ORIGINAL ARTICLES ALIMENTARY TRACT

Transcription

1 CLINICAL GASTROENTEROLOGY AND HEPATOLOGY 2009;7: ORIGINAL ARTICLES ALIMENTARY TRACT A Case-Control Study of Sociodemographic and Geographic Characteristics of 335 Children With Eosinophilic Esophagitis JAMES P. FRANCIOSI,*, VICKY TAM, CHRIS A. LIACOURAS,*, and JONATHAN M. SPERGEL, *Division of Gastroenterology, Hepatology and Nutrition and Division of Allergy and Immunology, The Children s Hospital of Philadelphia; and School of Medicine and School of Social Policy and Practice, University of Pennsylvania, Philadelphia, Pennsylvania Podcast interview: See Editorial on page 370. Background & Aims: The epidemiology of pediatric eosinophilic esophagitis (EE) is poorly characterized. In this study, we aimed to determine demographic, socioeconomic, and geographic characteristics of our cohort of EE children. Methods: A case-control design was used to compare 335 EE subjects with control subjects from gastroenterology (GI) and allergy clinics as well as 2000 U.S. census data. Results: EE subjects were significantly different than the greater Philadelphia population as well as control subjects from our gastroenterology and allergy clinics. EE subjects were 83.6% Caucasian, compared with 70.9% of GI control subjects (odds ratio [OR], 2.17; P <.001; confidence interval [CI], ), 64.9% of allergy control subjects (OR, 2.83; P <.001; CI, ), and 73.0% of the greater Philadelphia population. EE subjects were 75.8% male, compared with 48.0% of GI control subjects (OR, 3.39; P <.001; CI, ), 60.4% of allergy control subjects (OR, 1.62; P <.001; CI, ), and 48.0% of the greater Philadelphia population. We initially demonstrated that EE subjects are more affluent, more educated, and reside more often in suburban areas. However, Caucasian race was a significant confounding variable and accounted for socioeconomic or geographic differences among EE subjects and our control populations with one exception. A significant difference remained between suburban and urban residence in EE and allergy control populations. Conclusions: EE subjects are significantly different than control groups in their demographic characteristics of Caucasian race and male sex. EE subject socioeconomic and geographic characteristics are not different than our typical referral patterns to GI clinic when adjusted for race as a confounding factor. Eosinophilic esophagitis (EE) is an increasingly recognized disorder of esophageal inflammation. Currently, it is not known whether the rising number of children diagnosed with EE is due to greater physician awareness of a genetically predisposed condition or due to environmental influences. 1 5 Evidence for a genetic predisposition has been supported by an association with the eotaxin-3 gene, a familial pattern in a subset of patients, and its demographic characteristics. 6 8 In several uncontrolled studies, EE is consistently reported to be 80% 95% Caucasian and 70% 80% male, which is distinct from typical allergy demographics. 3,5 However, because EE is a diagnosis that requires upper endoscopy with biopsy and recognition of its existence by treating physicians, it is possible that demographic characteristics in uncontrolled studies at tertiary referral centers might have a referral bias. It has also been suggested that environmental factors might contribute to the pathogenesis of EE. Hygiene has been proposed as a possible environmental etiology in allergic conditions. The hygiene hypothesis in relation to socioeconomic factors states that individuals of higher socioeconomic status have cleaner environments with less exposure to environmental pathogens at an early age making them more susceptible to allergic, autoimmune, and chronic intestinal inflammatory disorders Bernstein et al 9 have shown that subjects with IBD have higher socioeconomic status, fewer enteric infections, are more concentrated in suburban settings, and have higher rates of geographic clustering compared with the general population. These findings support the hygiene hypothesis as a contributing environmental factor in the pathogenesis of IBD. 10 Currently, the socioeconomic and geographic characteristics of EE subjects are unknown. Our hypothesis is that EE subjects appear to be predominantly male Caucasians from suburban regions and higher socioeconomic classifications. In this study, we aimed to determine demographic, socioeconomic, and geographic characteristics of our cohort of EE children limited to the greater