Treating, Masking, or Mismeasuring? A Measurement-First Reframing of GLP-1 Receptor Agonists Across the Eating-Disorder Spectrum

Abstract

Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) and the dual GIP/GLP-1 co-agonist tirzepatide are increasingly prescribed for obesity and type 2 diabetes in populations enriched for binge-spectrum eating pathology. A recurring controversy asks whether these agents treat eating pathology or merely mask it. This critical narrative review argues that, posed as a simple binary, the question is under-specified, and reframes it around three problems. The central is a measurement problem: several intended effects of GLP-1 RAs are scored as improvement by eating-disorder instruments, creating a diagnostic blind spot. In the one disorder where a weight-acting drug has been tested against a placebo, efficacy on weight coincided with no effect on the psychological core—the dissociation that confounds measurement during GLP-1 RA therapy. The second is a phenotype problem: because BED subtypes already moderate response to drug versus psychological treatment, an agent acting on appetite and reward should not act uniformly, though this remains a hypothesis. The third concerns the post-discontinuation trajectory, and a therapy–harm asymmetry emerges across the spectrum. A factorial, phenotype-stratified trial with disorder-core endpoints and drug-free follow-up could resolve these questions; meanwhile, prudent practice combines uncontaminated pre-treatment screening, monitoring not reliant on confounded self-report, discontinuation planning, and integration with psychological care.

1. Introduction

Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) and the dual glucose-dependent insulinotropic polypeptide/GLP-1 co-agonist tirzepatide have transformed the pharmacotherapy of type 2 diabetes (T2D) and obesity. Their use is expanding rapidly, including substantial off-label prescribing, telehealth-mediated supply, and compounded-product use, often accompanied by minimal baseline assessment and follow-up.

These agents are being introduced into populations enriched for binge-spectrum eating pathology. Binge eating disorder (BED) is the most prevalent eating disorder in adults and is markedly over-represented among individuals seeking weight management, although it still affects a minority of those with obesity [1]. Bulimia nervosa (BN) is less prevalent but carries serious morbidity; anorexia nervosa (AN), although less prevalent still, has among the highest mortality of any psychiatric disorder. How frequently these conditions occur among the growing number of people receiving GLP-1 RAs, particularly outside structured clinical settings, is not well characterized.

A recurring controversy is sometimes posed as a binary: do GLP-1 RAs treat eating pathology, or merely mask it? This review does not claim that the field as a whole holds this naive position. On the contrary, recent reviews and clinical commentaries explicitly recognize that benefits and risks vary by diagnosis, presentation, mechanism and treatment context [2,3,4,5,6]. The problem is not that the field believes the question is binary; it is that this shared recognition of heterogeneity has not yet been translated into an operational framework that specifies how the heterogeneity should be measured, stratified and tested. In its absence, individual studies and guidelines continue to report average effects on binge frequency or weight, which is precisely the form of evidence least able to distinguish treatment from suppression.

This review supplies such a framework. Recent contributions identify the relevant clinical priorities—screening, risk stratification, monitoring, discontinuation planning—and several propose them as agendas for the field [5,6]. What they do not provide is an account of why average-effect evidence is structurally unable to settle the underlying question, nor a single design that resolves it. The present contribution separates the treat-versus-mask question into three distinct problems of unequal evidential weight. The central and most firmly grounded is a measurement problem: eating-disorder instruments score several intended pharmacological effects of GLP-1 RAs as clinical improvement, so they cannot, in principle, distinguish drug-induced suppression from recovery—a claim anchored in the one disorder where a weight-acting agent has been tested against a placebo. The second is a phenotype problem: an agent acting on both homeostatic and hedonic systems should not act uniformly across patients, so the tractable question is not whether GLP-1 RAs treat or mask, but whom they treat and whom they mask; because BED subtypes already moderate response to drug versus psychological treatment in randomized trials, this phenotype-differential effect is a supported but still untested hypothesis. The third is a temporality problem: the decisive variable is not the drug’s effect while taken, but the disorder’s trajectory after withdrawal, governed less by pharmacology than by whether evidence-based psychological treatment is delivered during the pharmacological window. A factorial, phenotype-stratified trial with disorder-core endpoints and mandatory post-discontinuation follow-up operationalizes all three into a single testable design.

Once these three problems are separated, the dichotomy dissolves into testable questions. The remainder of this article maps the eating-disorder spectrum, sets out the neurobiological substrate, reviews the clinical evidence and its structural limits, develops the three axes, situates the argument in the history of weight-centered pharmacology, specifies a resolving trial design, and translates the framework into practice. Throughout, the type and certainty of evidence is stated explicitly, because the core contribution is conceptual rather than empirical.

Accordingly, the aims of this review are fourfold: (i) to show why the treat-versus-mask question, posed as a simple binary, cannot be answered by the currently available average-effect evidence; (ii) to reframe it as three separable and individually tractable problems—measurement, phenotype and temporality; (iii) to derive from this reframing a single trial design capable of resolving all three; and (iv) to translate the framework into practical, appropriately hedged guidance for screening, monitoring, discontinuation planning and interdisciplinary care in the settings where GLP-1 RAs are actually prescribed. The review does not aim to establish the efficacy or harm of GLP-1 RAs in any eating disorder, which the present evidence cannot support, but to specify how that question should be measured, stratified and tested.

2. Materials and Methods

This is a critical narrative review, not a systematic review, and no meta-analysis was performed; consequently, PRISMA reporting is not applicable. To address a common and legitimate criticism of narrative reviews, the search approach is stated explicitly so that it can be appraised and reproduced in outline.

Sources and search: PubMed/MEDLINE was searched, supplemented by the publications and position statements of relevant professional bodies (eating-disorder, obesity-medicine, diabetology and biological-psychiatry societies) and by hand-searching reference lists of retrieved reviews. Search terms combined GLP-1-related terms (“glucagon-like peptide-1”, “GLP-1 receptor agonist”, semaglutide, liraglutide, dulaglutide, tirzepatide) with eating-disorder terms (“eating disorder”, “binge eating”, “bulimia nervosa”, “anorexia nervosa”, “disordered eating”) and, for the access-context literature, with terms capturing the prescribing channel (“telehealth”, “compounded”, “off-label”); material on prescribing channels derives substantially from clinical commentary rather than from controlled studies, and is identified as such in the text.

