According to our findings, there has been a significant increase in emergency department (ED) visits, especially for patients with COVID-19 pneumonia who are critically ill. The entire period lasted about 2 months, a total of 25423 patients visited the ED, including 631 critically ill patients. Initially, there was a significant increase in both general and critical emergencies, with general cases mainly involving internal ailments and critical cases mainly involving severe COVID-19 pneumonia. The peak lasted for approximately 20 days, and daily consultations reached a high of 668 on December 30 and remained above 500 for 14 days. As internal emergencies decreased and surgical cases rose, visit numbers returned to normal by late January 2023. This surge underscores the ongoing challenges in healthcare systems in managing pandemic-related cases alongside routine emergencies.
In consistent with the health and medical care situation faced by other countries and regions, ED has faced a variety of serious stiff issues during the whole break period, including the shortage of medical resources, the infection of the medical health team, overloaded and sustained high intensity work state, effort-reward imbalance.22,23 As previously mentioned, bed occupancy, oxygen and ventilators shortage are both highly intense in the ED. Under this broad environment of epidemic, our hospital has taken a series of immediate measures to alleviate the scarce resources, including bed utilization, ICU bed expansion, establishment and conversion of COVID-19 units, the assignment and supportment of ED staff. These adjustments have revealed the adaptation and responsibility of healthcare systems is crucial when facing a major public health crisis.
In addition to the overall macroscopical controlment, it is also vital to seek a reasonable and reliable rapid diagnosis and prediction tool, for the optimized diagnosis and therapies of the total critical patients. Computational assistance has indicated its advantage to ED care providers, who under high tension and exhaustion might lead to a decline in the accuracy of clinical judgment, resulting in increased mortality rates.24,25 Beyond the general benefits of data-driven decision-making, AI could help clinical professionals determine who needs a critical level of care more precisely.
In this study, we employed a novel machine-learning method using routine clinical and laboratory data to predict mortality in critically ill patients in the Emergency Department during an epidemic outbreak. This is the first study of its kind to develop a mortality prediction risk model for Chinese ED patients, predominantly affected by COVID-19. Meanwhile, the six machine learning models were trained and tested to find the optimal model and used the gini coefficient to screen risk factors to increase the interpretability of the model. Furthermore, we developed and validated a software-based risk calculator t perform risk scoring and predict the development of critical illness among ED patients. The performance of this risk score was satisfactory with accuracy based on a reliable AUROC as 0·8385 and 15 variables were included, which has the capability to accurately discern patients' likelihood of survival or mortality. The predictive model developed in this study demonstrates high accuracy, suggesting its potential as a valuable asset in emergency care protocols. It is found that all critical emergencies were elder people predominantly, and age is a risk factor for poor prognosis, which is consistent with previous studies.24,26–28 Patients with lower SpO2 and treated with noninvasive ventilation and mechanical ventilation tended to have a higher mortality for the severe respiratory dysfunction.24,28,29 Elevated direct bilirubin has also been found to be associated with poor studies in several studies, which is consistent with our study.26
Cytokine storms are well-known to play an important role in the pathophysiological process of COVID-19.27,30,31 Persistent high inflammatory state and immune dysfunction have been found to be associated with increased mortality in numerous studies.30–33 The proinflammatory state can be manifested as a significant elevation in WBC, ferritin, CRP, IL-6, IL-1 and other inflammatory cytokine. On the other side, the immune dysfunction is accompanied by a decrease in WBC and lymphocyte counts, which leads to an increase in secondary and opportunistic infections.33,34 This phenomenon has also been observed in this study.
The disturbance of coagulation system is a common and critical complication in COVID-19 patients.35–37 All coagulation and fibrinolysis parameters can be affected by COVID-19, including APTT, PT, fibrinogen, FDP, D-dimer and platelet count, which is called COVID-19 induced coagulopathy in the early stage, and with the progression of the disease, it can develop into DIC, resulting in a substantial morbidity and mortality.35–37 In this study, it could be observed that coagulation dysfunction including fibrinolytic hyperfunction and DIC has a higher occurrence in the death group. Accordingly, prolonged APTT and PT, alteration of fibrinogen and decreased platelets levels are both correlated with poor in-hospital prognosis through machine learning.
A key innovation of our study is the application of machine learning techniques to identify mortality predictors in COVID-19 patients. This approach not only contributes to the existing body of knowledge but also offers a practical tool for healthcare practitioners in emergency settings. Based on the above variables we found, a user-friendly GUI is established, which requires further clinical validation. The GUI developed as part of this study serves as a user-friendly platform for rapidly assessing patient risk, aiding in timely and informed decision-making. The current development of wearable systems makes continuous and predictive monitoring possible. However, in a clinical setting, the monitoring of patients still relies on intermittent observations of vital signs combined with the use of EWS. The objective of this study was to develop an approach for real-time outcome prediction based on continuous vital signs monitoring and advanced machine learning techniques.
The potential limitations of this study merit consideration. First, the data is confined to a single institution and a specific post-lockdown period, which may affect the generalizability of the findings. Future research should focus on replicated this study in various locations and for extended periods to validate and enhance the predictive model. Second, the omission of certain prognostic indicators like IL-6 and Lac due to excessive missing data may impact the study's comprehensiveness. Third, the sample size for both constructing and validating the risk score is small, and the data being exclusively from China could limit its applicability globally. The model's reliability and applicability across diverse populations requires further validation studies, especially from outside China.
In summary, this study provides valuable insights into the increased emergency department visits at a top teaching hospital in China following the COVID-19 lockdown. The importance of machine learning in healthcare is underlined, especially through the creation of a predictive model with a graphical user interface. By using initial hospital admission data, this innovation aids in accurately assessing the risk of critically ill patients. These advancements are essential for improving treatment decisions and efficiently allocating medical resources in emergency care, which can help solve ongoing healthcare challenges.