Association of Emotion Regulation Trajectories in Childhood With Anorexia Nervosa and Atypical Anorexia Nervosa in Early Adolescence

Key Points

Question  Is emotion dysregulation in childhood associated with an increased risk of broad anorexia nervosa in adolescents?

Findings  In this cohort study of 15 896 participants from the Millennium Cohort Study, poor emotion regulation skills at 3 years of age were not associated with an increased risk of broad anorexia nervosa. However, lack of improvement in emotion regulation skills across childhood was associated with increased risk in this disorder.

Meaning  These findings suggest that failure to meet key emotional developmental milestones from 3 to 7 years of age could confer an increased risk for broad anorexia nervosa, and support with developing emotion regulation skills across childhood could be beneficial in preventing anorexia nervosa.

Importance  People with anorexia nervosa often experience difficulties regulating their emotions. There is no longitudinal evidence as to whether these differences are already present in childhood or when they begin to emerge.

Objective  To investigate the association between emotion regulation trajectories from 3 to 7 years of age and symptoms of anorexia nervosa and atypical anorexia nervosa in adolescence.

Design, Setting, and Participants  This cohort study included all children with complete exposure data in the Millennium Cohort Study, a UK general population birth cohort. Data were acquired from June 2001 to March 2016 and analyzed from June to November 2020.

Exposures  Mothers reported on their children’s emotion regulation skills at 3, 5, and 7 years of age using the Children’s Social Behavior Questionnaire. Multilevel models were used to derive early childhood emotion regulation scores (ie, predicted intercept) and within-child changes in emotion regulation scores from 3 to 7 years of age (ie, predicted slope).

Main Outcome and Measures  Symptoms consistent with a DSM-5 diagnosis of anorexia nervosa or atypical anorexia nervosa at 14 years of age, defined using a range of questions relative to body image, weight perception, and dieting behaviors (hereinafter referred to as broad anorexia nervosa). Univariable and multivariable logistic regression models tested the association between exposures and outcome. Regression models were adjusted for child and family sociodemographic and socioeconomic characteristics and mental health difficulties, prenatal and perinatal factors, child’s cognitive development, and maternal attachment.

Results  A total of 15 896 participants (85.7% of total sample; 51.0% boys; 84.5% White individuals) had complete data on the exposure and were included in the main analyses. Among those with complete exposure and outcome data (9912 of the analytical sample [62.4%]), 97 participants (1.0%; 86 [88.7%] girls and 85 [87.6%] White individuals) had symptoms consistent with a diagnosis of broad anorexia nervosa at 14 years of age. No evidence suggested that children with lower emotion regulation ability at 3 years of age had greater odds of later reporting symptoms of broad anorexia nervosa (odds ratio [OR], 1.21; 95% CI, 0.91-1.63). However, children whose emotion regulation skills did not improve over childhood and who had greater problems regulating emotions at 7 years of age had higher odds of having broad anorexia nervosa at 14 years of age (OR, 1.45; 95% CI, 1.16-1.83).

Conclusions and Relevance  These findings suggest that difficulties in developing age-appropriate emotion regulation skills in childhood are associated with experiencing broad anorexia nervosa in adolescence. Interventions to support the development of emotion regulation skills across childhood may help reduce the incidence of anorexia nervosa.

Anorexia nervosa is a psychiatric disorder typically beginning in adolescence and characterized by restriction of food intake and severe anxiety regarding eating, body shape, and weight gain.¹ Anorexia nervosa, including its subthreshold presentations, affects approximately 1% of young people and is associated with mental and physical health comorbidities and high mortality.² Childhood interventions could help to prevent a number of anorexia nervosa cases, yet longitudinal etiological research remains scant, and targets for such interventions are elusive.

Emotion regulation, defined as the ability to both intrinsically and extrinsically monitor, appraise, and modify one’s emotional state,³ has been increasingly proposed as a potential target.⁴ In clinical samples, people with anorexia nervosa often show suboptimal emotion regulation and awareness compared with healthy controls, as well as greater difficulties tolerating distress and other maladaptive emotion regulation strategies such as rumination and suppression.^5-7 Poor emotion regulation can lead to emotional overcontrol, which is associated with cognitive rigidity, reward insensitivity, and perfectionism, all commonly observed in patients with anorexia nervosa.^8,9 It has been hypothesized that emotional avoidance might increase negative affect and the use of maladaptive coping strategies, such as disordered eating behaviors.^10-12 Although such maladaptive behaviors typically begin during adolescence, emotion regulation difficulties emerge throughout childhood¹³ and are associated with later psychopathology, including obsessive-compulsive disorder, depression, and anxiety,^14,15 common comorbidities^16,17 of and proposed risk factors for anorexia nervosa.^18,19

