Exacerbation predictive modelling using real-world data from the myCOPD app.

Area of Science:

• Digital health
• Machine learning in medicine
• Respiratory disease prediction

Background:

• Acute exacerbations of COPD (AECOPD) are critical events leading to hospitalization, disease progression, and mortality.
• Early detection and intervention are crucial for improving outcomes in COPD patients.
• Digital tools offer a promising avenue for capturing real-world patient data to monitor COPD.

Purpose of the Study:

• To develop and validate a machine learning model for predicting AECOPD 1-8 days in advance.
• To utilize routine patient-entered data from the myCOPD self-management app for predictive modeling.

Main Methods:

• Adaptations of the AdaBoost algorithm were employed for machine learning.
• A dataset of 506 patients and 55,066 stable COPD records, alongside 1263 AECOPD records, was used (2017-2021).
• Training data included COPD Assessment Test (CAT) scores, symptom scores, smoking history, and prior exacerbation frequency.

Main Results:

• The EasyEnsemble Classifier achieved 67.0% sensitivity and 65% specificity (PPV: 5.0%, NPV: 98.9%).
• An AdaBoost model with a cost-sensitive decision tree yielded 35.0% sensitivity and 89.0% specificity (PPV: 7.08%, NPV: 98.3%).

Conclusions:

• Machine learning applied to real-world digital therapeutic data shows potential for predicting AECOPD.
• The models can effectively rule out the risk of COPD exacerbations within an 8-day window.