Asthma in children is a complicated and heterogeneous disorder with distinct phenotypes. Using an unsupervised cluster analysis in children with a wide range of asthma severity characterized in the SARP network, we have identified four clusters of childhood asthma with shared phenotypic features. Similar to the previous SARP report that described increased allergic sensitization in clusters of adults with early-onset asthma (21
), clusters of childhood asthma were all atopic, although the magnitude of allergic sensitization differed between groups. Asthma duration, the number of asthma controller medications and baseline lung function were also major determinants of asthma phenotype in this cluster analysis. While children with ATS-defined severe asthma were present in all clusters, and no single cluster corresponded well to the definitions of asthma severity proposed in published guidelines (3
). This is likely due to overly stringent lung function requirements (i.e., FEV1
< 60%) for childhood severe asthma (12
), which were extrapolated from adult reference norms (3
). These findings highlight the complexity and unique differences of childhood asthma and emphasize the need for unbiased approaches to refine current guidelines for asthma diagnosis and treatment in children.
In a previous cluster analysis of adults enrolled in SARP, Moore et al. (19
) observed five distinct clusters of asthma that differed primarily in the age of asthma onset, allergic sensitization, baseline lung function, bronchodilator reversibility, medication usage, and healthcare utilization. Two of these clusters were associated with early-onset atopic asthma and normal or relatively mild airflow obstruction, while two others were associated with airflow obstruction that displayed different degrees of bronchodilator reversibility (19
). Using a similar characterization method, we have identified four similar clusters of asthma in children, although the degree of lung function impairment was significantly less. Whereas baseline FEV1
percent predicted values were 75–84% in Clusters 3 and 4, clusters of adults with early-onset atopic asthma had baseline FEV1
percent predicted values of 43–57% (19
). Similarly, the magnitude of FEV1 bronchodilator administration was significantly greater in children and suggests that “fixed” airflow limitation is not a distinguishing feature of severe asthma in this age group. Interestingly, children in Clusters 3 and 4 did have evidence of hyperinflation (air trapping) both at baseline and after bronchodilator administration, but to a much lesser extent than what has been previously reported in adults (19
While the stability of airflow obstruction and hyperinflation in childhood asthma is not entirely clear, there is increasing evidence that an important sub-group of children with persistent wheezing and asthma symptoms acquires significant baseline airflow limitation by the early adult years (29
). In the Melborne birth cohort study (32
), children with severe asthma at 10 years of age had the lowest FEV1
/forced vital capacity ratios throughout the first 42 years of life (32
). Thus the magnitude of airflow limitation in childhood asthma may represent an important marker of progressive asthma that worsens and results in more severe disease in adults over time. Even in children with mild-to-moderate asthma, approximately 30% have declines in the post-bronchodilator FEV1
percent predicted value of more than 1% per year regardless of treatment with ICS (33
). This observation may be related to impaired lung growth (34
), which could result in accelerated lung function decline in the adult years. Further study is needed to understand how lung function changes and evolves in these clusters with age.
Unlike previous cluster analyses of asthma in adults (18
), healthcare utilization was not a robust discriminator of cluster assignment in children. Although children in Cluster 4 had the highest degree of healthcare utilization, the majority of children in each cluster had physician contact for an asthma exacerbation within the previous year. While this observation may be an artifact of the study sample since children in SARP were recruited from academic medical centers, this finding is also consistent with the episodic nature of childhood asthma. Indeed, there is an important distinction between the severity of exacerbations and overall asthma control (10
). Whereas asthma severity refers to the required level of therapy during active treatment of asthma symptoms (i.e., the magnitude of disease activity), asthma control refers to the extent to which asthma symptoms are alleviated by treatment (36
). Although asthma control often predicts the risk of future exacerbations (37
), children can have severe exacerbations despite limited symptoms and normal lung function prior to the event (38
). These children are difficult to evaluate, because many are not symptomatic between exacerbations and medications may be discontinued. Future revision of definitions of asthma severity may need to take this observation into account, since the intensity of treatment in these children may not be the best indicator of impairment and future risk.
An important strength of this study is that cluster analysis, by definition, is unsupervised and thus the identified clusters conform to shared phenotypic features and not a priori
severity assignments. This study nonetheless does have limitations. First, it is unclear whether children enrolled in SARP differ systematically from children who refused participation. Although selection bias is a concern in all observational studies, this bias may influence the conclusions drawn and the generalization of our results, particularly since the SARP sample was enriched for children with difficult asthma who are evaluated at academic medical centers. However, the clinical characteristics associated with asthma severity in this sample, including lung function measures, markers of allergic sensitization and exhaled nitric oxide values, are similar to what has been previously reported in other samples of children with severe asthma (5
). Regardless, our sample may not accurately identify different phenotypes of milder asthma severity that are likely encountered in clinical practice. Thus expansion of our study to children with more mild intermittent forms of asthma would likely have resulted in additional subjects and therefore sub-clustering within Clusters 1 and 2. Second, while enrollment of additional non-Hispanic white subjects would have led to a more geographically representative sample, the disproportionate grouping of blacks in Clusters 3 and 4 likely reflects important ethnic differences in asthma phenotypes. Because healthcare utilization was highly prevalent in each cluster, the disproportionate racial distributions are not solely attributable to healthcare access. Indeed, other genetic-based studies have shown that black subjects with asthma have the earliest age of asthma onset, the strongest family history of asthma and the lowest baseline FEV1
percent predicted values compared to white and Hispanic subjects (39
). Third, it is also important to note that the results obtained from cluster analysis may be dependent on the cluster technique used. Because a cluster analysis will always find patterns in data, regardless of the organization of the dataset, there is not a single “best” method for performing the analysis. Thus the inclusion of more children would likely have resulted in further sub-clustering within our four identified clusters. For this reason, these results must be interpreted within the larger clinical context. While all children in this study were stable at the time of assessment, the stability of these clusters over time and in response to different or novel asthma interventions (including pharmacologic therapies) is unknown. Thus the predictive aspects of these clusters are also unclear and will require validation in future longitudinal studies of childhood asthma. A separate validation in a different and perhaps larger sample of children with severe asthma would also be useful to better understand the heterogeneity of the disorder.
In conclusion, we have identified four clusters of childhood asthma in the NIH/NHLBI SARP. Foremost, these data emphasize that asthma, particularly severe asthma, is a highly heterogeneous disorder. Importantly, no identified cluster corresponded in entirety to definitions of severe asthma proposed by national and international guidelines or the ATSWhile this may reflect our variable selection, the consensus-based definitions of severe asthma may also require further validation in children. Whereas the GINA and NAEPP criteria for severe asthma are based primarily on symptoms and lung function, our pediatric asthma clusters were determined as much by the magnitude of atopy and duration of asthma as by airflow limitation and hyperinflation. Exhaled nitric oxide concentrations and the age of asthma symptom onset were also differentiating features of the clusters, while healthcare utilization was a lesser determinant. These data highlight the complexity and heterogeneity of childhood asthma and support the need for additional studies, including validation of these clusters in other samples of children with severe asthma. If these clusters are indeed clinically meaningful, then cluster analysis and other unsupervised approaches may ultimately assist with the refinement of current guidelines for asthma diagnosis and treatment in children.