Open questions — anorexia nervosa¶
Last curated: 2026-09-30 (first periodic literature sweep; previously 2026-09-02 independent audit, then same-day depth pass). Status: draft. Tier 1 questions could change practice and are designable now; Tier 2 questions require enabling methods, larger networks or prior answers. Stable identifiers are retained across future updates.
Depth-pass note (2026-09-02). A subsequent depth pass re-searched every wiki page and widened the evidence base from 113 to 278 distinct records. Two consequences for this file: one asserted absence in OQ-8 was stale (a small direct comparison of involuntary and voluntary pathways exists) and has been rewritten; and eight new questions (OQ-18 to OQ-25) and four new junctions (D11–D14) were opened by material that had not previously been cited here.
Audit note (2026-09-02). Every "no study has…" assertion below was re-tested by live PubMed search during the independent audit. Two were stale and have been rewritten: the one-year outcome of higher- versus lower-calorie refeeding has been published (and is null), and the long-term-outcome question about compulsory treatment now has a large register cohort behind it, though not a comparative one. Questions that survive re-testing now carry an explicit dated absence statement rather than a bare gap claim.
Depth-pass audit note (2026-09-02, auditor: Codex; additions author: Claude). The new absence statements were re-searched live. The audit found an older 162-patient comparative mortality follow-up after compulsory treatment, a completed 371-dyad transition RCT, a published core-outcome-set development review, a placebo-controlled FMT protocol, and newer controlled rTMS evidence. The affected gaps are narrowed below; no claim now treats a protocol as a reported result.
Sweep note (2026-09-30, Claude; not yet audited). The first periodic sweep changed seven questions (OQ-1, OQ-6, OQ-10, OQ-16, OQ-17, OQ-20, OQ-24) and two junctions (D1, D5). OQ-16 and junction D5 asserted that weight-indexed refeeding had never been randomized; a trial doing exactly that had been registered since 2021 and posted results on 2026-09-25, so both are rewritten. OQ-17 asserted that no cohort reports course data specific to atypical AN; a re-search found a one-year clinical follow-up and a college-cohort trajectory study from 2024 that earlier passes missed, so the absence is narrowed to mortality and outcomes beyond one year. OQ-1, OQ-10, OQ-20 and OQ-24 gain one new result each, and OQ-6 now points to OQ-16 for the weight-indexed comparison. No question was closed and none was opened.
Dots not yet connected¶
| # | Dot A | Dot B | The missing junction | Powers |
|---|---|---|---|---|
| D1 | Genetic risk now maps to named metabolic and synaptic genes (LPL, BDNF, VAMP2) and to limbic/striatal GABAergic neurons, and six of the eight 2019 loci were significant again in 24,223 cases (Watson 2019, PMID 31308545; Song 2026, PMID 41927769; Termorshuizen 2026, PMID 42718672) | Treatment remains psychological and nutritional; the two drug attempts since 2019 were oxytocin (null) and olanzapine in young people (under-recruited) (Maguire 2024, PMID 38520886; Filiz 2026, PMID 42116676) | No effector gene at any genome-wide significant AN locus has been confirmed functionally, so no target exists to engage | OQ-1, OQ-2 |
| D2 | Family approaches have adolescent RCT support (Lock 2010, PMID 20921118) | Adult therapies show no consistent winner (Solmi 2021, PMID 33600749) | No dismantling/developmental trial identifies what stops transporting across age | OQ-3 |
| D3 | Mortality is 5.1/1,000 person-years; 20% of deaths are suicides (Arcelus 2011, PMID 21727255) | Trials center weight and symptom scales | Registry-linked comparative treatment studies rarely measure the defining outcome | OQ-4 |
| D4 | Atypical AN has comparable psychopathology and many medical complications (Walsh 2023, PMID 36508318) | Services often operationalize severity through current weight | The causal path from weight-based classification to delayed care and outcomes is unquantified | OQ-5 |
