Social Disparity and Treatment-Seeking in Eating Disorders During Childhood and Adolescence: A Nationwide Population-Based Cohort Study

Abstract

Objective

Advantaged socioeconomic circumstances are linked to a higher risk of diagnosed eating disorders (EDs). However, evidence on non-diagnosed cases, that is, individuals reporting ED symptoms without formal diagnosis, remains limited, suggesting detection and treatment-seeking barriers. In this study, we assessed associations between socioeconomic circumstances and clinically diagnosed and non-diagnosed EDs in childhood and adolescence.

Method

This study used data from the Danish National Birth Cohort, linking registry-based diagnosed EDs (participant ages 6-18 years) with self-reported ED symptoms at age 18 years. Data included 523,148 individuals born in Denmark between 1996 and 2003, with 44,552 providing information on self-reported ED symptoms used to classify non-diagnosed EDs. Socioeconomic indicators, including parental education and household income at age 6 years, were analyzed using quasi-Poisson and multinomial logistic regression to estimate incidence rate ratios (IRRs) and relative risk ratios (RRRs).

Results

Diagnosed EDs were more common among individuals with higher parental education, following a stepwise pattern. Compared to short education (∼12 years), long education (∼17-20 years) was associated with higher risk (IRR = 1.35, 95% CI = 1.21-1.52), whereas basic education (∼10 years) was linked to lower risk (IRR = 0.70, 95% CI = 0.59-0.83). The pattern reversed for non-diagnosed EDs, with individuals whose parents had basic education showing a higher risk (RRR = 2.19, 95% CI = 1.83-2.63) compared to individuals whose parents had short education. No clear pattern was observed for household income.

Conclusion

This study highlights socioeconomic disparities in ED diagnosis and treatment. Individuals from educationally advantaged backgrounds were more likely to receive a diagnosis, whereas those from educationally disadvantaged backgrounds had a higher risk of non-diagnosed EDs.

Study registration information

Parental Social Disparity and Adolescent Eating Disorders; https://osf.io/dm24e/.

Plain language summary

Eating disorders represent serious mental health conditions, but not all young people with symptoms receive a diagnosis. Using data from more than half a million children in Denmark, this study found that adolescents from more advantaged families were more likely to be diagnosed, while those from disadvantaged families often had symptoms without a diagnosis. These results highlight social inequities in recognition of symptoms and access to care, even within a universal health system.

Eating disorders (EDs) are serious psychiatric illnesses marked by chronicity, relapse, distress, functional impairment, and increased mortality.^1,2 Many individuals with EDs never seek treatment, and those who seek treatment have often been ill for years before doing so.3, 4, 5 This presents a particular challenge, as the sooner an ED is identified and treatment initiated, the better the prognosis for full recovery.^2,6 According to a systematic review, including data up to 2018 on EDs fulfilling diagnostic criteria, the lifetime prevalence of any ED is 8% for women and 2% for men, with increasing point prevalences across time.⁷ Data from the Danish registers also show an increasing incidence of diagnosed EDs over calendar time and a higher incidence for more recent birth cohorts.⁸ Adolescence is a critical period for the onset of EDs, with about half of all lifetime cases being diagnosed before the age of 19 years.^1,9 According to the Global Burden of Disease study, EDs are the mental disorder with the largest increase in prevalence and years lived with disability, and the only disorder for which there has been an increase in years of life lost among young people in Europe since 1990.¹⁰ Studies based on diagnosed cases of EDs, including nationwide register-based studies, show a higher incidence among adolescents from advantaged socioeconomic circumstances (SECs), in contrast to other mental illnesses, which are more prevalent among individuals from disadvantaged SECs.11, 12, 13, 14 Here, SECs refer to parental socioeconomic position, such as education and household income, reflecting the socioeconomic environment in which children and adolescents are raised.¹⁵ Although EDs are often perceived as more common in advantaged SECs, this pattern largely reflects treatment-seeking, with individuals from more advantaged SECs more frequently identified and treated, even though EDs are present across all socioeconomic groups.¹⁶ Population surveys that collect self-reported information on EDs in populations outside clinical treatment suggest that EDs are more common than once believed,^2,17, 18, 19 including in children and adolescents.²⁰ However, only about 20% of affected individuals present for treatment.3, 4, 5^,21, 22, 23

Denmark has a universal tax-funded health care system in which treatment, including specialized secondary care in hospitals, is provided free at the point of use.²⁴ This makes it an optimal setting to examine treatment barriers, as financial constraints are presumed to be less important in Denmark compared to many other countries. An additional advantage of this system is the existence of comprehensive administrative registers containing individual-level data on all recorded treatments.²⁴ This enables the identification of all patients with a hospital contact for an ED diagnosis, allowing for nationwide studies of diagnosed ED cases.

In this study, we linked these administrative data to a nationwide cohort followed from prenatal life, with self-reported ED symptom data collected at age 18 years. We aimed to investigate whether SECs were associated with the risk of diagnosed EDs in childhood and adolescence, and to quantify whether these associations were similar for self-reported ED symptoms, that is, non-diagnosed EDs.

