Portal de Periódicos da CAPES

Patterns and predictors of atopic dermatitis disease control past childhood: An observational cohort study

2017; Elsevier BV; Volume: 141; Issue: 2 Linguagem: Inglês

10.1016/j.jaci.2017.05.031

1097-6825

Katrina Abuabara, Ole Hoffstad, Andrea B. Troxel, Joel M. Gelfand, Charles E. McCulloch, David J. Margolis,

Food Allergy and Anaphylaxis Research

Atopic dermatitis ([AD], synonymous with eczema or atopic eczema) is a common and burdensome condition occurring in patients of all ages. Most research has focused on pediatric disease because symptoms often present early in life, and little is known about disease course past childhood.1Weidinger S. Novak N. Atopic dermatitis.Lancet. 2016; 387: 1109-1122Abstract Full Text Full Text PDF PubMed Scopus (1170) Google Scholar We report on the results of a longitudinal cohort study using the Pediatric Eczema Elective Registry (PEER), which follows individuals for up to 10 years during the transition to early adulthood. Our objectives were to define patterns of AD disease control by age and to examine whether patient characteristics are predictive of these patterns. We studied 6720 patients with mild-to-moderate AD who had used pimecrolimus for >6 weeks. They were enrolled by physicians from across the United States and subsequently followed via biannual mail surveys. The primary outcome used to define patterns of disease control was complete control without treatment (yes/no) for each 6-month period. This was derived from 2 repeating questions about disease control and treatment use. Latent class analysis was used to identify patterns of disease control. We included age at survey as a predictor of the outcome, and we controlled for age at cohort enrollment. As predictors of the latent class, we included all the covariates available at baseline hypothesized a priori to explain variation in persistence: sex, race, baseline family income, history of atopy, and age at disease onset. After performing the latent class analysis, we compared the proportion of individuals in each subgroup with filaggrin and filaggrin-2 null mutations. These were not used as predictors in the model because they were only available for a subset of the population who sent biosamples. In addition, we compared history of treatment use, history of allergies, and history of smoke exposure at baseline between the subgroups from the model. Additional details regarding the methods are available in this article's Online Repository at www.jacionline.org. A total of 65,237 surveys (mean, 10.4) were returned by 6,270 racially diverse participants ages 2 to 26 years, comprising 33,074 person-years of follow-up (Table I). The primary outcome, complete control without treatment, was reported on 9% of all surveys.Table IDemographic characteristics of cohort and results of latent class analysisPatient characteristics included in the modelClass 1Class 2OR (95% CI)∗Odds of membership in the "persistently active" class from the 2-class model that included age at survey and age at enrollment as predictors of outcome and all of the listed patient characteristics as predictors of class.OverallResolvingPersistently activeN = 6270n = 702n = 3905n (%)%%Female sex6270 (53)43551.63 (1.35-1.95)Income <$50,000/y4639 (75)57781.69 (1.37-2.08)Nonwhite race6269 (65)47732.38 (1.94-2.91)History of atopy4682 (75)72751.36 (1.10-1.67)Age at eczema onset (y), mean ± SDn = 6239, 2.3 ± 3.02.9 ± 2.52.0 ± 2.90.91 (0.88-0.93)Additional patient characteristics compared by model classificationP value†Pearson chi-square or t-test.Genetics Any filaggrin null mutation829 (17)1816.596 Any filaggrin-2 null mutation337 (45)3646.294Mean proportion of visits with frequent treatment use before age 8 Topical calcineurin inhibitor4010 (80)5785.000 Topical steroids4000 (54)3756.000 Any prescription med for eczema4010 (91)6996.000History of allergies at enrollment Animals6262 (22)1823.003 Foods6265 (24)1825.000 Meds6264 (12)1311.135Smoker in the home at enrollment Yes6213 (16)1115.001OR, Odds ratio.∗ Odds of membership in the "persistently active" class from the 2-class model that included age at survey and age at enrollment as predictors of outcome and all of the listed patient characteristics as predictors of class.