Abstract
Introduction: Clustering approaches to asthma phenotyping using clinical parameters have facilitated the identification of various asthma phenotypes. However, several challenges remain unresolved: 1) The characterizing parameters are only assessed at a single point in time, yielding only “snapshot” phenotypes. 2) Some of the biomarkers utilized in previous studies can be difficult to obtain in a clinical setting. 3) Available clinical and biomarker data are often incomplete.
Objective: To develop a clustering approach to asthma phenotyping solely based on low-cost and readily obtainable biomarkers measured during a pre-established time window that systematically copes with incomplete data sets.
Methods: The EFRAIM/PASTURE study is a cohort of asthmatic and healthy children. At age 6, N=604 children were subject to twice daily PEF and FEV1 measurements during 4 weeks. We compared the individual PEF and FEV1 distributions among all children using a metric assessing differences in the mean value, and in the magnitude and frequency of fluctuations around the mean. Using hierarchical clustering we identified clusters of children sharing similar distributions. We assessed the stability of the identified clusters upon random data removal.
Results: Tolerable levels of missing data were found via cluster stability analysis. We clinically validated the method within the cohort by means of enrichment analysis of clinical characteristics. We found a good agreement between the most relevant features of some of the clusters identified and well-known clinical characteristics of asthma.
Conclusion: Our method may be useful in the context of personalized asthma care.
- Copyright ©ERS 2015