| D5 | Higher-calorie monitored refeeding shortens stabilization but does not improve 1-year remission or readmission (Garber 2021, PMID 33074282; Golden 2021, PMID 33753542) | Fixed kcal/day prescribing underfeeds atypical AN by ~25% per kilogram and delays its stabilization (Garber 2024, PMID 38179719) | Weight-indexed (50 kcal/kg/day) and fixed (2,000 kcal/day) prescribing have now been randomized against each other in 77 hospitalized patients with atypical AN, but only registry results exist — 7.83 vs 7.21 days to medical stability, no inferential analysis, no delivered-dose data, no 12-month outcomes (NCT04966858, posted 2026-09-25) — and no trial has enrolled below 60% median BMI | OQ-6, OQ-16 |
| D6 | Capacity may be affected through illness-shaped valuation (Jefferson 2024, PMID 39463268) | Post-compulsion mortality is now quantified but confounded by indication (Bager 2026, PMID 42383339) | No study pairs capacity-assessment reliability with the outcomes of the decisions those assessments authorize | OQ-7, OQ-8 |
| D7 | Lived experience emphasizes identity/control (Espíndola 2009, PMID 19225241) | Trials measure weight/symptoms | Identity change has rarely been tested as mediator or harm | OQ-9 |
| D8 | DBS/rTMS/psilocybin studies target circuits/flexibility (Gallop 2022, PMID 35179712) | State effects make biomarkers non-diagnostic (Himmerich 2024, PMID 38331700) | Experimental trials lack validated target-engagement chains | OQ-10 |
| D9 | Fracture excess is established at population scale — osteoporotic-fracture HR 7.50 (5.62–10.01), and present in men as well as women (Cyrenne-Dussault 2026, PMID 41109616) | Every randomized bone study in AN reports BMD, microarchitecture or turnover surrogates in samples of 21–110 (Robinson 2017, PMID 28554377; Amorim 2026, PMID 41591404) | The outcome that has been measured epidemiologically has never been the endpoint of a trial | OQ-11 |
| D10 | Recovery rises with follow-up to 67% at ≥10 years across EDs (Solmi 2024, PMID 38214616) | A register-based severity index scored on admissions, treatment length and duration predicts cause-specific mortality (Larsen 2025, PMID 40670905) | A prognostic index validated against death exists, but no one has tested whether it — or any duration threshold — discriminates well enough to guide an individual decision | OQ-12 |
| D11 | Germ-free mice colonized with AN patient microbiota reproduce food restriction, hyperactivity, anxiety-like behaviour, inflammation, liver dysfunction and disrupted ovarian follicles (Hata 2019, PMID 31504398; Gabriel-Segard 2025, PMID 41239983) | The only human trial is an open-label single-FMT feasibility study with no psychopathology endpoint met, and a rat study found no behavioural effect (Panah 2026, PMID 41535289; Kooij 2024, PMID 38934721) | Nobody has run a controlled human trial of repeated microbiome-directed intervention with a clinical endpoint, so the strongest causal evidence in AN remains species-limited | OQ-20 | | D12 | 70–93% of the economic burden of eating disorders is indirect — lost productivity and unpaid care — out of PPP-USD 70.5 billion annually (Ahmed 2025, PMID 39542867) | Cost-effectiveness analyses and service planning are conducted from a health-system perspective, where inpatient bed-days dominate (Le 2017, PMID 29044637; Guarda 2017, PMID 28130794) | No analysis has evaluated an intervention against the indirect costs that constitute most of the burden | OQ-24 | | D13 | Cognitive remediation was built to correct set-shifting and central-coherence inefficiencies, which are reliably present at g = −0.38 and −0.53 (Keegan 2021, PMID 33305366) | Two meta-analyses find CRT does not improve those constructs or eating-disorder outcomes, and poorer baseline central coherence predicted faster improvement on standard therapy (Hagan 2020, PMID 32408265; Alserihi 2024, PMID 39540007; Keegan 2022, PMID 35715854) | The mechanism-to-treatment chain has been tested and failed, and no one has revised the model that generated it | OQ-22 | | D14 | At least 60 people with eating disorders underwent assisted dying between 2012 and 2024 on rationales of irremediability, terminality and voluntary request, with most government reports too sparse to verify the psychiatric condition (Roff 2024, PMID 39143961) | Recovery in AN rose from 31.4% at 9 years to 62.8% at 22 years, about half of those unrecovered at 9 years recovering by 22, which Eddy's authors read as "arguing against the implementation of palliative care for most individuals with eating disorders" (Eddy 2017, PMID 28002660; Robison 2024, PMID 38619462) | CLOSED 2026-10-07 — this junction was already made, by Dot A's own paper. Roff 2024 evaluates the irremediability and treatment-resistance rationales against exactly this long-term