Published ahead of print on March 20, 2008, doi:10.1164/rccm.200709-1419OC
© 2008 American Thoracic Society doi: 10.1164/rccm.200709-1419OC
Dimensions of Respiratory Symptoms in Preschool ChildrenPopulation-based Birth Cohort Study1 Respiratory Research Group, 2 Divisions of Psychiatry and 3 Biostatistics, Health Methodology Research Group, University of Manchester, Manchester, United Kingdom Correspondence and requests for reprints should be addressed to Dr. Jaclyn A. Smith, M.D., Ph.D., University of Manchester, ERC Building, Second Floor, Wythenshawe Hospital, Manchester M23 9LT, UK. E-mail: jacky.smith{at}manchester.ac.uk
Rationale: A focus on distinctive collections of symptoms may be more informative of the probability of respiratory disease than individual and possibly transient phenotypes. However, such collections or components of overall symptomatology need to be valid, and their relationship established with the known risk factors and physiologic measures. Objectives: To analyze detailed parentally reported respiratory symptoms by principal components analysis and derive symptom components; to examine the relationship of such components with measures of lung physiology and atopy. Methods: An unselected, population-based birth cohort (n = 946). Measurements and Main Results: Interviewer-administered questionnaires, lung function (specific airway resistance [sRaw]), airway reactivity (dry air challenge), and atopic status were obtained at ages 3 and 5 years; principal components analysis and multivariate analysis of variance were used to analyze the data. The four-component solution (wheeze, cough, colds, chronic symptoms) explained 53.2% of the variance in symptoms at age 3, and the five-component solution (wheeze, wheeze with irritants, wheeze with allergens, cough, chest congestion) explained 49.8% of variance at age 5. The multivariate analysis revealed novel relationships between symptoms, risk factors for asthma, and measures of lung function. At age 3, sRaw and the interaction between maternal asthma and child's atopy were not only related to wheeze but also independently to the cough component. At age 5, overall wheeze and allergic wheeze were related to lung function and airway reactivity; child's atopy was only related to symptoms when considered as a continuous trait. Conclusions: Our analysis supports the need to move beyond the presence or absence of individual symptoms. Syndromes of coexisting symptoms more likely reflect underlying pathophysiologic processes. Clinical trial registered with http://www.controlled-trials.com (ISRCTN72673620).
Key Words: principal components analysis • wheeze phenotypes • childhood asthma • cough
Childhood asthma is a clinical syndrome in which the diagnosis is often based on a history of wheezing and/or other associated respiratory symptoms (e.g., cough), coupled with the objective measurements of lung function and/or airway inflammation (1). In contrast, in epidemiologic studies, children are often assigned a phenotype based on the response to a single question (e.g., presence/absence of wheeze or doctor-diagnosed asthma), despite the fact that more detailed information has been collected using extensive questionnaires (2). Reshaping the wealth of epidemiologic data into a valid but comprehensible form to inform the clinical decision-making process would be of great value. One means to do this, which has been used for over 20 years in other branches of medicine and psychology, is principal components analysis (PCA) (3). PCA is a quantitative method for data reduction to extract simplified patterns from correlated multi-item data, and has relatively established criteria for deciding what may be common underlying components that best explain a dataset. It distills the mass of information from many questionnaire items into a small number of components, which explain the overall variance in item scores, and can generate coherent subsets of items. These components should be more reliable and valid than single items. Subsequent analysis may also reveal relationships between components and objective measures of disease. On the few occasions that PCA has been used in asthma and allergy research, symptoms formed a single component separate from lung physiology (4), atopy (5), systemic/airway inflammation (6), and quality of life (7), providing evidence of a disjunction between etiologic factors, symptoms, and consequences of symptoms. The fact that most cases of persistent wheezing and asthma begin in early childhood provides an added level of complexity. Our understanding of the nature of childhood wheezing illness has been augmented by the characterization of distinct wheeze phenotypes, based on the temporal patterns of wheezing (early, late onset, persistent) (8). Given this fragmentation of the trajectory and pathophysiology of asthma symptoms, it is not surprising that there have been calls to abandon the concept of asthma as a single disease (9). We hypothesized that, rather than relying on the presence or absence of an individual symptom (e.g., wheezing) or physician diagnosis (e.g., asthma), more useful information may be obtained by examining syndromes of coexisting symptoms derived from the collected data. Such syndromes may better reflect the underlying pathophysiologic processes and predict not only the presence but also the persistence of chronic respiratory disease, and hence be useful in both diagnosis and treatment decision making. Therefore, in a setting of a population-based birth cohort, we used PCA to analyze detailed parentally reported respiratory symptoms at ages 3 and 5 years and derived representative symptom components. We tested the validity of these components by examining their relations to measures of lung physiology and child's atopy at each age, and between components from one time point to the other. These data have been presented in abstract form (10).