† Pearson chi-square or t-test. Open table in a new tab OR, Odds ratio. Latent class analysis suggested there were 2 subgroups with clinically distinct patterns of disease control: a group more likely to report periods of complete control without treatment as they aged (termed "resolving") and a group more likely to report "persistently active" disease past childhood (Fig 1). History of atopy, female sex, family income under $50,000/year at enrollment, nonwhite race, and younger age at eczema onset were significant predictors of membership in the "persistently active" subgroup (Table I). Blacks made up nearly half of the cohort, and the results of sensitivity analyses using alternate categorizations of race (eg, black vs white) resulted in even higher odds of membership in the "persistently active" subgroup (odd ratio, 3.23, 95% CI, 2.54-4.10) (see Table E2 in this article's Online Repository at www.jacionline.org). Income data was limited to self-reported family income at enrollment and was not adjusted for family size. To address this potential source of measurement error, we conducted a sensitivity analysis without income and found similar results (Table E2). When comparing the classes produced by the model, we found no significant differences in the proportion of patients with filaggrin or filaggrin-2 null mutations. Patients classified into the "persistently active" class were significantly more likely to have reported frequent use of eczema medications at younger ages. Results are presented for treatment before age 8 years; additional comparisons varying the age cut point and found similar results (data not shown). Finally, we found that those in the "persistently active" group were significantly more likely to report a history of animal allergies, food allergies, or presence of a smoker in the home. Using a latent class analysis, which is well suited to identify trajectories of heterogeneous patterns of discrete outcomes, we demonstrate the existence of subgroups of patients with clinically distinct disease courses and found that standard patient characteristics are highly predictive of persistently active disease. These are helpful for clinicians to consider when counseling patients and identifying who may benefit from more intensive follow-up and intervention. This article builds on a prior publication that found similar factors are associated with symptom-free periods at any age, but that publication did not address disease course or answer the crucial and more clinically relevant question of which children are most likely to have persistent disease.2Margolis J.S. Abuabara K. Bilker W. Hoffstad O. Margolis D.J. Persistence of mild to moderate atopic dermatitis.JAMA Dermatol. 2014; 150: 593-600Crossref PubMed Scopus (198) Google Scholar Our findings that female sex and history of allergies are associated with more persistent disease corroborate those from other studies with limited follow-up.3Ballardini N. Kull I. Soderhall C. Lilja G. Wickman M. Wahlgren C.F. Eczema severity in preadolescent children and its relation to sex, filaggrin mutations, asthma, rhinitis, aggravating factors and topical treatment: a report from the BAMSE birth cohort.Br J Dermatol. 2013; 168: 588-594Crossref PubMed Scopus (66) Google Scholar, 4Garmhausen D. Hagemann T. Bieber T. Dimitriou I. Fimmers R. Diepgen T. et al.Characterization of different courses of atopic dermatitis in adolescent and adult patients.Allergy. 2013; 68: 498-506Crossref PubMed Scopus (191) Google Scholar, 5Illi S. von Mutius E. Lau S. Nickel R. Grüber C. Niggemann B. et al.for the Multicenter Allergy Study GroupThe natural course of atopic dermatitis from birth to age 7 years and the association with asthma.J Allergy Clin Immunol. 2004; 113: 925-931Abstract Full Text Full Text PDF PubMed Scopus (609) Google Scholar We also report a number of novel findings. Notably, we found that income and race were strongly associated with AD persistence. In multivariate models, the magnitude of association with income <$50,000/year and black race was larger than the association with history of atopy, previously thought to be among the strongest predictors of AD persistence.6Apfelbacher C.J. Diepgen T.L. Schmitt J. Determinants of eczema: population-based cross-sectional study in Germany.Allergy. 