Secondary Logo

Journal Logo

Original Articles: Gastroenterology: Celiac Disease

Comorbidities in Childhood Celiac Disease: A Phenome Wide Association Study Using the Electronic Health Record

Prinzbach, Ariana; Moosavinasab, Soheil; Rust, Steve; Boyle, Brendan∗,‡; Barnard, John A.∗,‡; Huang, Yungui; Lin, Simon∗,†

Author Information
Journal of Pediatric Gastroenterology and Nutrition: October 2018 - Volume 67 - Issue 4 - p 488-493
doi: 10.1097/MPG.0000000000002020

Abstract

See “Electronic Medical Records Enable Precision Medicine Approaches for Celiac Disease” by Kosti and Sirota on page 434.

What Is Known

  • Pediatric celiac disease is associated with a variety of extraintestinal autoimmune and inflammatory findings.
  • Characterizing the comorbidities and symptoms of celiac disease is important for identification of the disease and preventing long-term sequelae.
  • The expanding implementation of electronic health records for research purposes permits efficient analysis of disease comorbidities with high statistical power.

What Is New

  • Screening for eosinophilic disorders of the esophagus should be considered for pediatric celiac disease patients.
  • A Phenome Wide Association Study method enables effective secondary use of electronic health records to characterize a disease clinically and discover new associations due to its unbiased screening of associated symptoms, medical encounters, and comorbidities.

Celiac disease (CD) is a chronic systemic immune disorder triggered by the consumption of gluten. The global prevalence is increasing and an estimated 1% of the population is affected in the United States and Europe (1,2). Typical CD is characterized by chronic small intestinal inflammation and malabsorption, commonly manifesting in childhood, whereas atypical CD presents often with extraintestinal complaints that may precede or overshadow gastrointestinal symptoms, resulting in delays in diagnosis and treatment (3). CD results from a complex interaction between consumption of gluten and environmental, genetic, and immune factors. An association with autoimmune disorders such as type 1 diabetes, autoimmune thyroiditis, lupus, psoriasis, and inflammatory bowel disease is well described (1,4–6). Non-autoimmune associations include Down syndrome and Turner syndrome, and certain neurological disorders, including attention deficit hyperactivity disorder (ADHD) (3,6,7). The strength of many of these associations is unclear and limited by methodological challenges (4,5).

Herein, we examine, in a statistically robust and unbiased manner, various associations of CD in a large pediatric hospital. Use of International Classification of Diseases, 10th revision (ICD-10) diagnostic codes from electronic health records (EHRs) in our screening analysis not only permitted identification of comorbidities, but also significant symptoms and manifestations that could allow for better diagnosis and treatment.

We also demonstrate the effectiveness of an unbiased screening methodology using EHRs and a Phenome Wide Association Study (PheWAS) method. PheWAS studies have previously been valuable in genetics and clinical research, where all possible phenotypes from EHR-linked data are scanned for associations with a single genetic or clinical condition (8,9). In the current study, we simultaneously screen a large number of phenotypes, including comorbidities and symptoms identified by ICD-10 codes, for association with CD. Although straightforward in design, this unbiased screening approach has not been used previously to identify comorbidities in CD (10). Advantages include statistical power, cost-effectiveness, feasibility, and timeliness—characteristics not found in cohort and follow-up study designs. Although limitations may arise due to errors in ICD-10 coding, when paired with additional chart review to confirm results, this method is powerful and efficient (9). Confirmation of suspected comorbidities and the finding of new associations expands clinical understanding of CD in children and improve diagnosis and treatment.

METHODS

The Nationwide Children's Hospital (NCH) Institutional Review Board approved this study. We used 2 approaches: a non-hypothesis driven, phenome-wide screening for comorbidities through association modeling and an a priori hypothesis testing of 14 diseases putatively associated in the literature with CD (Supplemental Table 1, Supplemental Digital Content, https://links.lww.com/MPG/B383). These include: type 1 diabetes, autoimmune thyroiditis, thyroid disorders, selective IgA deficiency, Down syndrome, psoriasis, inflammatory bowel disease, dermatitis herpetiformis, Turner syndrome, alopecia areata, ADHD, systemic lupus erythematosus, autoimmune hepatitis, and sarcoidosis.

