Association Between Atopic Dermatitis and Educational Attainment in Denmark

Key Points Question Is atopic dermatitis (AD) associated with lower educational attainment? Findings This nationwide matched cohort study of 61 153 Danish children found evidence of reduced attainment of lower and upper secondary education in 5927 children with a hospital diagnosis of AD compared with 55 226 matched children from the general population; however, absolute differences were less than 3.5%. Estimates were less pronounced in a secondary analysis comparing patients with their full siblings. Meaning Hospital-diagnosed AD is associated with reduced educational attainment, but the clinical importance is uncertain owing to small absolute differences and possible confounding by familial factors.


Study Population 28
The study will include an exposed cohort of persons with atopic eczema and two comparison 29 cohorts with matched persons from the general population and full-siblings, respectively. The 30 Danish Health Data Authority has already performed the initial sampling of these cohorts, which 31 are described in detail below. 32 33

Eczema exposed 34
The Danish Health Data Authority has sampled an exposed cohort, using the following criteria: For the current study, we will restrict the study population defined by the Danish Health Data 46 Authority as above further as follows:  We will exclude eczema patients (together with their matched comparators) included by 48 means of an ICD-8 diagnosis recorded together with an additional ("modification") code 49 indicating that it potentially an uncertain/unverified/working diagnosis (c_diagmod=1-7). 50  We will exclude patients (together with their matched comparators) with inconsistencies in 51 registration of status and residence in the Civil Registration System (recorded as alive but 52 not living in Denmark but without an emigration date), as we assume that these patients are 53 not living in Denmark and thus not eligible. 54  We will limit to patients who are diagnosed with eczema prior to age 13 years and are born 55 on 30 June 1987 or earlier. These eligibility criteria will ensure a minimum possible attained 56 age of 30 years at end of follow-up (30 June 2017). In a sensitivity analysis, we will require 57 that persons are born on 30 June 1982 or earlier, to ensure a higher minimum attained 58 age (35 years) to account for the fact that some persons may not finish their final education 59 until after age 30 years. If we experience problems with power in the main analysis, we 60 will change the requirement for the highest possible attained age to 26 year. 61

Unexposed general population cohort (for main analysis) 62
The Danish Health Data Authority has sampled an unexposed general population cohort, as follows: 63  Up to 10 persons matched to each eczema patient by sex and birth year. 64  Comparators have to: (1) be born in Denmark, (2) be alive and living in Denmark on the 65 eczema diagnosis date of their matched eczema-exposed individual, and (3) have no 66 previous diagnosis of eczema. 67  A person with atopic eczema can be included in the comparison cohort until he/she is 68 diagnosed with eczema (and included in the exposed cohort). 69

