Disease Progression Modeling in Chronic Obstructive Pulmonary Disease.

Area of Science:

• Pulmonary Medicine
• Radiology
• Computational Biology

Background:

• Chronic obstructive pulmonary disease (COPD) progression is complex and challenging to track over time.
• Identifying distinct patient subtypes with unique disease trajectories is crucial for effective management.
• Current methods struggle to capture the heterogeneity in COPD progression patterns.

Purpose of the Study:

• To utilize a novel machine learning tool, Subtype and Stage Inference (SuStaIn), to identify distinct patient subtypes in COPD.
• To evaluate SuStaIn's utility in stratifying COPD patients based on their disease progression patterns.
• To analyze longitudinal progression patterns using cross-sectional computed tomography (CT) imaging markers.

Main Methods:

• Applied SuStaIn to CT imaging data from 3,698 COPDGene study participants (GOLD stages 1-4) and 3,479 controls.
• Confirmed identified subtypes and progression patterns using independent ECLIPSE study data.
• Assessed SuStaIn's stratification utility by correlating baseline subtypes/stages with longitudinal follow-up data.

Main Results:

• Identified two COPD progression subtypes: 'Tissue→Airway' (70.4%) and 'Airway→Tissue' (29.6%).
• Baseline stage in both subtypes correlated with future FEV1/FVC decline, indicating predictive value.
• SuStaIn identified 30% of smokers with normal lung function as having early imaging changes consistent with early COPD.

Conclusions:

• Two distinct COPD progression patterns, potentially representing different endotypes, were identified using SuStaIn.
• SuStaIn demonstrates utility for patient stratification and identifying individuals at risk for COPD.
• Detectable imaging changes in healthy smokers suggest a potential new biomarker for early COPD detection.