Selection: Priority was given to systematic reviews and meta-analyses, RCTs, and multi-society guidance, with emphasis on material dated 2023–2026; foundational older sources were retained only where they remain the primary reference for an instrument, a construct, or a historical episode. Selection was purposive, and the synthesis is therefore susceptible to selection and publication bias, as noted in the limitations.

Inclusion and exclusion criteria: Sources were eligible if they were English-language, peer-reviewed publications (or formal guidance and position statements of recognized professional societies) that addressed at least one of: (i) the effect of GLP-1 RAs or the GIP/GLP-1 co-agonist tirzepatide on eating behavior, disordered eating, or a diagnosed eating disorder; (ii) the neurobiology of GLP-1 and GIP signaling relevant to appetite and reward; (iii) the assessment, screening, or phenotyping of eating disorders; or (iv) the clinical or historical context needed to interpret weight-acting pharmacotherapy. Foundational older sources (for example, the original description of a screening instrument or a construct) were included only where they remain the primary reference for that item. Sources were excluded if they were not retrievable in full text; were non-peer-reviewed except where a society guideline or a clinical commentary was the most authoritative available source on prescribing channels (in which case the source is identified as commentary in the text); addressed pharmacotherapy unrelated to appetite, reward, or weight; or duplicated a more complete or more recent report of the same data. No restriction was placed on study design; rather, each retained source was graded by evidence tier as described above, so that lower-tier evidence informs the argument only in proportion to its strength. Because this is a critical narrative rather than a systematic review, these criteria describe the basis for selection and appraisal and are not applied as a formal screening algorithm with counts at each stage.

Evidence grading: Empirical claims are flagged and graded: tier 1—RCTs and meta-analyses of RCTs; tier 2—non-randomized comparative studies and cohorts; tier 3—case series and reports; guidance—clinical practice guidelines and multi-society position statements; and hypothesis or expert opinion. A conceptual thesis is appropriate in an evidence-poor field because it cannot be refuted by the mere absence of trials, but it can be judged on coherence, on consistency with available data, and on clinical utility.

3. The Eating-Disorder Landscape and Rationale for Focus

The current classification groups feeding and eating disorders into categories that differ in prevalence, in how strongly they intersect with the obesity and T2D populations that receive GLP-1 RAs, and in how their mechanisms relate to drug action. These three dimensions, not diagnostic tradition alone, determine which disorders are analyzed in depth (

Table 1. The feeding and eating disorder landscape mapped against GLP-1 receptor agonist therapy.

Disposition in this Review	Mechanistic Relation to Drug Action	Prevalence and Relation to Obesity/T2D Population	Disorder or Category
Restrictive pole; analyzed as the harm counterpart.	Appetite suppression and weight loss are directly aligned with, and may reinforce, the core pathology. No therapeutic rationale.	Lifetime prevalence ~1–2% in women [1]; among the highest mortality of any psychiatric disorder. Not the typical obesity/T2D population, although the binge-eating/purging subtype overlaps the binge spectrum.	Anorexia nervosa (restricting and binge-eating/purging subtypes)
Analyzed within the AN section; sharpest case of iatrogenic masking.	Patients are prime candidates yet least likely to be recognized as having an eating disorder; the drug enables restriction.	Common, possibly as prevalent as full AN; by definition normal or high weight, frequently with an obesity history.	Atypical anorexia nervosa (OSFED)
Core analytic focus; mid-spectrum archetype.	Appetite and reward suppression may reduce the binge component; gastrointestinal effects may reproduce purging sensations; compensatory and cognitive core untouched.	Lifetime prevalence ~1–1.5% [1]; over-represented at higher weight, including non-purging presentations.	Bulimia nervosa
Core analytic focus; bulk of direct evidence.	Mechanism aligns with principal drivers (craving, reward, homeostatic hunger); the main candidate indication.	Most prevalent eating disorder in adults (~2–3.5%) [1]; strongly over-represented among those seeking weight management.	Binge eating disorder
Folded into the nearest archetype.	Varies; subthreshold binge presentations behave mechanistically like BED.	Often the largest eating-disorder category in clinical samples; many in obesity care fall here.	Other specified feeding or eating disorder (subthreshold BN/BED, purging disorder, night eating syndrome)
Outside core analysis; flagged as an emerging, evidence-free concern.	Distinct mechanism; however, gastrointestinal effects may resemble and might theoretically trigger restriction of this type.	Better recognized in children but occurs in adults; restriction driven by sensory aversion, low interest or fear, not weight/shape concern; not characteristically tied to obesity.	Avoidant/restrictive food intake disorder
Outside scope.	Mechanisms unrelated to appetite or reward regulation; no plausible interaction with drug action.	Ingestion of non-nutritive substances or repeated regurgitation; occur across the weight spectrum.	Pica and rumination disorder

Note: OSFED = other specified feeding or eating disorder; T2D = type 2 diabetes; AN = anorexia nervosa; BED = binge eating disorder; BN = bulimia nervosa. Prevalence figures are approximate, population-dependent, and drawn principally from the systematic review by Galmiche et al. [1]; they should be interpreted as indicative ranges.

).

AN is defined by restriction of intake relative to requirements leading to significantly low weight, intense fear of weight gain or persistent behavior interfering with weight gain, and disturbance in the experience of weight or shape, with restricting and binge-eating/purging subtypes [7]. Atypical AN, classified under other specified feeding or eating disorder (OSFED), is the full clinical picture of AN in an individual whose weight is not low. Avoidant/restrictive food intake disorder (ARFID) involves restriction driven by sensory aversion, low interest or fear of aversive consequences, not weight or shape concern. BED is defined by recurrent binge episodes with loss of control and marked distress, without regular compensatory behaviors. BN adds recurrent inappropriate compensatory behaviors and self-evaluation unduly influenced by weight and shape. Subthreshold and atypical binge presentations fall under OSFED, often the largest eating-disorder category in clinical samples. First-line treatment across the binge spectrum and for AN is psychological [8,9].

Read across Table 1, the landscape resolves into a clinically important asymmetry. For binge-spectrum pathology (BED, BN and subthreshold presentations), GLP-1 RAs are a candidate therapy: their mechanism aligns with the disorder’s principal drivers, and the open questions are efficacy and durability. For restrictive pathology (AN, including atypical AN), GLP-1 RAs are a candidate harm: their mechanism—appetite suppression and weight loss—aligns with and may reinforce the disorder’s core, and the open question is iatrogenic risk. The same prescription that may help one patient may damage another.