outcome literature in its Discussion §5.1.2–5.1.3, concluding the rationales "lack validity and do not cohere with empirical understanding"; Crow 2023 (PMID 37057340) and Robison 2024 (PMID 38619462) cover the same ground. The row previously asserted the evaluation had "never" been done while citing the paper that does it. The distinct question that remains is now D15 | OQ-25 | | D15 | Roff's review identifies more than 60 assisted-death cases involving eating disorders, but only 19 have detailed case summaries; those 19 vary in diagnosis, age, illness duration, treatment history and use of a "treatment-resistant" rationale (Roff 2024, PMID 39143961) | In a long-term AN cohort, recovery rose from 31.4% at 9 years to 62.8% at 22 years; about half of those unrecovered at 9 years recovered by 22 (Eddy 2017, PMID 28002660) | Opened 2026-10-07 after D14 closed. Roff compares rationales with recovery evidence, but does not estimate prognosis for patients meeting a prespecified irremediability profile. Can a clearly defined combination of duration, prior treatment and clinical status predict long-term non-recovery well enough to inform an individual prognosis? This requires individual-level longitudinal data with those variables and external validation; existing cohorts have not been shown to contain every field. A group recovery probability alone cannot settle an ethical judgement | OQ-25 |
Tier 1 — practice-changing and designable now¶
OQ-1. Which metabolic phenotype is causally downstream of AN genetic liability?¶
The 2019 GWAS found correlations with glycaemic, lipid and anthropometric traits independent of common BMI variants, but correlation does not establish direction or tissue (Watson 2019, PMID 31308545). The 24,223-case follow-up reproduced the negative correlations with metabolic and anthropometric traits and estimated that a factor shared with BMI accounts for 10% (SE 1.4%) of AN genetic variance, leaving 90% non-BMI; its Mendelian randomization supported effects in both directions between BMI liability and AN risk, but results for the non-BMI component were inconsistent across methods — so direction remains open at a larger sample size (Termorshuizen 2026, PMID 42718672). A preregistered multi-ancestry study combining fine-mapping, colocalization, longitudinal metabolites and causal sensitivity analyses is feasible.
OQ-2. Can a metabolic target improve outcome when added to specialist care?¶
No clinical implication follows until a target changes and mediates a patient-important endpoint. A platform trial could require target engagement before expansion, while retaining nutrition and psychotherapy in every arm (Bulik 2022, PMID 35524137).
OQ-3. Which active ingredient in adolescent family treatment is developmentally portable?¶
FBT beats adolescent-focused therapy on some remission outcomes, while parent-focused and conjoint variants converge by follow-up (Lock 2010, PMID 20921118; Le Grange 2016, PMID 27453082). The same convergence pattern recurs with multifamily therapy, which beat single-family therapy at end of treatment (OR 2.55, 95% CI 1.17–5.52, p = 0.019) but not at 18 months (Eisler 2016, PMID 27881106) — three format comparisons producing faster rather than better recovery. The one randomized attempt to add a component for early non-responders (three sessions of intensive parental coaching) did not improve remission except where baseline parental self-efficacy was low (Lock 2024, PMID 38142046). Adding a component is not dismantling one: as of September 2026 no dismantling or component trial of family-based treatment for AN has reported. A factorial trial across older adolescents and emerging adults could separate meal support, caregiver empowerment and developmental work.
OQ-4. Can treatment effectiveness be linked to suicide and all-cause mortality?¶
The event rate is too low for ordinary trials, but national registries and target-trial emulation can link treatment exposure, confounders and cause-specific mortality (Arcelus 2011, PMID 21727255). The design must avoid immortal-time and severity-confounding biases.
OQ-5. Does weight-based diagnostic delay mediate worse atypical-AN outcomes?¶
Comparative evidence establishes substantial morbidity but little course evidence (Walsh 2023, PMID 36508318). A prospective restrictive-ED cohort should measure prior weight trajectory, referral delays, dismissive encounters, instability, treatment dose and recovery.
