Study population and research design

Our investigation comprised a retrospective analysis of 506 individuals diagnosed with COVID-19 who sought medical attention at the Emergency Department of Beijing Youan Hospital, from May 1, 2022, to May 31, 2023. Diagnostic classification was performed in accordance with the guidelines delineated in the ninth iteration of the “COVID-19 Diagnosis and Management Protocol” promulgated by the National Health Commission of the People’s Republic of China⁴. SARS-CoV-2 infection was verified using polymerase chain reaction (PCR) assays. During our study period, the included patients were all infected with the Omicron variant. Patient severity classification into mild/moderate or severe/critical groups was conducted following the COVID-19 clinical management recommendations set forth by the National Institutes of Health (NIH)³⁸. Patients were treated according to the guidelines, with a fixed team of doctors and nursing staff under the guidance of senior physicians, ensuring that patients received appropriate treatment.

This research endeavored to assess the prognostic efficacy of the PIRO Score upon hospital admission for COVID-19 patients. The primary endpoint was defined as mortality within a 28-day period post-admission. The study protocol received approval from the Ethics Committee of Beijing Youan Hospital (protocol identifier: LL-2023-006-K). Research procedures adhered to the ethical guidelines stipulated in the Declaration of Helsinki. This study is based on a research cohort of COVID-19 patients from the entire hospital. All subjects provided written informed consent prior to enrollment. To ensure participant privacy, data were de-identified and stored securely in accordance with institutional protocols.

Inclusion and exclusion criteria

This investigation incorporated subjects who fulfilled the diagnostic criteria delineated in the ninth iteration of the “COVID-19 Diagnosis and Management Protocol,” as promulgated by China’s National Health Commission⁴. Enrollment was contingent upon obtaining informed consent and voluntary participation. Exclusion parameters encompassed refusal to participate, individuals below 18 years of age, gravid females, mortality within 48 h post-admission, and cases with substantial data deficiencies or inaccessible records. Missing important data refers to the absence of data required for calculating the scores or the absence of information about the primary outcomes. The aforementioned criteria were meticulously applied to ensure a representative cohort for analysis, while adhering to ethical research practices and maintaining data integrity. This rigorous selection process aimed to minimize confounding factors and enhance the validity of subsequent findings.

Data collection

Clinical data were extracted from electronic health records, encompassing demographic characteristics, medical history, baseline parameters, vital signs, arterial blood gas analysis results, laboratory findings, and patient outcomes. Demographic variables included sex and age. The medical history focused on comorbidities such as hypertension, diabetes mellitus, coronary artery disease, cerebrovascular disorders, chronic obstructive pulmonary disease (COPD), hepatic conditions, and malignancies.

Vital sign assessment comprised body temperature (°C), respiratory rate (breaths/min), heart rate (beats/min), and systolic blood pressure (mmHg). Arterial blood gas analysis evaluated pH, partial pressure of carbon dioxide (PaCO₂), partial pressure of oxygen (PaO₂), peripheral oxygen saturation (SpO₂), and the oxygenation index (PaO₂/fraction of inspired oxygen [FiO₂] ratio).

The laboratory panel included markers of infection (procalcitonin [PCT, ng/mL] and C-reactive protein [CRP, mg/L]), hematological parameters (hemoglobin, total leukocyte count, neutrophil and lymphocyte counts), coagulation indices (international normalized ratio [INR] and D-dimer [mg/L]), and biochemical indicators (alanine aminotransferase [ALT, U/L], aspartate aminotransferase [AST, U/L], albumin [g/L], total and direct bilirubin [µmol/L]).

Composite inflammatory markers were derived from these laboratory parameters, including neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), lymphocyte-to-CRP ratio (LCR), CRP-to-albumin ratio (CAR), and systemic inflammation index (SII). The SII, calculated as (neutrophil count × platelet count)/lymphocyte count, along with other derived indices, provides critical insights into the severity and nature of the inflammatory response associated with the studied condition, as detailed in Table 1.

Definition of clinical scoring systems

The PIRO score incorporates factors such as comorbidities, advanced age, Infection/Insult, Response, Respiration and lactate (Additional file: Table S1)³⁹.

Organ dysfunction assessment tools, including SOFA, qSOFA, eSOFA, and sSOFA, were utilized to gauge the extent of organ impairment in critically ill patients. These instruments have demonstrated efficacy in predicting in-hospital mortality among adult patients with suspected infection in intensive care unit (ICU) settings (Additional file: Table S2)^40,41.

The CURB-65 score, a validated pneumonia severity index, evaluates five parameters: confusion, blood urea nitrogen, respiratory rate, blood pressure, and age ≥ 65 years (Additional file: Table S3)^42,43.

The PSI, comprising 20 independent risk factors, stratifies patients into five risk classes based on the cumulative score (Additional file: Table S4)^44,45.

The COVID-GRAM model, developed by Chinese researchers in 2020, integrates 10 independent predictors to estimate the probability of progression to severe illness in COVID-19 patients (Additional file: Table S5)²⁶.

The RAPS assesses pulse rate, mean arterial pressure, respiratory rate, and GCS score (Additional file: Table S6)^46,47.

Lastly, NEWS2 evaluates respiratory rate, oxygen saturation, supplemental oxygen requirement, heart rate, level of consciousness, and temperature to detect early signs of clinical deterioration (Additional file: Table S7)^48,49.

Statistical analysis

Normality of continuous variables was assessed via the Shapiro–Wilk method. Data following normal distribution were summarized as mean ± SD, with between-group differences evaluated by independent t-tests. Non-parametric variables were described using median and IQR, and analyzed with the Mann–Whitney U test. For categorical variables, frequencies and percentages were calculated, and comparisons were conducted using either Pearson’s chi-square or Fisher’s exact test, as appropriate. Multiple group analyses employed the Kruskal–Wallis test. The discriminative ability of prognostic models for 28-day mortality in COVID-19 patients was examined through ROC curve analysis. The selection of the optimal threshold, or Cutoff Values, is based on the Youden Index, where Youden Index = Sensitivity + Specificity − 1. A higher Youden Index indicates better predictive performance, and in this study, the selection of the best threshold uses the Cutoff Values with the highest Youden Index. Clinical utility assessment of these models utilized decision curve analysis (DCA). k-Fold Cross-Validation is used to assess the stability of scoring for patient prognosis prediction. The dataset is randomly divided into 10 equally sized subsets (referred to as ‘folds’), with one fold selected as the validation set in each iteration, and the remaining 10 − 1 folds used as the training set. This training and validation process is repeated 10 times, and the average of the 10 results is taken as a robust estimate of model performance. Calibration plots and Brier scores were further used to evaluate the calibration and accuracy of different scoring systems in predicting the 28-day mortality risk of COVID-19 patients. Calibration plots are utilized to assess the consistency between the model’s predicted probabilities and the actual observed outcomes. For instance, whether the observed mortality rate approximates 30% for patients with a predicted 30% probability of mortality. The Brier score measures the accuracy of predicted probabilities, with lower values indicating better accuracy. A score of 0 represents perfect prediction, while 0.25 is akin to random guessing for binary classification tasks such as life or death. Comparisons of AUC values across different scoring systems were performed using DeLong’s method. A p-value < 0.05 was considered statistically significant. Statistical analyses were conducted using R (version 4.2.1; R Foundation for Statistical Computing) software and SPSS (version 22.0; IBM Corp.). Visualization of data was accomplished using GraphPad Prism 9 (GraphPad Software Inc.).