The detailed analysis concentrates on three archetypes that bind the clinically relevant space. BED is the principal candidate indication and carries the bulk of direct evidence; BN is the mid-spectrum case, combining binge behavior with compensation; AN is the restrictive pole and the clearest setting of iatrogenic harm. OSFED presentations are folded into the nearest archetype. Pica and rumination disorder are excluded because their mechanisms are unrelated to appetite or reward. ARFID warrants more than a passing mention as an emerging, mechanism-plausible concern. The gastrointestinal effects of GLP-1 RAs—early satiety, nausea, delayed gastric emptying and, in some patients, food aversion—overlap phenomenologically with the drivers of ARFID-type restriction (fear of aversive consequences, low interest in eating, sensory aversion). It is therefore biologically plausible that, in susceptible individuals, sustained GLP-1 RA exposure could precipitate or entrench ARFID-like avoidant restriction that is distinct from weight- or shape-driven pathology and would not be captured by instruments designed around shape and weight concern. No direct evidence yet addresses this possibility; it is flagged here as a hypothesis meriting prospective surveillance rather than an established risk.

4. The Neurobiological Substrate: Two Systems, One Drug

GLP-1 is an anorexigenic peptide of the gut–brain axis, released peripherally by enteroendocrine L-cells and centrally by preproglucagon neurons of the nucleus tractus solitarius. GLP-1 receptors are expressed both in hypothalamic nuclei governing homeostatic energy balance and in mesolimbic structures—the ventral tegmental area and nucleus accumbens—that mediate food reward and incentive salience [10].

This dual distribution is the conceptual foundation of the argument: a drug acting on two partly independent systems should not be assumed to exert a uniform effect across patients whose pathology is weighted differently towards each. Preclinical studies show that GLP-1 RAs reduce food-motivated behavior, attenuate dopaminergic signaling in reward circuits, and decrease the reinforcing efficacy of palatable food and of several drugs of abuse; human neuroimaging studies report altered food-cue responses during exposure [10,11,12]. These findings establish mechanistic plausibility for an effect on hedonic eating and motivate the phenotype argument below.

Critically, GLP-1 RA action is concentrated on appetite and reward. Eating disorders, however, are not disorders of appetite alone: their diagnostic core includes overvaluation of weight and shape, body dissatisfaction, and the use of eating for affect regulation. A pharmacological effect on appetite and reward can plausibly reduce the behavioral load of a disorder without reaching its cognitive-affective core, a distinction that recurs throughout this review and that, as Section 5 shows, is also visible in the one disorder where a weight-acting drug has been tested in RCTs.

It is worth making explicit how these peripheral and central actions translate into the subjective experiences that define the proposed phenotypes, because that translation is the mechanistic bridge between a drug that acts on the gut and hormones and a psychology expressed as craving, satiety and perceived control. GLP-1 RAs act through both a peripheral and a central route. Peripherally, they slow gastric emptying and, with the endogenous incretin axis, modulate the post-prandial secretion of gut-derived satiety and hunger signals—including peptide YY (PYY), which promotes satiation, and acylated ghrelin, the principal orexigenic signal, whose suppression reduces meal initiation and food preoccupation; GIP signaling contributes additionally through the co-agonist mechanism noted below. Centrally, GLP-1 receptor activation in the hypothalamus dampens homeostatic hunger, while activation in mesolimbic reward structures attenuates the incentive salience of palatable food [10,11,12]. The subjective consequences map onto the three phenotypes in a testable way: reduced hedonic reactivity should most directly lower the craving and cue-driven wanting that characterize the reward-dominant phenotype; enhanced and earlier satiation, together with reduced homeostatic hunger, should raise the threshold for the physiological deprivation that drives restraint-linked loss of control; and the affective use of eating, which is not itself an appetite signal, should be the experience least altered by these mechanisms—precisely the dissociation predicted for the negative-affect-dominant phenotype. This mechanistic account is a set of directional hypotheses, not established fact, but it specifies how measurable neuroendocrine changes could produce the psychological differences the framework relies on, and it motivates the search for peripheral markers discussed in Section 12.

Tirzepatide, a dual GIP/GLP-1 receptor co-agonist, warrants separate comment because its mechanism is not identical to that of pure GLP-1 RAs, yet almost no eating-disorder-specific data exist for it. GIP receptors are expressed in central nervous system regions that regulate energy balance, and GIP receptor signaling there is now considered essential for the appetite-suppressing action of co-agonists; preclinical work localizes this effect partly to inhibitory GABAergic neurons, and co-agonism produces greater weight loss and food-intake suppression than GLP-1 receptor agonism alone [13,14]. Two implications follow for the present argument. First, because the eating-disorder-relevant action of tirzepatide still runs predominantly through appetite and reward pathways, the three-axis framework developed here applies to it largely as it does to single agonists. Second, because co-agonism suppresses appetite more powerfully, the measurement and iatrogenic-restriction concerns may, if anything, be amplified rather than attenuated; greater pharmacological appetite suppression means a larger potential gap between scored improvement and the untouched cognitive-affective core. This is a mechanistic inference, not an empirical finding: direct evidence on tirzepatide in binge-spectrum or restrictive pathology is currently almost absent, which is itself a priority gap.

5. Current Clinical Evidence and Its Structural Limitations

5.1. Binge Eating Disorder

Enhanced cognitive behavioral therapy (CBT-E) is the first-line, best-evidenced treatment for BED, and lisdexamfetamine is, in the United States, the only agent specifically approved by the Food and Drug Administration for moderate-to-severe BED in adults (it is not approved for this indication in the European Union) [8]. The 2023 update of the World Federation of Societies of Biological Psychiatry (WFSBP) guidelines grades the pharmacological evidence across eating disorders and recommends lisdexamfetamine and topiramate for BED; it should be noted that, despite such recommendations, neither olanzapine nor topiramate currently holds marketing authorization for any eating disorder [15]. For GLP-1 RAs specifically, the direct evidence is limited and of low certainty: small pilot RCTs of liraglutide and dulaglutide [16,17,18], plus open-label and retrospective cohorts including a retrospective semaglutide series reporting reduced binge frequency and severity [19].