OQ-6. What refeeding protocol works at each physiological-risk stratum?¶
Partly answered, and the answer reframes the question. The 120-person StRONG trial supports a 2,000-kcal monitored start in its population (time to stability HR 1.67, 95% CI 1.10–2.53; 4.0 days shorter stay), but its 1-year follow-up found no difference in clinical remission (p=0.42) or medical rehospitalization (32.8% vs 35.4%) (Garber 2021, PMID 33074282; Golden 2021, PMID 33753542). The residual question is no longer "how many calories" but "calories relative to what": a secondary analysis showed the fixed kcal/day design delivered 32.4 kcal/kg to patients with atypical AN versus 43.4 to those with AN, and that dose gradient — not diagnosis — predicted slower heart-rate restoration, less weight gain and more hypomagnesaemia (Garber 2024, PMID 38179719). Systematic review still judges the evidence insufficient for severe malnutrition, and as of September 2026 no randomized refeeding trial has enrolled patients below 60% median BMI (Garber 2016, PMID 26661289). The weight-indexed comparison has since been run in atypical AN and awaits publication (NCT04966858; see OQ-16). What remains undesigned is the low-weight end: a risk-stratified adaptive trial comparing starting energy and phosphate policies against organ-level safety outcomes below 60% median BMI.
OQ-7. How reliable are capacity assessments in AN?¶
Independent assessors should evaluate the same decisions using structured and ordinary clinical methods, report agreement, and follow change after physiological stabilization (Jefferson 2024, PMID 39463268). Disagreement must be analyzed, not resolved by majority vote.
OQ-8. What are the long-term benefits and harms of compulsory treatment?¶
No longer unevidenced, but still unanswered. A Danish register cohort of 4,425 people diagnosed with AN found that the 821 (18.5%) who received involuntary treatment had adjusted hazard ratios of 2.21 (95% CI 1.54–3.17) for all-cause mortality, 3.88 (1.94–7.77) for external causes and 5.30 (2.07–13.54) for suicide (Bager 2026, PMID 42383339). The authors interpret involuntary treatment as a marker of vulnerability, not a cause. Two comparisons provide context. Twenty-three involuntarily and 25 voluntarily admitted patients differed on weight restoration at about four years but not on quality of life, BMI or mortality, and perceived necessity of involuntary treatment increased over time (Abry 2024, PMID 37690079). A historical comparison of 81 compulsory and 81 other specialist-unit admissions found that an earlier five-year mortality excess was no longer significant at about 20 years (Ward 2015, PMID 25545619). Neither comparison adequately controls indication or measures the full outcome set of survival, remission, trauma, trust and quality of life. Harmonized multi-jurisdiction prospective cohorts adjusting for indication remain the feasible design.
OQ-9. Does identity-focused change mediate recovery or treatment dropout?¶
Qualitative synthesis identifies identity/control as central, but quantitative trials rarely model it (Espíndola 2009, PMID 19225241). Longitudinal mediation studies can combine qualitative sampling with validated identity, autonomy and symptom measures.
OQ-10. Does experimental-treatment target engagement mediate durable remission?¶
Small neuromodulation and psychedelic studies need locked targets, credible controls, independent outcome assessment and ≥12-month follow-up (Gallop 2022, PMID 35179712). The scale of what exists: the entire patient-level DBS literature is 36 patients across 11 studies with no blinded stimulation-off period anywhere in it (Shaffer 2023, PMID 36724522); the largest sham-controlled rTMS study randomized 34 people to a feasibility endpoint and returned BMI d=0.2 (95% CI −0.49 to 0.90) (McClelland 2018, PMID 30012789); the psilocybin evidence is a 10-participant open-label phase 1 with safety endpoints only (Peck 2023, PMID 37488291). The one adjunctive tDCS trial to reach its planned sample illustrates the cost of reading early signals: an interim within-group improvement in 20 participants did not survive, and the final 40-participant analysis found no group × time interaction on the primary outcome (p = 0.450) (Rząd 2026, PMID 42739834). Target engagement cannot mediate an outcome that has not been measured.
OQ-11. Which bone strategy prevents fractures?¶
The fracture endpoint is no longer hypothetical. A matched cohort of 7,332 people hospitalized with AN and 73,215 controls followed to 34 years found fracture hospitalization rates of 47.6 vs 21.0 per 10,000 person-years in women and 74.0 vs 39.9 in men, with an adjusted osteoporotic-fracture hazard ratio of 7.50 (95% CI 5.62–10.01) (Cyrenne-Dussault 2026, PMID 41109616). Physiologic transdermal oestrogen raises BMD Z-scores in adolescents (Misra 2011, PMID 21698665), catch-up remains incomplete (Misra 2014, PMID 24731664), bisphosphonates raise BMD in adults while oral contraceptives do not (Robinson 2017, PMID 28554377), adding rhIGF-1 to oestradiol conferred no benefit (Singhal 2021, PMID 33693703), and teriparatide produced only trend-level structural change in 21 women (Amorim 2026, PMID 41591404). As of September 2026 no trial in anorexia nervosa has used incident fracture as a primary endpoint, and no agent is approved for bone loss in this population. A pragmatic registry trial could now be powered on the observed fracture rates.
