MEWS++: Enhancing the Prediction of Clinical Deterioration in Admitted Patients through a Machine Learning Model.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Prediction Models

Background:

• Early detection of patient deterioration is critical for timely intervention.
• Traditional early warning scores have limitations in predicting the timing of decline.
• Machine learning offers potential for more accurate and timely risk identification.

Purpose of the Study:

• To develop and evaluate a machine learning model (MEWS++) for predicting patient escalation of care or death 6 hours prior.
• To compare the performance of MEWS++ against the traditional Modified Early Warning Score (MEWS).

Main Methods:

• Retrospective cohort study of adult inpatients (July 2011-July 2017).
• Trained and tested three machine learning models: random forest (RF), linear support vector machine, and logistic regression.
• Compared model performance using sensitivity, specificity, AUC-ROC, and AUC-PR against MEWS.

Main Results:

• The RF model (MEWS++) demonstrated superior performance with 81.6% sensitivity and 75.5% specificity (AUC-ROC 0.85).
• MEWS++ significantly outperformed traditional MEWS, increasing sensitivity by 37% and AUC-ROC by 14%.
• The model successfully predicted escalation of care or death up to 6 hours before the event.

Conclusions:

• Machine learning models, like MEWS++, can accurately predict clinical deterioration using readily available data.
• MEWS++ provides a valuable tool for early risk identification, facilitating timely clinical decisions.
• This predictive capability can significantly improve patient management and outcomes in hospital settings.