Clinical characteristics of Coronavirus Disease 2019 outside Wuhan and development of early risk stratification tool

DOI: https://doi.org/10.21203/rs.3.rs-19481/v1

Abstract

Importance: Clinical characteristics of coronavirus disease 2019 (COVID-19) outside Wuhan are not well described, and there is no risk stratification tool for the prediction of COVID-19 outcome.

Objective: To describe the clinical characteristics of COVID-19 outside Wuhan and to develop a risk stratification tool for early risk stratification.

Design, Setting, and Participants: Single center, retrospective observational study conducted at Wenzhou, Zhejiang province (January 2020-February 2020) that included 140 patients with confirmed COVID-19. Clinical characteristics were described. A multivariate regression model was built to predict the risk of length of stay in hospital > 20 days (ProLOS). The last patient visit was on February 23, 2020.

Interventions: None

Main Outcomes and Measures: The primary outcome was ProLOS. Other outcomes included conversion to negative nucleic acid test, date of hospital discharge, vital status at discharge.

Results: A total of 140 patients were included during the study period. Lower lymphocyte count (1.0 [0.7, 1.3] vs. 1.3 [0.9, 1.72] × 109/𝐿; p = 0.008), lower ionized sodium (136 [134.6, 137.83] vs. 138 [135.28, 140.03] mmol/L; p < 0.001) and higher PaCO2 (40.82 ± 3.96 vs. 38.48 ± 5.48 mmHg; p = 0.007) were associated with higher risk of ProLOS. The median time from hospital admission to the first negative nucleic acid test was 13 days (7.25 to 17 days). There were 4 (3%) critically and 5 (4%) severely ill patients. A multivariate model included predictors of age, time from contact to confirmation, DBP, lymphocyte count, AST, ionized sodium, PaCO2 showed good calibration and discrimination with an AUC of 81.6% (95% CI: 74.4% to 88.8%).

Conclusions and Relevance: The study described clinical characteristics of COVID-19 outside Wuhan. The major difference of COVID-19 in other cities included low comorbidity burden, low prevalence of severe or critical cases and low mortality rate. The risk stratification tool can be used for medical decision making and resource allocation.

Authors Yucai Hong and Jianping Huang contributed equally to this work and are co-first authors.

Introduction

Novel coronavirus disease 2019 (COVID–19) is caused by the severe acute respiratory coronavirus 2 (SARS-CoV–2), which has become a global health emergency due to its human-to-human transmission and the large number of confirmed cases in China and all over the world 1. China is now taking strenuous efforts to combat this epidemic, which showed significant effect that the overall newly confirmed cases declined steadily. The most affected region in China is Wuhan, Hubei province, with the number of confirmed cases reaching above 47,000 on February 26, 20202,3. There have been many clinical studies describing the clinical characteristics of COVID–19 in Wuhan 4–8. The cases ranged from mild to severe, and even death. The mortality rate is estimated to be around 2%. The clinical onset can be atypical, with only half of the patients reported to have fever 9,10. However, the report outside Wuhan is still rare 11. There has been evidence showing that the clinical course of COVID–19 from outside Wuhan can be different from that from Wuhan 12. For example, the crowdsourcing data from epidemic monitoring center showed that the mortality rate in other regions is much lower than that in Wuhan 13. The first aim of the study was to describe clinical characteristics of COVID–19 outside Wuhan city.

Another important issue in managing COVID–19 is early risk stratification. To the best of our knowledge, there is no prediction tool for the prediction of clinical outcome for COVID–19. The development of a prediction tool can assist medical decision making and resource allocation. Since the majority of our patients were mild, we employed length of stay in the hospital as the primary end point. A multivariate model was developed to predict patients who were at risk of prolonged length of stay (> 20 days) in the hospital (ProLOS).

Methods

Study design and setting

The study was conducted in a tertiary care hospital in Wenzhou, Zhejiang province from January 15 to February 24, 2020. The hospital was a designated hospital for the treatment of COVID–19. Medical records of all patients with confirmed COVID–2019 were retrospectively reviewed. All patients were followed up until February 24, 2020. Patients were divided into two groups according to whether they were stayed in the hospital for longer than 20 days or not. Those with hospital LOS > 20 days were classified as the ProLOS group, and those discharged hospital before 20 days were classified into the non-ProLOS group.