Patient Population

This retrospective cohort study was conducted using the Epic EHR at NCH, one of the largest pediatric hospitals in the United States. Records of 433 confirmed CD cases from January 2012 to May 2016 were obtained from a CD registry maintained by our CD clinic. The diagnosis was defined by a positive serological test for anti-tissue transglutaminase and confirmed by an abnormal duodenal biopsy. All cases were under 21 years of age and the demographics shown in Table 1 reflect the current epidemiology of the disease (2).

T1
TABLE 1:
Demographics of celiac disease cases

The control population was defined as having no diagnosis of CD and was also drawn from our EHR containing 704,399 unique patients in the period 2007–2016. A total of 4,330 controls were selected by 1:10 case:control matching for gender, race, date of birth (±365 days), and presence of primary care at NCH. Eight cases did not have 10 matched controls. For these, the race matching criterion was removed.

Data Preparation

For each study subject, ICD-10-CM codes from the EHR diagnoses and problem lists throughout the study period were collated and duplicates removed. To perform the screening analyses, we created a combined frequency list of all ICD-10 codes present among our population (433 cases and 4330 controls). Due to the vast quantity and specificity of ICD-10 codes, we grouped the codes into their ICD-10 Hierarchy Level 1 groups by disregarding designation after the decimal. This abstraction resulted in a list of 1034 unique level 1 ICD-10 codes. We aggregated the ICD codes to the Level 1 Hierarchy to enable sufficient statistical power since many ICD-10 codes are highly specific and likely to be underrepresented in our data set (see Discussion). We further narrowed the list of Hierarchy Level 1 ICD-10 codes to 625 by excluding any code with a frequency less than 5 in cases and controls combined, thus ensuring sufficient statistical power. SQuirrel SQL Client Version 3.7 was used to extract data from Epic's Clarity database, and IntelliJ Version 14.0.03 was used for Java programming language to clean data and generate case-control population.

Data Analyses

To conduct an unbiased screening of all clinical phenotypes using the PheWAS method (9), Fisher's Exact tests with Bonferroni correction were conducted for each of the 625 codes, using the exact2x2 package and function for R programming language. R Studio Version 0.99.473 was used for statistical analysis. The prevalence within cases and controls, odds ratio (OR), and confidence interval (CI) for the odds ratio were calculated for each code found to have a statistically significant association with CD. Of the Hierarchy Level 1 ICD-10 codes identified in the screening analysis, 15 were deemed too nonspecific for accurate interpretation and were broken down into sets of more specific ICD-10 codes with their corresponding frequencies (Supplemental Table 2, Supplemental Digital Content, https://links.lww.com/MPG/B383). The 14 a priori hypotheses were also tested by Fisher's Exact test with a Bonferroni correction, and the prevalence, odds ratios, and confidence intervals for odds ratios were calculated for each comorbidity analyzed.

Further investigation by EHR review was undertaken to expand interpretation of several results, including Type 2 diabetes (E11) and Esophagitis (K20). The significance of Eosinophilic Esophagitis (EoE; K20.0) was examined by Fisher's Exact test with a Bonferroni correction. We utilized standardized diagnostic criteria for EoE: symptoms consistent with EoE; at least 3 esophageal biopsies in both the distal and proximal esophagus with 15 or more eosinophils per high-power field; and exclusion of other disorders with similar clinical or endoscopic characteristics. Since clinical recognition and guidelines relating to proton pump inhibitor-responsive eosinophilia (PPI-REE) remain a subject of controversy (11), and no corresponding ICD-10 code exists, in this study it is possible that ICD-10 coded cases of EoE represent cases of PPI-REE. Likewise, the result of “Overweight and obesity” (E66) provoked further analysis, and the values of the exact body mass indexes (BMIs) for cases and controls were obtained from the EHRs. After dividing the cohort and control population into 4 age groups (ie, 0 < y < 5, 5 < y < 10, 10 < y < 15, and 15 < y), the average BMI was calculated per group.