Tier 2 — enabling evidence required¶
OQ-12. Can severe/enduring AN be defined without embedding therapeutic nihilism?¶
Blocked by weak individual prognosis, inconsistent definitions and unequal access to adequate treatment (Dalle Grave 2020, PMID 32400903; Crow 2023, PMID 37057340). One route is now open: a Danish register-based severity index built from onset age, admissions, outpatient courses, cumulative treatment length and illness duration predicted AN-specific, suicide, alcohol-related and all-cause mortality among 9,167 people, with the top quintile at markedly elevated risk (Larsen 2025, PMID 40670905). That establishes group-level prognostic signal from routine data. It does not establish discrimination or calibration adequate for an individual, and as of September 2026 no severity definition for AN has been evaluated for incremental predictive value over clinical judgement or for clinical utility.
OQ-16. Does weight-indexed refeeding close the atypical-AN outcome gap?¶
Added by the independent audit, 2026-09-02. Within a randomized inpatient trial, patients with atypical AN stabilized more slowly and gained less weight than those with AN, and the difference tracked kcal/kg received rather than diagnosis; every 10 kcal/kg was associated with heart-rate restoration 1.7 days faster (95% CI 1.0–2.5) and 1.6 %mBMI more weight gain (0.8–2.4) (Garber 2024, PMID 38179719). This is a within-trial observational gradient, not a randomized contrast.
Randomized, not yet answered (sweep, 2026-09-30). The previous version of this question said such a trial was "designable now". It had in fact been registered since July 2021 and is complete: i-STRONG randomized 77 hospitalized, medically unstable 12–24-year-olds with DSM-5 atypical AN at two US centres to 50 kcal/kg/day (capped at 3,000 kcal/day) or 2,000 kcal/day starts, and posted results on 2026-09-25 (NCT04966858, retrieved from the ClinicalTrials.gov v2 API on 2026-09-30). In 74 analysed participants mean time to medical stability was 7.83 days (SE 1.54) with weight-indexed dosing versus 7.21 days (SE 0.58) with the fixed start; low electrolytes were uncommon in both arms (hypophosphataemia 0 vs 2, hypomagnesaemia 5 vs 4) and serious adverse events occurred in 4 of 38 versus 6 of 36. The registry record posts no statistical comparison, no delivered kcal/kg and no data for the pre-specified 12-month %mBMI and EDE-Q outcomes, and a PubMed search on 2026-09-30 located no publication. The posted numbers do not favour weight-indexed dosing on the primary outcome, but they cannot establish equivalence. The question therefore narrows to three things the publication must supply: the inferential result, whether the intended dose contrast was achieved, and whether one-year weight suppression — the variable on which atypical AN diverged from AN in the earlier cohort (Golden 2024, PMID 37507351) — differs by arm.
OQ-17. What is the course of atypical anorexia nervosa?¶
Added by the independent audit, 2026-09-02. Two systematic reviews three years apart — 24 comparative publications in 2023, 64 in 2026 — both close by stating that longitudinal course, outcome and treatment-response evidence is missing (Walsh 2023, PMID 36508318; Lee 2026, PMID 42557659). Sharpened by the 2026-09-30 sweep: the absence claimed here was too broad. Two course studies published in 2024 were located by a title-restricted re-search. One year after medical hospitalization, 48 participants with atypical AN had regained less %mBMI (p = .02) and remained more weight-suppressed (p = .002) than 63 with AN, while EDE-Q improvement, resumption of menses (80% vs 76.9%) and rehospitalization (29.2% vs 37.9%) did not differ (Golden 2024, PMID 37507351). In a prospective college cohort, 125 students with a history of atypical AN had symptoms above controls but below AN and bulimia nervosa, with rising anxiety and stable binge eating (Hagan 2024, PMID 38303677). As of September 2026 no cohort study reports mortality, or recovery and relapse rates beyond one year, specific to atypical AN, and no population-based incidence estimate exists; the searches behind that statement are listed in the 2026-09-30 log entry. The design need is unchanged — a prospective restrictive-spectrum cohort with multi-year follow-up — but its first question is now specified by data: whether the persistent weight suppression seen at one year predicts later relapse.
