Study design and patient population
This retrospective study included 202 confirmed COVID-19 patients (≥18 years of age) admitted to the Fifth Affiliated Hospital of Sun Yat-sen University and Shiyan Taihe Hospital between January 17 and April 30, 2020. A confirmed case of COVID-19 was defined as; positive SARS-CoV-2 virus nucleic acids on the nasal and pharyngeal swab specimens by real-time reverse-transcriptase polymerase chain reaction (RT-PCR) assay.
COVID-19 was diagnosed and classified clinically according to the new coronavirus pneumonia diagnosis and treatment plan (trial version 7) [5] drafted by the National Health Committee of the People's Republic of China. Clinical classification of COVID-19 was; (1) mild, with mild symptoms and no obvious signs of pneumonia on imaging, (2) moderate, with fever, respiratory-tract symptoms and obvious signs on imaging indicating pneumonia, (3) severe, with one of the following; (a) respiratory rate ≥30 beats per min (bpm), (b) mean oxygen saturation in the resting state ≤ 93%, (c) ratio of arterial oxygen partial pressure (PaO2) to the fraction of inspiration (FiO2) ≤300 mmHg (1 mmHg = 0.133 kPa), or (d) pulmonary imaging showing an increase in manifestations of >50% within 24~48 h, (4) critical, with one of the following, (a) respiratory failure requiring mechanical ventilation, (b) shock, or (c) intensive-care unit (ICU) admission due to multiple-organ failure. The non-severe group comprised mild and moderate cases, whereas the severe group comprised the severe and critical cases. The inclusion criteria were: 1) Admission within 7 days from the onset of symptoms, 2) completed laboratory or medical examinations and questionnaires, 3) presence of the first lung CT examination. All the participants were followed until the end of the disease course; cure or death. This retrospective observational study was approved by the Research Ethics Committee of The Fifth Affiliated Hospital of Sun Yat-sen University (Approval Series No. K153-1). The need for informed consent was waived due to the retrospective study design.
Clinical data
Clinical data, including basic demographics, symptoms, vital signs, clinical classification, and complications, were extracted from electronic medical records. Laboratory evaluations included total blood cell count, coagulation function, liver and kidney function, electrolyte levels, lactate dehydrogenase (LDH), creatine kinase (CK), creatine kinase isoenzyme MB (CK-MB), alpha-hydroxybutyrate dehydrogenase (α-HBDH), C-reactive protein (CRP), blood gas analysis and D-dimer if the patient was breathing room air. Categorical variables were presented as frequency and percentages, whereas continuous variables as mean (±SD, standard deviation) and median (interquartile range (IQR) values.
CT scanning protocol
Each patient was placed in the supine position on the CT machine (uCT760 or Umi780, United Imaging, Shanghai, China; Precison32, Campo imaging, Shenyang, China) and scanned during the inspiratory phase. Images were reconstructed with a slice thickness of 1 mm and an interval of 1mm.
Lung CT images were screened by three imaging physicians who were blind to the RT-PCR results and clinical information. The CT images were independently read by two radiologists with more than 5 years’ experience in the diagnosis of chest CT scans. In case of dispute, they discussed and reached a consensus that was reviewed by a senior imaging physician with more than 10 years of experience.
Feature Selection and Model Establishment
Patients were randomly assigned to the training dataset (n = 163, with 43 in the severe group) or the validation dataset (n = 39, with 10 in the severe group) at a ratio of 8:2. Univariate analysis was applied to select candidate features with significant differences (p < 0.05) between non-severe and severe groups. Best subset selection via an exhaustive algorithm was then performed to establish the predictive model.
The features were selected using the leaps and rms package in R (version 3.6.2) which were used to fit the logistic regression model and nomogram, respectively. A decision curve analysis was performed by calculating the net benefits for a range of threshold probabilities in the whole cohort to assess the clinical efficiency of the nomogram. The prediction performance of the logistic regression model was evaluated based on sensitivity, specificity, and area under the receiver operator characteristic (ROC) curve (AUC).