Eating disorders (EDs) are severe psychiatric disorders associated with high levels of morbidity [1], mortality [2, 3], and social, psychological and physical impairment [4]. The Diagnostic and Statistical Manual for Mental Disorders 5^th edition (DSM-5) [5] recently broadened ED diagnostic criteria, aiming to reduce the number of individuals with an ED who do not fit full-threshold diagnostic categories. Binge-eating disorder (BED) was introduced as a diagnostic category, and the criteria for anorexia nervosa (AN) and bulimia nervosa (BN) were broadened. Although previously considered low prevalence disorders, this broadening of the diagnostic criteria in DSM-5 has yielded preliminary evidence that ED are more common than once thought. A small number of community-based studies have investigated the prevalence of DSM-5 EDs. Lifetime prevalence estimates of DSM-5 EDs have varied dramatically across studies [6, 7] and the same applies to period-prevalence estimates [8, 9]. The remarkable variability across studies is likely due to sample size, differences in study design (questionnaire only assessment vs. two-phase studies), and a focus on adolescents/young adults only (reflecting the peak age of onset of AN and BN), thus highlighting a need for further large studies. No previous studies have investigated the period or lifetime prevalence of EDs amongst women in the fourth and fifth decade of life, after most individuals would be considered to have passed through the primary window of risk. We recently highlighted a wide gap in access to healthcare amongst adults with ED in a UK population-based sample [8], and we therefore sought to replicate and extend our findings.

A long-term perspective offers the unique opportunity to investigate EDs and relevant precursors/risk factors using a disease life-course approach. Few studies have investigated risk factors for EDs using a longitudinal prospective design [10], and the majority have focused on treatment seeking samples and full threshold EDs. Prior evidence from our group [11, 12] and others [13, 14] points to childhood experiences and personality as important risk factors for EDs; however, there is a relative lack of population-based studies investigating these in relation to ED. Thus, our aims were (1) to determine the lifetime and 12-month prevalence of DSM-5 EDs in mid-life in women from a population-based cohort using a two-phase design and to explore healthcare access, as well as (2) to investigate associations between lifetime ED, risk factors (personality characteristics (personality, locus of control); early childhood experiences (sexual abuse, maternal care, carer/parent death, parental separation/divorce and being under local authority care)) and fixed factors (intelligence quotient (IQ)).

In this large sample of UK women in mid-life DSM-5 EDs were common. The lifetime prevalence of DSM-5 AN was higher than previously reported for DSM-IV AN but comparable to prior estimates of ‘broad’ DSM-IV AN [34] and expected given the removal of the amenorrhea criterion in DSM-5 [34], and the older age of our sample. The lifetime prevalence of BN and BED were also in line with previous community studies [35], although surprisingly, BED was less common during lifetime compared to AN and BN. This might be due to a higher percentage of highly educated women participating in Phase 1, and the ethnic composition of ALSPAC [15]. EDs other than AN, BN, and BED, now subsumed under OSFED, were common in this sample (7.6%), in particular the residual unspecified category (other OSFED). This suggests that, despite efforts in DSM-5 to reduce the prevalence of the ‘unspecified’ category (a goal of the revisions to DSM-IV), as previously shown [4, 6, 9], many individuals in the community experience EDs other than AN, BN, and BED. The relatively large subset of women presenting with ‘other OSFED’ (27.6% of all OSFED) is in line with our own [4] and others’ research [36].

EDs in the year prior to interview were more common than expected, no previous study – to our knowledge – has investigated the period prevalence of DSM-5 ED in a community sample in mid-life. OSFED was the most common ED, accounting for almost half of all prevalent ED cases and BED was the most common full-threshold disorder. These findings highlight, for the first time, that EDs are not confined to earlier decades of life and that both chronic and new onset disorders are apparent in this stage of life.

Although our data cover a wide time lag (last 40 years) and might therefore reflect past rather than current lack of identification of EDs and related healthcare provision in the UK, it is nevertheless surprising that, across their lifetime, very few women had sought or received treatment for EDs.

Our investigation of risk factors of lifetime EDs revealed important findings. Childhood sexual abuse, unhappiness, and low parental care were associated with binge and/or purge-type ED (AN-BP, full and sub-threshold BN and BED). The association between childhood sexual abuse and binge and/or purge-type ED is consistent with previous retrospective studies [14, 37, 38], and extends this evidence to sub-threshold ED. In line with our recent meta-analysis [11] and previous hypotheses that parenting risk factors and parental influences might act differently across the ED diagnostic spectrum [14], parental overprotection and low maternal care were associated with binge and/or purge disorders, but not AN. We recently showed that retrospectively reported parental influences (including poor parenting and overprotection) predicted body dissatisfaction in women with BN and AN-BP but not AN-R [39]. This association with binge/purge type disorders maybe mediated via negative affect, low self-esteem, or body dissatisfaction developmentally, as might be the case with sexual abuse. Further longitudinal studies are required to empirically test these pathways.