OQ-13. Is any “terminal AN” classification empirically defensible?¶
Largely resolved in the negative, though the underlying clinical problem is not. The criteria proposed in 2022 were AN diagnosis, age over 30, previous participation in high-quality care, and a capacitous, consistent determination that further treatment is futile (Gaudiani 2022, PMID 35168671). An empirical evaluation applied the first three to 782 patients in US treatment facilities: 51 met them, 16 also met a proxy for the fourth, and both groups showed substantial internal variability and an overall trend of improvement across physiological and self-report measures — evidence against specifying the diagnosis (Robison 2024, PMID 38619462). In 2025 the proposal's lead author expressly disavowed the concept and the phrase (Gaudiani 2025, PMID 40361218). A narrative review of 57 articles maps three arguments for and four against and concludes that the sides lack shared understanding of the value judgements involved (Connor 2026, PMID 42312722), while a stakeholder study found most respondents critical of three of four criteria, most sharply the age threshold (Robb 2026, PMID 42410953). The residual empirical question is narrow: the one direct test drew on people in active treatment and could not observe those who had disengaged from care. The residual clinical question — how to support people for whom repeated treatment has not worked — is untouched by the disavowal and is the more important one (Westmoreland 2021, PMID 34763793; Crow 2023, PMID 37057340; Bauschka 2025, PMID 40760030).
OQ-14. Which biomarkers distinguish liability, state and scar?¶
Blocked by cross-sectional designs, starvation confounding and small samples. Repeated-measures, recovered and familial-risk cohorts with external replication are prerequisites (Himmerich 2024, PMID 38331700).
OQ-15. What outcome set defines recovery across age and weight spectrum?¶
Recovery estimates change with definition and follow-up; pooled recovery rose from 42% before two years to 67% at ten years or longer (Solmi 2024, PMID 38214616), while the largest adult trial's 5-year assessment classified 41% as fully recovered, 41% partially and 18% still fully symptomatic on observed data (Herzog 2022, PMID 35294860). Quality of life is measurably below population norms across every eating disorder, but the meta-analysis establishing that could not distinguish between diagnoses and rests on seven studies (Winkler 2014, PMID 24857566). Stakeholder work shows what such a set would have to contain and where the groups diverge: interviews with 24 patients aged 12–23, 20 parents and 34 clinicians produced four recovery domains — psychological well-being, eating-related behaviours/attitudes, physical markers and self-acceptance of body image — with clinicians significantly more likely than patients and parents to endorse psychological well-being (χ² = 9.90, p = .007) and physical markers (χ² = 6.42, p = .04), and no group difference on eating behaviours or body acceptance (Richmond 2020, PMID 32453448). A 2024 review then documented 196 outcome measures across 91 eating-disorder RCT reports and proposed a Delphi process rather than delivering a completed core set (Brieva-Toloza 2024, PMID 38389169). Consensus still needs to retain weight trajectory, core symptoms, function, quality of life, safety and patient-valued autonomy.
OQ-18. Does screening for eating disorders change outcomes?¶
Opened by the depth pass, 2026-09-02. The SCOFF performs adequately as a test — pooled sensitivity 0.86 (95% CI 0.78–0.91) and specificity 0.83 (0.77–0.88) across 25 validation studies, and 84%/80% in adults in an independent review — but its accuracy is highest in case-control samples of young women with AN and bulimia nervosa and lower in community samples, in men and where binge-eating disorder is included; no included study used a reference standard covering all DSM-5 eating disorders (Kutz 2020, PMID 31705473; Feltner 2022, PMID 35289875). The decisive gap is downstream: across 57 studies (N=10,773) no study directly evaluated the benefits and harms of screening, and none of 40 intervention RCTs enrolled a screen-detected population, which is why the USPSTF issued an I statement (Davidson 2022, PMID 35289876). A screening trial with a treatment pathway and clinical endpoints is designable now and has never been run. Note that this is a different question from reducing delay in people who have already presented, where quasi-experimental early-intervention evidence does exist (Flynn 2021, PMID 33112472).