We hypothesized that children and adolescents from disadvantaged SECs would have a higher risk of non-diagnosed EDs than those from advantaged SECs, reflecting gaps in detection and treatment-seeking.

Method

The preregistered study protocol is available on Open Science Framework.²⁵

Study Design and Population

This register-based cohort study includes all individuals born in Denmark between 1996 and 2003. The study combines national register data with self-reported data collected at age 18 years and 3 months in a subset of participants from the Danish National Birth Cohort (DNBC).²⁶ The DNBC is a population-based cohort established between 1996 and 2002, during which approximately 100,000 pregnant women were recruited. ED symptoms were assessed during the 18-year follow-up (DNBC-18), which includes 44,552 participants. The cohort began as a pilot project in one Danish county in 1996 and expanded to nationwide recruitment in early 1999. To define the DNBC background population, we accounted for this stepwise expansion by including all children whose mothers were eligible for invitation during the recruitment period (Figure 1). This background population forms a subset of all children born in Denmark between 1996 and 2003 and was used to construct sample weights for analyses involving the DNBC-18.

Figure 1.

Flowchart of the Danish National Birth Cohort (DNBC) as a Nested Subset of All Individuals Born in Denmark Between 1996 and 2003

Note:Box sizes are not proportional to population size. The total population comprises all children born in Denmark in 1996 to 2003. The background population comprises children born in the same years who lived in areas with active recruitment and were eligible for enrollment in the Danish National Birth Cohort (DNBC)²⁶; this group served as the reference population for inverse probability weighting.

In Denmark, all residents are assigned a unique personal identification number, enabling individual-level linkage to various administrative registers.²⁷ This has allowed for the integration of DNBC data with register-based information on health characteristics, socioeconomic indicators, and ED diagnoses, creating a comprehensive dataset for analysis.

For this study, we restricted the sample to children who survived to age 6 years, resided in Denmark at some point between age 6 and 18.25 years, and had complete information on covariates, ED items, height, and weight. This final register-based cohort comprised 523,148 individuals (total population), of whom 44,552 had self-reported ED symptom information (DNBC-18).

Socioeconomic Circumstances

SECs were measured using parental education and household income at age 6, obtained from the Attainment Register and the Income Statistics Register. Parental education was defined as the highest level of education attained by either parent. Education levels were categorized according to International Standard Classification of Education (ISCED) 2011 classifications^28,29 and grouped into 4 categories, as done in previous research¹³: (1) basic (primary and lower secondary education, ∼10 years); (2) short (upper secondary education, ∼12 years); (3) medium (tertiary education, including a bachelor’s degree, ∼15 years); and (4) long (master’s degree or PhD, ∼17-20 years) (Table S1). Household income was derived from household disposable income, adjusted for family size, and categorized into quartiles based on the total population for the corresponding year. In cases in which data on the highest attained parental education or household income at the child’s age of 6 years were missing, information from the birth year was used. The highest parental education level at birth included ongoing education, recognizing that many individuals have children before completing their formal education.

Outcomes

For the ED outcome, we used 2 complementary sources: (1) clinically diagnosed EDs recorded in national registers between ages 6 and 18 years; and (2) self-reported ED symptoms at age 18 years, collected in the DNBC-18, referring to eating behaviors and concerns during the past year.

Diagnosed Eating Disorders

Information on clinical diagnosis from secondary care treatment was sourced from the Psychiatric Central Research Register and the National Patient Register using the International Classification of Diseases, Tenth Revision (ICD-10) codes, including both primary and secondary ICD-10 diagnoses of anorexia nervosa (AN) (F50.0, F50.1), bulimia nervosa BN (F50.2, F50.3), and eating disorder not otherwise specified (EDNOS) (F50.8, F50.9). We obtained data from the National Patient Register up to 2021 and included all hospital contacts from both private and public hospitals. As binge eating disorder (BED) is not formally classified in ICD-10, it could not be captured through register-based diagnoses. Consequently, BED may be recorded as EDNOS (F50.8-F50.9), and self-reported BED symptoms are therefore not expected to correspond directly to register-based ED diagnoses. Only diagnoses from age 6 years were included to exclude diagnoses related to early childhood feeding disorders. The admission date of the first in- or outpatient contact with an ED diagnosis was used as the index date, and overall diagnosed EDs were defined as any of the above ED subtypes given at this first contact.

Self-Reported Eating Disorder Symptoms

The DNBC-18 included up to 20 items referring to behaviors and concerns during the past year drawing surveys, including the McKnight Risk Factor Survey and the Youth Risk Behavior Surveillance System.^30,31 These instruments have been shown to have acceptable psychometric properties for assessing eating-related risk factors, weight and shape concerns, and behavioral symptoms such as binge eating and compensatory behavior in adolescent populations.³¹ In addition, selected behavioral items have been validated against interview-based assessments for screening binge eating and purging behaviors.³² EDs were categorized according to adapted DSM-5 criteria, following approaches used in previous population-based studies for AN, BN, and BED (Table 1).^33,34 Consistent with our previous research,³⁵ the definitions were refined in collaboration with a psychiatrist specialized in EDs. Compared with previous studies,^33,34 we aligned the definitions more closely with those of ICD-11 by incorporating BMI thresholds specific to each ED subtype to improve differentiation between diagnostic categories. We also included excessive exercise as a compensatory behavior for BN. For BED, fewer symptom indicators were available, and the definition therefore did not include items related to eating when not hungry or eating alone because of embarrassment.

