Skip to content

Epidemiology and incidence

TL;DR — Frequency estimates vary because case finding, age range and diagnostic rules differ more than the second decimal place suggests. AN commonly begins in adolescence, but onset and persistence occur across the lifespan and in all genders. Service samples underrepresent people in higher-weight bodies, males, gender-diverse people and populations with limited specialist access. Pandemic-era increases in presentations do not by themselves establish a matching increase in population incidence.

What is being measured

Measure Numerator Denominator Principal bias
Point prevalence Current cases Population at one time Misses remitted and concealed cases
12-month prevalence Cases during one year Population Recall and diagnostic access
Lifetime prevalence Ever met criteria Population Recall; changing criteria
Incidence New cases Person-time Onset date and detection differ
Treated incidence First service contact Catchment population Access and referral, not disease alone
Source Population and design AN estimate
van Eeden 2021, PMID 34419970 Narrative review of recent epidemiology Lifetime prevalence up to 4% of females and 0.3% of males; overall incidence stable over decades but rising in those aged <15; five-fold or greater mortality risk
Silén 2022, PMID 36125216 Worldwide review of DSM-5 studies, 2013–22, young people Lifetime AN 0.8–6.3% of women and 0.1–0.3% of men, within a total DSM-5 eating-disorder burden of 5.5–17.9% of young women and 0.6–2.4% of young men by early adulthood
Udo 2018, PMID 29859631 NESARC-III; 36,306 US adults, structured interview, 2012–13 Lifetime AN 0.80% (SE 0.07%); 12-month AN 0.05% (SE 0.02%)
Smink 2012, PMID 22644309 Earlier review Overall incidence stable over preceding decades, with an increase in the high-risk group of 15–19-year-old girls

The estimates are not interchangeable and the spread is instructive: a nationally representative structured-interview survey of adults returns a lifetime figure (0.80%) at the very bottom of the range that youth-focused studies report (0.8–6.3% of women). Age band, instrument, informant and diagnostic era all move the number more than any plausible real difference between these populations. Two reviews independently note that AN incidence has been broadly stable while rising among the youngest, and both decline to say whether that reflects earlier detection or earlier onset (Smink 2012, PMID 22644309; van Eeden 2021, PMID 34419970).

Register incidence: a number that partly measures psychiatry

The clearest test of whether AN incidence is rising uses a whole-country diagnostic register. In the Danish Psychiatric Central Research Registry, first-time diagnosed AN in people aged 4–65 rose from 6.4 to 12.6 per 100,000 person-years between 1995 and 2010 (N=5,902 incident AN cases), while bulimia nervosa rose only from 6.3 to 7.2 (N=5,113). The male-to-female ratio in 2010 was 1:8 for AN and 1:20 for BN, age at onset was earlier for AN than BN, and age at incidence fell over the period for AN but not BN. The decisive qualification is the authors' own: a sizeable part of the increase, particularly in the younger AN age groups, was attributable to a general increase in the total number of people receiving any first-time psychiatric diagnosis (249,607 over the same window) (Steinhausen 2015, PMID 25809026). A doubling of diagnosed incidence in a register is therefore compatible with little or no change in disease incidence.

Pandemic-era presentations

The pandemic produced the largest short-term movement in these numbers and the clearest example of why presentations are not incidence. A systematic review of 53 studies covering 36,485 individuals with eating disorders found pooled hospital admissions rose on average 48% during the pandemic relative to pre-pandemic timepoints (591 → 876 across 10 studies), with 36% of studies documenting increased eating-disorder symptoms and patterns varying by diagnosis and by lockdown timing (Devoe 2023, PMID 35384016).

A national analysis with an explicit exposure measure sharpens this. Across 11,289 Canadian eating-disorder hospitalizations from April 2016 to March 2023 — 77% of them in females aged 12–17 — a 10% increase in the Bank of Canada public-health stringency index was associated with higher hospitalization rates in every region analysed (adjusted rate ratios 1.05 in Quebec and Ontario, 1.08 in the Prairies, 1.11 in British Columbia). Excess hospitalizations peaked at the one-year mark, with rate ratios of 2.02–2.44 across regions (Roumeliotis 2024, PMID 38976259). Stringency tracked admissions, but this observational association cannot separate restriction effects from viral, service and other concurrent changes. It remains a health-system measure: nothing here establishes that population incidence of AN doubled.