RESULTS

The PheWAS method found 45 (7.2%) of the 625 analyzed ICD-10 codes to be associated (Bonferroni corrected, P < 0.05) with CD. These codes represent associated diseases, symptoms, and medical services, and fell into 3 groups: expected comorbidities, expected symptoms, and potential novel findings (Table 2).

T2
TABLE 2:
Level 1 International Classification of Diseases, 10th revision codes found to be significantly associated with celiac disease in the overall screening analysis

Expected Comorbidities

Of the 45 statistically significant codes, 13 are previously established comorbidities of CD. These include type 1 diabetes, gastritis and duodenitis, vitamin D deficiency, gastroesophageal reflux disease, thyroiditis, Crohn disease, Down syndrome, psoriasis, and iron deficiency anemia.

Because we initially collapsed ICD-10 codes into the broadest Level 1 Hierarchy, 5 of the codes were further evaluated by examining the frequency of the actual code entered in the EHR (Supplemental Table 2, Supplemental Digital Content, https://links.lww.com/MPG/B383). ICD-10 code K31 consisted mainly of the more specific code K31.9 (Disease of stomach and duodenum, unspecified). The Level 1 code K29 (Gastritis and duodenitis) consisted of 36 cases of K29.80 (Duodenitis without bleeding), and 11 cases of K29.70 (Gastritis without bleeding), as well as a few other less common variations (12). Further investigation of Level 1 ICD-10 code D80 (Immunodeficiency with predominantly antibody defects) revealed only 1 specific code, D80.2 (Selective deficiency of IgA), a known association with CD (13).

Expected Symptoms

Nine of the CD-associated ICD-10 codes were expected symptoms including abdominal and pelvic pain, other symptoms involving the digestive system, symptoms concerning fluid and food intake, other functional intestinal disorders, lack of expected normal physiological development in childhood and adults, nausea and vomiting, malaise and fatigue, flatulence and related conditions, and aphagia and dysphagia (Table 2). Code R19 (Other symptoms involving the digestive system) was due primarily to diarrhea. Code K59 (Other functional intestinal disorders) comprised the codes: K59.00 (unspecified constipation), K59.09 (Other constipation) and K59.01 (Slow transit constipation). Constipation and diarrhea are known symptoms of pediatric CD (13). Collectively, these findings validate our experimental approach.

Potential Novel Findings

The remaining 23 significant ICD-10 codes were potential novel findings (Table 2). Eight of the presenting Level 1 ICD-10 codes were further analyzed (Supplemental Table 2, Supplemental Digital Content, https://links.lww.com/MPG/B383) and 2, K20 (Esophagitis) and E66 (Overweight and obesity) were examined by chart review. Investigation of the ICD-10 code K20 consisted of 2 codes: K20.0 (EoE) and K20.9 (Unspecified esophagitis). The relationship of EoE and CD is contested in literature (14,15). Analysis of the code breakdown found that twelve of our patients had co-occurring EoE and CD. Chart review verified all twelve had EoE confirmed by biopsy (Supplemental Table 3, Supplemental Digital Content, https://links.lww.com/MPG/B383). Statistical analysis found this comorbidity was significant in the CD patients (P = 5.81E-05, OR = 9.6, CI = 4.26–21.50). Although others have found a link between CD and EoE, ours is the first using matched controls (14–16). Because no ICD-10 code exists for PPI-REE and the definition of this disorder and its distinction from EoE remains controversial, it is unclear how many of our cases represent PPI-REE distinct from EoE.

The code for R74 (Abnormal serum enzyme levels) consisted primarily of R74.0 (Nonspecific elevation of transaminases and lactic acid dehydrogenase). Transaminitis is a well-known, associated laboratory finding in children with CD (17,18). Our finding further supports screening for liver involvement in CD. Analysis of the codes Z23 (Encounter for immunization) and Z88 (Allergy status to drugs, medications, and biologics), and M25 (Other joint disorder, not elsewhere classified) identified potential new associations, which we explore further in the discussion.