Baseline demographic and clinical characteristics were compared between the two groups. A multivariate regression model was built to predict the risk of ProLOS. The study was approved by the ethics committee of Wenzhou Central Hospital (NL2020–01–011) and was conducted in accordance to the Helsinki Declaration. Informed consent was waived by the institutional review board due to retrospective nature of the study. The study was reported according to the STROBE checklist 14.

Participants

According to the General Office of National Health Committee, suspected cases were defined with clinical features and epidemiological risk. The clinical features included fever, imaging features of pneumonia, normal or reduced white blood cell count, or reduced lymphocyte count in the early stages of the disease onset. The epidemiologic risks included 1) a history of travel to or residence in Wuhan city, China or other cities with continuous transmission of local cases in the last 14 days before symptom onset; 2) contact with patients with fever or respiratory symptoms from Wuhan city, China or other cities with continuous transmission of local cases in the last 14 days before symptom onset; or 3) epidemiologically connected to SARS-CoV–2 infections or clustered onsets. Suspected cases who also fulfilled the following pathogenic evidence was considered as confirmed cases: (1) positive for the 2019-nCoV by the real-time PCR test for nucleic acid in respiratory or blood samples; 2) viral gene sequencing shows high homogeneity to the known SARS-CoV–2 in respiratory or blood samples 15. Patients were excluded if 1) they had signed do-not-resuscitate order; 2) had severe burn or trauma and 3) end-stage malignancy.

Variables

Variables related to demographics, signs and symptoms, comorbidities, laboratory findings, complications and clinical outcomes were extracted from electronic medical records in a standardized case report form (CRF). Medical history included the date of the first contact with infection source, the suspected infection source, the time from contact to diagnosis.

Comorbidities included hypertension, diabetes mellitus, chronic kidney disease (CKD), chronic liver disease, cardiovascular disease, chronic pulmonary disease and malignancy. Clinical signs and symptoms included cough, fever, rhinorrhea, fatigue, anorexia, pharyngalgia, myalgia, headache, abdominal pain, vomiting and diarrhea. Laboratory tests included blood routines, biochemical profiles, electrolytes, coagulation profile and procalcitonin. Clinical complications included liver failure, respiratory failure, progression to severe or critical cases. Treatment included mechanical ventilation, oxygen therapy, and non- invasive mechanical ventilation.

The clinical outcomes included conversion to negative nucleic acid test, date of hospital discharge, vital status at discharge. LOS in the hospital was defined as the difference between hospital discharge and admission. For those stayed in the hospital on February 24, 2020, they were censored at that day. The censoring was considered to be non-informative (i.e. random censoring) that it would not bias the stochastic distribution of the hospital LOS.

Statistical analysis

Descriptive analyses were performed for all variables for our cohort. The comparisons between ProLOS and non-ProLOS groups were performed with standard methods. Normal quantitative data were expressed as mean and standard deviation (SD) and were compared using t test. Skewed variables were expressed as median and interquartile range (IQR) and were compared using rank-sum test. Qualitive data were expressed as the number and percentage and were compared between groups using Chi-square test or Fisher’s exact test as appropriate. The CBCgrps package was used for the analyses 16.

Variables with p < 0.1 in univariate analysis were included to build a full model. Then the stepwise model selection was performed by Akaike Information Criterion (AIC), aiming to obtain a model with lowest AIC 17. The model performance was assessed using discrimination and calibration. An area under curve (AUC) of greater than 80% was considered as good for clinical utility. The use of the model was demonstrated using nomogram 18. All statistical analyses were performed using R (version 3.6.1). A conventional p value less than 0.05 were considered as statistical significance.

Results

Participants

A total of 140 patients were included during the study period. No subjects were excluded due to prespecified exclusion criteria. Sixty-four (45.7%) of the 140 patients had stayed in the hospital for more than 20 days (the ProLOS group).

