Skip to content

Mortality and long-term outcome

TL;DR — AN has markedly elevated mortality, from both medical causes and suicide. The 2011 meta-analysis estimated 5.1 deaths per 1,000 person-years and an SMR of 5.86 across 166,642 AN person-years, with one in five deaths by suicide (Arcelus 2011, PMID 21727255). Two independent 2026 syntheses on much larger samples reproduce that magnitude rather than revising it: an all-cause mortality risk ratio of 5.52 (95% CI 4.47–6.82) and suicide RR of 9.86 (95% CI 5.63–17.27) for AN across 83 studies (Semchishen 2026, PMID 41536100), and a pooled SMR of 5.06 (95% CI 3.47–7.38) from 30 studies (Lai 2026, PMID 41277145). Recovery accumulates over long follow-up, so short studies overstate chronicity; across eating disorders, pooled recovery rose from 42% before two years to 67% at ten years or longer (Solmi 2024, PMID 38214616). Definitions of recovery and relapse remain inconsistent.

Mortality estimates

Source Design Exposure/follow-up Estimate Constraint
Arcelus 2011 36-study meta-analysis 166,642 AN person-years 5.1 deaths/1,000 person-years; SMR 5.86 Studies through 2010; heterogeneous cohorts
Arcelus 2011 Cause synthesis Deaths in AN cohorts 20% of deaths were suicide Proportion among deaths, not patient risk
Solmi 2024 415 studies, all EDs Mean follow-up 72.2 months for mortality analyses Pooled mortality 0.4% (95% CI 0.2–0.7); 5.2 deaths/1,000 person-years (95% CI 4.4–6.1) in observational studies Transdiagnostic; follow-up varies
Semchishen 2026 83 studies; 307,710 ED patients vs 15.7 million controls; mean follow-up 11.96 yr Relative risk vs general population Any ED all-cause RR 4.92 (4.03–6.00); AN 5.52 (4.47–6.82); AN suicide RR 9.86 (5.63–17.27) Cohort/case-control designs; 13.3% rated poor quality
Semchishen 2026 Meta-regression Aggravating factors Higher all-cause mortality with lower BMI, shorter follow-up, male sex, and comorbid psychiatric, substance, alcohol, mood or personality disorder Ecological/study-level moderators, not individual predictors
Lai 2026 30 studies, 33,176 AN patients; 22 pooled All-cause SMR SMR 5.06 (3.47–7.38); suicide 21% and cardiac causes 19% of deaths Studies with zero deaths excluded; specialist-clinic samples dominate
Lai 2026 Sex-specific subgroups Male vs female samples Male SMR 3.47 (1.60–7.52) vs female 3.86 (1.82–8.20) — overlapping Male samples few and small
Bager 2026 Danish register cohort, 4,425 people with AN, 2000–2016 First involuntary-treatment event vs none All-cause HR 2.21 (1.54–3.17); external causes 3.88 (1.94–7.77); suicide 5.30 (2.07–13.54) Confounded by indication: compulsion marks vulnerability

These quantities answer different questions. The mortality rate uses person-time; the SMR compares observed deaths with population expectations; a risk ratio compares a cohort with matched or general-population controls; cumulative mortality depends on follow-up; and the proportion of deaths due to suicide does not equal suicide incidence (Arcelus 2011, PMID 21727255). Three meta-analyses spanning 2011 to 2026, using different pools and different comparators, report a roughly five-fold all-cause excess (Arcelus 2011, PMID 21727255; Semchishen 2026, PMID 41536100; Lai 2026, PMID 41277145). Because their cohorts and periods overlap and heterogeneity is substantial, that agreement does not by itself establish whether mortality changed over calendar time.

Both 2026 syntheses converge on cause structure as well: suicide accounts for about one fifth of deaths (21% in Lai 2026), and Lai and colleagues separately identify cardiac causes at 19% — a proportion that no earlier AN-specific meta-analysis had isolated (Lai 2026, PMID 41277145).

A dissenting estimate and an intermediate one

Not every synthesis returns the same magnitude, and the disagreement is methodological rather than about the data. A re-analysis of 41 cohorts from 40 studies published 1966–2010 — the same era Arcelus 2011 covered — applied double data extraction and log-linear regression with an over-dispersed Poisson model and obtained an all-cause SMR of 5.2 (95% CI 3.7–7.5) and a suicide-specific SMR of 18.1 (11.5–28.7) against 15–34-year-old females. Both were lower than previously published figures — by 10% for all-cause and by 49% for suicide — and the authors argued the earlier estimates were inflated by method rather than by cohort selection (Keshaviah 2014, PMID 25214371). The all-cause figure is essentially unchanged; the suicide figure is halved. Anyone quoting a suicide SMR for AN should say which model produced it.