Philadelphia area in a case-control fashion. Methods Study Setting This study was conducted at The Children s Hospital of Philadelphia, which represents the largest pediatric hospital in the 9-county greater Philadelphia area. Geographically, the greater Philadelphia area is defined as Philadelphia, Bucks, Chester, Delaware, Montgomery, Burlington, Camden, Gloucester, and Mercer Counties. Philadelphia County and the City of Camden were defined as urban, and the remaining 7 counties in addition to the remainder of Camden County were defined as suburban. Census data were used to define the 9-county region (Figure 1). Population density sampling has been used previously as a measure of urban- Abbreviations used in this paper: CI, confidence interval; EE, eosinophilic esophagitis; GI, gastroenterology; MHHI, median household income; MHV, median home value; NNI, nearest neighbor index; OR, odds ratio by the AGA Institute /09/$36.00 doi: /j.cgh

2 416 FRANCIOSI ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 7, No. 4 Figure 1. EE cohort within the 9-county greater Philadelphia area. Geographically, the greater Philadelphia area is defined as Philadelphia, Bucks, Chester, Delaware, Montgomery, Burlington, Camden, Gloucester, and Mercer Counties. Philadelphia and Camden cities were defined as urban (shaded region), and the remaining 7 counties in addition to the remainder of Camden county were defined as suburban on the basis of incidence density sampling. Within the greater Philadelphia area, EE subject and 2000 U.S. census data were compared in terms of Caucasian race, male sex, MHV of 2000 U.S. census block groups, MHHI of 2000 U.S. census block groups, and household college education of 2000 U.S. census block groups. ization and was used to dichotomize regions of greater Philadelphia as urban compared with suburban. 11 Data Sources Our EE cohort was obtained from The Children s Hospital of Philadelphia Center for Eosinophilic Disorders database. 12 This is an active database with subjects diagnosed through December 31, 2007 that requires all subjects identified as having EE to have had a localized esophageal eosinophilic infiltrate with 15 eosinophils per high-power field that did not respond to acid suppression therapy and consistent clinical symptoms. Because there is controversy as to the exact number of eosinophils used to define EE, only subjects with 20 eosinophils per high-power field who did not respond to acid suppression therapy were included in this study. EE cohort demographic, socioeconomic, and geographic characteristics limited to the 9-county greater Philadelphia region were first characterized and compared with U.S. census data. Next, control groups were carefully selected from The Children s Hospital of Philadelphia Allergy and Gastroenterology (GI) clinics. First, referral patterns for the last year were determined by using all subspecialty care centers for each clinic. A random sample of non-ee subjects was obtained in a 1:1:1 case to control ratio evenly distributed throughout the calendar year. Given the previous associations determined by Bernstein et al 9 with IBD subjects and high socioeconomic status as well as geographic clustering, all IBD subjects were excluded. Geocoding Geocoding was performed by using ArcView 9.0, ESRI, Redlands, CA, Geographic Information Systems (GIS) technology and has been described previously Briefly, geocoding is a procedure in which a spatially referenced data layer is created from a tabular data set such as street address. This layer can then be displayed as point features within a GIS map. This method compares the input address to the attribute information of a reference street file. If an address falls within the address range of the reference data, the spatial coordinates of the location will be recorded, thereby matching the record. If the automated process is unable to return matched points, these records will require further investigation and manual rematching. All initial address data from EE cases and controls were delinked from unique subject identifiers in accordance with our Institutional Review Board specifications. Street address information was assigned to a particular 2000 U.S. census geographically coded division term block groups. 16 The decreasing order of U.S. census geographic divisions includes state, county, census tract, block group, and block. U.S. census block groups represent the smallest level of division for which socioeconomic data are available. 