, 39 factors were associated with outcomes of patients with COVID-19, when P < 0.05 was used as the selection criterion. We made adjustments to combine various factors likely sharing a common mechanism as a single factor (acute respiratory distress syndrome and respiratory failure were merged into pulmonary insufficiency; myocardial injury, arrhythmia, and cardiac insufficiency were combined into heart dysfunction; the septic shock was merged with a lung infection for concurrent infection; coronary heart disease and cerebrovascular disease were combined as cardiocerebrovascular diseases). The remaining 32 factors were closely related to the prognosis of patients with COVID-19, including age (juvenile, youth, middle-aged, senior), sex (male, female), clinical type (mild, severe, critical), underlining diseases (hypertension, diabetes, cardiocerebrovascular diseases, chronic diseases of digestive system, chronic respiratory diseases, chronic diseases of urinary system, blood diseases, malignant tumor, autoimmune diseases), and complications (hypoproteinemia, electrolyte disturbance, liver failure, anemia, disseminated intravascular coagulation, pleural effusion, digestive tract bleeding, acidosis, hyperlacticemia, ascites, renal insufficiency, pulmonary insufficiency, heart dysfunction, coinfection).

Screening of key prognostic factors

First, based on the 32 factors identified as related to patient outcomes, a univariate Cox regression analysis was performed on the training data to determine the risk rate and statistical significance of each factor. Twenty-three statistically significant factors were initially identified (P < 0.05). Among them, protective factors include a mild clinical classification, the patient being young (18–40) or middle-aged (41–65), and being female (HR < 1). Risk factors (HR > 1) included the patient being male, showing critical symptoms, and having complications or underlying diseases [Table 2].

Table 2 - Univariate Cox regression analysis on the training set (n = 2649)
Type Variables HR β P
Protective factors Mild 0.03 −10.04 <0.001
Youth 0.12 −2.14 0.032
Middle-aged 0.25 −4.33 <0.001
Female 0.43 −2.96 0.003
Risk factors Male 2.30 2.96 0.003
Senior 6.40 6.03 <0.001
Critical 182.84 18.42 <0.001
Underlying diseases Hypertension 2.15 2.87 0.004
Malignant tumor 4.62 3.26 0.001
Chronic respiratory diseases 4.62 4.21 <0.001
Cardiocerebrovascular diseases 4.71 5.35 <0.001
Chronic diseases of urinary system 5.20 4.73 <0.001
Blood diseases 11.52 3.39 <0.001
Complications Liver failure 4.69 3.37 <0.050
Anemia 5.09 4.03 <0.001
Hypoproteinemia 12.54 7.25 <0.001
Electrolyte disturbance 12.56 6.96 <0.001
Heart dysfunction 23.69 10.91 <0.001
Digestive tract bleeding 49.61 6.56 <0.001
DIC 58.16 5.63 <0.001
Ascites 97.47 4.50 <0.001
Pulmonary insufficiency 98.62 17.09 <0.001
Acidosis 1240.64 7.96 <0.001
β: regression coefficient; DIC: Disseminated intravascular coagulation; HR: Hazard rate.

The data of 3,974 patients were randomly divided into two groups (two thirds [2649] were assigned to the training set and one third [1325] to the test set). A randomized survival forest model was built based on the results of a univariate Cox regression analysis in the training set, incorporating 23 statistically significant factors as input variables. The input data had no missing values and thus were not processed for missing values. First, 4,000 trees were constructed using the training data set, and the error rate of the test set was the output. We found that using approximately 1,000 trees stabilized the error rate at 6.7% (the accuracy rate was 93.3%; Figure S1A, https://links.lww.com/IDI/A20) to obtain the important ranking of variables in the training set. Subsequently, the variables with scores ≤0 were excluded, and the data set of significant variables was extracted for validation in the validation set. It was found that with a total of 4,000 trees, the error rate was stable at 4.1% and approximately 1,500 trees and was performing well. The remaining 18 factors had COVID-19 prognostic significance [Figure 2].

F2
Figure 2:
Random forest analysis. The order of importance of 23 factors, excluding the 18 factors with a score of ≤0.

Finally, multivariate Cox regression analysis was used to evaluate the synergy among the 18 factors screened. Age, clinical classification of severe and critical, complications of pulmonary insufficiency and hypoproteinemia, and underlying blood diseases were found to be statistically significant risk factors [Table 3].

Table 3 - Multivariate Cox regression analysis on the training set (n = 2649)
Variables HR β P
Sex 1.68 0.52 0.114
Age 1.04 0.03 0.005
Severe 5.79 1.76 <0.001
Critical 55.78 4.02 <0.001
Underlying diseases Blood diseases 5.67 1.73 0.040
Malignant tumor 0.67 −0.40 0.452
Complications Hypoproteinemia 3.14 1.14 0.004
Digestive tract bleeding 0.53 −0.64 0.365
Pulmonary insufficiency 5.13 1.63 <0.001
Heart dysfunction 1.03 0.03 0.951
β: Regression coefficient; HR: Hazard rate.