Two unexpected associated codes, E66 (Overweight and obesity) and E11 (Type 2 diabetes) were found to be inaccurate. Chart review found type 2 diabetes cases were miscoded cases of type I diabetes—a known CD association. The obesity code is likely the result of a quality improvement initiative in the gastroenterology clinic that obliged physicians to more actively code the presence of BMI in their patients. Physicians caring for controls did not participate in this initiative. Detailed analysis found the CD patients had a lower BMI in all age groups, as expected (Supplemental Table 4, Supplemental Digital Content, https://links.lww.com/MPG/B383).

Finally, we investigated both established and inconclusive comorbidities previously reported in the literature by analyzing 14 a priori hypotheses. Eight were significant (Table 3). Type 1 diabetes, autoimmune thyroiditis, thyroid disorders (including non-autoimmune causes of hypothyroidism and hyperthyroidism), selective IgA deficiency, Down syndrome, psoriasis, inflammatory bowel disease, and dermatitis herpetiformis were found to be significant, confirming what is well-established in literature (5,13,19–23). Dermatitis herpetiformis was present in only the CD cases, supporting it as a manifestation of CD (2). Turner's Syndrome, alopecia areata, ADHD, lupus, autoimmune hepatitis, and sarcoidosis were not significantly associated.

T3
TABLE 3:
Individual a priori analyses, listed in order of P value

DISCUSSION

With the increasing use of the EHR, the PheWAS method employed in this study is easily undertaken for other disorders as a means to screen for possible disease associations and comorbidities. In our study, 704,399 patients were used for statistical sampling, a size not practical in other types of studies. With the development of large EHR-derived, multi-institutional data resources like PEDSnet (24) this method could be applied to discover rare associations not likely found at a single institution.

A PheWAS method has not previously been utilized to examine comorbidities of pediatric CD. Since this method identified 13 established comorbidities and nine expected symptoms, the approach appears to be facile and valid. Further, since ICD-10 codes also include medical services, screening all possible associated codes permits better characterization of clinical service trends. For example, the code “Encounter for immunization” (Z23) was found significant in our analysis, and although its appearance in our screening could have numerous explanations (eg, children with CD have more regular hospital visits), this relationship would not have been tested if we examined specific hypothesized comorbidities. This demonstrates the PheWAS method allows for investigation of the clinical practice and management of a disease. Additionally, our data are derived from a large EHR-based data source, allowing us to eliminate low frequency ICD-10 codes and perform 10:1 control matching to strengthen statistical power. In our study, we chose to group ICD-10 codes at the Level 1 Hierarchy to better examine overall disease comorbidity, as subsequent hierarchy levels examined sub-classifications of diseases too specific for our purposes. If further specificity in associations is desired, a small adjustment in the analysis is needed for rapid results.

The association between EoE and CD is controversial (14,25,26). Recent studies conclude that further investigation is necessary (14,26). Not only did we find a significant association between EoE and CD, but we also identified 12 cases, the largest number of validated cases of EoE found in a pediatric CD population (16,27). Additionally, our study is the first to have found a significant association with use of a matched control population. Prior studies compared their study prevalence to the general population or measures of association (14–16,28,29). Despite these findings, the relationship between these 2 disease remains a topic of debate (14,26). Recent investigations reach opposing conclusions (16,27,30). Based on our work, we believe the association exists. Recognition of this association will impact clinical practice by improving adherence to biopsy recommendations during upper endoscopy to evaluate for either CD or EoE (31). It is currently believed EoE is a continuum of disorders that includes PPI-REE (11,32). Although classic EoE and PPI-REE share common genetic, molecular, pathologic, and endoscopic features, the management and prognosis may differ (11). Our Phe-WAS analysis utilizes ICD-10 codes as data and a single code for EoE (K20.0) exists that does not distinguish between these variations. We are not able to distinguish these 2 separately in our study and it is possible both entities exist within our 12 confirmed cases. Thus, our finding can be interpreted as a significant association between CD and the continuum of esophageal eosinophilic syndromes.