Table 1.

Self-Reported Eating Disorder Symptoms Definitions in the DNBC-18, Including Anorexia Nervosa (AN) Bulimia Nervosa (BN), and Binge Eating Disorder (BED)

AN	BMI <18.5 and (high weight/shape concern or at least monthly fasting or engaged in excessive exercisea)
BN	BMI ≥18.5 and weekly binge eating and (weekly purgingbor engaged in excessive exercisea)
BED	Weekly binge eating and eating until stomach hurt or felt sick and feeling guilt after binge eating and (absence of purgingband fasting and excessive exercisea)

Note: BMI = body mass index; DNBC-18 = Danish National Birth Cohort, 18-year follow-up.

^aExcessive exercise was defined as weekly exercise to lose or avoid weight gain, combined with: (i) interference with school or work, and (ii) either guilt when skipping exercise or never skipping planned exercise.

^bPurging includes both self-induced vomiting and/or the use of laxatives.

We defined a composite measure of EDs by integrating register-based diagnoses and self-reported ED symptoms into a mutually exclusive categorical variable. Specifically, “diagnosed EDs” included all individuals with any register-based ED diagnosis, regardless of self-reported ED symptoms. “Non-diagnosed EDs” included individuals who met threshold definitions of an ED based on self-reported symptoms in the DNBC-18 but had never received a register-based ED diagnosis, and “no EDs” included individuals with neither a diagnosed nor non-diagnosed ED. In addition, we defined “any EDs” as a binary outcome capturing individuals with either a diagnosed or non-diagnosed ED, with those classified as having no ED serving as the reference group.

Statistical Analysis

Population characteristics were examined by parental education and household income, with variable definitions provided in Table S2. Quasi-Poisson regression models were used to estimate the association between parental education and diagnosed EDs (yes vs no). The effect estimates were presented as incidence rate ratios (IRRs) with corresponding 95% confidence intervals. Risk time was measured from age 6 years until the earliest occurrence of death, emigration, ED diagnosis, or reaching the age of 18.25 years. Household income was analyzed in separate, similar models. Potential confounders included maternal and paternal age at birth, birth year, parental psychiatric diagnosis before the child was 6 years of age, and parental education when the child was 6 years of age (in the model of household income). Confounder selection was guided by a Directed Acyclic Graph (Figure S1) informed by prior evidence.^8,13 Analyses were conducted in the total population, the background population, those invited to participate in DNBC-18, and the DNBC-18 population to examine whether associations varied across populations. In the total population, we additionally adjusted for parental origin as a potential confounder. This adjustment was not feasible in the DNBC-18 population because of the small number of individuals of non-Danish origin.²⁶

To distinguish between diagnosed and non-diagnosed EDs in the DNBC-18 analyses, we applied multinomial logistic regression to the 3-category outcome (ie, diagnosed, non-diagnosed, and no ED), estimating relative risk ratios (RRRs) with corresponding 95% CIs. A likelihood ratio test was conducted to assess whether the 3-category model provided a better fit than the binary “any ED” outcome, modeled with logistic regression to estimate odds ratios (ORs) and 95% CIs. We conducted ED subtype analyses for AN and BN using 3-category outcomes, as done for overall EDs. Individuals could receive both an AN and a BN diagnosis over time and therefore contribute to both subtype-specific analyses. BED was modeled with logistic regression, as only self-reported information was available. Associations with parental education and household income were modeled separately.

We used inverse probability weighting to account for non-random selection into DNBC-18. Propensity scores were estimated using logistic regression in the background population, including all individuals regardless of missing values on predictors (N = 341,170). The outcome was DNBC-18 participation (yes vs no), and the explanatory variables comprised sex, household income, parental education, maternal and paternal age, urbanicity, parity, parental cohabitation, out-of-home placement, and child and parental psychiatric diagnoses. Variable definitions are provided in Table S2. To address missing data in covariates used for inverse probability weighting estimation, we included a distinct category for missing values on each variable. This allowed us to retain individuals with incomplete data and to adjust for differential participation. We stabilized the weights by multiplying them by the marginal probability of participation in DNBC-18 to prevent excessively high weights.

Sensitivity analyses were conducted to assess whether the results changed when restricting the sample to female individuals, given the low number of male ED case individuals, and when stratifying by birth period, comparing those born between 1996 and 1999 and between 2000 and 2003 to investigate potential trends in EDs.⁸ In addition, we performed an analysis excluding individuals with negative tax-reported incomes, as recommended by previous research, as these mainly reflect losses from interest, self-employment, or investments rather than actual low income.³⁶ As a post hoc analysis, we examined the distribution of sociodemographic characteristics among individuals with self-reported ED symptoms, distinguishing between diagnosed and non-diagnosed cases, as well as a diagnosed-only group (ie, diagnosed ED without self-reported symptoms), to explore potential treatment barriers. Finally, we analyzed all individuals with self-reported ED symptoms, regardless of diagnostic status, to assess whether associations differed when considering the entire group.