An update covering 2010–2024 across four databases pooled 74 effect sizes and returned a weighted SMR for any eating disorder of 3.39 (95% CI 2.90–3.95) with severe heterogeneity (I² = 95.1%), and AN-specific SMR of 5.21 from 30 studies — above eating-disorder-not-otherwise-specified (2.51), bulimia nervosa (2.20) and binge-eating disorder (1.46) (Krug 2025, PMID 39889307). Placing four syntheses side by side (Arcelus 5.86; Keshaviah 5.2; Krug 5.21; Lai 5.06), the pooled AN all-cause SMR repeatedly falls near five. This consistency describes the available cohorts; it is not a secular-trend analysis.

When the deaths occur

Mortality in AN is not evenly distributed over the illness. In a treatment-seeking cohort of 246 women interviewed every six months for a median of 9.5 years and followed for vital status to a median of 20 years, there were 16 deaths (6.5%). The all-cause SMR was 4.37 (95% CI 2.4–7.3) for lifetime AN, but risk peaked within the first 10 years of follow-up, where the SMR was 7.7 (95% CI 3.7–14.2). SMR also varied by illness duration: 3.2 (0.9–8.3) for 0–15 years of lifetime AN (4/119 died) versus 6.6 (3.2–12.1) for >15–30 years (10/67 died). Multivariate predictors of death were alcohol abuse, low BMI and poor social adjustment (Franko 2013, PMID 23771148). Duration and the early post-presentation decade both carry risk, which is why a single pooled SMR conceals the shape of the hazard.

The palliative-care argument meets its longest follow-up

The claim that AN of more than a decade's duration warrants palliative management is directly testable against cohorts that ran that long, and the longest one contradicts it. In the Massachusetts General Hospital longitudinal study, 228 women with DSM-III-R/DSM-IV AN or bulimia nervosa were assessed at 9 years and again at a mean of 22.1 years (77% re-interviewed; multiple imputation for the remainder). Recovery rose from 31.4% at 9 years to 62.8% at 22 years in AN, while bulimia nervosa was 68.2% at both timepoints. About half of the AN participants who had not recovered by 9 years went on to recover by 22. Early recovery strongly predicted long-term recovery in AN (OR 10.5, 95% CI 3.77–29.28, p < .01) but carried no information in bulimia nervosa (OR 1.0, 95% CI 0.49–2.05) (Eddy 2017, PMID 28002660). The authors state the implication directly: the trajectory argues against implementing palliative care for most individuals with eating disorders. This is the single most load-bearing dataset in the terminal-AN dispute — see severe and enduring illness and compulsory treatment.

A clinical cohort of a different design gives a more sober figure from the severe end. Following 1,693 consecutively admitted inpatients of a specialized hospital over 25 years (mean 10 years' follow-up, with a 112-patient 20-year subsample), remission was 30% in the total sample and 40% in the 20-year subsample; BMI, eating behaviour and general psychopathology improved but did not reach healthy-control levels; crossover to binge-eating disorder or obesity was rare, while crossover to eating disorder not otherwise specified was the main transition. Predictors of poor course were lower admission BMI, higher Eating Disorder Inventory Maturity Fears score, fewer follow-up years and higher age at admission — and notably not psychiatric comorbidity. Motherhood was associated with better outcome (Fichter 2017, PMID 28644530).

Both cohorts point the same way on the central question and disagree on the level. Recovery keeps accruing for decades; what fraction reaches it depends heavily on how severe the starting sample was, whether recovery is defined behaviourally or as full remission, and whether the analysis imputes for those lost.

The historical baseline

Reading these figures against the 20th-century baseline is what makes the absence of progress visible. A review of 119 study series covering 5,590 patients found less than half of survivors recovered on average, one third improved and 20% remained chronically ill; longer follow-up and younger onset age predicted better outcome; and there was "no convincing evidence that the outcome of anorexia nervosa improved over the second half of the last century" (Steinhausen 2002, PMID 12153817). Vomiting, bulimic features, purgative abuse, chronicity and obsessive-compulsive personality features were the clearest unfavourable prognostic markers. The 2024 transdiagnostic pooled recovery of 46% is not distinguishable from that historical figure.

Causes and pathways

Medical pathways include complications of malnutrition, electrolyte disturbance, cardiac dysfunction, infection and organ failure (Westmoreland 2016, PMID 26169883). Suicide is a distinct and substantial pathway requiring direct assessment rather than inference from weight. The comorbidity signal is now quantified rather than assumed: in the 2026 meta-regression, male sex and comorbid psychiatric, substance-use, alcohol-use, mood and personality disorders were each significantly associated with higher all-cause mortality across eating disorders, and mortality rose as mean BMI fell (Semchishen 2026, PMID 41536100).