70
As with eczema patients, we will additionally apply the following restrictions: 71 We will exclude matched comparators with inconsistencies in registration of status and residence in 72 the Civil Registration System (recorded as alive but not living in Denmark but without an 73 emigration date). 74 Persons in the atopic eczema cohort are eligible to be selected as comparators from birth until 75 diagnosis of eczema. If a person from the comparison cohort is diagnosed with eczema during We will follow the cohort to determine the highest level of educational achievement by age 30 108 programs/activities in four dimensions (main area; type of education; education level (variable 121 NIVEAU); and subject area). The main area dimension follows that of the Danish education 122 system. 9 The first two digits specifies the overall group of education programs, e.g. basic 123 schooling. It has no association with the ISCED. Education type has a similar structure. 124 The ISCED 10 was developed in 1976 by UNESCO with the aim to perform international 125 comparisons within education. It was revised in 1997 and 2011. The structure of ISCED is 126 developed in a collaboration between UNESCO, OECD, EU and all member states. It consists of a 127 7-digit code, where the first level digit describes the level/education level. 128 In our study, we defined study categories of the highest completed education based on the 129 main groups of the Danish nomenclature for education. These main areas correspond largely to 130 levels of ISCED-2011 with some minor differences for included subgroups, namely that: (1) We 131 include information on type of youth education and (2) Medium cycle higher education programs in 132 Denmark include "Professional bachelor's degrees" which internationally are classified as ISCED 133 level 6 ('Bachelor or equivalent') together with Univeristy Bachelor's degree. The table below  134 shows the categories for the current study, including registry codes, corresponding ISCED-2011 135 levels, a text description with examples of the educational activities/programs and jobs that it can 136 lead to, and approximate length of education.
For the analysis, we will generate a number of binary/dummy variables (one for each 138 education level) coded as "1" for those not achieving/failing to achieve that specific education level 139 and "0" for those who have achieved the given level. 140 Each main educational level is conditional on completing preceding levels. That is, youth 141 education is conditional on completing basic schooling and higher education is conditional on youth 142 education. To account for this and for changes in covariables over time, we will define three cohorts 143 for each of the main education levels: (1) "Main cohort": For lower secondary education as the 144 outcome, we used the main cohort of patients with AD and comparators described above and 145 baseline date was the 13 th birthday; (2) "Lower secondary education cohort": For upper secondary 146 education as the outcome, we restricted the study sample to include only those from the main cohort 147 who attained lower secondary education prior to or at age 30 years and baseline was the time for 148 graduation from upper secondary school; (3) "Upper secondary education cohort": For higher 149 education as the outcome, we restricted the sample further to include those who attained upper 150 secondary education prior to or at age 30 years and baseline was the time for graduation from upper 151 secondary school. Covariables will be redefined at baseline for each cohort. Comparators who had 152 eczema diagnosed before or on the baseline for each cohort were excluded. In the sibling analysis, 153 we will have corresponding three cohorts and will only include those who have no record of eczema 154 (diagnosis or treatment) at baseline for each cohort. 155 We will presume that no record of a given education level means that the specific level has 156 not been achieved. Thus, if none of the education levels are recorded, we will assume they have not 157 finished lower secondary school (i.e., basic schooling). We will exclude persons with non-158 consecutive recording of education, e.g. those who have lower secondary and higher education but 159 not upper secondary education recorded, as we presume they have missing data and thus cannot 160 follow them complete for the outcomes. A minor proportion of persons also may have died or 161 emigrated by age 30 years and these persons will be classified with the highest education level 162 achieved at the time of loss to follow-up. 163 Lower secondary education* "10" "10" 2 (Lower secondary education)

Definition of educational achievement 165
Lower secondary education, which is compulsory in Denmark.

Secondary analysis incorporating age in the education level 178
With the main outcome definition, we examine if a person has achieved a certain educational level 179 at age 30 years. This definition may not take into account that some persons may complete a 180 specific education level but need more time to do so (i.e., the rate of achievement is lower). In a 181 secondary analysis, we will aim to look into this in more detail by analyzing prevalence of each 182 education level by age for eczema exposed and the comparison cohort. 183 184 185 Covariables 186 We will define various variables for the main and sensitivity analyses, as shown in the following 187  1976-1980, 1981-1985, 1986-1990, 1991-1995, 1996   Population) and define the three cohorts. We will note the number of persons excluded at each 209 stage in this process, so that we can make a study flowchart. Also, we will look at basic 210 descriptive statistics (calendar period, age, and sex and index date) for persons who are 211 excluded because they have non-consecutive records of education. 212 2. We will compute summary statistics for covariables (at baseline for each cohort) for exposed 213 (eczema) patients and comparison cohorts. This will make Table 1  3. We will examine how many persons are lost to follow-up by age 30 years overall and for 218 eczema patients and comparison cohorts. This will make table 2.  sex, calendar period of index date). These results will make Table 3. 232