Results

Diagnosed EDs were present in 1% of both the DNBC-18 and the total population (Figure S2, available online), with AN being the most prevalent subtype (Figure S2a, available online). The incidence of AN and EDNOS peaked in mid-adolescence, whereas BN was diagnosed later, contributing to its lower prevalence (Figure S2b, available online). Non-diagnosed EDs were more than twice as prevalent as diagnosed cases, with BED as the most prevalent subtype. Both diagnosed and non-diagnosed EDs were more frequent in female (1.8% and 4.9%) than in male individuals (0.2% and 0.5%).

The population was evenly distributed between male and female individuals, with more than 90% of the total population and 99% of DNBC-18 having parents of Danish origin (Table 2). The lower representation of adolescents born between 1996 to 1999 in DNBC-18 reflects the absence of nationwide recruitment before 1999. Longer parental education was associated with higher parental age, Danish origin, and the absence of parental psychiatric diagnoses. Shorter parental education was more common among individuals with an unknown father. Similarly, higher education levels were more prevalent in higher income quartiles (Table S3, available online).

Table 2.

Characteristics of the Total Population (N = 523,148) and the Participants in the Danish National Birth Cohort, in the 18-year follow-up (DNBC-18) (N = 44,552) According to Parental Education

	Total			Basic			Short			Medium			Long
	total	DNBC-18		total	DNBC-18		total	DNBC-18		total	DNBC-18		total	DNBC-18
	N (%)	n (%)	%a	n (%)	n (%)	%a	n (%)	n (%)	%a	n (%)	n (%)	%a	n (%)	n (%)	%a
	523,148	44,552		49,443 (9.5)	1,314 (3.0)	8.0	229,431 (43.9)	15,379 (34.5)	42.4	166,495 (31.8)	18,074 (40.6)	33.5	77,779 (14.9)	9,785 (22.0)	16.1
Sexb
Female	254,665 (48.7)	26,122 (58.6)	49.0	24,011 (48.6)	790 (60.1)	50.2	111,874 (48.8)	9,242 (60.1)	49.8	80,915 (48.6)	10,480 (58.0)	48.0	37,865 (48.7)	5,610 (57.3)	48.0
Male	268,483 (51.3)	18,430 (41.6)	51.1	25,432 (51.4)	524 (39.9)	49.8	117,557 (51.2)	6,137 (39.9)	50.2	85,580 (51.4)	7,594 (42.0)	52.0	39,914 (51.3)	4,175 (42.7)	52.0
Maternal age, y b
<25	79,425 (15.2)	3,704 (8.3)	13.8	19,628 (39.7)	407 (31.0)	39.1	44,718 (19.5)	2,068 (13.5)	18.2	12,529 (7.5)	983 (5.4)	7.2	2,550 (3.3)	246 (2.5)	3.5
25-29	189,770 (36.3)	17,106 (38.4)	35.5	14,584 (29.5)	430 (32.7)	29.5	88,726 (38.7)	6,102 (39.7)	36.2	62,770 (37.7)	7,248 (40.1)	37.7	23,690 (30.5)	3,326 (34.0)	32.2
30-34	177,301 (33.9)	16,867 (37.9)	34.7	9,920 (20.1)	314 (23.9)	20.3	69,106 (30.1)	5,131 (33.4)	31.4	63,926 (38.4)	7,044 (39.0)	38.3	34,349 (44.2)	4,378 (44.7)	43.3
35+	76,652 (14.7)	6,875 (15.4)	16.0	5,311 (10.7)	163 (12.4)	11.1	26,881 (11.7)	2,078 (13.5)	14.2	27,270 (16.4)	2,799 (15.5)	16.8	17,190 (22.1)	1,835 (18.8)	21.0
Paternal age, yb
<25	36,113 (6.9)	1,703 (3.8)	6.4	10,210 (20.7)	215 (16.4)	20.2	19,912 (8.7)	958 (6.2)	8.3	5,037 (3.0)	418 (2.3)	3.1	954 (1.2)	112 (1.1)	1.3
25-29	138,745 (26.5)	11,803 (26.5)	25.8	13,473 (27.3)	368 (28.0)	28.2	68,409 (29.8)	4,417 (28.7)	27.4	41,417 (24.9)	4,765 (26.4)	25.1	15,446 (19.9)	2,253 (23.0)	21.4
30-34	190,156 (36.4)	17,622 (39.6)	36.6	12,258 (24.8)	371 (28.2)	24.3	80,294 (35.0)	5,735 (37.3)	35.3	66,068 (39.7)	7,411 (41.0)	39.7	31,536 (40.6)	4,105 (42.0)	39.9
35+	154,048 (29.5)	13,260 (29.8)	30.4	11,967 (24.2)	337 (25.7)	24.6	59,301 (25.9)	4,205 (27.3)	28.2	53,207 (32.0)	5,428 (30.0)	31.7	29,573 (38.0)	3,290 (33.6)	36.7
Unknown	4,086 (0.8)	164 (0.4)	0.8	1,535 (3.1)	23 (1.8)	2.7	1,515 (0.7)	64 (0.4)	0.8	766 (0.5)	52 (0.3)	0.4	270 (0.4)	25 (0.3)	0.7
Parental originb
Danish	475,348 (90.9)	44,312 (99.5)	99.3	37,130 (75.1)	1,293 (98.4)	98.4	208,430 (90.9)	15,313 (99.6)	99.5	156,220 (93.8)	17,981 (99.5)	99.4	73,568 (94.6)	9,725 (99.4)	99.3
Other	47,800 (9.1)	240 (0.5)	0.7	12,313 (24.9)	21 (1.6)	1.6	21,001 (9.2)	66 (0.4)	0.5	10,275 (6.2)	93 (0.5)	0.7	4,211 (5.4)	60 (0.6)	0.7
Birth year
1996-1999	265,643 (50.8)	15,527 (34.9)	35.7	26,721 (54.0)	473 (36.0)	36.3	122,802 (53.5)	5,627 (36.6)	37.0	80,839 (48.6)	6,153 (34.0)	34.7	35,281 (45.4)	3,274 (33.5)	33.6
2000-2003	257,505 (49.2)	29,025 (65.2)	64.4	22,722 (46.0)	841 (64.0)	63.7	106,629 (46.5)	9,752 (63.4)	63.0	85,656 (51.5)	11,921 (66.0)	65.3	42,498 (54.6)	6,511 (66.5)	66.4
Parental psychiatric diagnosisc
None	448,743 (85.8)	39,351 (88.3)	83.9	34,874 (70.5)	902 (68.7)	63.3	194,241 (84.7)	13,169 (85.6)	82.1	148,347 (89.1)	16,228 (89.8)	87.5	71,281 (91.7)	9,052 (92.5)	91.2
One or both	74,405 (14.2)	5,201 (11.7)	16.1	14,569 (29.5)	412 (31.4)	36.7	35,190 (15.3)	2,210 (14.4)	17.9	18,148 (10.9)	1,846 (10.2)	12.5	6,498 (8.4)	733 (7.5)	8.8
Diagnosed EDsd
No	518,955 (99.2)	44,070 (98.9)	99.0	49,180 (99.5)	1,302 (99.1)	99.4	227,723 (99.3)	15,205 (98.9)	98.9	165,037 (99.1)	17,884 (99.0)	99.1	77,015 (99.0)	9,679 (98.9)	98.8
Yes	4,193 (0.8)	482 (1.1)	1.0	263 (0.5)	12 (0.9)	0.7	1,708 (0.7)	174 (1.1)	1.1	1,458 (0.9)	190 (1.1)	0.9	764 (1.0)	106 (1.1)	1.2
Any EDse
Diagnosed		482 (1.1)	1.0		12 (0.9)	0.7		174 (1.1)	1.1		190 (1.1)	0.9		106 (1.1)	1.2
Non-diagnosed		1,226 (2.8)	2.7		64 (4.9)	5.2		438 (2.9)	2.6		464 (2.6)	2.3		260 (2.7)	2.3
No		42,844 (96.2)	96.4		1,238 (94.2)	94.1		14,767 (96.0)	96.4		17,420 (96.4)	96.8		9,419 (96.3)	96.6

