Phenotypic Clusters Predict Outcomes in a Longitudinal Interstitial Lung Disease Cohort

Area of Science:

• Pulmonary Medicine
• Data Science
• Clinical Phenotyping

Background:

• Current interstitial lung disease (ILD) classification systems present challenges due to overlapping clinical features and outcomes.
• Cluster analysis is a powerful statistical method for identifying distinct patient subgroups in heterogeneous diseases.
• The application of cluster analysis to ILD phenotyping remains underexplored.

Purpose of the Study:

• To apply cluster analysis to a longitudinal ILD cohort to identify novel clinical phenotypes.
• To compare the predictive accuracy of these novel phenotypes against current ILD classification criteria for clinical outcomes.

Main Methods:

• Utilized cluster analysis on baseline data from 770 subjects in a longitudinal ILD cohort.
• Identified four distinct phenotypic clusters based on demographic, clinical, and physiological variables.
• Stratified outcomes by identified phenotypic clusters and compared them with subgroups defined by current American Thoracic Society/European Respiratory Society ILD criteria.

Main Results:

• Four distinct ILD phenotypes were identified: Cluster 1 (younger, white, obese females with high FVC/Dlco), Cluster 2 (younger African-American females with low FVC), Cluster 3 (elderly white male smokers with emphysema), and Cluster 4 (elderly white male smokers with honeycombing).
• Phenotypic cluster stratification revealed significant differences in monthly FVC decline (-0.30% in Cluster 4 vs. 0.01% in Cluster 2).
• Cluster-based stratification independently predicted progression-free and transplant-free survival.

Conclusions:

• Cluster analysis successfully identified four distinct clinical phenotypes in a diverse ILD cohort.
• These novel phenotypes demonstrate superior predictive value for meaningful clinical outcomes compared to existing ILD diagnostic criteria.