Author s response to reviews Title: The epidemiologic characteristics of healthcare provider-diagnosed eczema, asthma, allergic rhinitis, and food allergy in children: a retrospective cohort study Authors: David Hill (hilld3@email.chop.edu) Robert Grundmeier (grundmeier@email.chop.edu) Gita Ram (ramg1@email.chop.edu) Jonathan Spergel (spergel@email.chop.edu) Version: 2 Date: 10 May 2016 Author s response to reviews: Comments to the Reviewers: Reviewer 1: 1. The topic may be of interest, even if not original; however, I have some concerns regarding the methods, that can lead to multiple bias. Considering the limitation of the retrospective study, it is very important that an adequate definition of allergic diseases is made to obtain suitable information and subsequent conclusions. We thank the Reviewer for their time and suggestions regarding our manuscript. We agree that accurate identification of subjects is of paramount importance in a retrospective cohort study, and we went to great lengths to ensure accuracy of our identification methods including performing a manual chart review for a sample of children in which we compared diagnosis codes in the EMR to disease definitions derived from consensus guidelines published by the American Academy of Allergy, Asthma, and Immunology. In this review, we found an overall accuracy of our identification methods to be >92% (Supporting Table 4). In our experience, and in comparison to studies utilizing similar methods, this degree of accuracy is very high. We also performed a sensitivity analysis in which we found little effect of alternative disease definitions on our outcomes of interest (Page 6, Lines 127-130; Page 10, Lines 201-207). Never the less, we agree that inherent biases exist in any study as a result of the inclusion and exclusion criteria utilized. We have added this new text to clarify our study population and methods, and the resulting biases that exist in our analysis (Pages 4 & 5, Lines 98-102; Page 6, Lines 130-135 & 138-142; Page 7, Lines 147-152; Pages 16 and 17, Lines 319-340).
2. Food allergy definition: The authors found rates of food allergies to peanut, milk, egg, shellfish and soy higher in their patient population. In the section "Limitations" (Lines 314-315) they state that "Diagnoses corresponding to both IgE and non-ige-mediated food allergies were included in our analyses, and not all food allergy diagnoses were confirmed by food challenge". As not all food allergies were confirmed with food challenge, that is the gold standard for the diagnosis, an overestimated prevalence can occur, as it happens in this study. We agree with the reviewer that this is a limitation of our study, albeit a necessary one. In our practice, we often diagnose children with food allergy based on skin prick testing or laboratory analysis, in the absence of a food challenge. Additionally, while we have a very large allergy practice, we cannot accommodate all of the allergy subspecialty needs of our primary care network, which includes 31 primary care sites, more than 300,000 patients, and spans the geographic region of three states. As such, many patients in our primary care network are evaluated by private allergists and their testing results are not readably accessible through our EMR. To address the Reviewers concerns as to our ability to accurately identify food allergic patients in our cohorts, we performed a chart review of patients in our cohort with food allergy. Of these patients, the diagnosis of food allergy was supported by evaluation of history and testing by a sub-specialist in 88% of patients (Supporting Table 4). We have revised our manuscript to more clearly communicate these findings, and to include the potential for overestimation of food allergy in our study (Page 8, Lines 173-175; Page 17, Lines 331-334). 3. Asthma definition: Line 127-128: "To minimize the likelihood of including non-allergic wheeze in our analysis, we excluded all reactive airway disease diagnosis codes and any asthma diagnoses made before the age of 1 year of life": can the authors exactly explain what diagnosis codes they excluded? The diagnosis codes excluded from our analysis are listed in a new supplemental table included in the revised manuscript (Supporting Table 3) and are clarified in new text (Page 6; Lines 130-133). 4. Moreover, I suppose the authors excluded any asthma diagnosis before the age of 1 year to minimize the risk of post viral wheezing. However, it is well known that post viral wheezing may occur also over the first year of life; conversely an asthma diagnosis can be made in the first years of life in case of intercritical symptoms, family history etc. Then, these exclusion criteria make a bias.