2011; 66: 206-213Crossref PubMed Scopus (79) Google Scholar Our results differ from many studies that have found that higher income is associated with higher AD incidence or prevalence.7Uphoff E. Cabieses B. Pinart M. Valdes M. Anto J.M. Wright J. A systematic review of socioeconomic position in relation to asthma and allergic diseases.Eur Respir J. 2015; 46: 364-374Crossref PubMed Scopus (113) Google Scholar We hypothesize that drivers of disease onset may be different from drivers of disease persistence; additional research is needed to test this hypothesis and to examine how changes in income or socioeconomic status over time relate to eczema disease control. The strengths of our study include a large and diverse cohort with repeated measures of disease control and treatment use. The primary limitation relates to the generalizability of the findings. Because patients were required to use pimecrolimus for ≥6 weeks prior to enrollment, the PEER cohort is likely biased toward more persistent disease (because those who had early clearance would be less likely to have required pimecrolimus at later ages). Therefore the proportion of individuals in each disease class may not be representative of the general AD population. The goal of our study was not to describe how many individuals are likely to have each disease control pattern. Rather, our focus was to identify predictors of each subgroup, which are more likely to be generalizable to AD population seeking treatment. It is also possible that use of pimecrolimus, corticosteroids, or other topical treatments could alter the skin barrier and predispose to more persistent disease, but sensitivity analyses stratifying by history of treatment use during follow-up did not reveal any differences in our results. An additional limitation, common to nearly all large longitudinal studies, is attrition; 35% of individuals were lost to follow-up. There was only a median of 0 missing surveys in sequence (interquartile range, 0-1), and although sensitivity analyses do not exclude the possibility of bias, they did not reveal any difference in the results when we included only individuals with longer-term or full follow-up (Table E2). Our results highlight the importance of characterizing AD as a chronic condition and the necessity of longitudinal studies that permit the analysis of disease control over time. Future research should examine mechanisms for these differences. We would like to acknowledge Luke Keele (PhD, Associate Professor, Georgetown University) for his uncompensated input regarding the statistical analysis. PEER began enrollment in 2004 and follows a diverse sample of US children and young adults with physician-diagnosed AD (based on the UK Working Party CriteriaE1Williams H.C. Burney P.G. Hay R.J. Archer C.B. Shipley M.J. Hunter J.J. et al.The U.K. Working Party's Diagnostic Criteria for Atopic Dermatitis. I. Derivation of a minimum set of discriminators for atopic dermatitis.Br J Dermatol. 1994; 131: 383-396Crossref PubMed Scopus (848) Google Scholar) for up to 10 years. Participants return surveys that include information on self-reported disease control and medication use at 6-month intervals. Subjects were enrolled by >1000 pediatricians, allergists, or dermatologists, and represent all major ethnic groups and geographic areas in the United States. The cohort was designed to test whether there is an increased risk of malignancy associated with the use of pimecrolimus, a topical calcineurin inhibitor, and all participants used 1% topical pimecrolimus cream for at least 42 of the 180 days prior to enrollment. After enrollment, data collection continued via mail surveys and was independent of physician visits and treatment decisions. Subjects were not required to continue to see the enrolling provider or continue using pimecrolimus, and in fact, use declined to a minority of participants over time.E2Kapoor R. Hoffstad O. Bilker W. Margolis D.J. The frequency and intensity of topical pimecrolimus treatment in children with physician-confirmed mild to moderate atopic dermatitis.Pediatr Dermatol. 