Note: DNBC-18 = Danish National Birth Cohort, 18-year follow-up; ED = eating disorder.

^aWeighted estimates for the DNBC-18 population (weighted N = 43,978).

^bMeasured at birth.

^cMeasured up to the child’s age 6 years.

^dDiagnosed between 6 and 18.25 years of age.

^eDiagnosis measured between 6 and 18.25 years of age, and non-diagnosed at age 18.25 years.

The weighting effectively balanced the characteristics, with standardized mean differences below 0.1 (Table S4, available online). Notable differences were observed when comparing the proportion of male individuals and income quartiles between the weighted and unweighted DNBC-18 populations.

Parental Education and Income and Diagnosed Eating Disorders

In the total population, parental education was associated with the risk of diagnosed EDs in a stepwise pattern, with higher education levels associated with higher incidence (Figure 2a; Table S5, available online). Compared to short education, individuals with long parental education had the highest risk (IRR = 1.35, 95% CI = 1.21-1.52), followed by medium education (IRR = 1.18, 95% CI = 1.08–1.30), whereas those with basic education had a lower risk (IRR = 0.70, 95% CI = 0.59-0.83) (Figure 2a; Table S5, available online). Individuals from the lowest income quartile had a lower incidence of diagnosed EDs (IRR = 0.80, 95% CI = 0.71-0.90) (Figure 2c; Table S5, available online) compared to those in the third quartile. When restricting the population to DNBC-18, overall differences were less pronounced (Figure 2; Table S6, available online). However, individuals with parents who had basic education showed a lower incidence of diagnosed EDs than those with short parental education (IRR = 0.59, 95% CI = 0.34-0.94) (Figure 2b; Table S6, available online). A similar pattern was observed in the lowest income quartile, where the incidence was lower compared to that in the third income quartile (IRR = 0.81, 95% CI = 0.57-1.13) (Figure 2d; Table S6, available online).