More recent syntheses sharpen rather than overturn this cautious reading. A 2026 rapid review of GLP-1 RA effects on eating behaviors and eating-disorder risk screened 1597 records and included 25 studies (two in adolescents, n = 275; 23 in adults, n = 8722); binge-eating episodes and scores were reduced with liraglutide and semaglutide, and food cravings were generally improved or unchanged [20]. Crucially for the measurement argument developed below, the synthesis rested almost entirely on binge-frequency, craving and behavioral endpoints: a global eating-disorder risk score was reported in only a single study (which found a reduction with liraglutide but no between-group difference), and the authors explicitly concluded that comprehensive assessment of eating behaviors is still needed to understand the benefits and risks of treatment [20]. Narrative and empirical reviews reach a convergent conclusion: GLP-1 RAs may reduce weight and binge symptoms in some patients with BED, but the data do not yet support claims of durable, disorder-level remission, and concern remains about unintended harm or misuse in vulnerable individuals [3,4,5]. Controlled trials of semaglutide and tirzepatide in diagnosed BED are anticipated, but to our knowledge none had reported disorder-core outcomes with post-discontinuation follow-up at the time of writing; such claims should be treated as provisional until registered results are available.

5.2. Bulimia Nervosa

Evidence in BN is markedly weaker, essentially comprising case reports and small retrospective observations (tier 3), including a frequently cited case of sustained symptom resolution with liraglutide [21]. For comparison, the WFSBP guideline recommends fluoxetine and topiramate as the best-supported pharmacotherapies for BN, neither of which is a GLP-1 RA [15]. Several features of GLP-1 RA therapy intersect unfavorably with BN: gastrointestinal effects, particularly nausea and vomiting, may reproduce or normalize purging-type sensations and, by expert account, act as somatic triggers in individuals with a purging history, while appetite suppression and rapid weight loss may reinforce restraint and overvaluation of weight and shape [2,22]. Expert opinion is accordingly cautious about prescribing in active or historical BN outside research settings [2,5,23].

5.3. Anorexia Nervosa and the Question of Pharmacotherapy for the Core Disorder

Before proceeding, one point must be made explicit to avoid any misreading: we do not suggest, and no responsible clinician would propose, that GLP-1 RAs be prescribed to treat AN. An appetite-suppressing, weight-reducing agent is categorically contraindicated in a disorder defined by restriction and low weight. AN is analyzed here for two reasons that are independent of any therapeutic use. First, patients with restrictive pathology are nonetheless exposed to these agents in the real world—through off-label, telehealth and compounded channels where the eating disorder is unrecognized, through atypical AN presenting at normal or high weight, and through deliberate misuse to facilitate weight loss—so the clinically relevant question is not whether to prescribe but how to detect and prevent harm. Second, AN is the one disorder in which a weight-acting drug has been tested against a placebo, which makes it uniquely informative for the measurement argument that follows, irrespective of GLP-1 RAs. AN is treated here not as a candidate indication, which it is not, but because the question it poses is the inverse of the binge-spectrum question. It is important to state the pharmacological evidence precisely, because it bears directly on the measurement argument that follows. Olanzapine is the one agent with RCT-level support in AN: a multi-site, double-blind, placebo-controlled trial in adult outpatients (N = 152) demonstrated a modest but statistically significant effect on the rate of weight gain, with no significant benefit on obsessionality or other psychological symptoms [24]. Consistent with this, the WFSBP guideline assigns olanzapine only a limited recommendation in AN, explicitly because the available evidence is restricted to weight gain while its effect on psychopathology is less clear [15].

This distinction—an objectively measurable effect on weight, without a demonstrated effect on the cognitive core—is not a rhetorical device introduced to protect the present argument; it is the explicit basis on which the relevant guideline graded the evidence. Low body weight is a defining and clinically critical feature of AN, and a drug that reliably promotes weight restoration is valuable. The narrower, evidence-based point is that in the single disorder where a weight-acting agent has been tested against a placebo, the trial establishing efficacy on weight simultaneously documented the absence of a measurable effect on the disorder’s defining psychological features—the dissociation the measurement axis (Section 6) predicts for GLP-1 RAs, and the reason weight-based endpoints cannot alone adjudicate treatment versus suppression.

There is no trial evidence and no mechanistic rationale for GLP-1 RAs in AN: an appetite-suppressing, weight-reducing agent is mechanistically contraindicated in a disorder of restriction and low weight. The relevant literature concerns harm. Case reports and clinical commentary (tier 3; expert opinion) describe GLP-1 RA exposure—frequently off-label, telehealth-mediated or compounded—precipitating or unmasking restrictive pathology including atypical AN, and document concern that these agents are sought, and at times misused, to facilitate restriction and weight loss [2,5,6,23].

5.4. Why the Current Evidence Cannot Answer the Question

The decisive point is not that the evidence is sparse, but that it is structurally unable to answer the treat-versus-mask question. Almost all binge-spectrum studies use binge frequency or weight as endpoints rather than validated eating-disorder psychopathology; almost none follow patients after discontinuation; and trial populations are defined by obesity or T2D rather than by a characterized eating-disorder phenotype [20]. A literature built on these design choices cannot distinguish remission from suppression, identify for whom benefit accrues, or establish durability. This evidential asymmetry should be stated plainly: the evidence base is least sparse precisely where the therapeutic claim is most modest (BED), and sparsest—essentially case-level—where this review advances its strongest claims, namely the iatrogenic-harm argument in BN and restrictive pathology. The harm argument therefore rests on the mechanism, on the measurement logic developed below, and on tier-3 reports, not on controlled data; it is offered as a precautionary inference, and the asymmetry is itself part of the case for the surveillance and trial design proposed later. Section 6, Section 7 and Section 8 address each limitation in turn.

6. Axis A: The Measurement Problem and the Diagnostic Blind Spot

Screening and severity instruments operationalize eating pathology partly through items capturing reduced or controlled intake, fullness, weight change and preoccupation with food—precisely the dimensions a GLP-1 RA is designed to alter. During treatment, a reduction in such item scores is ambiguous: it may reflect genuine improvement, pharmacological suppression of behavior without change in the disorder, or iatrogenic restriction that the instrument misreads as recovery. The instrument cannot discriminate between these, because the drug effect and the scored construct overlap. A meaningful portion of the reported improvement in uncontrolled GLP-1 RA studies may therefore be a measurement artifact rather than a therapeutic effect, a possibility that uncontrolled, unblinded designs cannot exclude. The olanzapine precedent in Section 5.3 is instructive: even in a blinded RCT, a drug can move weight without moving the psychological core, so an instrument weighted toward intake and weight will overstate improvement.