13 Each block group consists of a number of smaller blocks that are similar to residential street blocks. On average, a block group has a population of approximately 1200 people. Previous studies have validated block group analysis to determine differences in race, gender, and socioeconomic factors such as median household income (MHHI). 13,14 In contrast to block group data, zip codes represent an average population of 30,000, do not represent census area boundaries, and have not been shown to accurately reflect socioeconomic gradients. 14,15,17 In this study, the 9-county greater Philadelphia area consists of 310 zip codes and 4577 block groups. Demographic and Socioeconomic Measures Basic demographic information of race and sex was available on all of our subjects by using a computerized medical record database. Socioeconomic measures of MHHI, median

3 April 2009 SOCIODEMOGRAPHIC CHARACTERISTICS OF CHILDREN WITH EE 417 Figure 2. EE cohort geographic clustering. A geographic model of our EE cohort block groups in the 9-county greater Philadelphia area, with the shaded areas representing clustering of EE subjects. Clustering was determined by using the NNI, where values of 1.0 represent significant geographic clustering. Areas of first and second order clustering are illustrated and represent hierarchical grouping of geographic points (address data) on the basis of spatial proximity. There was more clustering in the EE cohort than would be expected by chance alone in the greater Philadelphia region (NNI, ; P.001). home value (MHV), and percentage of household adults with a college education were identified from 2000 U.S. census data and compared among the groups limited to the 9-county region. Variables of race, sex, MHHI, MHV, and percentage of household adults with a college education were compared among the EE cohort, the GI control group, the allergy control group, and census data by using the appropriate analytical statistical measures. Caucasian race was prespecified as a confounding variable, and a logistic regression model was developed: y (EE) b0 b1(mhhi) b2(mhv) b3(college) b4(caucasian) b5(sex) b6(suburban). Geographic Measures Spatial statistical analysis was used to determine clustering relationships of EE subjects compared with a random sample of GI and allergy clinic subjects. Clustering was assessed by using the nearest neighbor index (NNI), which is a comparison of distance between nearest points (actual home address, not block groups) and distances that would be expected to occur by chance. An NNI of less than 1.0 suggests nonrandom clustering, 1.0 suggests that any associations are likely due to chance alone, and greater than 1.0 suggests dispersion. 18 Results Data Exploration Our initial EE cohort comprised 508 subjects through December 31, Each EE subject diagnosis was reviewed including pathology reports, and families with multiple subjects were counted as a single entity. With these restrictions, the earliest subject diagnosed was in 1996; however, 380 subjects were diagnosed within the past 5 years (from ), and 187 subjects were diagnosed within the past 2 years ( ). Next, the cohort of 508 EE subjects was geocoded to their respective U.S. census track block group. Once geocoded, subjects were classified as matched within the 9-county greater Philadelphia region, matched outside the 9-county greater Philadelphia region, or unmatched. There were 335 EE subjects geocoded within the 9-county Philadelphia region, 140 subjects outside the study region, and 33 subjects who were not able to be matched. EE Cohort Demographic, Socioeconomic, and Geographic Characteristics EE cohort basic demographic and socioeconomic differences for subjects within the 9-county Philadelphia area were compared with U.S. census data and are highlighted in Figure 1. EE subjects were 83.6% Caucasian and 75.8% male, which was in stark comparison to census estimates of Caucasian race (alone or in combination) and male sex of 73.0% (P.001) and 48.0% (P.001), respectively. Socioeconomic measures of MHHI, MHV, and percentage of household adults with a college education were also noticeably different between the EE cohort and census data. Figure 2 illustrates clustering of EE subjects in the 9-county Philadelphia region by using the NNI. There was more clustering in the EE cohort (NNI, 0.8; P.001) than would be expected by chance alone in the greater Philadelphia region. Case-Control Comparison of Demographic, Socioeconomic, and Geographic