Using machine learning to predict deterioration of symptoms in COPD patients within a telemonitoring program

Area of Science:

• Pulmonary Medicine
• Biomedical Informatics
• Machine Learning in Healthcare

Background:

• Chronic Obstructive Pulmonary Disease (COPD) exacerbations significantly impact patient health and healthcare systems.
• Predictive models for COPD exacerbations can enable proactive clinical interventions.
• The telEPOC telehealthcare program offers a valuable dataset for developing such predictive models.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting the probability of COPD exacerbations within the next 3 days.
• To leverage historical data from the telEPOC program for model training and validation.
• To identify key patient variables that are most indicative of impending exacerbations.

Main Methods:

• Data preprocessing, cleaning, and harmonization from the telEPOC program.
• Temporal splitting of data into training, selection, and evaluation subsets.
• Training and comparison of gradient tree boosting (CatBoost) and neural network models.

Main Results:

• The CatBoost algorithm demonstrated superior performance, achieving an Area Under the ROC Curve (AUC) of 0.91 and an Area Under the Precision-Recall Curve (AUPRC) of 0.53.
• Breathing rate, heart rate, and SpO2 were identified as the most significant predictors of exacerbations.
• The model operates effectively in a 50% recall and 50% precision setting, balancing sensitivity and specificity.

Conclusions:

• A machine learning model utilizing CatBoost can effectively predict COPD exacerbations with high accuracy.
• Vital signs like breathing rate, heart rate, and SpO2 are crucial indicators for early exacerbation detection.
• The developed model shows potential for practical application in daily clinical practice for proactive COPD management.