The retrospective study enrolled patients with COVID-19 pneumonia who first attended Shandong Provincial Hospital affiliated to Shandong First Medical University, between 2022.12.01 and 2023.01.10. Based on the World Health Organization (WHO) Interim Guidelines for the Treatment of Novel Coronavirus Pneumonia (10th edition) published by the National Health Commission of China, COVID-19 pneumonia is diagnosed. Patients meeting the following inclusion criteria were included in this study: (1). Positive SARS-COV-2 RT-PCR by pharynx swab specimen; (2). Typical chest CT presentation of combined COVID-19 pneumonia. Exclusion criteria: (1). patients with unconfirmed COVID-19 pneumonia; (2). incomplete baseline clinical information; and (3). patients with no clinical outcome (discharge/death) during the study period. Severe pneumonia caused by COVID-19 is characterized by the fulfillment of at least one of the following criteria: (1). a respiratory rate of 30 breaths per minute or higher accompanied by shortness of breath; (2). an oxygen saturation level of 93% or lower while at rest without supplemental oxygen; (3). an oxygenation index (the ratio of arterial partial pressure of oxygen to oxygen concentration) of 300 mmHg or less (with 1 mmHg equivalent to 0.133 kPa) or an arterial partial pressure of oxygen equal to or below 60 mmHg on room air at rest; (4). a noticeable deterioration in clinical symptoms along with lung imaging that reveals a significant increase of over 50% in the affected area within a time frame of 24 to 48 h.

A total of 1,739 patients diagnosed with COVID-19 pneumonia participated in this study, among whom there were 151 patients experiencing severe pneumonia (Figure 1). In Figure 2, the selection process of the study population comprising 151 patients diagnosed with severe COVID-19 pneumonia is illustrated. The patients were systematically randomized into two distinct cohorts: the training cohort and the validation cohort. This division was executed in a ratio of 7:3, ensuring that the majority of participants were allocated to the training group while a smaller subset was designated for validation purposes. This methodological approach facilitates a comprehensive analysis and validation of the findings derived from the research. The ethics committee of Shandong Provincial Hospital affiliated with Shandong First Medical University approved the investigation (ethics number: SWYX: No. 2022-593). The study adhered to the Statement of Standards for Reporting Diagnostic Accuracy Studies, and the hospital’s ethics committee waived the requirement for written informed consent from patients with COVID-19 pneumonia (15).

All patients were admitted to the Shandong Provincial Hospital affiliated with the Shandong First Medical University, and their pharyngeal swab specimens were tested for SARS-CoV-2; patients who tested positive by real-time RT-PCR were considered confirmed cases.

After thorough consultation with respiratory and imaging experts and an extensive review of the literature related to COVID-19 prognosis, we finally identified a set of clinical, laboratory and imaging baseline characteristics for inclusion in the study. A number of clinical trials have now confirmed that various factors, such as Hypertension (P = 0.038), higher neutrophil-to-lymphocyte ratio (NLR) (P = 0.001), increased NT-proBNP (P < 0.001), age ≥ 70 years (OR = 1.184, 95% CI 1.061–1.321), panting (breathing rate ≥ 30/min) (OR = 3.300, 95% CI 2.509–6.286), lymphocyte count < 1.0 × 109/L (OR = 2.283, 95% CI 1.779–3.267), and interleukin-6 (IL-6) > 10 pg/ml (OR = 3.029, 95% CI 1.567–7.116) are significantly associated with the poor prognosis of patients with COVID-19 (16, 17). Therefore, in this study, we collated and included laboratory and imaging indicators that had been mentioned in the literature and were clinically available. These included: routine blood (WBC, Neutrophil and Lymphocyte), inflammatory factors (C-reactive protein, SAA, PCT, IL-6 and NLR), cardiac enzyme profiles (Tn-T and Pro–BNP), coagulation (D-DIMER), liver and kidney function (AST, ALT, LDH, ALB, HLN, and GLU), blood biochemistry (Ca, K, Na, Cl), and imaging features (Nature of lesion, Bronchial wall thickening, Thickening of blood vessels, Tree bud levy, Thickened lobular septa, Intralobular septal thickening, Stretch bronchiectasis, Pleural effusion, Heart enlargement, Pericardial effusion and Widening of the pulmonary artery) (Figure 3). Patients with incomplete baseline clinical information were excluded, as stated in the exclusion criteria. Among the baseline clinical characteristics were basic patient information, major symptoms and comorbidities, all of which were obtained at the time of the patient’s first admission (Table 1). Baseline laboratory characteristics included: routine blood (WBC, neutrophils, lymphocytes), inflammatory factors (c-reactive protein, SAA, PCT, IL-6, NLR), cardiac enzyme profile (Tn-T, Pro- BNP), coagulation (D-DIMER), liver and kidney function (AST, ALT, LDH, ALB, HLN, GLU), blood biochemistry (Ca, K, Na, Cl); the Cut-off value of each index in predicting the prognosis of patients with COVID-19 pneumonia was used as the basis for grouping. Cut-off values for continuous variables were determined based on clinical standards and previous studies. For example, SpO2 ≤ 93% is a criterion for severe COVID-19 according to WHO guidelines, and GLU > 7.5 mmol/L is a common threshold for hyperglycemia. For imaging features like heart enlargement, it was defined as a cardiothoracic ratio > 0.5 on axial CT images, as assessed by experienced radiologists.