Register work has begun to convert "severity" from a clinical impression into a measurable exposure. The Danish AN register-based severity index (AN-RSI), built from age at onset, number of inpatient admissions and outpatient courses, cumulative treatment length and illness duration and scored five years after first diagnosis, predicted death from AN, suicide, alcohol-related causes and any cause in 9,167 diagnosed individuals (132 deaths: 17 from AN, 30 from suicide, 85 other); the top AN-RSI quintile carried markedly higher mortality (Larsen 2025, PMID 40670905). This matters for the severe-and-enduring debate because it is prognostic information derived from routinely collected data rather than from a duration threshold.

Recovery, improvement and chronicity

The 2024 meta-analysis included 88,372 participants across 415 eating-disorder studies. Overall recovery was 46% (95% CI 44–49); chronicity was 25% (95% CI 23–29). Recovery increased with time: 42% at <2 years, 43% at 2–<4, 54% at 4–<6, 59% at 6–<8, 64% at 8–<10 and 67% at ≥10 years (Solmi 2024, PMID 38214616).

Outcome domain Examples of definitions Why estimates diverge
Weight BMI threshold; % median BMI; expected weight Different cut-offs and developmental handling
Behaviour No restriction, binge/purge or compensatory exercise Recall and observation windows vary
Psychopathology EDE/EDE-Q within population range Instrument and threshold differ
Function School/work, relationships, quality of life Often omitted
Composite recovery Weight + symptoms ± function Stringency changes numerator
Quality of life SF-36 or equivalent Significantly below population norms in every eating-disorder group, but the meta-analysis establishing this rests on seven studies and could not distinguish between diagnoses (Winkler 2014, PMID 24857566)
Relapse Loss of remission, rehospitalization or renewed criteria Index remission and follow-up differ

Relapse after adolescent treatment

In an exploratory 2–4-year follow-up of 79 of 121 original trial participants, only two of those in full remission at one year relapsed (6.1%), while ten additional participants achieved remission; selective follow-up and small denominators limit inference (Le Grange 2014, PMID 25440306). The result supports measuring delayed recovery, not declaring treatment completion at one year.

Adult long-term evidence

ANTOP randomized 242 adult female outpatients (BMI 15.0–18.5 kg/m², mean 16.7) to focal psychodynamic therapy, CBT-E or optimized treatment as usual. All groups gained BMI during treatment (0.73, 0.93 and 0.69 kg/m² respectively, no between-group differences) and gained further by 12 months (1.64, 1.30, 1.22 kg/m²); attrition was 54/242 (22%) at end of treatment and 73/242 (30%) by 12-month follow-up (Zipfel 2014, PMID 24131861).

At a mean 5.96 years (SD 0.2) after randomization, 154 of 242 (64%) were reassessed. Estimated mean BMI had reached 18.64–18.99 kg/m² across arms with no significant between-group difference, and on observed data 41% (95% CI 33–49) had fully recovered, 41% (33–49) partially recovered and 18% (12–24) still met full-syndrome AN; one participant died by suicide between the 1-year and 5-year assessments. Baseline BMI (p=0.0021), illness duration (p=0.0004) and baseline depression (p=0.012) predicted 5-year BMI (Herzog 2022, PMID 35294860). Long-term naturalistic care and 36% loss to follow-up complicate causal attribution, but the pattern — continued group-level improvement with roughly one in five still fully ill at five years — is the most durable adult outcome estimate in the randomized literature.

Prognostic interpretation

Duration of illness, age, low weight, purging, comorbidity and previous treatment are repeatedly studied, and ANTOP confirmed baseline BMI, illness duration and depression as 5-year predictors at group level (Herzog 2022, PMID 35294860). Prediction for an individual nonetheless remains weak. This is central to the “terminal AN” dispute: population-level poor prognosis does not identify a person for whom further recovery is impossible (Crow 2023, PMID 37057340).

Atypical AN course data remain the single clearest gap in this page. The 2023 systematic review of 24 comparative publications found "little information … on the course, outcome, and treatment response" (Walsh 2023, PMID 36508318); its invited 2026 update, now covering 64 publications, still closes by calling for longitudinal studies to clarify illness course and outcomes (Lee 2026, PMID 42557659). As of September 2026 no cohort study has reported long-term mortality or recovery rates specific to atypical anorexia nervosa — three years and forty additional comparative publications have deepened the cross-sectional picture without producing a single course estimate. Absence of a low BMI should not be treated as evidence of benign outcome.