Figure 2.

Parental Education, Household Income, and Eating Disorders (EDs) in the Total Population (N = 523,148) and in Participants in the Danish National Birth Cohort, 18-Year Follow-Up (DNBC-18) (N = 44,552)

Note:The figure presents incidence rate ratio (IRR) and corresponding 95% CIs for diagnosed EDs vs no EDs (reference group) according to parental education (a and b) and household income (c and d). Estimates are presented as crude and adjusted for maternal and paternal age at birth, birth year, parental psychiatric diagnoses (measured up to the child’s age of 6 years), and parental origin (total population only). In models for household income, parental education (measured at the child’s age of 6 years, or highest/ongoing at birth if missing) was additionally adjusted for. Estimates for DNBC-18 are weighted. Results are also presented inTables S5andS6, available online.

When comparing estimates across the various populations, the stepwise increase in the incidence of diagnosed EDs for parental education was apparent in the background population, and, although much attenuated, remained evident among the participants invited to DNBC-18 (Table S7, available online).

Socioeconomic Circumstances and Diagnosed and Non-Diagnosed Eating Disorders

Compared to short parental education, individuals with parents who had a basic education level had a lower risk of diagnosed EDs (RRR = 0.59, 95% CI = 0.38-0.93) (Figure 3a; Table S8, available online) but a higher risk of non-diagnosed EDs (RRR = 2.19, 95% CI = 1.83-2.63) (Figure 3b; Table S8, available online). Analyses of any EDs, that is, either a diagnosis and/or a self-report, showed increased odds among individuals with basic parental education (OR = 1.71, 95% CI = 1.44-2.01) (Table S8; Figure S8, available online), compared to those with parental short education. No difference was observed for household income (Figure 3d-f; Table S9, available online).

Figure 3.

Association Between Parental Education, Household Income, and Eating Disorders (EDs) in the Danish National Birth Cohort, 18-Year Follow-Up (DNBC-18) (N = 44,552)

Note:Relative risk ratio (RRR) with corresponding 95% CI for the composite outcome separating diagnosed EDs (a and d), non-diagnosed EDs (b and e), and no ED (reference group) according to parental education and household income are presented. Odds ratio (OR) and 95% CI for any EDs (combining diagnosed and self-reported ED symptoms as one group) vs no ED (reference group) are also presented (c and f). Estimates are presented as crude and adjusted for maternal and paternal age at birth, birth year, and parental psychiatric diagnoses (measured up to the child’s age of 6 years). In models for household income, parental education (measured at the child’s age of 6 years, or highest/ongoing at birth if missing) was additionally adjusted for. Estimates for DNBC-18 are weighted. The p value for the likelihood ratio test comparing the models with the binary vs the 3-category outcome measure was <.001 for both parental education and household income. Results are also presented inTables S8andS9, available online.

Eating Disorder Subtypes

Anorexia Nervosa

Lower parental education was associated with a lower risk of being diagnosed with AN, for individuals with basic parental education (RRR = 0.49, 95% CI = 0.26-0.90) (Figure S3; Table S10, available online) compared to short education. In contrast, a higher risk of non-diagnosed AN was observed among individuals with basic parental education (RRR = 2.84, 95% CI = 2.12-3.79) (Figure S3; Table S10, available online). No difference was observed for income in relation to diagnosed AN, but a lower risk of non-diagnosed AN was observed in the lowest income quartile (RRR= 0.69, 95% CI = 0.50-0.94) compared to the third quartile (Figure S4; Table S11, available online).

Bulimia Nervosa

The number of diagnosed BN cases was low, limiting precision. No clear pattern was observed for diagnosed BN across parental education levels, although individuals with long parental education had a higher risk compared to those with short education (RRR = 2.46, 95% CI = 1.34-4.51) (Figure S3; Table S10, available online). For non-diagnosed BN, there were no substantial differences by parental education. Regarding household income, individuals in the lowest income quartile had a lower risk of diagnosed BN (RRR = 0.19, 95% CI = 0.06-0.56) (Figure S4; Table S11, available online) compared to the third quartile, while no consistent pattern was observed for non-diagnosed BN across income groups.

Non-Diagnosed Binge Eating Disorder

Individuals with basic parental education had more than twice the odds of non-diagnosed BED compared to individuals with parental short education (OR = 2.11, 95% CI = 1.63-2.73) (Figure S3; Table S10, available online). A weaker pattern was observed for household income, with those in the lowest income quartile having slightly higher odds compared to those in the third quartile (OR = 1.27, 95% CI = 0.99–1.64) (Table S11; Figure S4, available online).