2009; 26: 682-687Crossref PubMed Scopus (20) Google Scholar Participants were excluded if, at baseline, they had a history of lymphoproliferative disease, systemic malignancy, skin malignancy, or had used oral immunosuppressive medications such as cyclosporine, tacrolimus, or methotrexate. For this analysis, we included data from all subjects with ≥1 returned survey after the baseline questionnaire through December 2015. The sample size is larger than in prior publications because PEER is an ongoing study. All study procedures were approved by the institutional review board at the University of Pennsylvania (no. 706073). All participating patients (or their guardians) signed an informed consent form. The primary outcome we used to define patterns of disease control was a repeated composite variable: complete control without treatment (yes/no) for each 6-month period. It was derived from 2 survey questions: "During the last 6 months, would you say your (or your child's) skin disease has shown: complete disease control, good disease control, limited disease control, or uncontrolled disease?" and "Have you used a prescription cream or ointment to treat this child's eczema during the last 6 months?" These questions were chosen because they are easily interpretable, patient-centered, and comprehensive. The level of disease control question has been used in randomized controlled trials and has been shown to correlate with a well-validated objective severity measure, the Eczema Area Severity Index (Spearman correlation coefficient at 6 months, 0.794).E3Kapp A. Papp K. Bingham A. Fölster-Holst R. Ortonne J.P. Potter P.C. et al.for the Flare Reduction in Eczema with Elidel (Infants) Multicenter Investigator Study GroupLong-term management of atopic dermatitis in infants with topical pimecrolimus, a nonsteroid anti-inflammatory drug.J Allergy Clin Immunol. 2002; 110: 277-284Abstract Full Text Full Text PDF PubMed Scopus (237) Google Scholar, E4Barbier N. Paul C. Luger T. Allen R. De Prost Y. Papp K. et al.Validation of the Eczema Area and Severity Index for atopic dermatitis in a cohort of 1550 patients from the pimecrolimus cream 1% randomized controlled clinical trials programme.Br J Dermatol. 2004; 150: 96-102Crossref PubMed Scopus (127) Google Scholar Examples of individual participant response patterns are shown in Fig E1. Covariates included sex, race, income, history of atopy, and age of disease onset. Race was reported as ≥1 of the following: American Indian, Asian, black or African American, Hawaiian or other Pacific Islander, or white; and for the primary analysis it was modeled as a binary variable: white only (yes/no). Family income at baseline was reported in 6 categories, and for the primary analysis was modeled as >$50,000/year or <$50,000/year based on the distribution of responses. Data on income at disease onset and household size were not collected. Self-reported age at disease onset was modeled as a numeric variable. History of atopy was defined as a personal history of asthma or seasonal allergies if ≥4 years of age, or a family history of asthma, seasonal allergies, or AD in a parent or sibling if age <4 years. For a subset of the population who sent biosamples, filaggrin (R501X, 2282del4, R2447X, and S3247X) and filaggrin-2 null mutations (rs1256874 and rs1683397) were genotyped using custom-made TaqMan allelic discrimination assays (Applied Biosystems, Foster City, Calif).E5Margolis D.J. Apter A.J. Mitra N. Gupta J. Hoffstad O. Papadopoulos M. et al.Reliability and validity of genotyping filaggrin null mutations.J Dermatol Sci. 2013; 70: 67-68Abstract Full Text Full Text PDF PubMed Scopus (6) Google Scholar, E6Margolis D.J. Papadopoulos M. Apter A.J. McLean W.H. Mitra N. Rebbeck T.R. Obtaining DNA in the mail from a national sample of children with a chronic non-fatal illness.J Invest Dermatol. 2011; 131: 1765-1767Abstract Full Text Full Text PDF PubMed Scopus (9) Google Scholar Duration of follow-up was calculated from the date of enrollment to the last available survey date. Loss to follow-up was defined as an 18-month period without a returned survey. Using the generalized linear latent and mixed model framework, we performed a repeated measures latent class analysis to identify subgroups of individuals within the cohort who had different patterns of disease control over time.E7Skrondal A. Rabe-Hesketh S. Generalized latent variable modeling: multilevel, longitudinal, and structural equation models. Chapman & Hall/CRC, Boca Raton (FL)2004Crossref Google Scholar Each latent class is associated with a characteristic vector of responses. We chose this method because it accommodates change that is discrete or discontinuous (as is the case with an episodic condition like AD) and allows for heterogeneity in patterns among subgroups.E8Collins L.M. Lanza S.T. Latent class and latent transition analysis: with applications in the social behavioral, and health sciences. Wiley, Hoboken (NJ)2010Google Scholar It also has the capacity to accommodate differential missingness and covariate-dependent missingness at random. We fit latent class models with 2 to 5 classes. To determine the optimal number of classes, we examined model fit statistics including the log likelihood and Bayesian information criterion and considered the homogeneity, separation, and size of classes.E9Nylund K.L. Asparoutiov T. Muthen B.O. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study.Struct Equ Modeling. 2007; 14: 535-569Crossref Scopus (6291) Google Scholar As we added more classes to the model, the statistical fit improved and we found that a 4-class model had the lowest Bayesian information criterion. Specifying models with larger numbers of classes created intermediate groupings between the "persistently active" and "resolving" groups, but these represented a small proportion of the total population (3% to 14%) (see Table E1 and Fig E2). Therefore, we present detailed results for the 2-class model because it is the most parsimonious and clinically meaningful. We explored various options for modeling time, including whether there were any inflection points in the disease course, and we found that a simple linear variable using the age at survey (rounded to the nearest half-year) allowed for good model fit and interpretability. We included age at survey as a predictor of the outcome in our model and controlled for age at cohort enrollment. As predictors of the latent class, we included all the covariates available at baseline hypothesized a priori to explain variation in persistence: sex, race, baseline family income, history of atopy, and age at disease onset. We ran binary logistic latent class models using repeated measures for a 2-level measure of disease control (complete control without treatment vs all other levels of control) and ordinal logistic latent class models using a repeated measure with all 5 levels of disease control (poor, limited, good, complete control with treatment, and complete control without treatment) and found similar results. The binary logistic models were chosen for presentation because they are more easily interpretable. We indicated the number of missing variables in Table I and performed sensitivity analyses using subsets of the cohort without missing data and alternative categorizations of race. After performing the latent class analysis, we compared the proportion of individuals in each subgroup with filaggrin or filaggrin-2 mutations. Genetic data were not used as predictors in the model because they were only available for a subset of the data. Instead, using the model results, we compared the proportions of individuals with null mutations in each class/subgroup of individuals. In addition, we compared history of treatment use, history of allergies, and history of smoke exposure at baseline between the subgroups from the model. All analyses were performed with Stata software (version 14.1, StatCorp, College Station, Tex), and we followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for reporting of observational studies.E10Vandenbroucke J.P. von Elm E. Altman D.G. Gøtzsche P.C. Mulrow C.D. Pocock S.J. et al.for the STROBE InitiativeStrengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration.Epidemiology. 