Existing studies of emotion regulation and anorexia nervosa have predominantly used case-control designs within clinical populations.⁷ These studies have limitations. They are prone to reverse causality and cannot exclude the possibility that emotion regulation difficulties are a consequence rather than a cause of anorexia nervosa. For instance, some studies of people with anorexia nervosa have found that body mass index (BMI) or weight restoration did not affect emotion regulation skills,^20,21 whereas others have found that the latter improved with weight restoration.²² Clinical studies are also susceptible to selection bias because cases are commonly drawn from secondary care and controls from the general population samples; in addition, they can be an unrepresentative selection of all people with eating disorders, because the latter group often does not seek help.²³ Residual confounding and small sample sizes are also a concern regarding the existing literature.

Longitudinal general population studies can address many of these limitations. They can also detail how emotion regulation difficulties develop across childhood and how these changes might be implicated in the etiology of anorexia nervosa. This process allows identification of key developmental stages when preventative interventions could be more effective. To this end, we used data from the Millennium Cohort Study (MCS), a large UK longitudinal general population birth cohort, to investigate the association between trajectories of emotion dysregulation across childhood and broadly defined anorexia nervosa at 14 years of age.

The MCS is a longitudinal population-based birth cohort study of children born from September 1, 2000, to January 11, 2002, who were living in the UK at 9 months of age and their families (eMethods 1 in the Supplement).²⁴ The Multi-Centre Research Ethics Committee gave ethics approval for the MCS. Participants gave written consent. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.

In this study, we included children with complete data on exposure, imputing missing confounder and outcome data. In cases of twins or triplets, we selected 1 child at random to avoid overestimation or underestimation of effects resulting from shared genetic and environmental factors.

We used a set of questions broadly covering DSM-5 criteria for anorexia nervosa or atypical anorexia nervosa to identify participants who at 14 years of age had behaviors and cognitions consistent with these 2 diagnoses (hereinafter referred to as broad anorexia nervosa) (Table 1 and Table 2) These DSM-5 criteria include deliberate restriction of energy intake, fear of gaining weight, behaviors preventing weight gain or maintaining low weight, and disturbed body weight perception.¹ Low weight is a diagnostic criterion for anorexia nervosa, but not atypical anorexia nervosa. We defined adolescents as having broad anorexia nervosa if they met all the following criteria: report of lifetime dieting and exercising for weight loss, currently trying to lose weight, skipping breakfast every day, describing themselves as overweight despite a BMI in the underweight or normal weight range, and scoring below the median sample score on a question on body image. We calculated BMI from objective measures of weight in kilograms divided by height in meters squared taken by trained researchers when participants were 14 years old. We used the Stata egen function zbmicat()²⁵ to generate age- and sex-appropriate BMI categories, based on International Obesity Task Force growth charts.^26,27

We used a broad anorexia nervosa definition for 2 reasons. First, individuals who do not meet the low BMI criterion are less likely to be referred to specialist services and are thus often excluded from clinical studies. Including adolescents with symptoms typical of clinical diagnoses can help elucidate risk factors that operate across the spectrum of disorder severity. Second, low BMI results from protracted restrictive eating behaviors and weight and shape concerns. Some people, despite experiencing all other behavioral and cognitive symptoms of the disorder, might never reach an underweight BMI. Furthermore, at the early stages of anorexia nervosa (which we are likely to observe at this age), people with the condition may still be in the normal BMI range. For descriptive purposes, we also restricted our definition to underweight participants.

At the sweeps of data collection at 3, 5, and 7 years of age, mothers reported on their children’s emotion regulation abilities over the previous 6 months using 5 questions from the Child Social Behaviour Questionnaire (a modified version of the Adaptive Social Behavior Inventory).^28,29 These questions are scored on a 3-point Likert scale (eTable 1 in the Supplement) giving a total score ranging from 0 to 10, with higher scores indicating greater difficulties regulating emotions. Published studies^30,31 using MCS and other data sets have used this measure previously. Our main exposures were children’s predicted emotion regulation scores at 3 years of age (ie, intercept) and their within-person linear change in score from 3 to 7 years of age (ie, slope), derived using multilevel models. More details on how this variable was created are found in the Statistical Analysis section. Data were acquired from June 2001 to March 2016.