Computed tomography scans for each patient collected during their hospital stay were sourced from the cloud-based data storage system. The original chest radiograph and CT scan were characterized as those taken upon admission or within one week of the onset of symptoms. Follow-up chest radiographs were performed every 2–3 days until the patient was discharged (18).

All the patients’ CT features were independently assessed on all data sets by two radiologists with 10 years of experience in reading chest CT images, who had no knowledge of the clinical and laboratory results. The analysis focused on the extent and patterns of pneumonia observed in both initial and follow-up CT scans. The severity of pneumonia across all lung zones on the CT scans was rated on a scale from 0 to 2, where a score of 0 indicates no pneumonia, a score of 1 indicates 1%–25% involvement, and a score of 2 indicates more than 25% involvement (18, 19). Pneumonia patterns identified in the CT images were classified as typical, indeterminate, atypical, or negative, according to the RSNA Expert Consensus Statement (18, 20). A typical appearance was characterized by peripheral bilateral ground-glass opacities (GGOs) or multifocal round GGOs, either with or without consolidation, and possibly featuring intralobular lines or a reverse halo sign. An indeterminate appearance referred to the presence of GGOs, with or without consolidation, but lacking definitive typical characteristics. In contrast, an atypical appearance was noted when typical or indeterminate features were absent, while lobar and/or segmental consolidation was observed without the presence of GGOs, distinct centrilobular nodules, lung cavitation, or smooth interlobular septal thickening accompanied by pleural effusion (18). We have also found some other interesting imaging features, including bronchial wall thickening, thickening of blood vessels, tree bud levy, stretch bronchiectasis, heart enlargement, pericardial effusion and widening of the pulmonary artery.

After recruitment, all patients were randomly grouped into a training cohort and a validation cohort in a 7:3 ratio. A nomogram predicting survival in patients with severe COVID-19 pneumonia was constructed in the training cohort based on baseline characteristics, and then validated in a separate validation cohort.

Categorical variables are described using frequencies and proportions, and continuous variables are described using medians and quartiles. Univariate logistic regression analysis was used to identify clinically relevant variables associated with death in patients with severe COVID-19 pneumonia in the training cohort, variables showing univariate relationships associated with death in patients with severe COVID-19 pneumonia (p < 0.05) were entered into a multivariate logistic regression analysis, and improvements in goodness of fit were assessed by reducing the Akaike information criterion for backward stepwise selection. If the number of events was too small to calculate an OR, the variables were excluded; a final model was created by selecting baseline clinical indicators from the training cohort; and a nomogram was constructed from the overall data of the training cohort based on the selected final model. To assess the ability of the nomogram model to differentiate the risk associated with death in patients with severe COVID-19 pneumonia. The area under the subject’s working curve (AUC) and 95% CI were calculated and compared to the AUC curve and 95% CI for each independent variable of interest in the training cohort. To assess the agreement between nomogram predictions and actual observations in the training cohort, 1,000 resamples (with replacement) were set up and calibration curves were created. To assess the clinical applicability of the predictive nomogram, a DCA curve analysis was performed by quantifying the net benefit at different threshold probabilities of death in patients with severe COVID-19 pneumonia. The net benefit was defined as the proportion of true positives minus the proportion of false positives, as measured by the relative risk of false positive and false negative outcomes. To assess the internal validity of the model, the model was applied to an independent dataset from Shandong Provincial Hospital affiliated to Shandong First Medical University. The internal validity of the model was assessed using AUC, calibration and DCA curve analysis for the risk associated with death in patients with severe COVID-19 pneumonia, respectively. All statistical analyses were performed using R (version 4.2.1) and p < 0.05 was considered statistically significant.

A total of 1,739 confirmed cases of COVID-19 with viral pneumonia first attended at Shandong Provincial Hospital affiliated to Shandong First Medical University between 2022-12-01 and 2023-01-10 were enrolled in this study. A total of 151 patients were diagnosed with severe COVID-19 after screening according to stringent entry criteria, the rate of serious illness was 8.68% (Figure 1). Table 1 display the fundamental characteristics of seriously ill patients. In this study, the median age of the patients was 75 years (range: 36–96 years), and 106 (70.1%) were males. A total of 72 (47.6%) of these patients had previously received at least one dose of the COVID-19 vaccine. The majority of patients in the entire cohort experienced fever (68.8%) and chest congestion (60.9%), whereas 69 patients (45.6%) had a cough. The most prevalent comorbidity was hypertension (52.3%), followed by cardiovascular disease (43%). Diabetes, chronic pulmonary disease, and cancer accounted for 30.4%, 12.5%, and 11.1% of all patients, respectively. A total of 106 patients (70%) were allocated to the training cohort, while 45 patients (30%) were assigned to the validation cohort. All patient information was provided, and the majority of characteristics did not differ substantially between the two groups (Table 1).