We agree with the Reviewer that ensuring accuracy of asthma diagnoses (in part by excluding post-viral wheeze as a confounding diagnoses) is of paramount importance in our study. We took several steps to ensure accurate asthma diagnosis in our study including 1) excluding asthma diagnosis prior to 1 year of life, 2) requiring two asthma diagnosis codes be present during at least two separate care visits occurring at least six months apart (both of which minimize the likelihood of including post-viral or other causes of transient wheeze), and 3) requiring the prescription of asthma medications to the patient (increasing the likelihood of persistent asthma symptoms). Never the less, the Reviewer brings up the important bias that inclusion and exclusion criteria inherently make in any study. To ensure that readers are aware of such biases in our study, we have provided a considerable degree of transparency as to our methods including listing included and excluded diagnosis and allergy codes used in our analysis (Supplemental Table 1, Supplemental Table 2, Supplemental Table 3) and making our primary data set available in a public repository (Page 7, Lines 165-166). Finally, we have added new text to our discussion to ensure that the potential for bias is clearly stated (Pages 16 and 17, Lines 319-340). 5. Definition of asthma: asthma based on definitions that required either (a) occurrence of at least two visits with an asthma-related diagnosis code (ICD-9 occurring at least six months apart or medication for asthma on at least two separate dates implies a potential overestimation of asthma, as the authors found in their study, as following: considering simply the occurrence of "asthma diagnosis" at least six months apart, not considering symptoms and/or use of medication during the intercritical period implies that may be defined as "asthma" also a post viral wheezing. The identification of asthma in our study required that children 1) were greater than 1 year of age at initial asthma diagnosis, 2) had asthma diagnosis codes during at least two separate care visits occurring at least six months apart AND 3) were prescribed asthma medications. We then performed a chart review (in which we reviewed history and physical exam of the diagnosing provider) and showed that these criteria provided >90% accuracy for identifying patients with asthma. We have clarified our criteria for identifying patients with asthma in our methods (Page 6, Lines 130-133 and Page 7, Lines 147-152). 6. The same authors state that with the methodological limitations of the study "We cannot draw conclusions on changing prevalence from our analysis, and variations in prevalence between our study and other studies could be influenced by methods of data collection".
One cannot draw conclusions on changes in prevalence over time from our data as ours is not a longitudinal study. Comparisons to prior studies of disease prevalence in the United States must take into account differences in methodology. We have modified the above language to more accurately describe this limitation (Pages 16 and 17; Lines 334-338). 7. Maximum age of inclusion in our birth cohort was set at 5 years: the reason of this choice is not clear. The maximum age of inclusion in our birth cohort was set at 5 years to maximize the size of this cohort while capturing the most common age of diagnosis for most conditions studied. We have clarified our discussion section appropriately (Page 16, Lines 326-329). 8. The references referring to food allergy (especially concerning diagnosis) should be updated (eg Sampson HA, Aceves S, Bock SA, et al. Food allergy: a practice parameter update-2014. J Allergy Clin Immunol 2014, Sicherer SH, Sampson H; Food allergy: epidemiology, pathogenesis, diagnosis and treatment. J J Allergy ClinImmunol 2014;133:291-307). We have included the above references in our revised manuscript (Page 6, Lines 140-142). Reviewer 2: 1. This is an interesting use of electronic health records (a slight modification of which the European nordic countries have been doing for several years) to study prevalence and incidence rates of allergic diseases in a defined pediatric population. The methods are correct and the conclusions sound and well derived from results. We thank the Reviewer for their time and suggestions regarding our manuscript. We are pleased that they found our methods to be correct and our conclusions to be sound. 2. For those non-familiar with the EHR in the Philadelphia Children's hospital, it would be of great interest to know what is the coverage of the registry: has every child born in the area an EHR? We utilize EPIC as our EMR at The Childrens Hospital of Philadelphia. Every child in our care network is included in this EMR. Our primary care network includes 31 separate locations and spans the geographic region of three states. However, there are children born in our geographic region who choose to access primary care though private practices, or other hospital networks,
and as such are not captured in our EMR data. Nevertheless, our care network is by far the largest in the region and offers an unprecedented opportunity to study disease patterns across the North-East region of the United States. We have added clarifying text to our methods section to better describe our healthcare network (Pages 4 and 5, Lines 98-102). 3. What is the geographic area covered by the hospital? This, of course, would have implications in order to make comparisons with other studies. The Childrens Hospital of Philadelphia care network includes a 527-bed hospital in central Philadelphia, and more than 50 satellite offices that provide care for more than 1 million patient visits each year. Our hospital is a national and international referral center, however, this study is limited to our primary care network which covers a geographic area of three states: Pennsylvania, New Jersey, and Delaware. Prior studies have shown that our primary care network is representative of our community in general. We have included an expanded description of our care network in our methods section and discussed the difficulty of generalization of our data to non-similar populations (Pages 4 and 5, Lines 98-102; Page 17, Lines 338-340). 4. Please include the total number in the headings of each column in table 1. We have updated this table accordingly (Table 1). 5. It would be more straightforward to the reader if -once defined- the authors refer always to "the birth cohort" and the "the cross sectional cohort" when referring to the two populations of children. An example of this is in the results section of the abstract when authors refer to the birth cohort to "children observed continuously from birth until de 5th birthday". We have made this change throughout the text. Reviewer 3: 1. The strength of the study is that it provides useful data regarding the epidemiology of common diseases and will challenge readers to update their knowledge in this domain. We thank the Reviewer for their time and suggestions regarding our manuscript. We are pleased that they feel that our findings are of use.