Design The Manchester Asthma and Allergy Study is an unselected, population-based birth cohort (11). The study was approved by the South Manchester Local Research Ethics Committee, Manchester, UK. Written, informed consent was obtained from subjects' parents/guardians. We recruited participants during the first trimester of the pregnancy; children were monitored prospectively and reviewed at age 3 and 5 years. For detailed description of the cohort, procedures, and analysis, see the online supplement.
Procedures
Lung function.
Atopy.
Statistical Analysis We used the same procedures to analyze the symptom items at ages 3 and 5 years. The steps involved in the analyses were as follows:
Of 1,085 children born into the study, we reviewed 1,068 at age 3 years. Of these, 122 were prenatally randomized to an environmental control regime (11, 19, 20) and excluded from this analysis. Of the remaining 946 children, 904 attended the follow-up at age 5 years. Lung function data were available in 533 children at age 3 and 782 at age 5 years; 811 were skin tested at age 3, 798 at age 5, and 521 had IgE measured at age 5 years (Figure 1). Of the subjects, 20.2 and 27.1% were sensitized to at least one allergen at ages 3 and 5 years, respectively.
Component Generation Age 3 years. Twenty-one items were included in the PCA (Kaiser-Meyer-Olkin measure of sampling adequacy, 0.92; Bartlett's test of sphericity, <0.001); item selection is summarized in Table E1 of the online supplement. Four components with eigenvalues greater than 1 were identified with a subsequent break in the scree plot (Figure E1a). The four-component solution explained 53.2% of the variance in symptoms. The questionnaire items loading greater than 0.4 are shown in Table 1, together with the proportion of variance explained by each component.
Age 5 years. The age 5 questionnaires contained a section on allergic and irritant triggers of wheeze, which was not included at age 3; 32 items were included (Kaiser-Meyer-Olkin measure of sampling adequacy, 0.89, Bartlett's test of sphericity, <0.001), and item selection is summarized in Table E2. Six eigenvalues greater than 1 were found. These components explained 54.5% of the variance in respiratory symptoms at this age. Because there was a break in the scree plot between five and six components (Figure E1b), we also examined the five-component solution, which explained 49.8% of the variance (eigenvalues > 1.35). The questionnaire items loading onto each component and the variance they explain are shown in Table 2. We selected the five-component solution, because the six-component solution split the chest congestion component into two, describing congestion with colds and chronic congestion. These two components were not significantly discriminated in the subsequent MANOVA, and appear to be less valid.
Different rotation techniques were examined (oblique promax and orthogonal varimax) but produced extremely similar results, demonstrating the stability of the components.
Correlations between components. Correlations between the component scores at different ages provide a further indicator of validity, suggesting stability over time.
Symptom Components, Lung Physiology, and Atopy
Age 5. In a model that used atopic sensitization as a dichotomous variable, several physiologic variables were significantly related to symptom components: baseline lung function (P  0.001), airway reactivity (P = 0.016), the interaction between maternal asthma and child's baseline lung function (P = 0.007), and maternal asthma and child's airway reactivity (P = 0.047). Four of five symptom components were significant in this multivariate model (wheeze, P = 0.001; cough, P < 0.001; wheeze with allergens, P = 0.002; chest congestion, P = 0.001), but not wheeze with irritants (P = 0.29). However, atopy (either on skin tests or IgE) was not a significant predictor of any of the symptom components. In contrast, in a model using atopy as a continuous variable (i.e., the absolute level of specific IgE), the nature of these relationships changed. In this model, a sum of specific IgE (P < 0.001), baseline lung function (P = 0.001), the interaction between maternal asthma and sum of specific IgE (P = 0.003), and the interaction between maternal asthma and child's airway reactivity (P = 0.04) were significantly associated with the components. Again, four of five symptom components were significant in the multivariate model (wheeze, P < 0.001; cough, P < 0.001; wheeze with allergens, P < 0.001; chest congestion, P = 0.001), but not wheeze with irritants (P = 0.29). Table 4 summarizes the relationships between the significant predictors and each individual symptom component (post hoc analyses from each MANOVA).
Key Results We have demonstrated that PCA can be successfully used to produce respiratory symptom components in preschool-age children from validated questionnaires administered in a population-based birth cohort. As expected, questions pertaining to wheeze comprised a component that explained the largest proportion of the variance at both age 3 and 5 years. At age 3, other components included questions pertaining to cough, colds, and chronic symptoms. At age 5 years, in addition to the existing wheeze component, two further wheeze-related components emerged (wheeze with irritants and wheeze with allergens); these questions had not formed part of the assessment at age 3. The components we identified have face validity (as they seem intuitively correct) and content validity (as we observed significant relationships with objective measures of respiratory function and known risk factors for asthma, e.g., maternal asthma). In addition, we observed a degree of consistency over time, suggesting their stability. Viewing atopic sensitization as a dichotomous trait in its relationship to respiratory symptom components obscured the nature of their relationships, which only became apparent when atopy was considered as a continuous trait.
