Predicting mortality risk in hospitalized COVID-19 patients: an early model utilizing clinical symptoms

According to the World Health Organization (WHO), as of April 2023, there have been over 762 million confirmed cases of COVID-19 globally, resulting in more than 6.8 million deaths. In Vietnam, the WHO reported 11,527,326 confirmed cases of COVID-19 and 43,186 deaths [1].

Despite global efforts to control the COVID-19 pandemic, the emergence of new viral strains poses a continued threat. The number of cases remains high in many countries worldwide, and the global mortality rate for COVID-19 was estimated at 3.4%, surpassing previous estimates of approximately 2% [2]. Individuals infected with the delta variant of COVID-19 typically exhibited nonspecific clinical presentations, ranging from asymptomatic SARS-CoV-2 carriers to middle-to-moderate and critical COVID-19 cases, and in some instances, led to fatalities possibly due to COVID-19 [3, 4]. Symptoms spanned from mild fever, fatigue, cough, and shortness of breath to severe cases involving lower respiratory tract infections and pneumonia. A subset of these cases progressed to Severe Acute Respiratory Distress Syndrome (ARDS) or multiorgan failure, ultimately resulting in mortality. To effectively stratify patients, optimize resource allocation, and provide appropriate treatment and care, there is a pressing need for a simple and accurate prognostic tool capable of rapidly identifying individuals at high risk of mortality. Early prognosis is important because it enables the prediction of treatment outcomes and facilitates appropriate patient management.

This study aimed to develop an early and straightforward prognostic model for predicting mortality among COVID-19 patients. Given the ongoing pandemic situation leading to a surge in patient numbers across medical facilities and considering the existing limitations in both medical equipment and the professional capacity of healthcare staff, especially in primary medical facilities in Vietnam. We opted for a research model centered on easily identifiable and assessable clinical symptoms. This choice diverges from prognostic models that incorporate numerous clinical and paraclinical variables, which can be challenging to implement. The model was based on initial and fundamental clinical information, which can be easily employed in various medical facilities and is suitable for the current clinical pandemic environment in Vietnam. Additionally, we aimed to explore the applicability of this model in early prognosis for patients with other acute respiratory infections caused by different viral etiologies, such as influenza A, Rhino, and Adeno, as these diseases share certain similarities in terms of pathogenesis and microbiology. The successful implementation of this model could contribute to the adoption of appropriate treatment approaches and ultimately reduce the mortality rate among affected patients.

The clinical presentation of COVID-19 is highly diverse, and its rapid progression poses challenges for prognosis and classification in clinical settings. Our study revealed that age over 50, a requirement for prehospital oxygen therapy, and the presence of multiple comorbidities were closely associated with an increased risk of mortality. These findings were consistent with several other published studies [6‐8].

Symptoms such as fever, cough, and shortness of breath are commonly observed in patients and typically manifest soon after viral infection of the respiratory system. The severity of respiratory symptoms, ranging from dyspnea to acute respiratory failure, depends on the extent of lung damage. Acute respiratory failure and progressive multiorgan damage may lead to acute disturbances of consciousness. Our study identified fever, dyspnea, acute disturbances of consciousness, blood oxygen saturation below 94% (SpO2 < 94%), and the immediate need for oxygen support upon admission as factors significantly associated with the risk of mortality. These findings align with reports from other authors who have highlighted the prognostic significance of advanced age, dyspnea, impaired consciousness, and hypoxemia in predicting disease progression and mortality in hospitalized COVID-19 patients [9‐11].

Since the early stages of the pandemic, numerous studies have focused on developing prognostic models to stratify patients effectively and reduce the rates of severe disease progression and mortality. In a review conducted by C. Buttia et al. (2022), a total of 314 published studies from 40 countries were identified, including 152 studies on mortality prognosis and 35 studies on mortality prognosis and intensive resuscitation treatment. Factors found to have a significant impact on mortality risk included age, sex, hypoxemia, body temperature, pulse, underlying diseases, impaired consciousness, levels of C-reactive protein (CRP), urea, and D-dimer, neutrophil count, lymphocyte percentage, and platelet count. The area under the receiver operating characteristic (ROC) curve for mortality prognostic models ranged from 0.49 to 0.99, with sensitivity ranging from 15.4% to 100% and specificity ranging from 10.9% to 98.7% [12].