The Cut-off values of each index in the prognosis prediction of patients with COVID-19 were used as the basis for their grouping, respectively (Table 2).

Oxygen therapy was the most common treatment for patients hospitalized with severe COVID-19. All patients hospitalized with severe COVID-19 received oxygen therapy, which was mainly nasal cannula oxygen (83.4%), mask oxygen (35.0%) and mechanical ventilation (13.2%); followed by antibiotic therapy (96.6%) and glucocorticoid therapy (90.7%). A total of 78.1%, 20.5%, and 13.2% of patients received antiviral drugs, intravenous immunoglobulin and Chinese medicine, Antiviral drugs, intravenous immunoglobulin and herbal medicines were used in 78.1%, 20.5%, and 13.2% of patients. Antivirals mainly consisted of azelvadine tablets and nematovir/ritonavir tablet combination packs (Table 3).

Among patients with severe COVID-19, 92 (61%) were discharged after cure, with a mean hospital stay of 21 days; 59 (39%) died during their hospital stay, with a mean hospital stay of 19 days (Figure 4).

In this study, the relationship between 54 factors and were included in the univariate logistic regression analysis. The results showed that Age, Temperature, Heart rate, Pulse Oximetry, Lymphocyte, C- reactive protein, SAA, NLR, Tn-T, Pro–BNP, D–DIMER, ALT, ALB, GLU, HLN, Ca, LDH, PCT, IL-6, Pleural effusion and Heart enlargement were associated with poor prognosis in patients with severe Omicron pneumonia (P < 0.05). These factors were subsequently included in a multivariate logistic regression analysis (Table 4), which showed that Pulse Oximetry (p = 0.034, OR = 0.25, 95% CI: 0.07–0.9), LDH (p = 0.047, OR = 4.39, 95% CI: 1.02–18.93), SAA (p = 0.020, OR = 25.01, 95% CI:1.65–378.4), GLU (p = 0.022, OR = 5.49, 95% CI: 1.28–23.44), Ca (p = 0.029, OR = 0.23, 95% CI: 0.06–0.86), Pleural effusion (p = 0.008, OR = 8.04, 95% CI:1.73–37.46), and Heart enlargement (p = 0.008, OR = 8.94, 95% CI: 1.79–44.59) were independent prognostic factors affecting survival in patients with severe Omicron pneumonia. Multicollinearity among the variables included in the multivariate model was assessed using the variance inflation factor (VIF), and no significant multicollinearity was found (all VIF < 5). Patients with a combination of decreased pulse oximetry and Ca, elevated LDH, SAA, and GLU levels at baseline, as well as pleural effusion and heart enlargement tended to have a poorer clinical prognosis.

In the training cohort, 63 (59.5%) patients were eventually cured and discharged, with an average length of hospital stay of 20.9 days; 43 (40.5%) patients experienced death during treatment, with an average length of hospital stay of 18.3 days. A nomogram model was developed to predict the risk of in-hospital death during treatment of patients with severe COVID-19 pneumonia. The final model included independent prognostic influences that were significant after multivariate logistic regression analysis: Pulse Oximetry, LDH, SAA, GLU, Ca, Pleural effusion, and Heart enlargement (Figure 5). The predicted risk of in-hospital death during treatment was determined by incorporating patient information into a risk scale to produce an overall score.

In the validation cohort, 29 patients were eventually discharged with a cure, with a mean length of hospital stay of 21.7 days; 16 patients experienced death during treatment, with a mean length of hospital stay of 17.7 days. The AUC for the training cohort was 0.914 and for the validation cohort was 0.802 (Figure 6). A balance test indicated that most baseline characteristics were comparable between the training and validation cohorts, although some differences were observed (e.g., gender distribution, prevalence of hypertension). This is a common challenge in single-center studies with limited sample size and random split, and it may partially account for the observed drop in the validation AUC. We also compared the performance of our nomograms with the performance of single factor categories for prediction, such as baseline clinical characteristics (Pulse Oximetry), serology (LDH, SAA, GLU, and Ca) and imaging (Pleural effusion and Heart enlargement). The results show that combining multiple types of baseline features nomograms is more advantageous than using a single factor to predict patient prognosis (Figure 6). We also carried out the plotting of calibration curves for the training and validation cohort survival prediction (Figure 7). This was used to illustrate the agreement between the nomogram’s predictions of patient survival prognosis and actual observations. The results of the study show that the model has good overall predictive efficacy. The predictive performance of the full nomogram was significantly superior to that of the single strongest predictor, Pulse Oximetry, in the training cohort (DeLong test, p < 0.05).

An evaluation of the clinical applicability of the risk prediction nomogram was conducted using DCA grounded in net benefit and threshold probabilities. The DCA results indicated that our risk prediction nomogram provided a greater net benefit across a broad spectrum of threshold probabilities in both the training and validation cohorts (Figure 8). The DCA indicated that the nomogram provided a positive net benefit for threshold probabilities between approximately 10% and 65%. The consistent performance in both the training (a) and validation (b) cohorts underscores the robustness of the model for potential clinical application.