We used direct acyclic graphs (eFigures 1 and 2 in the Supplement) to guide our choice of confounders, based on assumptions informed by previous literature and clinical observations. According to these assumptions, to estimate the total effect of both exposures on broad anorexia nervosa at 14 years of age, it was necessary to adjust analyses for each child’s birthweight, gestational age, sex, ethnicity, breastfeeding status, cognitive self-regulation, and cognitive development (indexed in our sample by language development and school readiness) at 3 years of age and underlying genetic risk. It was also necessary to adjust for family socioeconomic status, maternal and paternal depression, maternal and paternal emotion regulation, maternal prepregnancy BMI, smoking habits in pregnancy, and maternal attachment to the child. We had data on all these hypothesized confounders except genetic risk and parental emotion regulation. Additional details are provided in eMethods 2 in the Supplement.

We also adjusted analyses of the association between emotion regulation slope and broad anorexia nervosa for child BMI and mental health difficulties at 3 years of age. We did not include these 2 measures in the intercept analyses because we hypothesized that they could be mediators of its association with the outcome. Longitudinal evidence suggests that more emotionally dysregulated children have higher BMI³⁰ and greater internalizing and externalizing symptoms.^32,33 Because measures of emotion regulation, mental health difficulties, and BMI were collected at the same time, it was difficult to disentangle temporality. However, emotion regulation difficulties can also occur as a result of neurodevelopmental disorders,³⁴ and some evidence suggests that emotion regulation improves with weight restoration.²² Hence, in sensitivity analyses, we also included mental health difficulties and BMI as potential confounders.

Data were analyzed from June to November 2020. To create the exposure variables, we used multilevel models to model within-child repeated emotion regulation scores from 3 to 7 years of age. We used linear and quadratic indicators of child’s age at assessment (3, 5, and 7 years) and included a random intercept for child and a random slope for linear time. For each child, we predicted an intercept and a linear slope value representing their predicted emotion regulation scores at 3 years of age and the within-child score changes from 3 to 7 years, respectively. To derive the child’s intercept, we centered the age variable at 3 years. To derive the slope, we centered the age variable at 5 years. In regression models, we standardized these values to have a mean of 0 and an SD of 1.

To investigate the association between emotion regulation exposures and broad anorexia nervosa, we used univariable and multivariable logistic regression models. First, we ran univariable models for each exposure. Subsequently, we ran a series of multivariable models progressively adjusting each exposure for child- and family-level confounders. Models testing the association between emotion regulation slope and broad anorexia nervosa were also further adjusted for emotion regulation intercept to test whether changes in emotion regulation are associated with broad anorexia nervosa regardless of baseline levels. We imputed missing confounder and outcome data using multiple imputation with chained equations (eMethods 3 in the Supplement).³⁵ In all analyses, we used survey and nonresponse weights and accounted for sampling strata.

As sensitivity analyses, we reran all models restricting the sample to participants with complete exposure and outcome data and imputed confounders and complete data on all variables. We then ran univariable and multivariable models using emotion regulation scales 3, 5, and 7 years of age separately as exposures. These analyses were based on participants with complete exposure and imputed confounders and outcome at each time point. Finally, we used linear mixed models to model trajectories of emotion regulation scores for adolescents with and without broad anorexia nervosa. Our models included a random intercept for child and a random slope for linear time. We ran univariable and multivariable models, adjusting for all previously identified confounders, and included interactions with age and age squared to test for differences in trajectories’ slopes centering age at 5 years. Analyses were run in Stata, version 16 (StataCorp LLC). P < .05 indicated statistical significance.

Of the 18 552 children included in the first wave of the MCS, 15 896 (85.7%) provided data for at least 1 emotion regulation assessment and were included in the analyses. Of these, 9255 (58.2%) had data on all 3 assessments, 4103 (25.8%) had data on 2 assessments, and 2538 (16.0%) had data on 1. Among children with complete exposure data, 5984 (37.6%) did not have outcome data. In eTable 2 in the Supplement, we present factors associated with missing outcome data. A total of 8110 participants (51.0%) were male and 7786 (49.0%) were female. Most participants were White (13 432 [84.5%]), had at least 1 parent in a nonmanual occupation (9572 [63.7%] among those with data available), and had a mother who only completed compulsory education (10 218 [64.5%] among those with data available). A complete overview of sample characteristics is provided in Table 3.

Mean (SD) emotion regulation scores were 4.39 (2.25) at 3 years of age, 3.61 (2.30) at 5 years of age, and 3.59 (2.37) at 7 years of age. Multilevel models also showed a decrease in scores from 3 to 5 years of age and a subsequent stabilization (eTable 3 and eFigure 3 in the Supplement). Overall, boys, children from more deprived backgrounds, and children whose parents had greater depressive symptoms had more difficulty regulating emotions. Children with mental health difficulties, lower cognitive development, and greater prenatal and perinatal adversities (eg, those born prematurely or at lower weight or exposed to maternal smoking in utero) also were more emotionally dysregulated (Table 3).

