Fast decliner phenotype of chronic obstructive pulmonary disease (COPD): applying machine learning for predicting...

Area of Science:

• Pulmonology and Respiratory Medicine
• Medical Informatics
• Data Science in Healthcare

Background:

• Chronic obstructive pulmonary disease (COPD) presents diagnostic and therapeutic challenges due to its heterogeneous nature.
• Identifying patient phenotypes, particularly those experiencing rapid lung function decline, is crucial for timely interventions and improved management.
• This study focuses on characterizing the 'fast decliner' phenotype and assessing its reproducibility and predictive value for lung function loss post-diagnosis.

Purpose of the Study:

• To identify distinct patient phenotypes in chronic obstructive pulmonary disease (COPD) using machine learning.
• To characterize the 'fast decliner' phenotype and evaluate its reproducibility after COPD diagnosis.
• To identify key risk factors influencing lung function decline within the identified phenotypes.

Main Methods:

• A prospective, 4-year observational study involving 13,260 patients from the UK Royal College of General Practitioners and Surveillance Centre database.
• Machine learning algorithms were employed to identify COPD phenotypes in a pre-diagnosis training dataset.
• Phenotype reproducibility was validated using a separate dataset post-COPD diagnosis.

Main Results:

• Three distinct COPD phenotypes were identified, including the 'fast decliner' phenotype (younger age, fewer exacerbations initially, but rapid lung function decline).
• Other identified phenotypes included patients with highest COPD severity and older male patients with significant comorbidities.
• Phenotype classification demonstrated 80% accuracy in the validation dataset, with gender, COPD severity, and exacerbations identified as key predictors of decline in the 'fast decliner' group.

Conclusions:

• Three COPD patient phenotypes were successfully identified prior to clinical diagnosis.
• The reproducibility of these phenotypes in a post-diagnosis validation dataset suggests their generalizability across diverse patient populations.
• These findings support the potential for machine learning-driven phenotyping in early COPD management and intervention.