AB073. Machine learning model enhances predictive capability of modified early warning system

Background: Early recognition of impending systemic failure and the provision of appropriate medical care are critical for timely clinical intervention. The modified early warning system (MEWS) is an essential tool for identifying patients at risk of clinical deterioration. However, with the increasing complexity of patient conditions—such as advanced age, multiple chronic diseases, and other risk factors—automated warning systems often generate excessively frequent alerts, resulting in a high rate of false alarms. By applying predictive modeling, critical risk factors can be detected at an early stage, thereby supporting clinical decision-making and improving the overall quality of patient care. The objective of this study is to develop and validate an enhanced predictive model that extends the traditional MEWS framework by incorporating demographic and clinical risk factors to improve early detection of impending systemic failure. We aim to identify critical predictors of patient deterioration at an earlier stage and to improve risk stratification accuracy compared with conventional MEWS-based alerts. By integrating predictive modeling results into clinical decision support, this approach seeks to reduce false alarms, support timely and informed clinical decision-making, and ultimately enhance the quality and safety of patient care.

Methods: Retrospective data of cases in the chest wards of a medical center about 4,108 patients from January to June 2021. Features used the patient’s intervention measures and comorbidity parameters during hospitalization to understand the patient’s physical condition, and used machine learning (ML) to let the computer learn and judge, and then predict the patient’s final condition. Four model evaluation indicators were used, and XGBoost was the best prediction model, so it was adopted as the final model for prediction in this study.

Results: The relationship between all feature values and the predicted target death is moderately correlated with signing do not resuscitate (DNR), the shift time of the nursing station, and the patient’s age. Among the comorbidities, diabetes, hypertension, heart failure, and renal failure have a greater impact. The staff accepts the MEWS easily because of the clear design. The incidence of cardiopulmonary resuscitation (CPR) was reduced from 2.42% to 1.58% after application of the MEWS in 2022. The MEWS helped the improvement of the clinical outcomes in our study.

Conclusions: The establishment of the MEWS system is the basis for improvement. Comorbidities are graded according to disease severity to more accurately predict the impact of comorbidities.

Keywords: Machine learning (ML); modified early warning system (MEWS); clinical decision-making