The word “masking” denotes at least three distinct phenomena, and conflating them obscures clinical reasoning. Symptomatic masking is suppression of the observable behavior while the cognitive-affective core persists, so symptoms re-emerge on withdrawal; this is a temporality phenomenon. Diagnostic masking is the failure of the presentation to cross the threshold of clinical suspicion or of an instrument, because drug effects neutralize the scored features, so the disorder becomes statistically invisible rather than absent; this is the measurement phenomenon. Iatrogenic masking is the concealment, and possible simultaneous promotion, of emerging restrictive pathology, so the appearance of success is itself the early presentation of a new or relapsing disorder; this is linked to the restraint-dominant phenotype and to AN.

No obesity- or diabetes-care guideline currently mandates a validated eating-disorder screening protocol before GLP-1 RA initiation. A 2025 multi-society nutritional advisory recommends baseline assessment of disordered eating but specifies no instrument or threshold [25]. Eating-disorder organizations and clinicians recommend routine pre-prescription screening and structured monitoring, and some advise against prescribing in individuals with an eating-disorder history [2,6,23]. The measurement argument sharpens this concern: screening performed after initiation is partially invalidated by the drug, so pre-treatment screening is uniquely informative—the one assessment window not yet contaminated by construct confounding. Individuals with atypical AN or non-purging BN may be misread as having BED or simple obesity, prescribed a GLP-1 RA that reinforces restriction, and then monitored with instruments that score their deterioration as improvement.

The channel through which these drugs are increasingly obtained magnifies every one of these risks. A large and growing share of GLP-1 RA and tirzepatide use occurs outside conventional specialist care—through telehealth platforms, direct-to-consumer prescribing, and compounded formulations supplied during shortages or for cost reasons. These pathways are precisely those least equipped to detect eating pathology: contact is often brief and remote, structured eating-disorder screening is inconsistent or absent, weight loss is the explicit and sometimes sole therapeutic goal, and continuity with a clinician able to recognize emerging restriction or compensatory behavior is limited [2,6]. Compounded products add a further layer of uncertainty around dosing and oversight. The population most likely to be harmed—patients with unrecognized restrictive or binge-purge pathology seeking rapid weight loss—overlaps substantially with the population most likely to use these channels, and commentators have specifically flagged self-directed and weight-loss-motivated use as a setting of elevated risk for misuse and for the precipitation or worsening of disordered eating [25]. This is not an argument against access, but an argument that screening, monitoring and escalation pathways must be built into these channels rather than assumed, because the diagnostic blind spot of Axis A is widest exactly where clinical contact is thinnest.

No instrument has been validated for eating-disorder assessment specifically before or during GLP-1 RA therapy. Brief screens such as the five-item SCOFF are feasible but have limited and variable sensitivity and specificity that depend on age and weight [26,27]. The Binge Eating Scale suits grading and monitoring binge severity but is not diagnostic [28]. The Eating Disorder Examination Questionnaire and its short form provide a global psychopathology score interpretable against norms, including for obesity, but are longer and derive from, rather than equate to, the Eating Disorder Examination interview [29,30]. A defensible strategy is tiered—a brief screen, then a fuller measure, then clinical interview—but every on-treatment score must be interpreted against construct confounding and supplemented by clinician assessment, weight-trajectory review, and explicit evaluation of cognitive features such as overvaluation of weight and shape, which appetite suppression does not directly produce.

A further dimension is poorly captured by any of these instruments: the quality of a person’s relationship with food, which is conceptually distinct from the frequency of disordered behaviors. GLP-1 RAs can reduce binge episodes and grazing while leaving—or even worsening—a disturbed relationship with eating: loss of interoceptive hunger and satiety cues as meaningful signals, eating governed by rule rather than appetite, anxiety around food, loss of the social and hedonic dimensions of eating, and rigid or moralized food choices. These features are the everyday substance of disordered eating as encountered in nutritional practice, yet they map onto few or no items in screens built around binge frequency, weight and shape. A reduction in measured symptoms can therefore coexist with a deteriorating relationship with food, which is precisely the kind of change a behavior-and-weight-weighted instrument is structurally unable to detect. This is the measurement blind spot expressed at the level of lived eating experience, and it is the dimension in which dietetic assessment is most informative.

This concern extends well beyond patients who meet diagnostic criteria. Disordered eating that falls below any eating-disorder threshold—emotional or stress-driven eating, chronic dieting and rigid restraint, “good-food/bad-food” thinking, chaotic or skipped meals, and weakened interoceptive awareness—is highly prevalent in the obesity and weight-management populations to which GLP-1 RAs are most often prescribed, and a weight-loss intervention can move it in either direction. A systematic review and meta-analysis of weight-loss interventions (38 studies, 3364 participants, including behavioral programs and licensed weight-loss pharmacotherapy) found that, on average, disordered-eating scores improved rather than worsened in adults with overweight or obesity, leading the authors to conclude that such interventions might not worsen, and may improve, disordered eating; substantial between-study heterogeneity nonetheless leaves room for deterioration in susceptible individuals [31]. GLP-1 RAs plausibly recapitulate this two-sidedness: by reducing hunger and reward they may ease reward-driven and emotional overeating, yet by enabling rapid, effortless restriction they may entrench rigid control, food avoidance, or a more anxious relationship with eating—none of which constitutes a diagnosable disorder, and none of which a symptom-count instrument is built to register. This subclinical zone is the natural domain of psychonutritional work: supporting regular, flexible, adequate eating; rebuilding hunger and satiety cues; and loosening rigid food rules. Crucially, it sits on the safe side of the competence boundary set out in Section 9.3, because it concerns the relationship with food rather than the entrenched cognitive-affective core of an eating disorder. Attending to it is both a means of preserving benefit and an early-warning system: a worsening relationship with food during otherwise “successful” treatment is often the first sign that a patient is drifting from subclinical disordered eating toward diagnosable pathology.

7. Axis B: The Phenotype Problem and a False Binary

Even with perfect measurement, asking whether the drug treats or masks binge eating has no single answer, because binge-type eating is not one phenotype. Empirical work on BED supports clinically meaningful subtypes. A widely replicated distinction separates a dietary-restraint-driven pattern from a pattern dominated by negative affect and emotion-regulation difficulty, the latter associated with greater psychopathology and poorer treatment outcome [32,33]. A partly orthogonal dimension is reward-driven eating, indexed by a reward-based eating drive of preoccupation, craving and loss of control [34].