Characteristics Next, EE subjects confined to the greater Philadelphia region were compared with a random sample of control subjects from GI and allergy clinics in 1:1:1 ratio. For GI and allergy controls, subjects with EE and IBD were excluded. Among 508 GI control subjects geocoded, 410 were within the 9-county Philadelphia area, 66 subjects were outside the study region, and 34 subjects were not able to be matched. Among 508 allergy control subjects, 29 were outside the study region, and 55 subjects were not able to be matched. EE cohort, GI control group, and allergy control group basic demographic and socioeconomic differences for subjects within the 9-county Philadelphia area were compared and are highlighted in Tables 1 and 2. Socioeconomic estimates of MHV, MHHI, and college education were initially analyzed as continuous variables and were all significantly different between EE and GI controls as well as EE and allergy controls (P.05). For the crude and

4 418 FRANCIOSI ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 7, No. 4 Table 1. EE, GI, and Allergy Cohort Demographic and Socioeconomic Characteristics Caucasian Male MHV MHHI College education Suburban EE 83.6% 75.8% $167,770 $68, % 87.5% GI 70.9% c 48.0% c $153,005 a $63,171 b 28.1% b 82.2% a Allergy 64.9% c 60.4% c $145,007 b $59,933 c 27.2% b 69.6% c NOTE. Statistical significance was determined with 2 test and Student t test for the respective dichotomous and continuous variables for the respective GI and allergy control groups compared with the EE cohort. a P.05. b P.01. c P.001. adjusted odds ratios (ORs) of MHV and MHHI, data were dichotomized by using established census cut points so that data could be more easily interpreted. MHVs were dichotomized into 2 groups according to established U.S. census cut points, less than $100,000 and greater than $100,000. There were significant differences between EE subjects and all control groups in the percentage of subjects with an MHV of less than $100,000: 19% of EE subjects versus 28% of GI control subjects (OR, 0.61; P.05; confidence interval [CI], ) and 19% of EE subjects versus 35% of allergy control subjects (OR, 0.45; P.001; CI, ). Similarly, MHHI was dichotomized into 2 groups according to established U.S. census cut points, less than $50,001 and greater than $50,001. There were significant differences between EE subjects and all control groups in the percentage of subjects with an MHHI of less than $50,001: 25% of EE subjects versus 33% of GI control subjects (OR, 0.65; P.05; CI, ), 25% of EE subjects versus 40% of allergy control subjects (OR, 0.47; P.001; CI, ), and 25% of EE subjects versus 58% of the greater 9-county area population. For purposes of ease of interpretation, college education crude and adjusted ORs were represented as changes of 10% per block group. Data were also restricted to EE subjects diagnosed within the past 5 years ( ) and within the past 2 years ( ), and similar trends were observed (data not shown). Next, we investigated for confounding by using a logistic regression model: y (EE) b0 b1(mhhi) b2(mhv) b3(college) b4(caucasian) b5(sex) b6(suburban). Within this logistic regression model, EE subjects remained significantly different than control groups in their demographic characteristics of Caucasian race and male sex, as shown in Table 2. Furthermore, similar trends were observed in Caucasian race and male sex when the model was stratified on the basis of socioeconomic variables. For example, in the adjusted analysis when comparing EE subjects with allergy controls, the odds of Caucasian race are In a stratified analysis comparing EE subjects with allergy controls, the odds of Caucasian race are 1.86 in subjects with an MHV of less than $100,000, and the odds of Caucasian race are 2.24 in subjects with an MHV of greater than $100,000. EE subjects were no different than GI and allergy control subjects in their MHHI, MHV, and percentage of household adults with a college education in the adjusted analysis, as shown in Table 2. A significant difference remained between suburban and urban residence in EE and allergy control populations. Clustering was compared between the EE cohort and the GI and allergy control groups by using the NNI. Although there was more clustering in the EE cohort (NNI, 0.8; P.001) than would be expected by chance alone in the greater Philadelphia region, this was not appreciably different than the GI (NNI, 0.7; P.001) and allergy (NNI, 0.7; P.001) control groups. Discussion In this study, we have described the demographic, socioeconomic, and geographic characteristics of our cohort of EE children in a case-control fashion. Demographic variables of Caucasian race (83.6%) and male sex (75.8%) in the EE cohort were significantly different than all control groups. Socioeconomic and geographic differences between EE and controls are explained by Caucasian race as a confounding factor. The fact that our percentage of subjects who are Caucasian is less than previous reports can be explained by the data sources and data classification for this study. 