2007; 18: 805-835Crossref PubMed Scopus (1388) Google ScholarFig E2Graphical comparison of results from 2 to 5 class models. Epanechnikov kernel-weighted local polynomial smoothed plots with 95% CIs in gray. The x-axis represents age in years, and y-axis represents the proportion of the cohort reporting complete control without treatment at each age. As more classes were added to the model, the "resolving" and "persistently active" subgroups, which included over 75% of the population, remained similar.View Large Image Figure ViewerDownload Hi-res image Download (PPT)Table E1Results of 2 to 5 class models from latent class analysis2-Class model3-Class model4-Class model5-Class modelModel fit statistics Log likelihood−10,875−10,279−10,148−10,108 Degrees of freedom11192735 BIC21,86920,76420,58920,595Distribution of individuals among subgroupsn (%)n (%)n (%)n (%)Persistently active class3,905 (85)3,622 (79)3,344 (73)3,343 (73)Intermediate class AN/A592 (13)615 (13)629 (14)Intermediate class BN/AN/A377 (8)276 (6)Intermediate class CN/AN/AN/A151 (3)Resolving class702 (15)393 (9)271 (6)208 (5)Odds of subgroup membership by patient factorOR (95% CI)OR (95% CI)OR (95% CI)OR (95% CI)"Persistently active" vs "Resolving" class Female1.63 (1.35-1.95)1.73 (1.36-2.19)1.68 (1.24-2.27)1.91 (1.30-2.79) Income <$50,000/y1.69 (1.37-2.08)1.68 (1.28-2.21)1.74 (1.23-2.45)1.55 (1.03-2.34) History of atopy1.36 (1.10-1.67)1.55 (1.19-2.02)1.76 (1.27-2.46)2.07 (1.40-3.07) Nonwhite race2.38 (1.94-2.91)2.94 (2.25-3.84)3.57 (2.54-5.01)4.24 (2.80-6.42) Age of onset (y)0.91 (0.88-0.93)0.88 (0.85-0.91)0.79 (0.75-0.85)0.81 (0.75-0.88)"Intermediate class A" vs "Resolving" class Female1.19 (0.88-1.62)1.47 (1.04-2.07)1.63 (1.08-2.46) Income <$50,000/y0.78 (0.55-1.10)0.97 (0.66-1.42)0.90 (0.57-1.43) History of atopy1.19 (0.85-1.66)1.28 (0.88-1.85)1.45 (0.93-2.24) Nonwhite race1.62 (1.14-2.28)1.67 (1.14-2.45)1.99 (1.26-3.14) Age of onset (y)0.97 (0.93-1.01)0.96 (0.91-1.01)0.98 (0.92-1.04)"Intermediate class B" vs "Resolving" class Female0.92 (0.61-1.39)1.24 (0.70-2.20) Income <$50,000/y0.77 (0.48-1.22)0.84 (0.45-1.55) History of atopy1.20 (0.77-1.89)1.21 (0.68-2.15) Nonwhite race1.71 (1.08-2.71)2.18 (1.18-4.04) Age of onset (y)0.97 (0.92-1.03)1.04 (0.97-1.11)"Intermediate class C" vs "Resolving" class Female1.03 (0.57-1.85) Income <$50,000/y0.45 (0.23-0.89) History of atopy2.42 (1.18-4.98) Nonwhite race1.73 (0.87-3.44) Age of onset (y)1.00 (0.92-1.09)BIC, Bayesian information criterion; N/A, not applicable. Open table in a new tab Table E2Results of sensitivity analyses for 2-class latent class analysisSensitivity analyses exploring race categorizationClass 1 "resolving"Class 2 "persistently active"OR (95% CI)∗Odds of membership in the "persistently active" class from a 2-class model that included age at survey and age at enrollment as predictors of outcome and all of the listed patient characteristics as predictors of class.Race modeled as any black vs any white (including Hispanic and mixed) No. in each subgroupn = 629n = 3628 Female sex45551.51 (1.25-1.84) Income <$50,000/y58781.52 (1.22-1.89) History of atopy74751.19 (0.95-1.48) Any black35652.83 (2.27-3.53) Age at eczema onset (y), mean2.92.00.90 (0.87-0.92)Race modeled as black only vs white only No. in each subgroupn = 573n = 3268 Female sex44551.53 (1.25-1.88) Income <$50,000/y56781.44 (1.14-1.83) History of atopy76761.11 (0.88-1.41) Black only34673.23 (2.54-4.10) Age at eczema onset (y), mean2.92.00.90 (0.87-0.93)Sensitivity analyses exploring missing and limited variablesModeled without income variable (because only available at enrollment, not adjusted for family size, and 26% had preferred not to answer) No. in each subgroupn = 1015n = 5215 Female sex45551.47 (1.25-1.71) History of atopy40701.26 (1.05-1.51) Nonwhite race74753.50 (2.99-4.09) Age at eczema onset (y), mean3.12.10.90 (0.88-0.92)Including only those not lost to follow-up (ie, survey within the last 18 mo) No. in each subgroupn = 562n = 2413 Female sex44551.55 (1.26-1.90) Income <$50,000/y55761.75 (1.39-2.21) History of atopy45711.31 (1.04-1.65) Nonwhite race72762.27 (1.8-2.85) Age at eczema onset (y), mean2.82.00.90 (0.87-0.93)Including only those with follow-up over 5 y No. in each subgroupn = 522n = 2130 Female sex43551.62 (1.32-2.01) Income <$50,000/y49701.79 (1.42-2.26) History of atopy41671.41 (1.12-1.78) Nonwhite race71762.29 (1.82-2.89) Age at eczema onset (y), mean2.71.90.91 (0.88-0.94)Values are percentages unless otherwise indicated.∗ Odds of membership in the "persistently active" class from a 2-class model that included age at survey and age at enrollment as predictors of outcome and all of the listed patient characteristics as predictors of class. Open table in a new tab BIC, Bayesian information criterion; N/A, not applicable. Values are percentages unless otherwise indicated.