Genetic liability interacts with the epidemiology

Register-linked genetic data allow risk factors to be examined jointly rather than in isolation. Using Danish nationwide registers with 7,003 individuals with AN and 45,229 without, AN polygenic risk score was associated with urbanicity, parental ages, genitourinary tract infection and parental socioeconomic factors, and the expected associations between 22 register risk factors and AN were only slightly attenuated by adjusting for parental psychiatric history and/or AN polygenic score. Interaction analyses found the effect of polygenic liability differed by level of sex, maternal age, genitourinary tract infection, caesarean section and parental socioeconomic and psychiatric factors (Papini 2024, PMID 38347808). Epidemiological risk factors for AN are therefore not simply proxies for inherited liability, and the two are not additive.

Age and developmental distribution

A meta-analysis of 192 epidemiological studies (n=708,561) placed the peak age at onset for feeding and eating disorders at 15.5 years, with a median of 18 (IQR 15–23) across 11 contributing studies: 15.8% had onset before age 14, 48.1% before 18 and 82.4% before 25. These are ICD-11 diagnostic-block estimates covering feeding and eating disorders together, not AN specifically (Solmi 2022, PMID 34079068). The comparison worth holding is that the same analysis put the peak for any mental disorder at 14.5 years — eating disorders are a youth-onset condition but not an unusually early-onset one within psychiatry. Puberty, growth velocity, school transitions and increasing autonomy matter for detection and service design. Adult-onset and later-life presentations exist; an adolescent prototype should not become an exclusion criterion.

Sex, gender and ascertainment

Women and girls constitute most diagnosed cohorts, but men and boys are not protected. Diagnostic stereotypes, muscularity-oriented symptoms, sport context and lower help-seeking can reduce detection in males. Gender-diverse populations are poorly represented in legacy cohorts. The nationally representative US adult study by Udo and Grilo provides DSM-5-defined estimates beyond specialist clinics, and found the odds of lifetime and 12-month diagnoses of AN, BN and BED all significantly greater for women than men after adjustment for age, race/ethnicity, education and income; adjusted odds of lifetime AN were significantly lower for non-Hispanic Black and Hispanic respondents than for white respondents (Udo 2018, PMID 29859631). Whether that last finding measures incidence or measures who gets identified as a case is precisely the ascertainment problem this section is about; country and survey design further limit transferability.

Quantitative data on gender-diverse populations remain scarce and design-dependent. In 10,415 people identifiable as transgender in a 2018 US commercial claims database (all receiving some form of gender-affirming care), 2.43% (95% CI 2.14–2.74%) had a diagnosed eating disorder, of which anorexia nervosa accounted for 0.84%; prevalence was 5.60% among those aged 12–15 and 0.52% among those aged 45–64. The authors emphasize that this is lower than self-reported estimates in transgender populations, and that identifying transgender status through gender-affirming-care claims selects people with medical access (Ferrucci 2022, PMID 35524487). Claims-based and self-report-based estimates in this population differ by an order of magnitude, and neither is clearly the better measure of underlying frequency.

Under-ascertained group Mechanism of undercount Better measurement
Males/boys Feminized stereotype; measures centered on thinness Include muscularity and sex-neutral screening
Higher-weight people Low-weight gate; clinician weight bias Record trajectory and atypical AN
Racialized minorities Access barriers; culturally narrow instruments Validate instruments and sample communities
Gender-diverse people Small cells or binary recording Self-described gender and powered subgroup analysis
Low-resource regions Sparse surveillance and specialist services Population sampling, not clinic counts

Geographic and temporal variation

GBD 2019 modelled burden across 204 countries and territories, 23 age groups and 1990–2019, using Bayesian meta-regression over systematically reviewed prevalence, incidence, remission, duration, severity and excess-mortality data. Model completeness depends on underlying observations and diagnostic comparability. One structural feature deserves emphasis: anorexia nervosa and bulimia nervosa were the only mental disorders identified as underlying causes of death in GBD 2019, so they are the only ones with years of life lost — yet the resulting eating-disorder figure of 17,361.5 YLLs (95% UI 15,518.5–21,459.8) is described by the authors themselves as "extremely low" and as not reflecting premature mortality in these populations (GBD 2019 Mental Disorders Collaborators 2022, PMID 35026139). A modelled burden estimate that concedes it under-counts death cannot be used to rank AN against other conditions. The most defensible presentation is therefore source-specific estimates side by side, with age, year, case definition and ascertainment retained; see statistics.

Apparent temporal change can reflect broader criteria, increased awareness, altered referral thresholds, disruption of routines, social-media exposure, genuine incidence change, or all of these. Administrative presentations are a health-system measure and should not be relabelled as population prevalence.