With the aim of developing a clinical symptom-based prognostic tool for early risk classification upon hospital admission, we focused on selecting objective and easily assessable clinical factors associated with mortality risk. Eight clinical factors were identified from the univariate analysis as simple, objective, and significantly associated with mortality risk. These factors were further analyzed using multivariate analysis, revealing age over 50 years, multiple comorbidities, dyspnea, acute disturbance of consciousness, SpO2 below 94%, and oxygen demand exceeding 5 L per minute upon admission as independent predictors of mortality. A logistic regression algorithm was applied to construct a predictive model for mortality in COVID-19 patients, demonstrating favorable performance across all evaluation measures. This model facilitates risk stratification of patients and has the potential to reduce mortality rates within hospitals [13]. Prognostic models generated through multivariate analysis hold significant value, particularly in individualized predictions, by providing insights into a patient's likelihood of survival [14]. In a previous study, we utilized the multivariate analysis approach to assess the prognostic capability of the CURB-65 scale, which comprises five variables (consciousness disorder, uremia, tachypnea, low blood pressure, and age ≥ 65 years), for mortality prediction in hospitalized pneumonia patients with COVID-19. Our findings indicated that the CURB-65 scale exhibits promising prognostic ability, with a CURB-65 score of ≥ 2 points demonstrating a sensitivity of 82% and specificity of 83% in predicting mortality [15].

The findings from our study demonstrated a proportional increase in the mortality rate and MH175 score. When the MH175 score reached ≥ 3 points, the mortality rate exceeded 39.8%. The MH175 score exhibited a good prognostic capability for mortality, as indicated by an area under the AUROC of 0.87. The optimal threshold value for predicting mortality using the MH175 score was ≥ 3 points, yielding a sensitivity of 96.1%, specificity of 63.4%, positive predictive value of 58%, and negative predictive value of 96.9%.

In a meta-analysis conducted to assess the predictive performance of four common prognostic scores (ISARIC-4C, COVID-GRAM, qCSI, and NEWS) for in-hospital mortality in COVID-19 patients, a favorable predictive value for mortality risk was observed. The ISARIC-4C score exhibited the highest AUROC at 0.799, followed by COVID-GRAM with 0.785, NEWS with 0.764, and qCSI with 0.749 [16].

Although the complexity of the pandemic and variations in healthcare systems across different countries may introduce biases in prognostic models, the ISARIC-4C mortality prognostic model, developed based on a large database from the UK, has demonstrated a relatively low risk of bias. The 4C scale incorporates eight variables: age, sex, respiratory rate, oxygen saturation, comorbidities, state of consciousness, blood urea, and C-reactive protein [6, 17]. The clinical variables included in the MH175 scale were comparable to those in the 4C scale. However, a major distinguishing feature of the MH175 scale is that its variables can be promptly and easily assessed upon patient admission to the hospital, rendering the MH175 scale practical and feasible for prognosis and patient classification in primary healthcare settings.

Currently, the COVID-19 pandemic continues to pose a significant threat to human health despite efforts to control its spread. Mortality rates remain high, particularly among elderly patients who are immunocompromised or have many comorbidities. Additionally, the potential for new strains and outbreaks adds to the ongoing risk. To address these challenges, our study proposes an additional tool for the early prediction of mortality risk in COVID-19 patients. This tool aims to facilitate efficient patient stratification, improve treatment efficacy, and ultimately reduce mortality.

However, it is important to acknowledge the limitations of our study. First, as a single-center study, the generalizability of the findings may be limited. Second, the sample size used in the study was relatively small, necessitating further investigation with larger sample sizes. Moreover, the study's limitation lies in assigning a score to each clinical variable without precisely assessing its clinical significance. Additionally, regarding the MH175 score's prognostic ability for risks such as HFNC ventilation, NIV, and ICU admission, we did not conduct further analysis in this study, which we acknowledge as an additional limitation. Last, in the context of a pandemic, there is a lack of consensus regarding the initial diagnosis and treatment protocols for patients prior to admission, which could potentially influence the risk of mortality.

It is crucial to recognize that all prognostic models have inherent limitations when applied in clinical practice, particularly in the context of a pandemic with diverse patient characteristics across different countries and races, as well as variations in healthcare systems. Ideally, highly effective prognostic models should undergo rigorous evaluation across various populations and countries before being widely implemented in clinical practice. This approach helps mitigate errors in decision-making, patient classification, and treatment choices [18].

Based on our study involving 375 hospitalized COVID-19 patients at Military Hospital 175, we have drawn several key findings. First, the MH175 scale exhibits a favorable ability to predict the risk of mortality, as indicated by its area under the receiver operating characteristic (AUROC) value of 0.87. Additionally, there is a direct correlation between the MH175 score and mortality, with an increasing score corresponding to an elevated mortality rate. Moreover, a threshold value of ≥ 3 points on the MH175 scale emerges as an effective indicator for predicting mortality risk, demonstrating a sensitivity of 96.1%, specificity of 63.4%, positive predictive value of 58%, and negative predictive value of 96.9%. These promising results suggest that the MH175 scale can be implemented in clinical settings to enable early identification of mortality risk, facilitate patient stratification, and aid in selecting appropriate treatment interventions, ultimately reducing the likelihood of death among COVID-19 patients. Additionally, the score scale can serve as a tool to assist in classifying patients based on their risk levels, facilitating appropriate patient allocation within the treatment hierarchy.