We also identified a potential association of juvenile arthritis with CD. Initial screening for the ICD-10 code of M25 (Other joint disorder) was statistically significant. Further analysis of secondary codes demonstrated pain in joints such as the knees, ankles, and hips (Supplemental Table 2, Supplemental Digital Content, https://links.lww.com/MPG/B383), and suggested a potential association of CD with juvenile arthritis. Juvenile arthritis in CD patients has been examined on a limited basis and several published studies note an association (33), including 2 that show an increased prevalence of CD in children with juvenile arthritis, but not the inverse relationship (34,35). Using the ICD-10 code M08 (Juvenile arthritis) we found a total of 7 patients with concurrent juvenile arthritis and CD. Although not significant in the initial PheWAS screening (P = 0.0013 without correction), the presence of 7 co-occurring cases merits further investigation.

The association with code Z88 (Allergy status to drugs, medications, and biologics) was also significant. These allergies were primarily penicillin antibiotics and narcotics. Due to the common underlying pathophysiology of T-cell mediated immune response in CD and immune-related drug allergies (36), the increased prevalence of drug allergies in our case population suggests the need for additional investigation.

The association of ADHD and CD is controversial (7,37). A recent literature review concludes no definitive association exists but called for further investigation (7). Our work further suggests no association between these 2 common childhood disorders.

Examining the demographics of our cohort, 92% of our patients identify as white (Table 1), reflecting population diversity similar to other studies investigating CD (2,38). The primary NCH service area of Columbus, OH, is, however, 61.5% white and 28.0% black, comparable to other urban areas (39). Due to our utilization of case-control race matching, our demographics can likely be explained by the higher prevalence of CD among whites in the patient population. This is consistent with a recent study conducted by the National Health and Nutrition Examination Survey that identified the prevalence of CD as 4 to 8 times higher in non-Hispanic whites (40).

Our study has several limitations of EHR and diagnostic coding. Entry of diagnostic and problem list ICD-10 codes may be erroneous, and coding habits vary across providers and hospitals. An example is our finding of the erroneous entry of type II diabetes ICD-10 codes (E11), due to a mistranscription. We also found the ICD-10 code E66 (Overweight and obesity) inaccurate due to a coding quality improvement project being conducted specifically within the gastroenterology clinic. Further investigation found a lower average BMI in all 4 age subgroups in the CD population, as suspected from the pathophysiology of CD (Supplemental Table 4, Supplemental Digital Content, https://links.lww.com/MPG/B383). Coding accuracy rates are improving (41), and according to the Health Information Management, coding of endocrine, nutritional and metabolic diseases has been an area of coding strength for ICD-10 coding (42). With supplemental rigor by chart reviews and analysis of clinical relevance, the PheWAS method can overcome some shortcomings of EHR data.

In the future, a PheWAS method can be applied to other systemic and chronic diseases. With expansion of EHRs and the emergence of large national databases of healthcare information, these data can be effectively utilized to rapidly examine and expand the sophistication of known phenotypes and comorbidities of diseases in both pediatric and adult populations.

Our study demonstrates the PheWAS method enables effective secondary use of EHRs to more deeply characterize a disease clinically, due to its unbiased screening nature of analyzing all possible associated symptoms, medical encounters, and comorbidities. In addition, this approach strengthens the statistical associations of chronic diseases and permits prompt investigations of new associations in a cost effective, timely, and efficient manner. Clinically, our study further elucidates the relationship of several controversial comorbidities of CD, including eosinophilic disorders of the esophagus and ADHD. Our finding of a significant association with validated eosinophilic disorders of the esophagus supports recommendations for screening considerations for this disorder amongst CD patients.

Acknowledgments

The authors acknowledge Richard Hoyt for EHR data extraction, Katherine Strohm for project management, and Melody Davis for copy editing.