Combined with the previously mentioned clinical observations of patients with COVID-19 and the reported literature at this stage, we finally determined nine key prognostic factors (sex, age, clinical typing, underlying diseases including blood diseases, malignant tumors, and complications including hypoproteinemia, digestive tract bleeding, pulmonary insufficiency, and heart dysfunction) as parameters for the subsequent establishment of a scoring system.

Establishment of an early prognosis evaluation scoring system for patients with COVID-19

A nomogram algorithm was used to establish a COVID-19 death risk assessment scoring system for the nine key prognostic factors. Figure 3A shows the results, where each indicator for the patient corresponds to a point for each part, and the sum of the points is the patient’s risk assessment score. The scoring system had a C-index of 0.9655 and the correction curve is shown in Figures S2A and B, https://links.lww.com/IDI/A21. The median score of the 2,649 patients who died, who were included in the training set, was calculated to be 117.77, with a score ≥117.77 being defined as high risk and <117.77 as low risk. Using this cutoff value, survival images of 2,649 patients in the training set were drawn. These results were statistically significant [P < 0.001, Figure 3B]. An interactive webpage for the point prediction system is available online at: https://the-second-medical-center-of-pla-hospital.shinyapps.io/COVID-19_DRAS_E/.

F3
Figure 3:
COVID-19 risk nomogram, training set score, and validation. A, Each predictor is assessed a score on each axis. The sum of all scores for all predictors is calculated and expressed as the total score. B, The COVID-19 death risk assessment score system, where a score ≥117.77 indicates high risk and a score <117.77 indicates low risk. Survival analysis for the 2,649 patients in the training set after distinguishing the low- and high-risk groups according to the score threshold; the difference was statistically significant (P < 0.001). C, Validation results of 1,325 patients in the test set for the point system. Survival curves for these patients are divided into high- and low-risk groups according to cutoff value. The two groups are significantly different (P < 0.001).

Evaluation of the integral system

The scoring of 1,325 patients in the test set verified the point system. The calculated test set C-index was 0.9665 and the correction curve is shown in Figures S3A and B, https://links.lww.com/IDI/A22. The test-set patients were classified as high or low risk using the established cutoff value. The survival curves of the two groups were significantly different [P < 0.001, Figure 3B]; therefore, the effectiveness of the scoring system was initially verified.

Finally, to compare the evaluation efficiency of the proposed scoring system with the current COVID-19 clinical classification guidelines, we separately calculated their C indexes (0.95 and 0.89, respectively), which suggested that the proposed system is better than the current clinical classification.

Discussion

This study analyzed the clinical data of 3,974 patients from the Fire God Mountain Hospital and the Maternal and Child Hospital of Hubei Province. The nomogram identified key prognostic factors to establish a COVID-19 patient death risk assessment scoring system. According to their clinical and sociological characteristics, patients with COVID-19 were divided into high- or low-risk groups at the early stage of admission, which is of great significance for the selection of follow-up treatment options and the prevention and treatment of related complications. This study found that being a senior and male with severe or critical clinical type adversely affected the prognosis, consistent with previous research.[9] In terms of complications, pulmonary insufficiency, hypoproteinemia, heart dysfunction, and digestive tract bleeding adversely affect the prognosis. A high correlation between pulmonary insufficiency and adverse clinical events in patients with COVID-19 has been reported.[10] Although digestive tract bleeding is rarely seen clinically, previous studies have indicated that it is correlated with an increased likelihood of disease aggravation, which may be closely related to gastrointestinal injury caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) but requires confirmation by further study.[11] This study found an adverse effect of hypoproteinemia on the prognosis of patients, which suggests that liver function reserve plays an important role in the immune function of patients with COVID-19. Another finding is that most cardiac dysfunction occurs in patients with a history of coronary heart disease or severe COVID-19, and these patients often have a poor prognosis, which may be closely related to the cardiac injury caused by SARS-CoV-2 reported in previous studies.[12] Furthermore, our study found that diseases of the hematopoietic system were also important factors affecting the prognosis of patients with COVID-19. Although there have been no reports of SARS-CoV-2 harming hematopoiesis, clinical observation and our previous studies have found that coronaviruses, including the severe acute respiratory syndrome coronavirus and SARS-CoV-2, can damage the hematopoietic system after infection, which may partially account for patients with COVID-19 presenting with anemia, thrombocytopenia, and lymphocytopenia.[13–16]

This study has a few limitations. The factors considered in this study were obtained from electronic records without any direct quantitative index of coronavirus and immune function. This drawback could be partially compensated for by clinical typing, in which the clinical type of COVID-19 is closely related to the comprehensive effects of coronavirus quantity and immune function.

In summary, this study used machine learning to model and analyze a large amount of clinical data and identify important factors affecting the prognosis of patients with COVID-19. Compared with general statistical methods, this method can quickly and accurately screen the relevant factors affecting prognosis, assign them an order of importance, and establish a scoring system based on nomograms, which is of great clinical significance. The scoring system developed in this study allows patients with COVID-19 to be classified as high or low risk according to their sociological and clinical characteristics at the time of admission and thus helps clinicians to choose the optimal treatment plan. Meanwhile, early identification and intervention of critically ill patients with COVID-19 can reduce their mortality.