3,5 To compare data from our EE cohort and control groups, race data from each group were abstracted from our computerized medical records and classified as Caucasian, African American, Hispanic, Native American, Asian, other, or unknown. In addition to 83.6% of subjects who were identified as Caucasian, 5.2% were classified as other or unknown. This data source also did not capture subjects of mixed race. Therefore, the estimation of Caucasian race is likely an underrepresentation of the true proportion of our EE cohort who are Caucasian. However, for our study we were interested in comparing the differences of our EE cohort compared with control populations, and therefore, it was impor- Table 2. EE, GI, and Allergy Cohort Crude and Adjusted Analyses Caucasian Male MHV $100,000 MHHI $50,001 College Suburban Crude OR (CI) EE-GI 2.17 ( ) 3.39 ( ) 0.61 ( ) 0.65 ( ) 1.14 ( ) 1.59 ( ) EE-allergy 2.83 ( ) 1.62 ( ) 0.45 ( ) 0.47 ( ) 1.18 ( ) 3.05 ( ) Adjusted OR (CI) EE-GI 1.90 ( ) 3.49 ( ) 0.80 ( ) 0.88 ( ) 1.05 ( ) 0.82 ( ) EE-allergy 1.98 ( ) 1.52 ( ) 1.06 ( ) 0.78 ( ) 0.96 ( ) 2.08 ( ) NOTE. Crude ORs and the appropriate CIs were obtained by using Stata version 8.0 (Stata Corporation, College Station, TX). Next, a logistic regression model was developed to determine adjusted ORs that investigated for confounding between the demographic and socioeconomic variables.

5 April 2009 SOCIODEMOGRAPHIC CHARACTERISTICS OF CHILDREN WITH EE 419 tant to use a consistent data source. Furthermore, despite the difficulty in obtaining accurate demographic information, each cohort was asked the same demographic questions, making the comparisons valid. In terms of socioeconomic and geographic factors, the results of the unadjusted analysis suggest that our EE subjects are more affluent, more educated, and are more often in the suburban areas of Philadelphia compared with our control populations. These results are consistent with previous work by Green et al 10 that used these data to suggest hygiene as an environmental factor that might contribute to the pathogenesis of IBD. However, we were concerned that Caucasian race might be a confounding variable. Therefore, we performed an adjusted analysis with a logistic regression model accounting for confounding variables. With this regression model, Caucasian race was found to be a significant confounding variable and, in the adjusted analysis, accounted for any differences of MHHI, MHV, percentage of household adults with a college education, and suburban residence among EE subjects and our control populations with one exception. A significant difference remained between suburban and urban residence between the EE and allergy cohorts. The finding of allergy control subjects residing more commonly in urban areas is in keeping with previous allergy literature findings of geographic and demographic associations. One potential limitation of our study is that it was not a population-based ecologic study and therefore does not account for other hospitals in the greater Philadelphia region. However, The Children s Hospital of Philadelphia is the largest pediatric hospital in the region, with the largest cohort of EE subjects in the country. Furthermore, a case-control design that was restricted to the 9-county greater Philadelphia region limits the potential referral bias to a large pediatric center. A case-control design also allows for analysis that accounts for confounding variables such as race in a logistic regression model. Another limitation of this study is that the data presented relied on census data of block groups rather than individual patient data. Although this methodology has been used in previous studies for similar investigations, it is possible that prospective studies that collect individual subject data might demonstrate socioeconomic and geographic differences that our analysis did not demonstrate. Another