Study Design, Participants, and data collection
This descriptive-analytical study was performed on 171 consecutive COVID-19 patients admitted to Bohlool hospital in Gonabad, Iran, between April 4 and June 5, 2021. Inclusion criteria included a definitive diagnosis of COVID-19 by RT-PCR test and a minimum age of 18 years. All patients were treated according to the physician's diagnosis, and interleukin 6 (IL-6), C - reactive protein (CRP), Lactate Dehydrogenase (LDH), and Ferritin tests were performed as a supplement for all patients. Other data were extracted using hospital information system (HIS) resources. Patients with incomplete records were excluded from the study.
The study was approved by the ethics committee of Gonabad University of Medical Sciences. The ethical principles of human medical research (Helsinki declaration) was observed in all phases of the study. Data of all patients were extracted confidentially and encrypted from the HIS system. The patient's diagnosis process was carried out under the guidelines published by the World Health Organization and the Iranian Ministry of Health.
Potential Predictive Variables
Potential predictor variables were: age, sex, pregnancy, use tobacco, use opium, history of COVID-19, inpatient department, Pao2, temperature, Computed Tomographic (CT) imaging score, signs, and symptoms at admission (fever, cough, muscular pain, respiratory distress, loss of consciousness, decreased sense of smell, decreased sense of taste, seizure, abdominal pain, nausea, vomiting, diarrhea, anorexia, headache, vertigo, paresthesia, paraplegia, chest pain, skin lesion, comorbidities (liver diseases, diabetes, hematologic diseases, HIV/ AIDS, autoimmune diseases, heart diseases, kidney diseases, asthma, chronic lung diseases, nervous diseases, hypertension, and other), laboratory values (WBC, NLR, CRP, IL-6, LDH, Ferritin), Clinical management & pharmacological treatment (tracheal intubation, O2 therapy, drugs).
Patients were divided into two groups based on the outcomes: 1) without disease progression: patients who fully recovered and discharged or showed stable symptomatic improvement, 2) with disease progression: patients who had progressed to severe illness and stayed in ICU or dead.
Data were analyzed by SPSS software version 16.
Descriptive statistics and bivariate analysis
The Kolmogorov-Smirnov test was used to examine if the normal distribution for quantitative variables, including age, Pao2, temperature, duration of hospitalization, and laboratory values, are met or not. The age distribution was normal and reported as the mean (standard deviation) and was compared between the two groups with and without disease progression using the independent t-test. Median (25th percentile, 75th percentile) and the Mann-Whitney test were used to describe and compare other quantitative variables between the two groups, respectively. Describing and statistical comparison for qualitative variables were carried out using number (percentage) and the Chi-square test, respectively. A two-sided P-value less than 0.05 was considered significant.
Development and validation of risk score prediction model
After randomly dividing data into two parts (discovery dataset (80% of data) and validation dataset (20% of data)), to develop a risk score prediction model for disease progression in patients with COVID-19, the prediction model was developed using a logistic regression model on the discovery dataset. The variables with a p-value less than 0.2 in the simple logistic regression model were entered into a multiple logistic regression model. We considered a backward removal method with p < 0.05 for entering variables and p < 0.1 for removing variables into/from the model. The coefficients obtained from the multiple logistic regression model were converted to an integer risk score. The highest value of sensitivity and specificity were used to determine the optimum cut-point for the risk score model. Assessment of model calibration was performed using the Hosmer-Lemeshow test with a P-value greater than 0.05, indicating acceptable goodness of fit to the data. The area under the receiver operator curve (AUC), with the minimum value of 0.70 as a desirable discrimination ability, was used to evaluate the discrimination ability of the risk score model in both discovery and validation datasets. The performance of the risk score prediction model was compared with each of the predictors of the model and also other statistically significant laboratory variables in bivariate analyses alone. The significance level was considered 0.05.