Predicting intermediate phenotypes in asthma using bronchoalveolar lavage-derived cytokines

Area of Science:

• Pulmonary Medicine
• Computational Biology
• Immunology

Background:

• Personalized medicine requires identifying disease subtypes.
• Bronchoalveolar lavage (BAL) cytokine patterns correlate with lung responsiveness.
• Asthma presents with diverse clinical and biological characteristics.

Purpose of the Study:

• To identify quantitative intermediate phenotypes of asthma.
• To evaluate machine learning methods for predicting these phenotypes using BAL cytokine data.

Main Methods:

• Analysis of physiological data from 1,048 subjects in the US Severe Asthma Research Program (SARP).
• Identification of four distinct intermediate phenotypes: eosinophilic/neutrophilic inflammation, bronchodilation, and methacholine sensitivity.
• Application of four statistical learning methods (logistic regression, MARS, etc.) to predict phenotypes using BAL cytokine measurements in a 76-subject subset.

Main Results:

• Logistic regression and Multivariate Adaptive Regression Splines (MARS) demonstrated the highest accuracy in predicting intermediate asthma phenotypes.
• Model performance was assessed using area under the ROC curve and overall classification accuracy.
• Distinct patient subgroups were identified based on inflammation profiles and treatment responses.

Conclusions:

• Machine learning models, particularly logistic regression and MARS, can accurately predict intermediate asthma phenotypes from BAL cytokine profiles.
• These predictive models offer a valuable tool for future translational research in asthma.
• Quantitative proteomics and computational methods advance the development of personalized asthma medicine.