Forecasting medical state transition using machine learning methods.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Cardiovascular Monitoring

Background:

• Early detection of circulatory failure is crucial for patient outcomes and reducing healthcare burden.
• Current methods may lack precision in predicting critical state transitions.
• Improving pre-warning abilities can mitigate medical fatigue and enhance patient care.

Purpose of the Study:

• To propose and evaluate a novel transformed state for early circulatory failure detection.
• To compare the performance of the transformed state against the original state using various machine learning models.
• To identify key demographic factors influencing model performance in circulatory failure prediction.

Main Methods:

• A transformed response variable was developed to represent state changes.
• The proposed method was compared with the original 0-1 state using logistic regression, AdaBoost, and XGBoost models.
• Performance was evaluated using metrics such as Area Under the Curve (AUC), F1-score, and Sensitivity.
• Subgroup analysis was conducted based on sex, age, weight, and height.

Main Results:

• The XGBoost model demonstrated superior performance, achieving AUC, F1, and Sensitivity values around 0.93, 0.91, and 0.90, respectively, at prediction gaps of 5, 10, and 20.
• The transformed response variable significantly improved model performance by 1-4% compared to the original variable under XGBoost (p < 0.01).
• Sex and age were found to have a more significant impact on model performance than weight and height, particularly at larger prediction gaps.

Conclusions:

• The transformed state approach, especially with the XGBoost model, offers enhanced accuracy for early circulatory failure detection.
• The findings highlight the importance of considering patient demographics, specifically sex and age, for optimizing predictive models.
• This research provides a valuable tool for improving pre-warning systems and clinical decision-making in critical care settings.