OQ-19. What is in the 80% of registered AN trials that never publish?¶
Opened by the depth pass, 2026-09-02. Of 201 AN trials registered on ClinicalTrials.gov between 2000 and 2018, 101 were completed, 41 were published, 8 showed evidence of prospective registration, and 7 had primary findings replicated (Murray 2020, PMID 31638722). Every synthesis in this knowledge base — including the network meta-analysis that screened 14,003 reports to find 16 trials (Solmi 2021, PMID 33600749) — operates downstream of that filter and cannot recover what it removed. The tractable version of this question is an audit that contacts investigators of completed-but-unpublished registrations and characterizes results and reasons for non-publication; it requires no new patients.
OQ-20. Is gut dysbiosis in AN a contributor or a consequence?¶
Opened by the depth pass, 2026-09-02. Two independent germ-free transfer experiments reproduce AN-like phenotypes in mice colonized with patient microbiota — reduced weight gain and food efficiency, anxiety-like and compulsive behaviour, lower brainstem serotonin (Hata 2019, PMID 31504398); food restriction, hyperactivity, inflammation, liver dysfunction and disrupted ovarian follicles (Gabriel-Segard 2025, PMID 41239983). A faecal transplant into rats did not alter behavioural flexibility (Kooij 2024, PMID 38934721), and the reported human study is an open-label single-FMT feasibility trial in which the microbiota shifted toward donor composition at one week with no change in psychopathology or appetite-related biomarkers (Panah 2026, PMID 41535289, NCT05834010). A placebo-controlled 20-participant adolescent FMT feasibility trial is now published as a protocol, but has no results (Couturier 2026, PMID 42009386). A second randomized pilot is registered but not yet recruiting: 36 patients allocated to psychotherapy as usual with or without a single FMT, unmasked, with feasibility as the primary outcome (NCT07574892, first posted 2026-05-08, retrieved 2026-09-30). A third, first posted 2026-10-16 and not yet recruiting, is double-masked and placebo-controlled — a crossover of capsule FMT in 60 females with AN in Denmark — but its primary outcome is change in microbiome composition at five weeks, not a clinical measure (NCT07877675, retrieved 2026-10-21). No placebo-controlled clinical outcome has therefore been reported, and the first placebo-controlled adult trial is designed to show engraftment rather than benefit.
OQ-21. Is diagnostic crossover in AN the rule or the exception?¶
Opened by the depth pass, 2026-09-02. Seven years of weekly prospective interview data in 216 women found that the majority of those with AN crossed diagnoses, more than half between AN subtypes, and concluded that the subtyping scheme is not supported (Eddy 2008, PMID 18198267). A Swedish register study of 9,622 treatment-seeking individuals seen at least twice found diagnostic instability common but transition between threshold diagnoses infrequent, and concluded there is more temporal continuity than previously reported (Schaumberg 2019, PMID 29911514). Both can be locally correct — a register records what was coded at a contact, an interview records what was true each week — but the field cannot currently say whether restricting and binge/purge subtypes designate anything stable, and cohorts continue to be recruited and compared on subtype. The resolving design is a register-linked cohort with periodic structured interview in the same individuals.
OQ-22. If cognitive remediation does not move its own targets, what does the neuropsychological model predict?¶
Opened by the depth pass, 2026-09-02. Set-shifting and central-coherence inefficiencies in AN are real but modest (g = −0.38 and −0.53) and are not specific: bulimia nervosa shows comparable deficits that do not significantly differ from AN, so they cannot be attributed to low weight (Keegan 2021, PMID 33305366). Two meta-analyses find CRT does not improve central coherence versus control (g = 0.25, 95% CI −0.35 to 0.85) or eating-disorder outcomes (Hagan 2020, PMID 32408265; Alserihi 2024, PMID 39540007). A secondary analysis of a three-arm psychotherapy trial found that participants with poorer baseline central coherence improved faster on standard therapy (Keegan 2022, PMID 35715854). Three independent results now point away from treating these constructs as therapeutic targets, and no revised model has been proposed. The same question applies to reward prediction error, which is elevated in AN but scales inversely with BMI across the whole eating-disorder spectrum (Frank 2021, PMID 34190963) — a trait marker and a starvation readout are not distinguishable in any published design.