Three idealized phenotypes are useful, recognizing that real patients are mixtures. The reward- or craving-dominant phenotype is driven by heightened hedonic reactivity to palatable food. The negative-affect-dominant phenotype is one in which binge eating functions as affect regulation. The restraint- or overvaluation-dominant phenotype is embedded in dietary restriction and overvaluation of weight and shape, shading into atypical AN and non-purging BN. These dimensions are not purely theoretical: each maps onto existing, validated self-report measures—for example the Reward-Based Eating Drive scale for the reward dimension [34] established negative-affect and emotion-regulation indices for the affective dimension [33], and the weight- and shape-concern subscales of the Eating Disorder Examination Questionnaire for the restraint/overvaluation dimension [29,30]. No instrument has yet been validated to assign a patient to a dominant phenotype for the specific purpose of GLP-1 RA decision-making, however, and developing one is itself part of the research agenda; the phenotypes are therefore presented as a research-ready hypothesis and a structure for clinical attention, not as a validated triage tool.

As an explicitly labeled hypothesis, not an established finding, because GLP-1 RAs act on appetite and reward but not on the cognitive-affective core, their effect should differ systematically by phenotype. This prediction has a precedent. In a randomized double-blind trial of CBT versus fluoxetine (N = 108), overvaluation of shape and weight was the most salient predictor and moderator of outcome: it predicted lower binge-eating remission overall (29% vs. 57% without overvaluation), the gap was most pronounced under medication alone (10% vs. 42%), and patients with overvaluation improved significantly more with CBT [35]. An earlier trial similarly found a dietary–negative-affect subtype reported more frequent post-treatment binge episodes than a pure-dietary subtype [36]. The restraint/overvaluation and negative-affect dimensions thus already differentiate who benefits from pharmacological versus psychological intervention. Extending that logic to GLP-1 RAs—whose action is even more narrowly confined to appetite and reward—three predictions follow. In reward-dominant binge eating, the drug acts on the principal driver, so genuine benefit is expected. In negative-affect-dominant binge eating, behavior may be suppressed while the affect-regulation deficit persists and is expressed through other channels, so benefit should be partial and unstable. In restraint- or overvaluation-dominant presentations, appetite suppression and weight loss amplify restraint and weight-shape salience, so the predicted effect is neutral-to-harmful and most prone to iatrogenic masking. This prediction is consistent with reports that GLP-1 RAs reduce hunger and binge frequency yet may reinforce rigid dietary control in vulnerable individuals [4,5].

One nuance must be stated honestly. Classic restraint theory holds that the binge–restrict cycle is driven partly by physiological deprivation [37]. GLP-1 RAs reduce hunger and therefore the deprivation signal even when intake is low—a state unlike voluntary, hunger-laden dieting—so for some restraint-driven patients they may be less destabilizing than unaided dieting. This does not rescue the binary; it reinforces that the effect is phenotype-contingent and cannot be predicted without characterizing the patient. Cognitive overvaluation of weight and shape, and the reinforcement provided by visible weight loss, remain unaddressed in all phenotypes.

Table 2. Phenotype-by-time conceptual model of predicted trajectories (hypothesized; no direct empirical data on GLP-1 RA effects by phenotype currently exist).

After Discontinuation, Drug Plus CBT-E	After Discontinuation, Drug Only	During Drug Exposure	Phenotype
Best-case durability; the pharmacological window is used to consolidate skills.	Partial relapse as reward drive returns; benefit not fully durable.	Substantial reduction in binge frequency and craving; plausible genuine benefit.	Reward- or craving-dominant
Conditional durability if therapy targets affect regulation.	Relapse likely; the core driver is never addressed.	Binge behavior suppressed; affect-regulation deficit untouched; risk of symptom substitution.	Negative-affect-dominant
Therapy may mitigate harm; drug use remains questionable in this phenotype.	Weight regain may trigger marked distress and a disordered response.	Restraint and weight-shape salience amplified; iatrogenic masking; risk of atypical AN or non-purging BN.	Restraint- or overvaluation-dominant

Note: AN = anorexia nervosa; BN = bulimia nervosa; CBT-E = enhanced cognitive behavioral therapy. Cell entries are hypothesized trajectories derived from the phenotype-differential hypothesis described in the text, not empirical results.

sets the three phenotypes against treatment status and against whether integrated psychological treatment is delivered. The cell entries are predictions derived from the hypothesis above, not trial findings.

The restraint- or overvaluation-dominant phenotype is the point at which the phenotype argument meets the analysis of AN, and the setting of the sharpest safety concern. Atypical AN is the paradigm case: these patients are not underweight, and many have an obesity history, so they are precisely those to whom a GLP-1 RA is most likely to be offered and least likely to be recognized as having an eating disorder. The drug then acts directly along the axis of the pathology—enabling and rewarding restriction, accelerating weight loss, and producing a clinical picture that, on routine instruments, is difficult to distinguish from therapeutic success. This is iatrogenic masking in its purest form: the drug conceals a disorder it is simultaneously feeding. Two consequences follow. Pre-treatment assessment must actively look for restrictive pathology in patients with obesity, rather than assuming that a high body mass index precludes an eating disorder. And in a patient with active or historical restrictive pathology, the default should be that GLP-1 RA therapy is contraindicated or deferred pending specialist assessment.

Sex and gender deserve explicit attention here, in line with SAGER reporting principles, because they interact with the recognition problem. Eating-disorder constructs, screening instruments and diagnostic thresholds were developed predominantly in girls and women, and eating disorders in boys and men are correspondingly under-recognized and under-diagnosed, often presenting with a focus on muscularity, leanness or “clean” eating rather than the weight-and-shape concerns that questionnaires emphasize. At the same time, the GLP-1 RA population is becoming more sex-balanced, with substantial and growing use by men for obesity and type 2 diabetes. The combination is hazardous: a male patient with restrictive or compensatory pathology is both less likely to be screened for an eating disorder and less likely to be detected by instruments calibrated to female presentations, so the diagnostic blind spot of Axis A is widened by sex bias. Pre-treatment assessment should therefore be applied irrespective of sex, with explicit awareness that male presentations may be atypical and may evade standard screens. Most published data, however, derive from predominantly female samples, so sex-disaggregated reporting of eating-disorder outcomes in GLP-1 RA studies is an important and currently unmet need.

