Prediction of Respiratory Decompensation in Patients Receiving Home Mechanical Ventilation: Machine Learning Model...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Chronic respiratory diseases necessitate long-term home mechanical ventilation (HMV).
• Telemonitoring of HMV patients offers real-time data but may miss early decompensation signs.
• Unnoticed deterioration leads to emergency visits and hospitalizations, increasing healthcare burdens.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting respiratory decompensation events in HMV patients.
• Utilize telemonitoring data to identify early signs of acute deterioration.
• Improve patient outcomes by enabling timely medical interventions.

Main Methods:

• Retrospective analysis of data from 482 HMV patients across three telemonitoring platforms.
• Data included device usage, compliance, mask leakage, and ventilator settings over a 5-week window prior to events.
• Trained and evaluated multiple machine learning models, prioritizing recall, using 10-fold cross-validation and SHAP analysis for interpretability.

Main Results:

• Logistic regression achieved high recall (0.94) but moderate accuracy (0.60).
• Random forest classifier offered the best balance: accuracy (0.66), recall (0.78), and F1-score (0.70).
• SHAP analysis identified increased device use, leakage, and compliance as key predictors of decompensation.

Conclusions:

• Predicting respiratory decompensation using HMV telemonitoring data is feasible.
• Tree-based models like random forests show promise for balanced performance and clinical insights.
• Further research with larger, multicenter datasets and additional physiological data is needed to enhance model robustness.