REFERENCES

1. Lauret E, Rodrigo L. Celiac disease and autoimmune-associated conditions. Biomed Res Int 2013; 2013:127589.
2. Green PHR, Christophe C. Celiac disease. N Engl J Med 2007; 357:1731–1743.
3. Ciccocioppo R, Kruzliak P, Cangemi GC, et al. The spectrum of differences between childhood and adulthood CD. Nutrients 2015; 7:8733–8751.
4. Viljamaa M, Kaukinen K, Huhtala H, et al. Coeliac disease, autoimmune diseases and gluten exposure. Scand J Gastroenterol 2005; 40:437–443.
5. Bardella MT, Elli L, De Matteis S, et al. Autoimmune disorders in patients affected by celiac sprue and inflammatory bowel disease. Ann Med 2009; 41:139–143.
6. Ediger TR, Hill ID. Celiac disease. Pediatr Rev 2014; 35:409–415. quiz 416.
7. Ertürk E, Wouters S, Imeraj L, et al. Association of ADHD and celiac disease: what is the evidence? A systematic review of the literature. J Atten Disord 2016; Epub ahead of print. doi:10.1177/1087054715611493.
8. Denny JC, Bastarache L, Ritchie MD, et al. Systematic comparison of phenome-wide association study of electronic medical record data and genome-wide association study data. Nature Biotechnology 2013; 31:1102–1110.
9. Doss J, Mo H, Carroll RJ, et al. Phenome-Wide Association Study of Rheumatoid Arthritis Subgroups Identifies Association Between Seronegative Disease and Fibromyalgia. Arthritis & Rheumatology 2017; 69:291–300.
10. Denny JC, Bastarache L, Roden DM. Phenome-wide association studies as a tool to advance precision medicine. Annu Rev Genomics Hum Genet 2016; 17:353–373.
11. Lucendo AJ, Molina-Infante J, Arias Á, et al. Guidelines on eosinophilic esophagitis: evidence-based statements and recommendations for diagnosis and management in children and adults. United European Gastroenterol J 2017; 5:335–358.
12. Lebwohl B, Green PHR, Genta RM. The coeliac stomach: gastritis in patients with coeliac disease. Aliment Pharmacol Ther 2015; 42:180–187.
13. Khatib M, Baker RD, Ly EK, et al. Presenting pattern of pediatric celiac disease. J Pediatr Gastroenterol Nutr 2016; 62:60–63.
14. Lucendo AJ, Arias Á, Tenias JM. Systematic review: the association between eosinophilic oesophagitis and coeliac disease. Aliment Pharmacol Ther 2014; 40:422–434.
15. Leslie C, Mews C, Charles A, et al. Celiac disease and eosinophilic esophagitis: a true association. J Pediatr Gastroenterol Nutr 2009; 50:1.
16. Dharmaraj R, Hagglund K, Lyons H. Eosinophilic esophagitis associated with celiac disease in children. BMC Res Notes 2015; 8:1.
17. Castillo NE, Vanga RR, Theethira TG, et al. Prevalence of abnormal liver function tests in celiac disease and the effect of a gluten-free diet in the US population. Am J Gastroenterol 2015; 110:1216–1222.
18. Anania C, De Luca E, De Castro G, et al. Liver involvement in pediatric celiac disease. World J Gastroenterol 2015; 21:5813–5822.
19. Cosnes J, Cellier C, Viola S, et al. Incidence of autoimmune diseases in celiac disease: protective effect of the gluten-free diet. Clin Gastroenterol Hepatol 2008; 6:753–758.
20. Iqbal T, Tariq I, Zaidi MA, et al. Celiac disease arthropathy and autoimmunity study. J Gastroenterol Hepatol 2012; 28:99–105.
21. Mårild K, Stephansson O, Grahnquist L, et al. Down syndrome is associated with elevated risk of celiac disease: a nationwide case-control study. J Pediatr 2013; 163:237–242.
22. Yang A, Chen Y, Scherl E, et al. Inflammatory bowel disease in patients with CD. Inflamm Bowel Dis 2005; 11:528–532.