8. Axis C: The Temporality Problem and the Pharmacological Window

What matters clinically is not the disorder’s state while the drug is taken, but its trajectory after the drug is withdrawn. That trajectory is determined less by pharmacology than by what is done during exposure. GLP-1 RA therapy can be reconceived as a time-limited period of pharmacological opportunity—referred to hereafter as the pharmacological window—during which the biological intensity of binge urges (craving, food-cue reactivity and homeostatic hunger) is reduced. By pharmacological window we mean the interval of active drug exposure during which these urges are pharmacologically dampened, creating a more favorable condition for psychological treatment; the term denotes an opportunity for intervention, not a treatment in itself, and its clinical value depends entirely on whether that opportunity is used.

As a hypothesis, a window left unused yields symptomatic masking: behavior is suppressed for the duration of exposure, the cognitive-affective disorder is unaddressed, and urges return on discontinuation, often compounded by weight regain. Expert commentary describes this pattern, including patients who report confidence in managing urges that were in fact pharmacologically absent—a false confidence exposed on cessation [2,23]. A window actively used—by delivering CBT-E or an equivalent during exposure—could in principle improve outcomes beyond either modality alone: reduced binge-drive may make cognitive and behavioral techniques more learnable, and skills consolidated during the window may persist after the drug is stopped. This integrated-care model is plausible and testable; it is not yet demonstrated.

Discontinuation is common—whether by choice, cost, supply or adverse effects—and is typically followed by weight regain. GLP-1 RA-associated weight loss includes loss of lean mass, and regain is fat-predominant. The resulting weight suppression and weight cycling are recognized risk factors for eating-disorder onset and poorer recovery [31,38]. An unused window may therefore leave the patient worse than at baseline, leaving the disorder intact and leading to a period of reinforced weight-shape focus and a destabilizing weight trajectory. This is why discontinuation planning is a core element of safe prescribing, and why the near-absence of post-discontinuation follow-up in the current evidence base is a decisive, not incidental, gap.

9. Discussion

The three axes developed in Section 6, Section 7 and Section 8 are not independent complaints but a single connected argument. The measurement problem explains why average-effect evidence systematically overstates benefit; the phenotype problem explains why a single average effect is the wrong estimand in the first place; and the temporality problem explains why even a correctly measured, phenotype-specific in-treatment effect cannot, on its own, distinguish recovery from suppression. Read together, they convert an unanswerable binary into a set of questions with defined measurements, defined moderators and a defined outcome horizon. This section situates that argument in the history of weight-centered pharmacology, states the research design it implies, and translates it into clinical practice; the limitations of the argument, its conclusions, and the priorities it sets for future work follow in Section 10, Section 11 and Section 12. Two themes recur throughout and are worth stating in advance: first, that objectively measurable change in weight or behavior can coexist with an untouched cognitive-affective core, so weight-based and symptom-count endpoints cannot adjudicate the underlying question; and second, that the gap between an idealized phenotype-informed framework and the real prescribing environment—telehealth, direct-to-consumer and compounded channels, where specialist assessment is often unavailable—is itself a central clinical problem rather than a caveat, and is addressed directly in Section 9.3.

9.1. Historical Context: A Recurring Pattern

The present situation has partial precedents, which should be read as cautionary rather than predictive. Several centrally acting weight interventions had their psychiatric or eating-related consequences recognized late, in part because evaluation prioritized weight and cardiometabolic endpoints: fenfluramine–phentermine (withdrawn 1997, cardiac valvulopathy), sibutramine (withdrawn 2010, cardiovascular risk), and most instructively rimonabant, a centrally acting weight drug whose serious psychiatric adverse effects emerged prominently only after marketing and led to withdrawal [39]. Bariatric surgery, although effective and not in question as unsafe, is associated with post-operative loss-of-control eating and, in a minority, new or recurrent disordered eating [40].

The lesson is methodological and deliberately limited in scope. These agents differ from GLP-1 RAs in mechanism, risk profile and magnitude of harm, so the analogy is not evidence about GLP-1 RA safety and should not be read as such. What the pattern does support is narrower: interventions evaluated mainly on weight and metabolic endpoints can leave eating-related and psychiatric outcomes undetected, which is a direct argument for measuring eating psychopathology, not only weight, and for following patients beyond drug exposure—precisely the design features advocated here on independent grounds.

9.2. A Research Agenda

The treat-versus-mask question becomes answerable only with a design that corrects all three axes simultaneously. Recent commentary in the eating-disorder field has converged on a similar list of clinical priorities—routine screening before and during treatment, risk stratification, validated monitoring tools, and discontinuation planning [6]. The contribution here is to specify how those priorities can be tested rather than merely asserted, by translating them into a single experimental design and an associated effect-modification hypothesis.

Such a study would enroll a phenotype-characterized population: adults with diagnosed BED, characterized at baseline along the reward-, negative-affect- and restraint-dominant dimensions using existing validated subscales as continuous measures—for example the Reward-Based Eating Drive scale, established negative-affect indices, and the weight- and shape-concern subscales of the EDE-Q [29,30,33,35]—rather than a single categorical phenotype label, since no validated phenotype-triage instrument yet exists. The phenotype would be pre-specified as an effect modifier. It would use factorial randomization—a 2 × 2 design of GLP-1 RA versus placebo and CBT-E versus control—to isolate the drug effect, the therapy effect and their interaction. It would adopt a disorder-core primary endpoint—the Eating Disorder Examination global score and its weight- and shape-concern subscales, rather than binge-day frequency alone—so that the cognitive-affective core is measured directly. Establishing and validating such a phenotype characterization is itself a deliverable of the program, not an assumption.

It would include a mandatory, protocolized post-discontinuation follow-up period—for example 6 to 12 months drug-free—to distinguish durable remission from symptomatic suppression, the single most important omission in the current literature. It would build in measurement safeguards, interpreting on-treatment self-report against construct confounding and supplementing it with clinician-rated assessment and body-composition measures, including lean mass. It would pre-specify an effect-modification analysis powered to test the phenotype-differential hypothesis, rather than to estimate an average effect that, if the phenotype argument is correct, would be clinically misleading. Finally, it would include prospective safety surveillance for emergent restrictive pathology and atypical AN, which efficacy-focused designs are not built to capture. Until such evidence exists, claims that GLP-1 RAs treat eating disorders, or that they merely mask them, outrun the data in both directions.