233
Regarding choice of regression model, we wanted to estimate risk ratios rather than odds 234 ratios because the outcome was common. We originally considered using log-binomial 235 regression rather than logistic regression. 11 However, we experienced issues with accounting 236 properly for the matching of AD patients and general population comparison cohort in the 237 main analysis and family in the sibling analysis. We therefore finally chose the conditional 238 Poisson regression as our model, as recommended by Cummings. 12 239 240 We will also present the prevalence for each education level graphically by age for the 248 eczema cohort and the comparison cohort. 249 250 Figure 1. Prevalence of each education level by age for eczema exposed and comparators 251 Secondary analyses (sibling comparison) 253 6. We will repeat the analyses (descriptive analyses, prevalence estimates and conditional 254 Poisson regression) using the sibling comparison cohort. We will condition on family to 255 ensure within-family comparisons in the unadjusted model. Then we will additionally adjust 256 for age at baseline and sex. This will make tables 4-6 and figure(s) for prevalence by age. 257  1976-1980 1981-1985 1986-1990 1991-1995 1996-2000 Birth year 1973-1977 1978-1982 1983-1987

Secondary analyses (subgroup/stratified analyses) 277
Note: Stratified and sensitivity analyses will be performed for overall education levels only. 278 279 7. We will explore whether the association between atopic dermatitis and educational 280 achievement differ by age. This will make Tables 7a and 7b. 281 . For the sibling comparison, we will explore whether the association between atopic 315 dermatitis and educational attainment differ by parents income and educational level (both 316 defined below). This will make Tables 9a and 9b. 317

Parental income
Categorized based on quartiles of the variable PERINDKIALT_13 (during 1987-) and PERINDKIALT (1980PERINDKIALT ( -1986 using the income for the parent with the highest income. Parental educational level at index date As defined for outcome (lower secondary education, upper secondary education, higher education) using the education level for the parent with the highest education. 11. We will repeat analyses after restricting further to patients who were born on 30 June 1982 351 or earlier to be able to determine outcome at age 35 years instead of 30 years. This will 352 make table 11. 353 12. We will repeat analyses after excluding individuals in both cohorts who were born preterm, who had a 5-min Apgar score <7 or 360 intrauterine/birth asphyxia, who had low birth weight or chromosomal abnormalities. This will make table 12. 361 362 Sensitivity analyses pertaining to the sibling design specifically 369 13. We will repeat the main analysis based on the subset of eczema patients (and their comparators) who are also included in the sibling 370 analyses (i.e., eczema patients who had at least one sibling in the dataset), in order to compare the results from the analysis for the 371 matched general comparison cohort and the sibling cohort. This analysis examines the assumption that results for sibling 372 comparisons will generalize to other samples (e.g., families with only one child, families without variability in the outcome). I.e., to 373 ensure that any difference between main and sibling comparisons, is not due to the exclusion of single-offspring families. This will 374 make Table 13. 375 14. The sibling design makes it possible to adjust for confounding by measured and unmeasured family-related factors, which are stable. 385 However, some factors of interest, e.g., parental income, parental educational level, and family structure (living with single parent), 386 may possible change over time and differ between siblings. In order to address this potential limitation, we will repeat the analysis 387 for siblings with an age difference of 3 years or less. This also limits any confounding by differences in calendar period (in case 388 there are any political changes with impact on educational attainment during the study). This will make 15. As atopic eczema aggregates within families, there is a greater risk of misclassification of eczema as non-eczema in the sibling 397 analysis. We will therefore do a sensitivity analysis where we also require that siblings have no prior prescription record (available 398 since 1995) of a topical steroid/calcineurin inhibitor (Prescription Registry ATC codes: "D07" "D11AH01" "D11AH02") at 399 baseline. This will make table 15. 400 401 examining the risk of not attaining vocational education would be a mix of persons who do not attain any type of upper secondary 419 education and persons attaining general upper secondary education. Thus, it because difficult to tell if any association reflects a failing level 420 of education or choosing another type of education. As we overall found no substantial association between eczema and the educational 421 outcomes, and because adjustments had limited impact on RR, we therefore chose a simple approach where we computed the probability of 422 subtypes of upper secondary education and higher education in children with atopic dermatitis, the matched general population comparison 423 cohort, and cohort of full-siblings.