In December 2019, there was an emerging viral infection outbreak in Wuhan, China. The pathogen was later identified as a new strain of coronavirus, severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), and the disease it caused was named coronavirus disease 2019 (COVID-19). The disease rapidly spread from Wuhan to the rest of the world (1).

The disease severity ranged widely, from asymptomatic or minor symptoms, such as rhinorrhea, productive cough, anosmia, ageusia, and fever, to more severe conditions, such as pneumonia, acute respiratory failure, and even acute respiratory distress syndrome (ARDS). It can progress rapidly (2, 3) and advanced life support with intensive care, such as oxygen therapy, non-invasive ventilation (NIV), and invasive mechanical ventilation (IMV), may be warranted. Excessive demand on healthcare services overwhelmed healthcare systems worldwide (4–7). Administration triage for optimized patient care became essential.

Clinical evaluation alone may lead to misjudgment, under- or overestimation of disease severity and result in suboptimal medical treatment and admission to an inappropriate setting (8). Disease severity scores have been proposed since the early 1980s to help physician decision-making and predict outcomes. For instance, the CURB-65 (confusion, uremia, respiratory rate, BP, age ≥ 65 years) score and qSOFA (quick sepsis-related organ failure assessment) score are clinically relevant predictive tools for community-acquired pneumonia and sepsis, but their risk prediction performance in COVID-19 is not satisfactory (9, 10).

Several predictive scores had been published (11–20), but only a handful of them had ever been validated externally (10). The first wave of the COVID-19 pandemic in Taiwan occurred with a delay of several months in comparison with the first waves in other countries; nevertheless, the waves had similar viral characteristics. During the first wave, all COVID-19 patients had to be admitted to a hospital for quarantine according to the Taiwan Centers for Disease Control (CDC) regulation (21), regardless of the severity. Therefore, our cohort may be more representative of the spectrum of COVID-19 disease and be a good cohort to validate the accuracy of the previous predictive scores. The Spanish Society of Infectious Diseases and Clinical Microbiology (SEIMC) score (11) and the Shang COVID severity score (12) were developed to predict mortality, while the COVID-intubation risk score-neutrophil/lymphocyte ratio (IRS-NLR) score (13), inflammation-based risk score (14), and ventilation in COVID estimator (VICE) (22) score were designed to predict the need for IMV. All five predictive scores are useful since only clinical parameters and commonly available laboratory results were included, but the accuracy of these scores has been uncertain. Herein, the primary aim of the present study was to validate these severity scores and predictive models to predict mortality and the need for IMV.

A total of 311 patients were enrolled in this study period and were followed until they were discharged from our hospital or died. A total of 26 patients who refused intubation during respiratory failure with DNI orders were excluded, leaving 285 patients for inclusion in the analysis.

This study evaluated the performance of the COVID-19 SEIMC score, the COVID-IRS-NLR score, the inflammatory-based risk score, the Shang COVID severity score, and the VICE prediction rule in predicting 30-day mortality and IMV requirements in hospitalized COVID-19 patients. In our cohort, the SEIMC score and Shang COVID severity score were good models for predicting 30-day mortality. For intubation prediction, the IRS-NLR, and VICE score prediction rules showed the best performance.

Our results reinforce the results of several previous studies that found specific initial parameters to be significant predictors of poor outcome in patients with COVID-19. Age (23, 24), lower SpO₂ (25), higher D-dimer, higher CRP, and hypoalbuminemia (26) increased the chances of mortality and ventilation requirements in our study. Previous studies found that an increase in D-dimer and fibrinogen is associated with an increase in COVID-19 severity and mortality (27–30). Increased D-dimer represents activation of coagulation cascades secondary to systemic inflammation and causes microthrombi formation inside the blood vessels that can induce disseminated intravascular coagulation (DIC) (31). CRP levels are reliable markers for prognostic factors (32). Taken together, these results indicate that COVID-19 virus-induced inflammatory and hypercoagulation responses drive the severity of disease (33).

Hundreds of prognostic scoring systems have been developed and studied during the COVID-19 pandemic to predict different outcomes, including mortality (11, 12, 15), severe illness (16–18), critical illness (34), intensive care unit (ICU) admission (19, 20, 35), or IMV use (13, 14). The definition of severe COVID-19 varies across different studies, and the hospitalization criteria may differ by disease prevention policy across countries. The complexity of the ICU admission criteria may fluctuate and be affected by demography and ICU bed scarcity (10). Therefore, the outcomes used in our study, 30-day mortality and IMV, are the most clinically relevant, while they have objective and reduced diversity.