Atypical AN and the denominator

DSM-5 recognition of atypical AN changes the observable spectrum by including restrictive illness without currently low BMI. The systematic review found a sizable post-2013 literature but only 24 eligible comparative publications and little course/outcome evidence (Walsh 2023, PMID 36508318); its 2026 update grew the eligible set to 64 publications and reports that people with atypical AN differ demographically from those with AN — which is exactly the finding that makes shared denominators unsafe (Lee 2026, PMID 42557659). Studies should publish results both for AN and for the broader restrictive spectrum rather than mixing them silently.

Familial aggregation and comorbidity, measured at population scale

Register linkage converts "AN runs in families" and "AN is comorbid with OCD" from clinical impressions into estimates.

A Danish three-generation study compared 2,370 people who had any psychiatric diagnosis before age 18 and developed AN at some point with 7,035 matched controls without a pre-18 psychiatric diagnosis, using first-degree relatives' diagnoses. AN occurred significantly more often in case families than control families. Risk factors included having a sibling with AN, affective disorders in family members, and comorbid affective, anxiety, obsessive-compulsive, personality or substance-use disorders; female sex and later birth year were also associated with AN. Urbanization was not related to family load, and case relatives did not develop AN earlier than control relatives (Steinhausen 2015, PMID 24777686).

A larger two-country analysis used Danish and Swedish national registers from 1972 to 2016 covering over 67,000 individuals with eating disorders, their first-degree relatives, and matched controls drawn from populations totalling 17 million. Population-level heritability was moderate — 36% for AN, 39% for bulimia nervosa and 30% for other eating disorders — with substantial genetic overlap between AN and obsessive-compulsive disorder (rg = 0.65) and moderate correlation with autism (rg = 0.36); significant genetic associations with cardiometabolic diseases replicated across both countries (Meijsen 2025, PMID 40615413). The 36% figure sits below the 50–60% twin band quoted in genetics. The designs and case definitions differ, so it should not automatically be treated as a lower bound on the same estimand.

The AN–OCD relationship has direction. Among 6,449 Danish individuals with AN and 9,352 with OCD, high birth weight uniquely predicted subsequent OCD in the AN group (HR 3.06), while in the OCD group a history of other eating disorders strongly predicted subsequent AN (HR 7.47) and anxiety disorders in first-degree relatives were protective against developing AN (HR 0.32; HR 0.22 for female first-degree relatives) (Zhu 2025, PMID 40525476). Which disorder comes first appears to select different risk pathways — an observation with no current mechanistic explanation.

Comorbidity and burden

Mood, anxiety and obsessive-compulsive symptoms are common in clinical cohorts; timing relative to malnutrition matters because state effects can inflate symptom burden. A quality-of-life meta-analysis restricted to a single instrument, the SF-36, found significantly lower health-related quality of life than population norms in every eating-disorder group. It rests on seven studies (AN: five studies, n=227; BN: four, n=216; EDNOS: two, n=166; BED: four, n=148) and was unable to establish any difference between diagnostic groups — an underpowered null, not evidence of equivalence (Winkler 2014, PMID 24857566). Mortality burden is treated separately because follow-up duration strongly determines observed deaths (Arcelus 2011, PMID 21727255).

Open questions

  • What is the population incidence of atypical AN when substantial weight loss is measured prospectively rather than recalled? As of September 2026 no population-based incidence estimate for atypical anorexia nervosa has been published; the 64 comparative publications assembled to date are clinical-sample cross-sections (Walsh 2023, PMID 36508318; Lee 2026, PMID 42557659).
  • How much of the sex ratio reflects biology versus measurement and referral processes (Udo 2018, PMID 29859631)?
  • Can comparable surveillance be established in regions absent from longitudinal outcome evidence (Solmi 2024, PMID 38214616)?
  • How much of the doubling of register-diagnosed AN incidence is diagnostic expansion? The Danish study attributes a sizeable part of it to a general rise in first-time psychiatric diagnosis, without quantifying the residual (Steinhausen 2015, PMID 25809026).
  • Did the pandemic increase incidence or only presentations? Admissions rose ~48% and tracked public-health stringency with a rate ratio of about 2 at one year, but no population-incidence study has separated the two (Devoe 2023, PMID 35384016; Roumeliotis 2024, PMID 38976259).
  • Why do claims-based and self-report estimates of eating-disorder frequency in transgender populations differ by an order of magnitude (Ferrucci 2022, PMID 35524487)?
  • Which epidemiological risk factors act independently of inherited liability? Register-linked analysis finds risk-factor associations largely unattenuated by polygenic score but interacting with it at several levels (Papini 2024, PMID 38347808).
  • Why is register-based heritability of AN (36%) consistently below twin-based estimates (50–60%), and which is the better guide for a population risk statement (Meijsen 2025, PMID 40615413; Bulik 2006, PMID 16520436)?
  • Why does the order of onset in AN–OCD comorbidity select different risk factors — high birth weight in one direction, familial anxiety protection in the other (Zhu 2025, PMID 40525476)?
  • What accounts for familial aggregation of AN once shared genetics are accounted for, given that urbanization was unrelated to family load (Steinhausen 2015, PMID 24777686)?