2. Allergic rhinitis symptoms arise as a result of inflammation induced by an immunoglobulin E (IgE) mediated immune response to specific allergens. The diagnosis of allergic rhinitis is based upon the coordination between a typical history of allergic symptoms and diagnostic tests (Skin Prick Tests and/or Rast tests) (Bousquet J, et al Allergy 2008: 63 (Suppl 86): 8-160. Therefore, a fair diagnosis requires allergy evaluation by SPTs and/or Rast tests. Were all the children under 5 years of age diagnosed with allergic rhinitis subjected to an evaluation for allergy? Unless they had, their symptoms may have causes other than allergy (vasomotor, infectious, etch.) It is risky to establish the diagnosis based only on clinical criteria in the absence of testing for allergy. Data regarding the epidemiology of allergic rhinitis should be limited to those who were tested for allergy and/or those who responded to antihistamines. The diagnostic criteria on which a diagnosis was established (eg. typical clinical symptomatology in combination with testing for allergy and/or response to antihistamines should be stated in the document). We agree with the Reviewer (and with the similar concerns raised by Reviewer #1) that the gold standard for diagnosis of allergic rhinitis is skin prick testing. However, while we have a very large allergy practice at our hospital, we cannot accommodate all of the allergy subspecialty needs of our primary care network which includes 31 separate locations, more than 300,000 patients, and spans the geographic region of three states. As such, many patients in our primary care network are evaluated by private allergists and their testing results are not readably accessible through our EMR and are therefore not able to be ascertained for this study. Additionally, it is the case that antihistamines are often not covered by insurance providers in the United States because they are available without a prescription. As such, correlation of allergic rhinitis diagnosis codes with antihistamine prescriptions would likely result in an underestimation of disease rates. As noted in our comments to Reviewer #1 previously, we performed a manual chart review for a sample of children in which we compared diagnosis codes in the EMR to disease definitions derived from consensus guidelines published by the American Academy of Allergy, Asthma, and Immunology. In this review, we found 100% accuracy for our ICD code data related to rhinitis (Supplemental Table 4). We have clarified the text of our revised manuscript to reflect these constraints and findings (Page 6, Lines 138-142; Page 8, Lines 173-175; Page 16, Lines 331-335). 3. Asthma and eczema diagnosis can be established based on clinical criteria. Please define those criteria. For example, was an asthma diagnosis established based on episodes of wheeze in the absence of fever, taking also into account the response to bronchodilators and/or glucocorticoids. Was the eczema diagnosis based on the criteria of the American Academy of Pediatrics, the Hanifin & Rajka Criteria, or other? Please specify.
During our manual chart review for a sample of children, diagnoses were compared to practice parameters and consensus guidelines from the Joint Task Force on Practice Parameters (representing the AAAAI, ACAAI, and the JCAAI). We have added this clarification to our methods section (Page 6, Lines 140-142). 4. On page 11-12 (Lines 234-235) does the statement suggest that an overall of 86% of children with food allergy will develop respiratory allergy (asthma and/or allergic rhinitis) later in life? Please clarify. We have significantly expanded our analysis of the extent to which food allergy predisposes to the development of respiratory allergy (see below). In this analysis, asthma and rhinitis were examined as independent outcomes. We have clarified this analysis in our methods and results sections (Page 7, Lines 155-162; Page 12, Lines 236-247). We have also separated these outcomes into two tables and incorporated them into the manuscript to further minimize confusion (Table 4 and Table 5). 5. In the population of children who were under 18 years of age what is the prevalence of asthma development if they had IgE-mediated food allergy to a specific food? What is the percentage of children who develop asthma after milk, egg or peanut allergy? Apart from the overall prevalence the risk of asthma development for each specific food allergen should be presented. We thank the Reviewer for this excellent suggestion. We have significantly expanded our analysis of the extent to which food allergy predisposes to the development of respiratory allergy (asthma and/or allergic rhinitis) in our revised manuscript. In new analyses presented in two new tables, we now show that food allergy to peanut, milk, and egg independently predispose to asthma, and rhinitis. Additionally, we have shown that children with multiple food allergies are at increased risk of respiratory allergy as compared to children with a signal food allergy. These new data are presented in two new tables (Table 4 and Table 5) and discussed in additional text (Page 7, Lines 155-162; Page 12, Lines 236-247; Pages 15 and 16, Lines 304-314).