Funding

This work was supported by National Key Research and Development Program of China (2020YFC2002706).

Author Contributions

Xue-Chun Lu and Fu-Sheng Wang investigated the present status of COVID-19 research and conceived the idea and provided data sources. Hao-Min Zhang, Lei Shi, Hao-Ran Chen, and Jun-Dong Zhang conducted the data collation and analysis and the evaluation of the results. Ge-Liang Liu, Zi-Ning Wang, Peng Zhi, Run-Sheng Wang, Zhuo-Yang Li, and Xi-Meng Chen were in charge of searching and organizing the information related to this study. All authors discussed the results, commented on the manuscript and approved the final manuscript.

Conflict of Interest

None.

Editor note: Fu-Sheng Wang is the editor of Infectious Diseases and Immunity. The article was subject to the journal’s standard procedures, with peer review handled independently by this editor and his research group.

Data Availability Statement

The data sets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

References

1. Huang C, Wang Y, Li X, et al. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. Lancet 2020;395(10223):497–506. doi: 10.1016/S0140-6736(20)30183-5.
2. World Health Organization. WHO Coronavirus (COVID-19) Dashboard. Available from: https://covid19.who.int/. Accessed August 26, 2022.
3. Ginestra JC, Mitchell OJL, Anesi GL, et al. COVID-19 critical illness: a data-driven review. Annu Rev Med 2022;73:95–111. doi: 10.1146/annurev-med-042420-110629.
4. Wang H, Paulson KR, Pease SA, et al. Estimating excess mortality due to the COVID-19 pandemic: a systematic analysis of COVID-19-related mortality, 2020–21. Lancet 2022;399(10334):1513–1536. doi: 10.1016/S0140-6736(21)02796-3.
5. The National Health Commission of the PRC. Guidance for Corona Virus Disease 2019. 7th ed. 2020. Available from: http://www.gov.cn/zhengce/zhengceku/2020-03/04/content_5486705.htm. Accessed March 4, 2022.
6. Ishwaran H, Kogalur UB, Blackstone EH, et al. Random survival forests. Ann Appl Statist 2008;2(3):841–860. doi: 10.1214/08-AOAS169.
7. Harrell FE Jr. rms: Regression Modeling Strategies. Available from: https://CRAN.R-project.org/package=rms. Accessed March 4, 2022.
8. Jalali A, Alvarez-Iglesias A, Roshan D, et al. Visualising statistical models using dynamic nomograms. PLoS One 2019;14(11):e0225253. doi: 10.1371/journal.pone.0225253.
9. Guan WJ, Ni ZY, Hu Y, et al. Clinical characteristics of coronavirus disease 2019 in China. N Engl J Med 2020;382(18):1708–1720. doi: 10.1056/NEJMoa2002032.
10. Guan WJ, Liang WH, Zhao Y, et al. Comorbidity and its impact on 1590 patients with COVID-19 in China: a nationwide analysis. Eur Respir J 2020;55(5):2000547. doi: 10.1183/13993003.00547-2020.
11. Jin X, Lian JS, Hu JH, et al. Epidemiological, clinical and virological characteristics of 74 cases of coronavirus-infected disease 2019 (COVID-19) with gastrointestinal symptoms. Gut 2020;69(6):1002–1009. doi: 10.1136/gutjnl-2020-320926.
12. Li B, Yang J, Zhao F, et al. Prevalence and impact of cardiovascular metabolic diseases on COVID-19 in China. Clin Res Cardiol 2020;109(5):531–538. doi: 10.1007/s00392-020-01626-9.
13. O'Donnell R, Tasker RC, Roe MF. SARS: understanding the coronavirus: apoptosis may explain lymphopenia of SARS. BMJ 2003;327(7415):620. doi: 10.1136/bmj.327.7415.620-b.
14. Yang M, Hon KL, Li K, et al. The effect of SARS coronavirus on blood system: its clinical findings and the pathophysiologic hypothesis. Zhongguo Shi Yan Xue Ye Xue Za Zhi 2003;11(3):217–221.
15. Zhang H, Chen H, Yang Y, et al. Bioinformatics prediction of molecular mechanism and intervention drugs of SARS-related immune injury and their significance for COVID-19 treatment. Zhonghua Wei Sheng Wu Xue He Mian Yi Xue Za Zhi 2020;40(3):165–173. doi: 10.3760/cma.j.cn112309-20200215-00064.
16. Chen XM, Cao F, Zhang HM, et al. Exploration of omics mechanism and drug prediction of coronavirus-induced heart failure based on clinical bioinformatics. Zhonghua Xin Xue Guan Bing Za Zhi 2020;48(7):587–592. doi: 10.3760/cma.j.cn112148-20200308-00172.
Keywords:

COVID-19; Machine learning; Prognosis model

Supplemental Digital Content

Copyright © 2022 The Chinese Medical Association, published by Wolters Kluwer Health, Inc.