Risk Stratification Scores for Hospitalization Duration and Disease Progression in Moderate and Severe Patients With COVID-19

Jiaqi Huang; Yu Xu; Bin Wang; Ying Xiang; Na Wu; Wenjing Zhang; Tingting Xia; Zhiquan Yuan; Chengying Li; Xiaoyue Jia; Yifan Shan; Menglei Chen; Qi Li; Li Bai; Yafei Li


BMC Pulm Med. 2021;21(120) 

In This Article



This retrospective, single-center cohort study was conducted at Huoshenshan Hospital, one of the largest special hospital of COVID-19, in Wuhan, China from January to April 2020. The patient inclusion criteria were at least 18 years old and confirmed with SARS-CoV-2 infection based on positive nucleic acid or antibody detection. Patients with unclassified diagnoses, and moderate patients who had not been discharged by the end of the study were excluded. As of April 14th, 2020, 2907 COVID-19 patients were screened, 265 patients did not meet the inclusion criteria, including 260 cases of negative nucleic acid or antibody test and 5 cases of children or adolescents. Meanwhile, a total of 217 cases were excluded: unclassified or mild type cases (n = 206), moderate patients still in hospital (n = 11). Finally, 2425 of 2642 patients (1681 moderate patients and 744 severely ill patients) were included (Figure 1). This study was approved by the Ethics Committee of Huoshenshan Hospital.

Figure 1.

Flow chart for the participants

According to "Diagnosis and Treatment Protocol for Novel Coronavirus Infection-Induced Pneumonia (Version seven)" published by the National Health Commission of China.[8] Mild cases were defined as having mild clinical symptoms (low fever, slight fatigue) and no evidence of pneumonia on imaging, most cases recovered after one week. Mild patients were not included in this study due to the mild symptoms, and majority of them do not need hospitalization. Moderate cases were defined as having symptoms such as fever and respiratory tract symptoms (cough, sore throat, runny nose, and sneezing), etc., with pneumonia. Some cases may have no clinical signs and symptoms, but imaging shows lung lesions. Adult severe cases were defined as meeting any of the following three criteria: (1) respiratory distress, respiratory rate (RR) ≥ 30 times/min; (2) oxygen saturation ≤ 93% at resting state; (3) arterial partial pressure of oxygen (PaO2)/oxygen concentration (FiO2) ≤ 300 mmHg. Critical cases were defined as meeting any of the following criteria: (1) respiratory failure and requiring mechanical ventilation; (2) shock; (3) with other organ failure require Intensive Care Unit (ICU) care. In this study, we combined severe and critical cases as severely ill patients.

The discharge criteria were defined as the following conditions: (1) body temperature returned to normal for at least three days; (2) respiratory symptoms improved obviously; (3) pulmonary imaging showed obvious absorption of inflammation; (4) nucleic acid test was negative for two consecutive times on respiratory tract samples, and the sampling interval was at least 24 hours.

MuLBSTA score, a previous scoring system for predicting the poor prognosis of viral pneumonia, was calculated using following factors: (1) Imaging multiple pulmonary infiltrations (5 points), (2) Lymphocyte counts ≤ 0.8 x10[9]/L (4 points), (3) Combined with bacterial infection (4 points), (4)acute-smoker (3 points)/quit-smoker (2 points), (5) History of hypertension (2 points), (6) Age ≥ 60 years old (2 points). The cut-off value for mortality risk stratification was 12 points.[9]

Data Collection

Demographic information, clinical characteristics, radiological data and treatment information of each patient were extracted through the electronic medical record system using a standardized uniform form. Most of treatment measurements were to reduce clinical symptoms and to provide supportive care, such as antibiotics, antiviral, corticosteroids, traditional Chinese medicine, oxygen therapy, etc. More than 85% of patients with SARS-CoV-2 infection are being treated with traditional Chinese medicine in China, such as Lian Hua Qing Wen Capsule, QingfeiPaidu decoction, Tan Re Qing injection,Xue Bi Jing injection, etc. These drugs have been recommended as general prescriptions in the diagnosis and treatment protocol of COVID-19.[8,10]