OQ-23. Should refeeding targets be indexed to premorbid rather than population weight?¶
Opened by the depth pass, 2026-09-02. In 201 adolescents with restrictive eating disorders across a wide BMI range followed for a year in a family-based programme, recovery was independently associated with lower presenting EDE-Q, higher weight gain at three months, and lower weight suppression at follow-up — the gap between premorbid and current BMI (Swenne 2017, PMID 28915806). This is the endpoint analogue of the dosing problem in OQ-16: fixed kcal/day underfeeds heavier patients (Garber 2024, PMID 38179719), and a population-referenced target under-restores patients whose premorbid trajectory was high. As of September 2026 no trial has randomized a premorbid-indexed weight target against a population-referenced one. Both arms are deliverable within existing protocols. Sharpened by the 2026-10-21 sweep. Two observational adolescent cohorts now bear on whether full premorbid restoration is even the right endpoint. In 99 girls followed after inpatient or family-based treatment, menses resumed at a median 92.0 %mBMI, 6.5 units below premorbid %mBMI (98.5; p < 0.001), and the authors conclude that full premorbid restoration is not necessary for menstrual recovery (Woltmann 2026, PMID 42314025). In 214 adolescents with AN or atypical AN after inpatient discharge, those with premorbid overweight or obesity were less likely to reach ≥99% of treatment goal weight at every follow-up, yet 19 of the 48 who did not reach it within a year (40%) met the study's a priori recovery criteria (Franklin 2026, PMID 42868898). The question is therefore not only premorbid versus population reference, but whether any single weight target should gate recovery. The absence statement was re-tested on 2026-10-21 in PubMed (anorexia nervosa × target/goal/premorbid weight × randomized; 5 records, none randomizing a target) and in ClinicalTrials.gov (condition AN × the same weight terms; 5 records, none randomizing a target); it holds in those two sources.
OQ-24. Can any intervention be shown to reduce the indirect burden that constitutes most of the cost?¶
Opened by the depth pass, 2026-09-02. The nationwide annual financial cost of eating disorders is estimated at PPP-USD 70.5 billion, of which 70–93% is indirect — lost productivity and unpaid care — alongside PPP-USD 355.6 billion in intangible burden (Ahmed 2025, PMID 39542867). Every cost-effectiveness analysis located here takes a health-system perspective in which inpatient bed-days dominate (Le 2017, PMID 29044637; Mayr 2025, PMID 39776084; Guarda 2017, PMID 28130794), so the modelled savings come from avoided admissions rather than from restored function or reduced caregiving. Carer-directed intervention is the one tested route into that territory, and it reduced caregiving time and inpatient bed-days with small patient effects (Hibbs 2015, PMID 27703724). What the indirect burden looks like at the level of a person is now partly described: in the Northern Finland Birth Cohort 1986, 165 people with AN had lower cumulative income in 2011–2016 than 4,884 cohort members in both sexes and more unemployment days among those with any unemployment, but no difference in sickness-absence days (Majuri 2026, PMID 40858910). If that holds elsewhere, the productivity loss in AN runs through earnings and employment rather than through recorded sick leave. A trial powered on productivity and caregiver-time outcomes has not been run.
OQ-25. What evidence standard should govern irremediability judgements in eating disorders?¶
Opened by the depth pass, 2026-09-02. At least 60 people with eating disorders underwent assisted dying between 2012 and 2024 across multiple jurisdictions, on rationales grouped as irremediability, terminality and voluntary request — with most government reports lacking data descriptive enough to verify the psychiatric condition, and the reviewers judging that in many cases the rationales "lack validity and do not cohere with empirical understanding" (Roff 2024, PMID 39143961). The empirical position against irremediability is not weak: AN recovery rose from 31.4% at 9 years to 62.8% at 22 years in the longest cohort (Eddy 2017, PMID 28002660), remission reached 30–40% even in a 1,693-patient severe inpatient series (Fichter 2017, PMID 28644530), the only direct test of proposed terminal criteria found improvement rather than decline (Robison 2024, PMID 38619462), and the proposal's lead author has disavowed the term (Gaudiani 2025, PMID 40361218). This is a question about the standard of proof required before an irreversible decision, and it is currently answered differently by different jurisdictions on records that cannot be audited. It is Tier 2 because the enabling step is reporting quality, not a study.