Identifying clinically important COPD sub-types using data-driven approaches in primary care population based...

Area of Science:

• Pulmonary Medicine
• Data Science
• Health Informatics

Background:

• Chronic Obstructive Pulmonary Disease (COPD) is a complex, heterogeneous condition with diverse etiological and prognostic profiles.
• Current classification systems inadequately capture the full spectrum of COPD heterogeneity.
• Subtyping COPD is crucial for understanding disease variations and improving patient outcomes.

Purpose of the Study:

• To discover, describe, and validate distinct subtypes of COPD.
• To utilize cluster analysis on electronic health record data for COPD subtyping.
• To investigate the clinical relevance of identified COPD subtypes.

Main Methods:

• Applied unsupervised learning algorithms (k-means, hierarchical clustering) to a large cohort of COPD patients (30,961).
• Utilized 15 clinical features from linked national electronic health records in England.
• Performed dimensionality reduction and validated cluster associations with exacerbations and mortality.

Main Results:

• Identified five distinct COPD patient clusters characterized by unique demographics, comorbidities, and risk profiles.
• Clusters were associated with anxiety/depression, severe airflow obstruction/frailty, cardiovascular disease/diabetes, obesity/atopy, and a low-comorbidity group.
• Demonstrated the utility of a decision tree classifier for assigning new patients to identified clusters.

Conclusions:

• COPD patients can be effectively sub-classified based on primary care record data.
• The identified clusters highlight differing risk factors, comorbidities, and prognoses.
• Anxiety and depression are significant drivers in specific COPD patient subgroups, particularly young females.