Among the 9912 participants with complete exposure and outcome data, 97 (1.0%; 86 girls [88.7%] girls, and 85 [87.6%] White individuals) had symptoms consistent with a broad diagnosis of anorexia nervosa in adolescence. Prevalence was lower in boys (11 of 4967 [0.2%]) than in girls (86 of 4936 [1.7%]) and was similar in White (85 of 8354 [1.0%]) and ethnic minority group (14 of 1558 [0.9%]) participants (eTable 4 in the Supplement). Seven of the 9912 participants (0.07%) met strict criteria for anorexia nervosa.

There was no evidence that children with higher intercept scores had increased odds of broad anorexia nervosa at age 14 years in the univariable model (odds ratio [OR], 1.19; 95% CI, 0.94-1.48). The association remained largely unchanged when including all hypothesized confounders (OR, 1.21; 95% CI, 0.91-1.63) (Table 4). Including child BMI and mental health difficulties in the sensitivity analyses did not alter these findings (OR, 1.15; 95% CI, 0.82-1.62).

Increases in emotion regulation slope were associated with greater odds of broad anorexia nervosa at 14 years of age in the univariable model (OR, 1.40; 95% CI, 1.13-1.74). This association persisted in models adjusted for all hypothesized confounders (OR, 1.45; 95% CI, 1.16-1.83) (Table 4).

When running models for participants with complete exposure and outcome variables and imputed confounders (n = 9912) (eTable 5 in the Supplement) and those based on complete cases (4004 for intercept and 3768 for slope) (eTable 6 in the Supplement), the results showed a reduced strength of association, although the 95% CIs overlapped with those of the main analyses. When we used individual measurements of emotion regulation, we found no evidence of an association between poorer emotion regulation at 3 years of age (fully adjusted OR, 0.99; 95% CI, 0.89-1.11) and 5 years of age (fully adjusted OR, 1.11; 95% CI, 0.95-1.31) and broad anorexia nervosa at 14 years of age. However, we found strong evidence of an association between poorer emotion regulation at 7 years of age and increased odds of broad anorexia nervosa at 14 years of age (OR, 1.19; 95% CI, 1.05-1.36) (eTable 7 in the Supplement). These findings were also mirrored by mixed model analyses. Here, we found no evidence of differences in score at 5 years of age between those with and without anorexia nervosa (mean difference, 0.25; 95% CI, −0.09 to 0.60). However, there was an interaction between broad anorexia nervosa and time (interaction coefficient, 0.20; 95% CI, 0.08-0.32; P = .001) indicating that, although emotion regulation scores were similar between groups at 3 years of age, trajectories began to diverge at 5 years of age, resulting in differences in scores by 7 years of age (eTable 8 in the Supplement and the Figure).

In this cohort study, we found that children whose emotion regulation skills improved less during the course of childhood and who had greater problems regulating emotions by 7 years of age had increased odds of broadly defined anorexia nervosa at 14 years of age. This association was independent of baseline levels of internalizing and externalizing symptoms, emotion, cognitive self-regulation skills, cognitive development, maternal attachment, BMI, and family characteristics and was consistent across main and sensitivity analyses. We did not find consistent evidence of an association across main and sensitivity analyses when investigating emotion regulation difficulties at 3 years of age as the exposure.

Clinical case-control studies⁷ have found that people with anorexia nervosa have emotion regulation difficulties. A previous longitudinal study³⁶ also found that adolescents with emotion regulation difficulties had greater disordered eating behaviors at a 7-month follow-up. Our study improves on these previous investigations by showing that emotion regulation difficulties not only precede the onset of anorexia nervosa, they also likely emerge during the course of early to middle childhood. We observed that although most children experienced improvement in their ability to regulate their emotions between 3 and 7 years of age, this was not the case for those who later developed symptoms consistent with a diagnosis of anorexia nervosa or atypical anorexia nervosa (Figure). This suggests that failure to meet key emotional developmental milestones, as opposed to having greater emotion dysregulation since early childhood, could confer an increased risk for anorexia nervosa. Emotion regulation difficulties might also be independent from low BMI and increase the risk of restrictive eating behaviors across a range of eating disorders. This hypothesis needs to be investigated further.