References

  1. van Eeden AE, et al. Incidence, prevalence and mortality of anorexia nervosa and bulimia nervosa. Curr Opin Psychiatry. 2021. PMID 34419970.
  2. Silén Y, Keski-Rahkonen A. Worldwide prevalence of DSM-5 eating disorders among young people. Curr Opin Psychiatry. 2022. PMID 36125216.
  3. Smink FRE, et al. Epidemiology of eating disorders: incidence, prevalence and mortality rates. Curr Psychiatry Rep. 2012. PMID 22644309.
  4. Solmi M, et al. Age at onset of mental disorders worldwide: meta-analysis of 192 epidemiological studies. Mol Psychiatry. 2022. PMID 34079068.
  5. Udo T, Grilo CM. Prevalence and correlates of DSM-5-defined eating disorders in US adults. Biol Psychiatry. 2018. PMID 29859631.
  6. GBD 2019 Mental Disorders Collaborators. Global burden of 12 mental disorders, 1990–2019. Lancet Psychiatry. 2022. PMID 35026139.
  7. Walsh BT, et al. A systematic review comparing atypical anorexia nervosa and anorexia nervosa. Int J Eat Disord. 2023. PMID 36508318.
  8. Winkler LA, et al. Quality of life in eating disorders: a meta-analysis. Psychiatry Res. 2014. PMID 24857566.
  9. Arcelus J, et al. Mortality rates in patients with anorexia nervosa and other eating disorders. Arch Gen Psychiatry. 2011. PMID 21727255.
  10. Solmi M, et al. Outcomes in people with eating disorders. World Psychiatry. 2024. PMID 38214616.
  11. Lee V, Hagan KE. An invited updated systematic review and meta-analysis comparing atypical anorexia nervosa and anorexia nervosa. Int J Eat Disord. 2026. PMID 42557659.
  12. Steinhausen HC, Jensen CM. Time trends in lifetime incidence rates of first-time diagnosed anorexia nervosa and bulimia nervosa across 16 years in a Danish nationwide psychiatric registry study. Int J Eat Disord. 2015;48:845-850. PMID 25809026.
  13. Devoe DJ, et al. The impact of the COVID-19 pandemic on eating disorders: a systematic review. Int J Eat Disord. 2023;56:5-25. PMID 35384016.
  14. Roumeliotis N, et al. Pandemic stringency measures and hospital admissions for eating disorders. JAMA Pediatr. 2024;178:879-887. PMID 38976259.
  15. Ferrucci KA, et al. Prevalence of diagnosed eating disorders in US transgender adults and youth in insurance claims. Int J Eat Disord. 2022;55:801-809. PMID 35524487.
  16. Papini NM, et al. Interplay of polygenic liability with birth-related, somatic, and psychosocial factors in anorexia nervosa risk: a nationwide study. Psychol Med. 2024;54:2073-2086. PMID 38347808.
  17. Steinhausen HC, et al. A nation-wide study of the family aggregation and risk factors in anorexia nervosa over three generations. Int J Eat Disord. 2015;48:1-8. PMID 24777686.
  18. Meijsen J, et al. Shared genetic architecture between eating disorders, mental health conditions, and cardiometabolic diseases: a comprehensive population-wide study across two countries. Nat Commun. 2025;16:6193. PMID 40615413.
  19. Zhu LY, et al. Predictors of anorexia nervosa and obsessive-compulsive disorder comorbidity and order of diagnosis in a Danish national cohort. Int J Eat Disord. 2025;58:1817-1829. PMID 40525476.
  20. Bulik CM, et al. Prevalence, heritability, and prospective risk factors for anorexia nervosa. Arch Gen Psychiatry. 2006;63:305-312. PMID 16520436.