Open questions

  • Which recovery definition best predicts sustained function, quality of life and survival (Solmi 2024, PMID 38214616)?
  • Can registry linkage make suicide and all-cause mortality feasible outcomes for treatment comparisons, rather than for exposure descriptions? Danish register work has now delivered both a severity index and an involuntary-treatment mortality estimate, but neither compares treatments (Larsen 2025, PMID 40670905; Bager 2026, PMID 42383339).
  • Which prognostic model is calibrated well enough for individual decisions in severe enduring illness (Crow 2023, PMID 37057340)?
  • Which suicide SMR is correct? Methodological re-analysis of the same 1966–2010 literature lowered the suicide-specific estimate by 49% while leaving the all-cause estimate essentially unchanged (Arcelus 2011, PMID 21727255; Keshaviah 2014, PMID 25214371).
  • Why has the AN all-cause SMR remained near 5 across four independent syntheses spanning three decades of literature — is this stable disease biology, stable ascertainment, or the absence of any intervention that targets mortality (Arcelus 2011, PMID 21727255; Keshaviah 2014, PMID 25214371; Krug 2025, PMID 39889307; Lai 2026, PMID 41277145)?
  • Is the hazard of death front-loaded? One cohort finds SMR 7.7 within the first 10 years of follow-up but also higher SMR with >15 years of illness; no synthesis has modelled the hazard shape directly (Franko 2013, PMID 23771148).
  • Does recovery continue to accrue after two decades, and does the 31%→63% AN trajectory replicate in a cohort recruited under DSM-5 criteria (Eddy 2017, PMID 28002660; Fichter 2017, PMID 28644530)?

References

  1. Arcelus J, et al. Mortality rates in patients with anorexia nervosa and other eating disorders: a meta-analysis of 36 studies. Arch Gen Psychiatry. 2011;68:724-731. PMID 21727255.
  2. Solmi M, et al. Outcomes in people with eating disorders: a systematic review, meta-analysis and meta-regression. World Psychiatry. 2024. PMID 38214616.
  3. Westmoreland P, et al. Medical complications of anorexia nervosa and bulimia. Am J Med. 2016. PMID 26169883.
  4. Le Grange D, et al. Relapse from remission at two- to four-year follow-up in two treatments for adolescent anorexia nervosa. J Am Acad Child Adolesc Psychiatry. 2014. PMID 25440306.
  5. Zipfel S, et al. ANTOP: focal psychodynamic therapy, CBT and optimized treatment as usual. Lancet. 2014. PMID 24131861.
  6. Herzog W, et al. ANTOP 5-year follow-up. Lancet Psychiatry. 2022. PMID 35294860.
  7. Crow SJ, et al. Terminal anorexia nervosa cannot currently be identified. Int J Eat Disord. 2023. PMID 37057340.
  8. Walsh BT, et al. A systematic review comparing atypical anorexia nervosa and anorexia nervosa. Int J Eat Disord. 2023. PMID 36508318.
  9. Winkler LA, et al. Quality of life in eating disorders: a meta-analysis. Psychiatry Res. 2014. PMID 24857566.
  10. Semchishen SN, et al. All-cause and cause-specific mortality risk in individuals with eating disorders: systematic review and meta-analysis of relative risk and aggravating or attenuating factors. World Psychiatry. 2026;25:125-140. PMID 41536100.
  11. Lai ET, et al. Systematic review and meta-analysis of mortality in patients with anorexia nervosa. Int J Eat Disord. 2026;59:424-449. PMID 41277145.
  12. Bager L, et al. Excess mortality among patients with anorexia nervosa treated involuntarily. Int J Eat Disord. 2026. PMID 42383339.
  13. Larsen JT, et al. Construction of a register-based severity index for anorexia nervosa in Denmark: association with overall and cause-specific mortality. Eur Psychiatry. 2025;68:e104. PMID 40670905.
  14. 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.
  15. Keshaviah A, et al. Re-examining premature mortality in anorexia nervosa: a meta-analysis redux. Compr Psychiatry. 2014;55:1773-1784. PMID 25214371.
  16. Krug I, et al. A meta-analysis of mortality rates in eating disorders: an update of the literature from 2010 to 2024. Clin Psychol Rev. 2025;116:102547. PMID 39889307.
  17. Franko DL, et al. A longitudinal investigation of mortality in anorexia nervosa and bulimia nervosa. Am J Psychiatry. 2013;170:917-925. PMID 23771148.
  18. Eddy KT, et al. Recovery from anorexia nervosa and bulimia nervosa at 22-year follow-up. J Clin Psychiatry. 2017;78:184-189. PMID 28002660.
  19. Fichter MM, et al. Long-term outcome of anorexia nervosa: results from a large clinical longitudinal study. Int J Eat Disord. 2017;50:1018-1030. PMID 28644530.
  20. Steinhausen HC. The outcome of anorexia nervosa in the 20th century. Am J Psychiatry. 2002;159:1284-1293. PMID 12153817.