Sensitivity and Post Hoc Analyses

When restricted to female individuals, in both the total population and the DNBC-18 sample, the results remained largely unchanged, with slightly stronger associations observed among female individuals only (Tables S12 and S13, available online). When stratifying by birth year, higher risks of diagnosed and non-diagnosed EDs by parental education were more pronounced among individuals born between 1996 and 1999, whereas the association with household income was more pronounced in those born between 2000 and 2003 (Tables S14-S16, available online). Excluding individuals with negative income did not alter the results (Tables S17 and S18, available online). When focusing solely on self-reported ED symptoms regardless of diagnosis, the findings were consistent with those observed for non-diagnosed EDs (Table S19, available online).

In a post hoc analysis, we compared sociodemographic characteristics of adolescents with self-reported ED symptoms in DNBC-18, distinguishing between those with and without a diagnosis and those with a diagnosis only (Table S20, available online). Across all groups, most individuals were female, particularly those with both self-reported and diagnosed EDs (96%). Diagnosed-only cases were broadly like those with both self-reported and diagnosed EDs, but had slightly lower parental education and income and were diagnosed at a younger age. Parental cohabitation was more common among adolescents with both self-reported and diagnosed EDs (85%) than among non-diagnosed individuals (75%). Urbanicity also varied, with non-diagnosed individuals more prevalent in the capital and suburban areas, and diagnosed individuals more often in provincial municipalities.

Discussion

This large-scale population-based study using nationwide register data and one of the world’s largest birth cohorts shows that the risk of diagnosed EDs increases with higher parental educational level, whereas individuals whose parents have basic education have the highest risk of non-diagnosed EDs. For diagnosed ED, we observed a clear educational gradient, with highest incidence among individuals from more educationally advantaged SECs and the lowest incidence among those with basic parental education. In contrast, this pattern was reversed for non-diagnosed EDs, where individuals with basic parental education had a higher risk than those with short education. No clear difference was observed in household income.

When comparing populations, the patterns for diagnosed ED were broadly similar across the total population and the DNBC-18, albeit attenuated in DNBC-18. Applying inverse probability weighting based on the background population moved the DNBC-18 pattern closer to that observed in the total population, although the pattern was still weaker. The contrasting patterns between diagnosed and non-diagnosed ED are consistent with previous research suggesting that only a minority of ED cases are detected by the health care system, leading to underrepresentation of individuals from less advantaged backgrounds in clinical samples.^3,13,16,35 These findings likely reflect both differences in treatment-seeking behavior across SEC groups and disparities in detection within the health care system, including missed identification in primary care. These factors may contribute to the underdiagnosis and undertreatment of EDs among individuals from more disadvantaged SECs. Previous studies have highlighted barriers such as stigma, limited health care access, and lack of knowledge about symptoms and severity as potential contributors to these disparities in ED diagnosis and treatment.^5,16,21,23

Female individuals had a markedly higher prevalence of both diagnosed and non-diagnosed EDs, consistent with prior studies and likely reflecting sex differences in treatment-seeking, and under-recognition of ED symptoms in male individuals.^21,37 This similar sex distribution across diagnosed and non-diagnosed cases may partly reflect our strict symptom-based definitions intended to approximate clinical threshold, but these criteria may have undercaptured male-typical manifestations, limiting sex-stratified analyses.

Differences between parental education and household income likely reflect that these indicators capture different aspects of SECs. Conceptual frameworks on pathways to inequalities in child health emphasize that SECs influence health through multiple mechanisms, including material conditions, psychosocial stress, and parental resources.¹⁵ Danish register-based evidence further shows that socioeconomic inequalities often follow similar patterns but differ in magnitude across diseases, depending on whether SECs are measured using education or income,³⁸ which may partly explain the differing findings by socioeconomic indicator observed in our study of EDs. Differences in associations with ED subtypes according to parental education and household income have also been reported previously in a Danish register-based study.¹³

Because Denmark has a tax-funded health care system with universal access, socioeconomic inequalities in diagnosed EDs may be smaller than in settings with greater financial barriers to care. Inequalities in other health care systems may therefore be more pronounced than those observed in this study.³⁹

When examining ED subtypes, these socioeconomic patterns were most pronounced for AN, with disadvantaged SECs, indicated by lower parental education and lower income, associated with a lower risk of being diagnosed with AN. This aligns with previous registry-based research showing that individuals from more advantaged SECs have a higher risk of receiving an AN diagnosis.¹³ However, our findings also suggest that non-diagnosed AN may be more prevalent among individuals from disadvantaged SECs, potentially reflecting barriers to health care access and recognition of symptoms in these populations. For BN, low case numbers limited precision, although diagnosed cases were again less common in the most disadvantaged income group. For BED, a strong association with lower parental education indicated that BED may be more prevalent in disadvantaged SEC groups.

