2.1 Patients
Ethical approvals by our institutional review boards were obtained for this retrospective study, and the need to obtain informed consent was waived. 1127 suspected patients were consecutively enrolled from January 15th to March 10th, 2020 in several hospitals in Jiangsu and Anhui provinces of China. 262 patients were hospitalized and had confirmed COVID-19 via laboratory testing with real-time reverse transcriptase polymerase chain reaction (RT-PCR) of respiratory secretions. A total of 217 COVID-19 patients (127 males and 90 females, mean age 46 years, age range 6-86 years) with chest CT abnormality were included in our study (Figure 1). According to the guidelines for the diagnosis and treatment of COVID-19 (trial version 7) developed by the National Health Committee of the People's Republic of China [[i]], confirmed patients are divided into mild, common, severe and critical types (Supplementary material). According to the clinical severity, the patients are divided into mild illness group (mild and common types) and severe illness group (severe and critical types) during the follow-up (Supplementary material). Of the 217 patients with COVID-19, 146 cases were in the mild group and 71 cases were in the severe group. According to a ratio of 7:3, patients were randomly assigned to the training cohort (102 mild cases and 50 severe cases) and validation cohort (44 mild cases and 21 severe cases).
2.2 Follow-up
All patients were followed for more than 30 days after admission. The patients underwent laboratory and CT examination, and symptoms, treatments and outcome events were recorded after admission. The initial clinical data analyzed were as follows: age, sex, symptoms, underlying diseases, laboratory results and days from illness onset to admission. The endpoint of this study was the development of severe illness.
2.3 CT examinations
All patients underwent chest CT examinations in the supine position by using the GE BrightSpeed Elite 16 scanner or GE LightSpeed VCT scanner (GE Healthcare, Milwaukee, USA) or Siemens SOMATOM Definition AS+ scanner (Siemens Healthineers, Milwaukee, Germany). The scanning range was from the apex to the bottom of the lung. The scanning parameters were as follows: tube voltage 120kV, tube current automatic mA, helical pitch 0.938, rotation speed 0.6 s, slice thickness and spacing 5mm. CT images were reconstructed with a slice thickness of 0.625-1.25 mm using a lung kernel as part of the reconstruction process.
2.4 CT image analysis
All images were reviewed by two chest radiologists with 5-15 years of experience by consensus. Further review was undertaken by a third radiologist with 20 years of experience if there was disagreement. The initial CT images were evaluated for each of the 152 patients. The radiologists recorded the following lesion features: location, distribution, morphology, density, vascular bundle thickening, air bronchogram sign, crazy-paving sign, fibrosis, mosaic perfusion sign, pleural effusion, thoracic lymphadenopathy (defined as lymph node size of ≥10 mm in short-axis dimension), number of segments involved and “severity score of lung”. Location was recorded as unilateral lung or bilateral lung. Distribution was defined as peripheral, peripheral with central or central. Morphology was described as nodular or patchy, nodular with patchy or patchy with segmental. Density was recorded as ground glass opacity (GGO), GGO with consolidation or consolidation. Each of the five lung lobes was assessed for degree of involvement and scored as 0 (0% involvement), 1 (1%-25% involvement), 2 (26%-50% involvement), 3 (51%-75% involvement), or 4 (76%-100% involvement) [6]. An overall “severity score of lung” was calculated by summing the five lobe scores (range of scores, 0-20) [6].
2.5 Statistical Analysis
Categorical variables were reported as frequency and proportions, and continuous variables were reported as the mean±standard deviation or median with interquartile range (IQR). The independent-sample t test or Mann-Whitney U test was performed to compare the quantitative parameters and Chi-square test to compare the qualitative features between the mild group and severe group. After potential risk factors were selected, multivariate logistic regression analysis was used to determine the independent risk factors associated with severe COVID-19. Selected variables were incorporated in the nomogram to predict severe COVID-19 using “rms” package in R software (version 3.5.1, http://www.rproject.org).
The performance of nomogram was assessed by discrimination and calibration. The discriminating ability of nomogram was evaluated using area under the receiver operating characteristic (ROC) curve (AUC). Calibration was evaluated using a calibration plot, a graphic representation of the relationship between the observed and predicted probability, with a bootstrapped sample of the study group for 1000 times. Further calibration of the nomogram was evaluated using the Hosmer-Lemeshow goodness of fit test. Decision curve analysis (DCA) was conducted to determine the clinical usefulness of the nomogram by quantifying the net benefits at different threshold probabilities and clinical impact curve to determine the influence on the outcome of patients. A risk chart plotted performance against caseload. The “rmda” package was used in DCA and “rattle” package in risk chart. Statistical analyses were performed with SPSS 20.0 and R software. All analyses were considered significant at P values of less than 0.05 (two-tailed).