Baseline characteristics

There was no difference between ProLOS and non-ProLOS groups in age, gender, epidemiological history and most comorbidities (Table 1). The ProLOS group showed significantly less days from contact to confirmation (median [IQR]: 6.5 [4, 9.25] vs. 7.5 [5, 14] days; p = 0.026). There was greater proportion of patients with malignancy in the ProLOS group (3% vs. 0%; p = 0.041). Most patients showed normal vital signs on admission. Interestingly, patients in the ProLOS group had higher SBP (131 [120, 147] vs. 126 [118, 134] mmHg; p = 0.037) and DBP (86.41 ± 11.21 vs. 80.17 ± 10.76 mmHg; p = 0.001). A large proportion of patients had abdominal symptoms on admission (39%).

Table 1 Baseline characteristics of patients and comparisons between ProLOS and non- ProLOS groups

Variables

Total (n =

140)

Non-ProLOS

(n = 76)

ProLOS (n =

64)

p

Age, Mean ± SD

45.66 ± 14.1

47.44 ± 15.42

43.54 ±

12.13

0.099

Age group, n (%)

 

 

 

0.078

> 60 years

17 (12)

13 (17)

4 (6)

 

< 60 years

121 (86)

62 (82)

59 (92)

 

Not reported

2 (1)

1 (1)

1 (2)

 

Male, n (%)

71 (51)

37 (49)

34 (53)

0.723

Medical worker, n (%)

1 (1)

0 (0)

1 (2)

0.457

Epidemiological history, n (%)

 

 

 

0.236

Traveling to Wuhan

46 (33)

23 (30)

23 (36)

 

Contact with patients from Wuhan

8 (6)

4 (5)

4 (6)

 

Contact with other patients

57 (41)

37 (49)

20 (31)

 

No known epidemiologic risk

29 (20)

12 (16)

17 (27)

 

Time From Contact To

Confirmation, Median (IQR)

7 (4, 12)

7.5 (5, 14)

6.5 (4, 9.25)

0.026

Smoking history, n (%)

8 (6)

7 (9)

1 (2)

0.07

Drinking, n (%)

4 (3)

1 (1)

3 (5)

0.332

Comorbidities

 

 

 

 

Hypertension, n (%)

36 (26)

17 (22)

19 (30)

0.285

Diabetes mellitus, n (%)

12 (9)

6 (8)

6 (9)

0.648

Chronic kidney disease, n (%)

2 (1)

2 (3)

0 (0)

0.166

Chronic liver disease, n (%)

9 (6)

4 (5)

5 (8)

0.134

Chronic cardiovascular disease, n

(%)

4 (3)

4 (5)

0 (0)

0.019

Chronic pulmonary disease, n (%)

4 (3)

2 (3)

2 (3)

0.368

Malignancy, n (%)

2 (1)

0 (0)

2 (3)

0.041

Other comorbidities, n (%)

8 (6)

5 (7)

3 (5)

0.36

On admission vital signs

 

 

 

 

Temperature (), Median (IQR)

36.9 (36.5,

37.4)

36.9 (36.5,

37.23)

37 (36.65,

37.5)

0.328

Heart rate (bpm), Mean ± SD

90.61 ±

12.35

88.11 ± 11.07

93.59 ± 13.2

0.009

Respiratory rate (rpm), Median

(IQR)

20 (20, 20)

20 (20, 20)

20 (20, 20)

0.478

SBP, Median (IQR)

128 (119.5,

139)

126 (117.75,

134.25)

131 (120,

147)

0.037

DBP, Mean ± SD

83 ± 11.36

80.17 ± 10.76

86.41 ±

11.21

0.001

Urine output, Mean ± SD

1511.82 ±

707.63

1268.57 ±

550.98

1937.5 ±

826.01

0.213

Symptoms on presentation

 

 

 

 

Cough, n (%)

114 (81)

60 (79)

54 (84)

0.545

Sputum, n (%)

101 (72)

51 (67)

50 (78)

0.208

Rhinorrhea, n (%)

11 (8)

7 (9)

4 (6)

0.739

Pharyngalgia, n (%)

17 (12)

8 (11)

9 (14)

0.705

Fatigue, n (%)

40 (29)

20 (26)

20 (31)

0.648

Myalgia, n (%)

8 (6)

3 (4)

5 (8)

0.469

Headache, n (%)

19 (14)

8 (11)

11 (17)

0.369

Nausea and vomiting, n (%)

20 (14)

15 (20)

5 (8)

0.077

Abdominal pain, n (%)

10 (7)

7 (9)

3 (5)

0.345

Diarrhea, n (%)

55 (39)

26 (34)

29 (45)

0.244

Dyspnea, n (%)

28 (20)

17 (22)

11 (17)

0.581

Abbreviations: ProLOS: prolonged length of stay in hospital; SD: standard deviation; IQR: interquartile range.

Laboratory findings

Laboratory tests were recorded on the day of hospital admission. Lower lymphocyte count (1.0 [0.7, 1.3] vs. 1.3 [0.9, 1.72] × 109/𝐿; p = 0.008), lower ionized sodium (136 [134.6, 137.83] vs. 138 [135.28, 140.03] mmol/L; p < 0.001) and higher PaCO2 (40.82 ± 3.96 vs. 38.48 ± 5.48 mmHg; p = 0.007) were associated with higher risk of ProLOS. PCT was not associated with ProLOS because most of them (99%) were within normal range. Other laboratory measurements were not significantly different between the two groups (Table 2).

Table 2. Laboratory findings and comparisons between ProLOS and non-ProLOS groups

Variables

Total (n = 140)

Non-ProLOS (n

= 76)

ProLOS (n =

64)

p

PCT (ng/ml), Median (IQR)

0.04 (0.02,

0.07)

0.04 (0.02,

0.07)

0.04 (0.02,

0.06)

0.914

CRP (mg/dl), Median (IQR)

10 (2.9, 25.68)

8.2 (1.67,

31.07)

11.55 (3.6,

24.67)

0.283

WBC (× 109/𝐿), Median (IQR)

4.6 (3.77, 5.8)

4.9 (3.9, 6.32)

4.35 (3.4, 5.62)

0.05

Neutrophil count (× 109/𝐿), Median (IQR)

3 (2.3, 3.9)

3.05 (2.5, 4.1)

2.9 (2.2, 3.7)

0.17

Lymphocyte count (× 109/

𝐿), Median (IQR)

1.1 (0.8, 1.6)

1.3 (0.9, 1.72)

1 (0.7, 1.3)

0.008

Hematocrit (%), Mean ± SD

40.03 ± 4.09

39.74 ± 4.43

40.38 ± 3.64

0.351

Hemoglobin (g/L), Mean ±

SD

135.46 ± 15.22

134.72 ± 16.76

136.33 ± 13.26

0.528

Platelet count (× 109/𝐿), Median (IQR)

182 (150.75,

228.5)

182.5 (151.5,

243)

181 (150.75,

214)

0.441

PT, Mean ± SD

12.71 ± 1.15

12.45 ± 1.06

13.03 ± 1.2

0.003

APTT, Mean ± SD

30.83 ± 3.27

30.3 ± 2.91

31.47 ± 3.59

0.041

D-dimmer (ng/ml), Median

(IQR)

134 (89, 188)

138 (87.5, 219)

124 (96.5,

162.25)

0.441

Fibrinogen (g/L), Median

(IQR)

3.72 (3.1, 4.13)

3.68 (3.12,

4.19)

3.72 (3.11,

4.11)

0.537

Albumin (g/L), Mean ± SD

41.94 ± 4.05

41.48 ± 4.44

42.49 ± 3.49

0.133

Globulin (g/L), Median

(IQR)

28.65 (26.4,

31.13)

28.05 (26.35,

30.6)

28.75 (27.4,

32.25)

0.102

ALT (U/L), Median (IQR)

22 (13, 32.5)

18.5 (12.75, 32)

22.5 (15, 36)

0.227

AST, Median (IQR)

24 (19, 34)

23 (18, 31.5)

27 (20.75, 34.5)

0.085

Total bilirubin (μmol/L), Median (IQR)

11.95 (8.67,

15.9)

12.3 (9.28,

16.47)

10.8 (8.38,

15.6)

0.344

Direct bilirubin (μmol/L), Median (IQR)

4.5 (3.2, 6.1)

4.65 (3.38,

6.32)

4.05 (3.08, 5.9)

0.22

BUN (mmol/L), Median

(IQR)

3.65 (3.1, 4.23)

3.75 (3, 4.6)

3.6 (3.1, 4.03)

0.358

Abbreviations: SD: standard deviation; IQR: interquartile range; PCT: procalcitonin; ALT: Alanine Aminotransferase; AST: Aspartate Aminotransferase; TB: total bilirubin; CKMB: Creatine Kinase Myocardial Band; CK: Creatine Kinase; PT: partial thrombin time; APTT: activated partial thrombin time; WBC: white blood count; LDH: Lactate Dehydrogenase.

Complications and outcomes

Seven patients (5%) developed respiratory failure and one patient (1%) had acute respiratory failure during hospital stay. Seventy-five (54%) patients required oxygen therapy via nasal tube, 7 patients (5%) required facial mask, and 4 patients (3%) required non-invasive mechanical ventilation. The median time from hospital admission to the first negative nucleic acid test was 13 days (7.25 to 17 days), and the second negative nucleic acid test was 15.5 days (10 to 19 days). Of note, two patients showed second positive nucleic acid test after negative nucleic acid tests. There were 4 (3%) critically and 5 (4%) severely ill patients (Table 3).

Table 3. Complications and outcomes of the overall population and comparisons between ProLOS and non-ProLOS groups

Variables

Total (n =

140)

Non-ProLOS

(n = 76)

ProLOS (n =

64)

p

Respiratory failure, n (%)

7 (5)

5 (7)

2 (3)

0.454

Liver injury, n (%)

2 (1)

0 (0)

2 (3)

0.207

Oxygen therapy via nasal canula,

n (%)

75 (54)

35 (46)

40 (62)

0.076

Facial mask, n (%)

7 (5)

5 (7)

2 (3)

0.454

Non-invasive mechanical

ventilation, n (%)

4 (3)

3 (4)

1 (2)

0.625

Fist negative nucleic acid test

day, Median (IQR)

13 (7.25, 17)

8.5 (4, 12)

18.5 (15, 20)

0.001

Second negative nucleic acid test

day, Median (IQR)

15.5 (10, 19)

12 (6.25, 15)

21 (18,

21.75)

0.001

Severity, n (%)

 

 

 

0.667

Critical

4 (3)

3 (4)

1 (2)

 

Mild

131 (94)

71 (93)

60 (94)

 

Severe

5 (4)

2 (3)

3 (5)

 

Mortality, n (%)

 

 

 

0.032

Alive

63 (45)

41 (54)

22 (34)

 

Follow

77 (55)

35 (46)

42 (66)

 

Hospital LOS, Median (IQR)

19.52 (16.44,

24.55)

16.51 (11.52,

18.29)

25.41 (22.6,

27.76)

0.001

HosLOS20, n (%)

 

 

 

0.001

Abbreviations: SD: standard deviation; IQR: interquartile range; LOS: length of stay.

Model selection with stepwise procedure resulted in a model with 7 predictors including age, time from contact to confirmation, DBP, lymphocyte count, AST, ionized sodium, PaCO2. These predictors were obtained on hospital admission (Table 4). The model was well calibrated and had a good prediction discrimination with an AUC of 81.6% (95% CI: 74.4% to 88.8%, Figure 1). The model could be used with a nomogram (Figure 2).

Table 4 Prediction model established by using stepwise model selection

Predictors

OR [95% CI]

p

Age (with each year increase)

0.95 [0.91, 0.99]

0.008

Time From Contact To Confirmation (with each day increase)

0.92 [0.85, 0.99]

0.040

DBP (with each 1 mmHg increase)

1.06 [1.02, 1.10]

0.004

Lymph Count (with each 1 × 109/𝐿 increase)

0.30 [0.13, 0.66]

0.004

AST (with each 1 IU/L increase)

1.02 [1.00, 1.05]

0.056

Sodium (with each 1 mmol/L increase)

0.88 [0.76, 0.97]

0.029

PaCO2 (with each 1 mmHg increase)

1.15 [1.06, 1.27]

0.002

Abbreviations: DBP: diastolic blood pressure; AST: Aspartate Aminotransferase.

Discussion

The study described clinical characteristics of COVID–19 outside Wuhan. The results showed that the prevalence of severe and critical cases was low, and there was no patient died during the follow up. This is an important feature of COVID–19 outside of Wuhan city. Initial reports from Wuhan showed that 17–29% patients developed acute respiratory distress syndrome and 11–12% died of multiple organ failure 4,5. However, the high mortality rate was only observed in Wuhan. A nationwide multicenter study showed that 5% patients required ICU admission and the overall mortality rate was 1.36% 19. Similarly, Studies from Zhejiang province reported that the prevalence of severe cases was very low and no death was reported11,20. The large difference in disease severity between Wuhan and other cities may reflect the shortage of medical resources in Wuhan when large number of cases emerged.

The median time from contact with infection source to case confirmation was 7 days. The median incubation period was reported to be 3 days (Range: 0 to 24 days) 19. It generally takes a couple of days for a patient be hospitalized after symptom onset. Thus, the median time from contact with infection source and case confirmation was 7 days (IQR: 4 to 12 days) in our cohort. To the best of our knowledge, there has been no study reporting the time for negative conversion of the nucleic acid test. Our study showed that it might take a median of 13 days for nucleic acid test to become negative. However, it should be noted that there were two cases whose second nucleic acid test return positive following the first negative test, which highlighted the importance of multiple nucleic acid tests before hospital discharge.

Our study for the first time developed a risk stratification tool for the prediction of ProLOS for a cohort with mild illness. The discrimination and calibration of the model was good. The predictors in the model were routinely available in clinical practice. Sophisticated stepwise model selection procedure was performed with the aim to minimize the AIC value 21. The model aimed to improve the predictive accuracy while keeping the number of predictors at a minimum level (i.e. the principle of parsimony).

Several limitations must be acknowledged in the study. First, the study was performed in a single center, and the external validity of the findings was not known 22. As has been discussed above, there was large difference between our cohort and studies from Wuhan. Since studies from outside Wuhan are still rare, more studies can be performed to elucidate the reasons underlying the disparity. Second, although our prediction model showed good discrimination and calibration in internal validation, it lacks external validation. In other words, the model carries the risk of overfitting 23. However, we have tried to minimize the problem of overfitting. The criteria for model selection also considered the degree of freedom, which helped to control the overfitting problem. Third, the majority of patients in our study was mild, and the risk stratification tool may not be applicable to severe or critical cases. For example, older age was identified as a protective factor for ProLOS, which is in contrast to the current knowledge that older age is a risk factor for adverse clinical outcome 24. Probability, patients in our study were not very old and comorbidity burden was very low. The adverse effect of older age is mediated via increased comorbidity burden. With low comorbidity burden, the mediation pathway was blocked and the casual linkage between age and adverse outcome was also changed.

In conclusion, our study described clinical characteristics of COVID–19 outside of Wuhan. The major difference of COVID–19 in other cities included low comorbidity burden, low prevalence of severe or critical cases and low mortality rate. Furthermore, we developed a multivariate regression model to predict the risk of ProLOS, which showed good discrimination and calibration.

Reference

  1. Kickbusch, I. & Leung, G. Response to the emerging novel coronavirus outbreak. BMJ368, m406 (2020).
  2. Wu, S. S. et al. Epidemiological Development of Novel Coronavirus Pneumonia in China and Its Forecast. medRxiv 02.21.20026229 (2020).
  3. Wu, Z. & McGoogan, J. M. Characteristics of and Important Lessons From the Coronavirus Disease 2019 (COVID-19) Outbreak in China: Summary of a Report of 72 314 Cases From the Chinese Center for Disease Control and Prevention. JAMA (2020). doi:10.1001/jama.2020.2648
  4. Chen, N. et al. Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study. Lancet (2020). doi:10.1016/S0140-6736(20)30211-7
  5. Huang, C. et al. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. Lancet (2020). doi:10.1016/S0140-6736(20)30183-5
  6. Li, Z. et al. Caution on Kidney Dysfunctions of 2019-nCoV Patients. medRxiv2020.02.08.20021212 (2020).
  7. Chen, et al. [Analysis of clinical features of 29 patients with 2019 novel coronavirus pneumonia]. Zhonghua Jie He He Hu Xi Za Zhi 43, E005–E005 (2020).
  8. Wang, D. et al. Clinical Characteristics of 138 Hospitalized Patients With 2019 Novel Coronavirus-Infected Pneumonia in Wuhan, China. JAMA (2020). doi:10.1001/jama.2020.1585
  9. Zhou, L. & Liu, H. G. [Early detection and disease assessment of patients with novel coronavirus pneumonia]. Zhonghua Jie He He Hu Xi Za Zhi 43, E003–E003 (2020).
  10. An, P. et al. Clinical Features of 2019 Novel Coronavirus Pneumonia Presented Gastrointestinal Symptoms But Without Fever Onset. (2020).
  11. Xu, X.-W. et al. Clinical findings in a group of patients infected with the 2019 novel coronavirus (SARS-Cov-2) outside of Wuhan, China: retrospective case series. BMJ 368, m606 (2020).
  12. Novel Coronavirus Pneumonia Emergency Response Epidemiology Team. [The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19) in China]. Zhonghua Liu Xing Bing Xue Za Zhi 41, 145–151 (2020).
  13. Sun, K., Chen, J. & Viboud, C. Early epidemiological analysis of the 2019-nCoV outbreak based on a crowdsourced data. medRxiv 01.31.20019935 (2020).
  14. Elm, von, E. et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational PLoS medicine 4, e296 (2007).
  15. Jin, Y.-H. et al. A rapid advice guideline for the diagnosis and treatment of 2019 novel coronavirus (2019-nCoV) infected pneumonia (standard version). Mil Med Res 7, 4–23 (2020).
  16. Zhang, Z., Gayle, A. A., Wang, J., Zhang, H. & Cardinal-Fernández, P. Comparing baseline characteristics between groups: an introduction to the CBCgrps package. Ann Transl Med 5, 484–484 (2017).
  17. Harrell, F. E. Regression Modeling Strategies. (Springer New York, 2001). doi:10.1007/978-1-4757-3462-1
  18. Zhang, Z. & Kattan, M. W. Drawing Nomograms with R: applications to categorical outcome and survival data. Ann Transl Med 5, 211–211 (2017).
  19. Guan, W.-J. et al. Clinical characteristics of 2019 novel coronavirus infection in China. medRxiv 2020.02.06.20020974 (2020).
  20. Qian, G.-Q. et al. Epidemiologic and Clinical Characteristics of 91 Hospitalized Patients with COVID-19 in Zhejiang, China: A retrospective, multi-centre case medRxiv 2020.02.23.20026856 (2020).
  21. Zhang, Z. Variable selection with stepwise and best subset approaches. Ann Transl Med 4, 136–136 (2016).
  22. Riley, R. D. et al. External validation of clinical prediction models using big datasets from e-health records or IPD meta-analysis: opportunities and challenges. BMJ 353, i3140 (2016).
  23. Zhang, Z. Too much covariates in a multivariable model may cause the problem of overfitting. J Thorac Dis 6, E196–7 (2014).
  24. Villar, J. et al. Age, PaO2/FIO2, and Plateau Pressure Score: A Proposal for a Simple Outcome Score in Patients With the Acute Respiratory Distress Syndrome. Care Med. 44, 1361–1369 (2016).