One major strength of this study is the integration of registry-based diagnosed EDs with self-reported ED symptoms in a large population-based cohort, followed from fetal development through entering adulthood. An important limitation of this study is the potential for self-selection and attrition bias, as DNBC-18 includes only a subset of the original cohort. Our analysis indicated that the association between SECs and diagnosed EDs varied depending on the sample, likely reflecting both initial selection bias from maternal participation in DNBC and subsequent attrition influenced by SECs and ED diagnosis. To address these biases, we applied sample weights based on the background population and register-based information, presenting both weighted and unweighted estimates for DNBC-18. Participants in DNBC and DNBC-18 with lower parental education and/or household income may also represent a more resourceful and health-literate subgroup compared to non-participants with similar SECs, potentially increasing their likelihood of seeking treatment if needed. Comparable patterns of selection have been observed in other Danish survey data, where individuals with mental disorder diagnoses were generally underrepresented relative to the general population, although this was not the case for diagnosed EDs.⁴⁰

A key strength of using self-reported measures is their ability to capture both diagnosed cases and individuals who do not engage with formal treatment, offering a broader picture of EDs. However, this approach has limitations. First, the operationalization of ED symptoms was based on self-administered questionnaire data without any structured diagnostic interviews, which may lead to misclassification. In particular, our categorization of the self-reported ED symptoms likely represents conservative estimates, as symptom threshold and strict operational criteria may fail to capture atypical ED presentations. Self-reported EDs reflected symptoms during the past year, whereas register-based diagnoses covered ages 6 to 18 years, which makes direct comparisons difficult. Although this time difference is a limitation, the 2 measures capture different but complementary aspects of the ED spectrum: register-based diagnoses reflect clinically identified and treated cases, whereas self-reported EDs capture current symptoms that may not have led to care. Evidence from Denmark suggests that children and adolescents who are referred to outpatient psychiatry with EDs often experienced symptoms for about 1 year before referral.⁴¹ A systematic review also found that the average duration of untreated EDs was about 2.5 years for AN, 4.5 years for BN, and nearly 6 years for BED, with shorter durations in younger age groups.⁴² Therefore, the first hospital diagnosis likely reflects detection rather than true onset. To account for this, we used a hierarchical classification in which register-based diagnoses overruled self-reported ED symptoms, creating mutually exclusive categories and reducing concerns about time-related misalignment. Another limitation is that ICD-10 and DSM-5 are not fully concordant. ICD-10 includes EDNOS, whereas BED is not coded in ICD-10 and is captured only in the self-report. This non-alignment may introduce outcome misclassification. Individuals exhibiting BED symptoms during the study period could have been diagnosed under ICD-10 as EDNOS (F50.8, F50.9).⁴³ As BED was not formally recognized as a separate diagnosis in ICD-10, and as ICD-11 use has not yet been implemented in Denmark, such cases were not systematically captured in the registers. The absence of a specific diagnostic category likely also limited clinical awareness and treatment availability. In Denmark, treatment for BED has been available primarily through project-based pilot programs over the past decade, and national guidelines or formal hospital-based treatment have not yet been implemented nationwide, although awareness has increased substantially.⁴⁴ To further address potential misclassification, we reported diagnosed and non-diagnosed outcomes separately and presented subtype-specific analyses. Finally, registry data do not capture care from private psychiatrists, psychologists, or therapists,⁴⁵ which may lead to misclassification of some treated individuals as non-diagnosed. As individuals from more advantaged backgrounds are more likely to use private care,⁴⁶ some diagnosed EDs in these groups may not be recorded in the registers, leading to an underestimation of socioeconomic inequalities in ED diagnosis.

Parental education and household income were measured at the child’s age of 6 years, which we treat as the latest time point before the start of the risk period for incident EDs, thereby reducing the risk of reverse causation. In Denmark, most parents complete their education before or soon after their first childbirth, and educational attainment is relatively stable thereafter; measuring income at age 6 years also reduces short-term fluctuations related to parental leave and early childcare. Household income data were obtained from tax records, which may lead to misrepresentation of financial capacity, particularly among households with significant assets but low reported income.^36,47 This discrepancy arises because tax systems often allow deductions for investment and business losses, a benefit more commonly used by wealthier individuals.^36,47 Consequently, some high-asset households may appear in the lowest-income categories, potentially underestimating the association between low income and mental health outcomes.^36,47 Income levels can also fluctuate over time, particularly based on parental age and family dissolution. Parental education includes only completed education and does not account for ongoing studies. If a parent continued their education after the child turned 6 years of age, this would not be reflected in the data. For cases in which parental education at age 6 years was missing, information from birth, including ongoing education, was used instead. However, if a parent later discontinued their education, this could have introduced misclassification.

In conclusion, our findings underscore the importance of considering socioeconomic disparities in ED diagnosis and treatment. Whereas individuals from more advantaged SECs had a higher risk of receiving an ED diagnosis, those from disadvantaged SECs had a higher risk of non-diagnosed EDs. These results point to social inequities and highlight the need for targeted interventions to improve early detection and access to care in disadvantaged SEC groups, ensuring that all individuals with EDs receive timely and appropriate treatment.

CRediT authorship contribution statement

Ella C. Hesse Krenchel: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Andrea Joensen: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Else Marie Olsen: Writing – review & editing, Validation, Supervision, Methodology, Investigation. Helen Bould: Writing – review & editing, Validation, Supervision, Methodology, Investigation. Amanda M. Hughes: Writing – review & editing, Validation, Methodology, Investigation. Claus Thorn Ekstrøm: Writing – review & editing, Supervision, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Katrine Strandberg-Larsen: Writing – review & editing, Supervision, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization.