Multiview Cluster Analysis Identifies Variable Corticosteroid Response Phenotypes in Severe Asthma.

Area of Science:

• Pulmonary Medicine
• Computational Biology
• Immunology

Background:

• Corticosteroids (CSs) are effective asthma treatments, but patient responses vary significantly.
• Systemic CS use carries risks of long-term side effects, necessitating precision management.
• Previous research identified factors in CS responsiveness, but lacked an integrated approach to define differential responses.

Purpose of the Study:

• To identify distinct asthma subphenotypes with differential responses to corticosteroid treatment.
• To utilize an unsupervised multiview learning approach for subphenotype identification.
• To enhance precision medicine strategies in asthma care.

Main Methods:

• Applied multiple-kernel k-means clustering to 346 adult asthma participants.
• Utilized 100 clinical, physiological, inflammatory, and demographic variables, including paired sputum data before and after triamcinolone administration.
• Employed machine learning to identify key baseline variables for predicting cluster assignment.

Main Results:

• Identified four distinct asthma clusters with varying corticosteroid responses.
• Cluster 3 showed late-onset asthma, low lung function, high eosinophilia, and greatest CS responsiveness.
• Cluster 4 comprised young, obese females with severe airflow limitation and least CS responsiveness.

Conclusions:

• Machine learning approaches revealed novel insights into corticosteroid responsiveness mechanisms in asthma.
• The identified subphenotypes offer potential for improved, personalized asthma treatment strategies.
• This study highlights the utility of advanced computational methods in understanding complex disease heterogeneity.