High interpersonal sensitivity was associated with all EDs. Interpersonal sensitivity has been described as sensitivity to other’s feedback and fear of social rejection [26], and it is characterized by misinterpretation of interpersonal behaviors, interpersonal avoidance and discomfort in the presence of others due to a sense of inadequacy. Our finding confirms and strengthens existing cross-sectional evidence that social impairment and interpersonal difficulties are common across EDs [11], and might contribute to their onset and maintenance [11, 40].

We replicated associations identified in clinical studies between high IQ and AN [41, 42], in a community setting (women with lifetime AN-BP had a total IQ on average 5 points higher than women with no EDs). Whether the higher IQ observed is secondary to higher levels of perfectionism, or indeed indexes specific cognitive strengths requires further study and elucidation.

This study is the first to investigate childhood risk factors for PD, a newly described ED. The only twin study of PD recently showed that non-shared environmental factors explained 56% of the variance for PD [43]; however, the study could not disentangle the effect of non-shared environment versus genetic factors. Our findings suggest a role for childhood experiences and parenting as risk factors for PD. Similarly, this is the first study to investigate risk factors for atypical AN, with initial evidence of a risk factor profile more similar to binge/purge type ED than AN-R. This is the first study to show a similar pattern of risk for full threshold and sub-threshold BN and BED. These findings, together with evidence of similar outcomes between threshold and sub-threshold BN and BED [4], confirm similarities between full and sub-threshold ED, in this case in relation to risk factors.

Few associations were identified between environmental risk variables and restrictive AN. This finding might reflect our hierarchical approach to lifetime diagnosis, in that to be included in this group, women had to have met criteria for AN-R only (and not other EDs). As such, our findings point to a smaller contribution of environmental risk to this phenotype [34].

Strengths of the study include a large community sample of women, overcoming bias introduced by studying treatment-seeking individuals. The two-phase epidemiological design, one of the best approaches to estimate prevalence of disease [44], and the survey analytical techniques allowed more accurate estimates to be obtained using our entire Phase I sample. We used a validated and reliable assessment for EDs and supplemented this with a longitudinal assessment of lifetime symptoms to obtain DSM-5 diagnoses. The availability of risk factor data independently collected 20 years prior to the current study allowed a less biased estimation of risk factors, although recall bias might explain some of our findings.

Limitations of the study include the nature of the ALSPAC cohort, i.e. women who were pregnant at a specific point in time in a defined geographic area. The sample is therefore likely to include women with ED who were able to become pregnant at least once and is therefore not representative of the general population. Nevertheless, the lowest ever self-reported BMI in this sample was 10.7, and the lowest measured BMI at mean age 48 years was 15.4, suggesting a range of ED severity within the sample. Participation in Phase 1 was selective; however, we were able to determine that more educated women and those with fewer children participated. Despite attrition between Phase 1 and 2, our analytical approach allows minimizing bias due to attrition. Moreover, risk factor analyses were controlled for socio-demographic factors associated with non-participation in Phase 1, therefore increasing generalizability of the findings. It is possible that women with higher levels of psychopathology were less represented in this study; however, levels of self-reported EDs at enrolment were comparable across participants and non-participants, therefore we are unlikely to have underestimated the prevalence of EDs. Small sample size in some diagnostic groups might account for false negatives. Similarly, chance might explain some of our positive findings. We could not directly investigate other psychiatric disorders and, therefore, the specificity of risk factors for ED versus other psychopathology needs elucidating further.

EDs are common across the lifespan and in mid-life. Poor healthcare access was evident in this sample of women. This has implications for service provision, which at present is not specifically geared towards women in mid-life, and in identification of women who might be misdiagnosed given the lack of awareness amongst healthcare professionals of ED presentations. Although some risk factors differed across ED subtypes, childhood sexual abuse and poor parenting were associated with binge/purge type disorders, whilst personality factors were more broadly associated with several diagnostic categories. Few risk factors were specifically associated with one diagnostic category. These patterns suggest shared environmental risk across the ED diagnostic spectrum, independent of full/sub-threshold symptoms. The evidence that lifetime and active EDs are common amongst women in mid-life, compounded by the lack of healthcare access and treatment, highlights the likelihood of high disease burden and unmet needs. Future studies should also aim to better characterize EDs in mid-life, and clarify their correlates in terms of physical and psychiatric comorbidities, as well as differences in precipitating factors leading to ‘late onset’ compared to adolescent/young adult onset.