Asthma clustering methods: a literature-informed application to the children's health study data

Area of Science:

• Pulmonary Medicine
• Biostatistics
• Epidemiology

Background:

• Asthma heterogeneity necessitates identification of distinct clinical phenotypes.
• Lack of standardized data analysis approaches hinders reproducibility in asthma clustering studies.
• Phenotyping is crucial for understanding asthma and developing targeted treatments.

Purpose of the Study:

• To identify common and effective data analysis practices in asthma clustering literature.
• To apply identified practices to a Southern California cohort of schoolchildren with asthma.
• To evaluate the clinical relevance and predictive value of identified asthma phenotypes.

Main Methods:

• Systematic review of 77 asthma clustering studies.
• Application of hierarchical clustering, k-medoids, and latent class analysis (LCA) to 598 schoolchildren.
• Utilized 12 input variables including spirometry and exhaled nitric oxide.

Main Results:

• Latent class analysis identified asthma clusters characterized by exhaled nitric oxide and spirometry measures (FEV1/FVC, FEV1% predicted).
• These LCA-derived clusters were predictive of asthma control at two-year follow-up.
• Clusters from hierarchical clustering and k-medoids were less clinically meaningful, primarily differing by sex and race/ethnicity.

Conclusions:

• Common data clustering approaches were identified in the asthma phenotyping literature.
• Latent class analysis, using exhaled nitric oxide and spirometry, revealed clinically relevant asthma phenotypes in the Children's Health Study cohort.
• The identified LCA clusters demonstrated predictive value for long-term asthma control.