Authors’ Response

Dear Drs Huh, Sullivan, and Huntington,

We are grateful for your thoughtful critiques and welcome the opportunity to clarify our research. Below, we address the concerns raised in turn.

The authors’ first contention is that our study is “not representative of most communities.” The Duke Preschool Anxiety Study recruited its sample from specific pediatric primary care practices in the Durham, North Carolina, area with patient populations that are representative of the surrounding community. Our 2013 article summarizes the sample and community demographics in Table 11 and states that our sample demographics are similar to those of the area but not the nation. Most of the major epidemiologic, longitudinal studies of child mental health of the last 30 years are community studies (eg, Great Smoky Mountain Study, Pittsburgh Youth Study, and Christchurch Health and Development Study). None of these studies are nationally representative, and all have yielded significant insights about child psychopathology.^2,3 We, like the authors, see the need for a nationally representative mental health study, but the absence of such does not negate the value of community studies in child psychopathology.

The authors also argued that our 2-stage sampling design oversampled for children at risk for anxiety disorders and is therefore not appropriate for a study of SE. This is not the case. Such designs are commonly used for epidemiologic samples when the prevalence of disorder is low in the population, and a simple random sampling framework would require a very large number of subjects.^4–6 All participants were given a weight inversely proportional to their probability of selection and all analyses were conducted with robust variance (sandwich type) estimates to adjust the standard errors for the stratified design effects. With such procedures, the sample can be thought of as “random sample” from the community and analyses will generate reasonably generalizable community estimates.¹

The authors also expressed concern about the rates of anxiety disorders identified in this study. Indeed, the rate of preschool anxiety disorders has varied across our studies, as well as across community studies of preschoolers that use the PAPA and other measures.^7–10 A review of these data can be found in our previous work.¹ The degree to which these differences represent unreliability within our measures or design is reflected in our confidence intervals (CIs). As noted in our previous work,¹ the overall prevalence of any anxiety disorder found in our sample (19.4% [CI: 15.3%–23.6%]) is similar to the rate found in Bufferd et al’s⁹ 2011 reported rate of 19.6% (CI: 16.3%–22.9%) in their community study of preschoolers using the PAPA. We are pleased that multiple groups around the world are studying this important area of mental health and expect our understanding to increase over the next decade.

The authors also argue that the definition of a “moderate selective eater” was unclear. To clarify, in the diagnostic interview employed, each individual symptom item has to cross a threshold of impairment across 2 domains to be endorsed; these determinations were made by a trained interviewer. A moderate selective eater only ate within a limited range of foods and the child’s eating caused impairment in 2 domains (eg, child required separately prepared meals and child would not eat food from a restaurant). To state confidently that our moderate SE group was or was not similar to typical picky eating would imply that there is a precisely operationalized definition of picky eating, which to our knowledge is lacking.

Related to that clarification, the authors argued that our recommendation for intervention at moderate levels of SE was not commensurate with our results. Our recommendations for intervention resulted from the impairment associated with SE inherent in our definition. However, this interpretation was also shaped from our clinical experiences: that current strategies for managing SE are inadequate and more precise tools are needed. These points are strongly supported by our data, which demonstrate that the clinical presentation of SE is more complex than previously thought. Previous research has established the association of anxiety with SE; our data highlights the role of food aversion, depressive symptoms, and sensory sensitivity as other crucial elements that need to be addressed in the clinical resources we provide to parents. Thus, we disagree with the authors’ formulation as it just focuses on the role of anxiety.

In closing, we wish to join with the authors in emphasizing the importance of the study of SE and in using well-designed sampling strategies to help characterize the boundary between typical development in eating behaviors and those that warrant clinical attention.