Clinical scoring systems are designed to aid decision making and add to clinical judgment in different healthcare services. Some simplified scoring methods are designed to frequently assess dynamic requirements for escalating levels of respiratory support and to rescore after interventions, such as the Brescia-COVID respiratory severity scale (BCRSS) (36) and the Quick COVID-19 Severity Index (qCSI) (37). These kinds of scoring systems are practically applicable in emergent department settings, which allows quick triage. Most COVID-19 risk scores aimed to predict ultimate outcomes by using the initial evaluation, while some scores use parameters that have not been routinely collected in our cohort [such as CT scan (38) or red cell distribution width (39)]. Further prospective studies are needed for further validation.

To date, the largest, multicenter cohort of 14,343 patients to validate systematically selected prognostic scores for 30-day in-hospital mortality had been demonstrated that 4C mortality (15) and ABCS score had modest utility (AUC > 0.75) (10). The SEIMC scores had low prediction in that French cohort but good prediction in our cohort. The Shang et al. (12) severity score possesses the highest AUC-ROC curve in our patient population. The mortality rate observed within our cohort was much lower than expected in the Shang et al. (12) cohort. The mortality rate was 81.1% among the high-risk group (above 2 points), and the observed mortality rate of the high-risk group in our cohort was only 19.6%. The differences between the two cohorts may be because they faced the first wave of COVID-19, and there were no well-established treatment strategies. In contrast, the SEIMC score predicted the mortality rate in our cohort more precisely (11), while incremental risk stratification represented increased mortality.

The COVID-IRS-NLR and VICE score showed the strongest prediction and greatest accuracy of IMV among our patient population. The intubation rate of the high-risk and very high-risk patients based on the COVID-IRS-NLR score and those who were in quintiles 4th and 5th based on the VICE score was high (13, 22), and these factors may warrant intubation. Early intubation could prevent patient self-inflicted lung injury (40). However, IMV is also associated with complications, such as prolonged sedation and paralysis, infection, and barotrauma (41), and it was reasonable to treat the patients in the low-risk group using a high-flow nasal cannula or awake proning rather than early intubation.

Triage is important in the face of the COVID pandemic (42). If we can foresee the possible progress of the patient’s condition during the first episode, different approaches, treatment options, and patient relocation can be arranged accordingly (43, 44). Patients who are predicted to be less likely to worsen can be treated with a step-down approach and at home without consuming medical resources (45). Patients with a high risk of death or requiring mechanical ventilation support should be admitted to the hospital ward or even the ICU. Therefore, we can reduce the possibility of missed diagnosis of severe COVID-19 and mortality (43).

The underlying reasons for the performance differences among these severity scores could be multifactorial. Patient characteristics (46), vaccination status (47), healthcare system (48), and the variables included in each of the score all contribute to the performance differences. Table 3 shows the different variables used in each scoring systems. The variables could be categorized into patient characteristics, symptoms/signs, laboratory data, and clinical parameters, such as respiratory rate, SpO₂/FiO₂. Age was included both in the SEIMC score and Shang COVID severity score which were the better predictive scores of 30-day mortality. These finding implied the importance of age in driving severity among hospitalized COVID-19 patients (49). The good predictive scores for IMV in our study were the COVID-IRS-NLR and VICE score, and the common variables were lactic dehydrogenase (LDH) and SpO₂/FiO₂. Elevated LDH was associated with poor outcome in COVID-19 patients (50). SpO2/FiO2 may substitute for PaO2/FiO2 as a diagnostic and prognostic marker in COVID-19 patients (51). The performance differences do exist in different race/ethnicity and regions; external validation is needed (52). Multi-center, cross-country studies to verify the accuracy of these severity scores need to be conducted.

This study had several limitations. First, it was conducted at a single center, and the results may not be generalizable worldwide. Second, the retrospective design with some missing values for several important variables, such as body mass index (BMI) and statin use, results in an inability to validate more scoring systems. Third, due to the small number of cases in our cohort, it may not be significant to identify new cut-off points in each score. Finally, our cohort was composed of COVID-19 alpha variant patients and most of them were unvaccinated. COVID-19 vaccination status is associated with the severity of COVID-19 illness. The prediction ability of the five risk scores for other variants or vaccinated populations is unknown. Further well-designed prospective studies are needed to validate these findings in the future.

The discriminative performance of the SEIMC score and Shang COVID severity score were good for 30-day mortality in COVID-19 hospitalized patients. For intubation prediction, the COVID-IRS-NLR, and VICE score prediction rules showed the best performance.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by the Institutional Review Board of MacKay Memorial Hospital. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

C-JW contributed to the conception and study design. H-PC and Y-HT were in charge of execution, acquisition of data, interpretation, and drafting. C-YC took charge of data analysis. C-HC, W-KC, K-CK, and Y-TC were in charge of execution, acquisition of data, analysis, and interpretation. J-CW and C-YL took charge of drafting, revising, and critically reviewing the article. All authors gave final approval of the version to be published, had agreed on the journal to which the article has been submitted, and agreed to be accountable for all aspects of the work.