Challenges in Identifying Asthma Subgroups Using Unsupervised Statistical Learning Techniques

Abstract

Rationale: Unsupervised statistical learning techniques, such as exploratory factor analysis (EFA) and hierarchical clustering (HC), have been used to identify asthma phenotypes, with partly consistent results. Some of the inconsistency is caused by the variable selection and demographic and clinical differences among study populations. Objectives: To investigate the effects of the choice of statistical method and different preparations of data on the clustering results; and to relate these to disease severity. Methods: Several variants of EFA and HC were applied and compared using various sets of variables and different encodings and transformations within a dataset of 383 children with asthma. Variables included lung function, inflammatory and allergy markers, family history, environmental exposures, and medications. Clusters and original variables were related to asthma severity (logistic regression and Bayesian network analysis). Measurements and Main Results: EFA identified five components (eigenvalues > 1) explaining 35% of the overall variance. Variations of the HC (as linkage-distance functions) did not affect the cluster inference; however, using different variable encodings and transformations did. The derived clusters predicted asthma severity less than the original variables. Prognostic factors of severity were medication usage, current symptoms, lung function, paternal asthma, body mass index, and age of asthma onset. Bayesian networks indicated conditional dependence among variables. Conclusions: The use of different unsupervised statistical learning methods and different variable sets and encodings can lead to multiple and inconsistent subgroupings of asthma, not necessarily correlated with severity. The search for asthma phenotypes needs more careful selection of markers, consistent across different study populations, and more cautious interpretation of results from unsupervised learning.

Asthma phenotypes in inner-city children

“…The inclusion of variables measuring longitudinally measured outcomes while receiving standardized, guidelines-based asthma and rhinitis management over a full year is another aspect of our study that addresses limitations of several other HCA. 1,19 Another strength of this work is the use of a tree-based clustering algorithm, which has major advantages over standard clustering algorithms that use Euclidean distance metrics. In particular, this method adapts well to the inclusion of both continuous and categorical variables in the clustering algorithm, is resistant to outliers, is unaffected by linear scaling, and lessens the effect of collinearity among variables.…”

“…32-35 Another potential limitation is that data reduction techniques may contribute to the generation of inconsistent clusters that are more sensitive to relatively conservative shifts in the specific variables selected. 19 However, at this time, there is no standard set of variables that can be universally recommended for inclusion in HCA. Finally, our findings may be specific for asthma phenotypes restricted to low-income urban children and not generalizable to other populations.…”

Systems Biology and Clinical Practice in Respiratory Medicine. The Twain Shall Meet

“…Powerful observerindependent statistical clustering methods have been used to phenotype patients with asthma on the basis of symptoms and biomarkers (81)(82)(83). All of these clustering methods, however, are population based, cross-sectional, and measured at one point in time.…”

Update in Asthma 2013

“…This study adds to our understanding of the durability of asthma phenotypes and the natural history of the poorly understood nonatopic phenotype. Prosperi and colleagues instead focused on methodological pitfalls of cluster approaches (2). They found that changes in variable definition led to multiple and inconsistent subgroupings of asthma and were more influential than the clustering method.…”