Several mechanisms could explain the association we observed. Emotion dysregulation might represent an early manifestation of genetic and neurobiological risk. Cognitive rigidity, reward insensitivity, and perfectionism typical of anorexia nervosa^8,9 could emerge in response to or accompany difficulties regulating negative emotions. Emotion dysregulation could also trigger environmental risk factors. For instance, children learn from a young age to avoid peers who display extreme emotions, and consequently children who are unable to regulate emotions often struggle to develop peer relations.³⁷ Poor social competence and emotional control can lead to being bullied,³⁸ a risk factor for eating disorders in young people.³⁹ The ability to regulate emotions also helps individuals cope with negative experiences and fosters resilience in stressful situations.^40-42 In adolescence, when peer relationships begin to play a key role in an individual’s life, the lack of such skills could lead to feelings of anxiety and consequently to developing coping strategies such as emotional avoidance, which are common clinical observations in this population.⁴³

This study has some limitations. The MCS did not collect information on binge eating and purging behaviors. Hence, we cannot exclude that some participants whom we classified as having symptoms of broad anorexia nervosa could have had bulimia nervosa or binge eating disorder, although at this age, this is likely to be a minority of adolescents.⁴⁴ Also, some adolescents with broad anorexia nervosa at 14 years could have later transitioned to a different eating disorder, as this is often observed in clinical samples. Future studies should aim at capturing whether emotion dysregulation is a shared or specific risk factor across diagnoses.

The question on weight perception does not include a qualitative element (ie, whether the adolescent is happy or not with their perceived weight) and that on body image is not weight or shape specific. To capture adolescents who were dissatisfied with their weight, we included only those who overestimated their weight and also scored higher than the median sample value on the body image question. This choice should have restricted our definition to adolescents with more severe weight and shape concerns.

We made our causal assumptions around confounder adjustment explicit by using direct acyclic graphs, and we were able to adjust for most variables that we had identified. However, we did not have data available on possible confounders such as parental emotion regulation and genetic factors. Residual confounding may explain our results. Although this can never be excluded entirely as a possibility, we observed minimal change in the magnitude and strength of the association between the univariable and the multivariable models, thus providing some reassurance around the robustness of the results.

As is common in longitudinal cohort studies, this study was affected by some degree of attrition. Nevertheless, our main analyses—based on participants with complete exposure—included most of the sample (85.7%). Furthermore, when we compared the results from these models with those of complete case analyses and analyses imputing only confounders, our results remained largely consistent.

This cohort study found that difficulties in developing age-appropriate emotion regulation skills in childhood were associated with experiencing broad anorexia nervosa in adolescence. If the associations we observed were causal, universal interventions fostering skills for emotion regulation in this age group, such as building tolerance for uncomfortable feelings and learning how to overcome frustration, could have a preventative role in the emergence of eating disorders and other mental health problems with an onset in adolescence.

Accepted for Publication: May 6, 2021.

Published Online: July 7, 2021. doi:10.1001/jamapsychiatry.2021.1599

Corresponding Author: Francesca Solmi, PhD, Division of Psychiatry, University College London, 148 Tottenham Court Rd, Maple House, 6th Floor, Wing A, W1T 7NF London, United Kingdom (francesca.solmi@ucl.ac.uk).

Author Contributions: Dr Solmi had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Henderson, Flouri, Srinivasan, Solmi.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Henderson, Flouri, Harrison, Gemma Lewis, Stafford, Solmi.

Critical revision of the manuscript for important intellectual content: Henderson, Bould, Flouri, Gemma Lewis, Glyn Lewis, Srinivasan, Warne, Solmi.

Statistical analysis: Henderson, Flouri, Gemma Lewis, Glyn Lewis, Solmi.

Obtained funding: Solmi.

Administrative, technical, or material support: Henderson.

Supervision: Harrison, Glyn Lewis, Stafford, Solmi.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by Sir Henry Wellcome fellowship 209196/Z/17/Z (Dr Solmi) and grant 211163/Z/18/Z (Dr Srinivasan) from the Wellcome Trust; grants MR/S019707 (Dr Harrison) and MR/S020292/1 (Drs Warne and Bould) from the Medical Research Council/Medical Research Foundation; and the University College of London Hospital National Institute for Health Research Biomedical Research Centre. The Millennium Cohort Study is funded primarily by the UK Economic and Social Research Council with cofunding by a consortium of UK government departments.

Role of the Funder/Sponsor: The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: The Millennium Cohort Study data are available free of cost to researchers from the UK Data Service website (https://www.ukdataservice.ac.uk/).

Additional Contributions: We thank the Millennium Cohort Study families who voluntarily participate in the study and a large number of stakeholders from academic, policy-maker, and funder communities and colleagues at the Centre for Longitudinal Studies involved in data collection and management of these cohort studies.