We also recorded the results of laboratory tests on the peripheral blood of patients within 48 hours after admission. The laboratory biomarkers included blood routine indices [leucocyte count, lymphocyte count, hemoglobin, platelet count, neutrophil-to-lymphocyte ratio (NLR)], infection/inflammation-related indices (CRP), blood biochemistry indices [alanine aminotransferse (ALT), albumin, blood urea nitrogen (BUN), creatinine, creatine kinase, LDH], blood coagulation indices (prothrombin time, D-dimer). All data were checked by two researchers (Yu Xu and Bin Wang) and any disagreement was reached by consensus or participation of third researcher (Li Bai).


For moderate patients, the length of hospital stay (discharge date minus admission date) was the primary outcome. We used the median of length as the cut-off point to divide moderate patients into short-stay and long-stay groups. For severely ill patients (including severe and critical type), the primary outcome was disease progression, meeting any of the following criteria: from severe to critical or death, from critical to death, or admission to ICU.

Statistical Analysis

Continuous variables were presented by medians with interquartile ranges (IQR), and categorical variables by numbers with percentages. Difference comparisons between groups were performed by a Mann-Whitney U test, Kruskal-Wallis H test or Chi-Square test.

A logistic regression analysis was performed to evaluate the independent factors associated with outcomes. In an univariate logistic regression, all laboratory biomarkers were brought in the form of continuous variables. Specific symptoms were replaced by the number of symptoms in this analysis. In a multivariate logistic regression, laboratory biomarkers were defined as categorical variables using the upper or lower limit of normal values (see Additional file 1: Table S1 for details). The cut-off point of NLR was defined by a receiver operator characteristic (ROC) curve (with largest Youden index). A multivariate logistic regression was performed with significant variables (p < 0.05) in the univariate logistic regression. Firstly, the variance inflation factor (VIF) was used to identify collinearity among the covariates. The collinearity was negligible cause the VIFs of variables were less than 2.5. Then three methods (entering, forward and backward for likelihood ratio test) were used to select the significant variables in the multivariate logistic regression. Variables retained in any one of the three method models (with p < 0.05) were used to construct the final model by an entering method (likelihood ratio test). In order to rule out the impact of death on the length of stay of moderate patients, sensitivity analysis was performed to exclude the dead patients. We estimated the goodness of fit of the final model using a Hosmer and Lemeshow test. Risk stratification scores were assigned by the weight of different levels of significant factors. The weighted point (λ) of each factor was simplified by the integer form of the quotient of one factor's regression coefficient and the lowest regression coefficient in the model as shown in Figure 2 (e.g., number of symptoms > 3 got one point because the quotient of its regression coefficient and LDH's regression coefficient equal to 1.29),[11] and total points were calculated by summing these weighted points.

Figure 2.

Factors associated with hospitalization duration in moderate patients and disease progression in severely ill patients. a Moderated patients. b Severely ill patients. CRP, C-reactive protein; LDH, lactate dehydrogenase; NLR, neutrophil-to-lymphocyte ratio. ORs (95% CI) and p-values were calculated using a multivariate logistic regression analysis adjusted the variables with p < 0.05 in the univariate logistic regression analyses. * indicates p < 0.05. β is the regression coefficient of the multivariate logistic regression model. Risk scores were assigned by the weighted point of factors which simplified by the integer form of the quotient of one factor's regression coefficient and the lowest regression coefficient

An internal validation was performed to estimate the predictive performance of risk scores by bootstrapping with 1000 replications of the derivation cohort. The discriminative ability was assessed using the area under the ROC curve (AUC). Discrimination between TRPNCLP and MuLBSTA score was also assessed by comparing AUC, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) for severely ill patients. The calibration for agreement was measured by a calibration-in-the-large (perfect = 0), calibration slope (perfect = 1), and calibration plot after deviation correction.[12] Statistical analysis was performed with SPSS (version 25.0; SPSS Inc., Chicago, IL, USA.) and R (version 3.5.4, R Foundation for Statistical Computing, Vienna, Austria), A two-tailed p-value < 0.05 was considered statistically significant.