limitation of our study would be that these results might not be generalizable to adult EE subjects and to children in rural environments. In summary, although several case series have reported the significance of Caucasian race and male sex in EE, our study marks a case-control study that accounts for GI and allergy clinic referral patterns to demonstrate these differences. Differences in Caucasian race and male sex between EE subjects and control groups support the belief that EE has a genetic predisposition. The lack of EE subject socioeconomic and geographic differences compared with control populations is an important finding. Further research is needed to explore genetic and environmental factors that might be involved in the pathogenesis of EE. References 1. Gill R, Durst P, Rewalt M, et al. Eosinophilic esophagitis disease in children from West Virginia: a review of the last decade ( ). Am J Gastroenterol 2007;102: Vanderheyden AD, Petras RE, DeYoung BR, et al. Emerging eosinophilic (allergic) esophagitis: increased incidence or increased recognition? Arch Pathol Lab Med 2007;131: Assa ad AH, Putnam PE, Collins MH, et al. Pediatric patients with eosinophilic esophagitis: an 8-year follow-up. J Allergy Clin Immunol 2007;119: Cherian S, Smith NM, Forbes DA. Rapidly increasing prevalence of eosinophilic oesophagitis in Western Australia. Arch Dis Child 2006;91: Liacouras CA, Spergel JM, Ruchelli E, et al. Eosinophilic esophagitis: a 10-year experience in 381 children. Clin Gastroenterol Hepatol 2005;3: Blanchard C, Wang N, Rothenberg ME. Eosinophilic esophagitis: pathogenesis, genetics, and therapy. J Allergy Clin Immunol 2006;118: Blanchard C, Wang N, Stringer KF, et al. Eotaxin-3 and a uniquely conserved gene-expression profile in eosinophilic esophagitis. J Clin Invest 2006;116: Katzka DA. Eosinophilic esophagitis: it s all in the family. Gastrointest Endosc 2007;65: Bernstein CN, Blanchard JF, Rawsthorne P, et al. Epidemiology of Crohn s disease and ulcerative colitis in a central Canadian province: a population-based study. Am J Epidemiol 1999;149: Green C, Elliott L, Beaudoin C, et al. A population-based ecologic study of inflammatory bowel disease: searching for etiologic clues. Am J Epidemiol 2006;164: Hall SA, Kaufman JS, Ricketts TC. Defining urban and rural areas in US epidemiologic studies. J Urban Health 2006;83: Spergel JM, Beausoleil JL, Mascarenhas M, et al. The use of skin prick tests and patch tests to identify causative foods in eosinophilic esophagitis. J Allergy Clin Immunol 2002;109: Chen FM, Breiman RF, Farley M, et al. Geocoding and linking data from population-based surveillance and the US Census to evaluate the impact of median household income on the epidemiology of invasive Streptococcus pneumoniae infections. Am J Epidemiol 1998;148: Krieger N, Chen JT, Waterman PD, et al. Race/ethnicity, gender, and monitoring socioeconomic gradients in health: a comparison of area-based socioeconomic measures the public health disparities geocoding project. Am J Public Health 2003;93: Krieger N, Chen JT, Waterman PD, et al. Painting a truer picture of US socioeconomic and racial/ethnic health inequalities: the Public Health Disparities Geocoding Project. Am J Public Health 2005;95: Census Bot census of population: social and economic characteristics. Washington, DC: Department of Commerce, Economics, and Statistics Administration, Krieger N, Waterman P, Chen JT, et al. Zip code caveat: bias due to spatiotemporal mismatches between zip codes and US census-defined geographic areas the Public Health Disparities Geocoding Project. Am J Public Health 2002;92: Bernstein CN, Wajda A, Blanchard JF. The clustering of other chronic inflammatory diseases in inflammatory bowel disease: a population-based study. Gastroenterology 2005;129: Reprint requests Address reprint requests to: James P. Franciosi, MD, MS, MSCE, Assistant Professor, Cincinnati Children s Hospital Medical Center, Division of Gastroenterology, Hepatology and Nutrition, 3333 Burnet Avenue, ML 2010, Cincinnati, Ohio jpfranciosi@yahoo.com; fax: (513) Acknowledgments We would like to acknowledge Douglas Wiebe, PhD, from the Department of Biostatistics and Epidemiology at the University of Pennsylvania School of Medicine for his invaluable contribution to study design and statistical analysis. Conflicts of interest The authors disclose the following: Funding was provided by the American Partnership for Eosinophilic Disorders (APFED), a nonprofit advocacy organization.