23. Cataldo F, Marino V, Bottaro G, et al. Celiac disease and selective immunoglobulin A deficiency. J Pediatr 1997; 131:306–308.
24. Forrest CB, Margolis PA, Bailey LC, et al. PEDSnet: A National Pediatric Learning Health System. J Am Med Inform Assoc 2014; 21:602–606.
25. Pellicano R, De Angelis C, Ribaldone DG, et al. 2013 update on celiac disease and eosinophilic esophagitis. Nutrients 2013; 5:3329–3336.
26. Watkins RD, Blanchard SS. Eosinophilic esophagitis and celiac disease: a true association or coincidence? J Pediatr Gastroenterol Nutr 2017; 65:1–2.
27. Hommeida S, Alsawas M, Murad MH, et al. The association between celiac disese and eosinophilic esophagitis: mayo experience and meta-analysis of the literature. J Pediatr Gastroenterol Nutr 2017; 65:58–63.
28. Ooi CY, Day AS, Jackson R, et al. Eosinophilic esophagitis in children with celiac disease. J Gastroenterol Hepatol 2008; 23 (7 Pt 1):1144–1148.
29. Thompson JS, Lebwohl B, Reilly NR, et al. Increased incidence of eosinophilic esophagitis in children and adults with celiac disease. J Clin Gastroenterol 2012; 46:e6–e11.
30. Ahmed OI, Qasem SA, Abdulsattar JA, et al. Esophageal eosinophilia in pediatric patients with CD. J Pediatr Gastroenterol Nutr 2015; 60:493–497.
31. Wallach T, Genta RM, Lebwohl B, et al. Adherence to celiac disease and eosinophilic esophagitis biopsy guidelines is poor in children. J Pediatr Gastroenterol Nutr 2017; 65:64–68.
32. Durrani S, Rothenberg M. Recent advances in eosinophilic esophagitis. F1000Res 2017; 6:1775.
33. De Maddi F, Pellegrini F, Raffaele CGL, et al. Celiac disease and juvenile idiopathic arthritis: a still enigmatic crossover. Scand J Gastroenterol 2013; 48:511–512.
34. Prignano F, Bonciani D, Bandinelli F, et al. Juvenile psoriatic arthritis and comorbidities: report of a case associated with enthesitis and celiac disease. Dermatol Ther 2010; 23:S47–S50.
35. Lepore L, Martelossi S, Pennesi M, et al. Prevalence of celiac disease in patients with juvenile chronic arthritis. J Pediatr 1996; 129:311–313.
36. Schnyder B, Pichler WJ. Mechanisms of drug-induced allergy. Mayo Clin Proc 2009; 84:268–272.
37. Dazy KM, Rubenstein JH, Holevinski L, et al. Sa1285 The Prevalence of ADHD in adults and children previously diagnosed with celiac disease: a hospital-based study. Gastroenterology 2013; 144:S251–S252.
38. Riddle MS, Murray JA, Porter CK. The incidence and risk of celiac disease in a healthy US adult population. Am J Gastroenterol 2012; 107:1248–1255.
39. United States Department of Commerce. Bureau of the Census. Census of Population and Housing, 2010 [United States]: National Summary File of Redistricting Data. ICPSR Data Holdings. http://dx.doi.org/10.3886/icpsr33442.v1.
40. Mardini HE, Westgate P, Grigorian AY. Racial differences in the prevalence of celiac disease in the US Population: National Health and Nutrition Examination Survey (NHANES) 2009–2012. Dig Dis Sci 2015; 60:1738–1742.
41. Burns EM, Rigby E, Mamidanna R, et al. Systematic review of discharge coding accuracy. J Public Health 2012; 34:138–148.
42. Butler M. Analyzing Eight months of ICD-10. J AHIMA 2016; 87:16–22.
Keywords:

comorbidity; eosinophilic disorders of the esophagus; EHR; unbiased screening

Supplemental Digital Content

Copyright © 2018 by European Society for Pediatric Gastroenterology, Hepatology, and Nutrition and North American Society for Pediatric Gastroenterology, Hepatology, and Nutrition