Limitations The different questionnaire items used at two time points may account for the differences in the symptom components identified. The majority of the additional questions at age 5 described wheezing episodes, and the circumstances in which they occurred. By age 5 years, children are more likely to have been exposed to a wider range of environmental factors that may precipitate wheezing. Moreover, the additional wheeze components identified (wheeze with allergens and wheeze with irritants) were extremely poorly correlated with the wheeze component at age 3 years, implying that these are truly independent emerging symptom components. PCA is an exploratory analysis and the inherent assumptions and limitations of PCA should be taken into account. Not all questions were included in the PCA; excluding repetitive and correlated questionnaire items presupposes that they will contribute nothing and the subjective selection of which items to remove may affect the results. The choice of rotation may also alter the outcome of the analysis, with some studies preferring orthogonal rotations such as the varimax (6, 21, 22). Orthogonal rotations are useful in that they maximize the differences between the components, but the resulting factors do not correlate, implying etiologies for each component involving distinct nonoverlapping risk factors and no correlation in exposure to them. Oblique rotations are generally considered more natural, because they generate components that may correlate. In the current study, the choice of rotation had no material effect on the results.
Interpretation The symptom components generated by the PCA at both age 3 and 5 years were clear in character. The age 3 components correlated little with each other, apart from the wheeze and cough components. At age 5, the wheeze and cough components again show moderate correlations, but chest congestion seemed to be a separate dimension. Similarly, there was a stability of these main components as the children aged, and the correlations over time were reassuringly specific, despite the differences in the questionnaires used (i.e., the correlations of corresponding symptom components at two time points were consistently stronger that those between the different components at the same age). Overall, the component correlation coefficients were not strong between the two time points studied, most likely due to the transient nature of symptoms in some of the children. These components all remained separate when the varimax rotation of the model was used, confirming they are distinct and probably have different etiologies. This multivariate analysis has revealed novel relationships between symptoms, known risk factors for asthma, and objective measures of lung function and atopy. For example, at age 3 years, sRaw was not only related to wheeze but also independently to the cough component. Moreover, the interaction between maternal asthma and child's atopy was independently related to both cough and wheeze. We have previously demonstrated a significant effect of the interaction between child's atopy and maternal asthma on lung function among asymptomatic children at age 3 years (14), and suggested that these features affect the development of wheeze via their effect on lung function. However, it was apparent in the current analysis that lung function was related to symptom components independent of the interaction between child's atopy and maternal asthma, raising questions about the underlying pathophysiologic mechanisms. Analysis at age 5 years suggests different associations, implying different pathologies for each symptom component. When atopy was considered as a dichotomous variable, the symptom components were not related to child's atopic status. However, using atopy as a continuous trait revealed the associations between symptom components, lung function, airway reactivity, and level of allergen-specific IgE. Our findings suggest that end-organ susceptibility and systemic immune responses contribute independently to distinct symptom components and that their effect differs between components. The additional questions at age 5 years enabled characterization of three separate wheeze components (wheeze, wheeze with allergens, and wheeze with irritants), each with different relationships to physiologic hallmarks of asthma. This raises a fundamental question as to whether all early childhood wheezing should be treated by the same antiasthma drugs (23–25).
Generalizability It would be of value to explore similar symptom components in the other large birth cohort studies. This could be achieved either by each study generating its own "syndromes" from its datasets and comparing those between the studies, or by testing the components described here within the other datasets (provided the questionnaires used are identical). Another approach would be to identify such components in a pooled dataset, allowing a more generalized definition of syndromes, which can be used as endpoints in etiologic analyses or identification of treatment groups. We acknowledge the relative complexity of this approach to the analysis of datasets collected using standardized questionnaires with multiple questions. However, once determined, the components become continuous clinical outcome variables in their own right, which can be used in further analyses. The advantage of this approach is that these components retain the appropriately weighted information collected from numerous questions, while reducing the problem of multiple testing when each individual question is used as a dichotomous variable.
Conclusions
The authors thank the Manchester Asthma and Allergy Study (MAAS) children and their parents for their continued support and enthusiasm. They also acknowledge the dedication of the MAAS study team.
Supported by Asthma UK grant 04/014 and Moulton Charitable Trust. This article has an online supplement, which is accessible from this issue's table of contents at www.atsjournals.org Originally Published in Press as DOI: 10.1164/rccm.200709-1419OC on March 20, 2008 Conflict of Interest Statement: None of the authors has a financial relationship with a commercial entity that has an interest in the subject of this manuscript. Received in original form September 25, 2007; accepted in final form March 19, 2008
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
AT A GLANCE COMMENTARY