9.3. Practical Implications, Including Real-World and Non-Specialist Settings

It is useful to state directly, while respecting the limits of the evidence, which patient is most likely to benefit. The most favorable candidate profile, on current reasoning, is an adult with binge eating disorder whose pathology is reward- or craving-dominant—driven by heightened hedonic reactivity to palatable food and loss of control over eating—without prominent overvaluation of weight and shape, without a history of restrictive pathology, and without active purging. In such a patient the drug acts on the principal driver of the disorder, and the predicted benefit is greatest. Two qualifications are essential. First, this benefit is a candidate, not an established, effect: efficacy on disorder-core psychopathology and, above all, durability after discontinuation remain unproven. Second, the benefit is conditional on use of the pharmacological window: the most defensible expectation is that a GLP-1 RA reduces the behavioral intensity of reward-driven binge eating and thereby makes psychological treatment more tractable, not that it resolves the disorder alone. Benefit becomes progressively less predictable in negative-affect-dominant presentations, where the affect-regulation deficit is untouched, and turns to candidate harm in restraint- or overvaluation-dominant presentations and in any restrictive pathology, where the drug acts along the axis of the disorder.

Pending such evidence, the reframing yields concrete, phenotype-informed guidance, set out in full in

Table 3. Phenotype-informed clinical framework for GLP-1 receptor agonist use when eating-disorder risk is relevant.

Rationale	Recommended Action	Phase
Pre-treatment is the one assessment window not confounded by drug effects; phenotype should inform the decision.	Structured baseline assessment using a tiered approach (brief screen, then fuller measure, then clinical interview for ambiguous cases); characterize the dominant phenotype and actively look for restrictive pathology.	Before initiation
Therapy–harm asymmetry; predicted harm in the restraint-dominant phenotype.	Greatest caution in restraint- or overvaluation-dominant presentations; therapy contraindicated or deferred in active or historical AN or BN; for reward-dominant BED, prefer integration with psychological treatment over pharmacotherapy alone; document informed consent including discontinuation expectations.	Decision-making
On-treatment self-report instruments are partially invalidated by the drug and must be supplemented.	Monitor eating cognitions, mood, nutritional adequacy, lean mass and rate of weight loss; do not rely on confounded self-report alone; use clinician assessment and weight-trajectory review; watch for restriction, body-image preoccupation and weight-checking.	During treatment
An unused pharmacological window may leave the patient worse than at baseline.	Establish a discontinuation and weight-regain plan at the outset; consider gradual tapering where appropriate; reinforce regular structured eating and behavioral skills.	Discontinuation planning
Phenotype-informed, integrated care is the operative variable determining outcome.	Interdisciplinary, weight-inclusive care with a registered dietitian and, where indicated, eating-disorder-trained mental-health input; ensure communication between prescriber and team.	Care model

Note: AN = anorexia nervosa; BN = bulimia nervosa; BED = binge eating disorder. The framework synthesizes current guidance and the argument of this article and has not been prospectively validated.

and consistent with current multi-society nutritional guidance and eating-disorder expert opinion [5,6,25]. Two points deserve emphasis beyond the table. First, pre-treatment is the one uncontaminated assessment window, so structured baseline assessment—tiered from brief screen to clinical interview, attentive to the dominant pattern of eating and actively looking for restrictive pathology—is uniquely informative; this attention sharpens clinical judgment rather than substituting for it, since no validated phenotype-triage instrument yet exists. Second, because on-treatment self-report is partially invalidated by the drug, monitoring must combine clinician assessment, weight-trajectory review and body-composition measures rather than relying on questionnaires alone, with a discontinuation and weight-regain plan established at the outset.

A fair objection is that detailed psychological phenotyping is unrealistic in the settings where most GLP-1 RAs are now prescribed. With the rapid expansion of telehealth and direct-to-consumer pathways, a large and growing share of patients are managed outside specialist eating-disorder services by clinicians without eating-disorder training and often without access to psychiatric or psychological input; many patients cannot accurately characterize their own psychopathology, and full phenotype triage of the kind set out in Table 3 is not feasible at the point of prescribing. We accept this constraint, and it sharpens rather than weakens the argument, because the framework is intended to be applied in tiers matched to available resources rather than as an all-or-nothing specialist protocol. At minimum—and this is achievable in any prescribing setting, including asynchronous telehealth—a brief validated screen for eating pathology should be completed before the first prescription, together with two structured questions that require no specialist skill: whether the patient has any history of an eating disorder or of driven, rule-bound dieting, and whether weight and shape dominate their self-evaluation. A positive response on any item, rapid or greater-than-intended weight loss on treatment, or intensifying dietary rigidity should trigger referral rather than continued unmonitored prescribing. Full phenotype characterization and integrated psychological care are the aspiration for adequately resourced services; the minimum safe standard for non-specialist and remote settings is pre-treatment screening, a small number of history questions, and a defined escalation pathway. Framed this way, the practical question is not whether every clinic can perform phenotyping, but whether the prescribing system can reliably identify the minority of patients for whom these agents are hazardous—and the biomarker research proposed in Section 12 is, in part, an attempt to make that identification feasible where subjective assessment is impractical.

Care should be interdisciplinary and weight-inclusive. Within it, the registered dietitian occupies a pivotal but bounded role. Because the dietitian conducts a structured dietary history and sees the patient’s actual eating, they are well placed to perform the baseline phenotype-aware assessment, to support regular structured eating that counteracts both restriction and post-discontinuation rebound, to monitor protein adequacy and lean-mass preservation during rapid weight loss, and—most relevant to the measurement argument—to detect deterioration in the relationship with food that questionnaires miss: emerging food avoidance, rigid rules, loss of hunger cues, or anxiety around eating. This is psychonutritional territory in which dietetic observation often precedes any change in a screening score.

The boundaries of this role must be equally explicit, because GLP-1 RA prescribing draws many patients into weight-management settings where eating-disorder expertise is uneven and professional training is frequently insufficient for safe identification and referral [41]. The dietitian’s scope is nutritional assessment, support of regular and adequate eating, and early detection and referral; it does not extend to treating the cognitive-affective core of an eating disorder—overvaluation of weight and shape, or the use of eating for affect regulation—which is the domain of eating-disorder-trained psychological therapy. The line is crossed when the work moves from supporting eating behavior and the relationship with food toward attempting to modify the entrenched cognitive and emotional drivers of the disorder; at that point the dietitian’s task is escalation, not treatment. Concrete triggers for escalation to eating-disorder-trained mental-health input include weight loss faster than intended, intensifying dietary rigidity or weight-checking, food avoidance or fear of eating, marked body-image preoccupation, and withdrawal from social eating. Defining this boundary protects patients from both under-recognition and over-reach, and it makes the dietitian a sentinel within the monitoring framework rather than a substitute for psychological care.

10. Limitations

Several limitations should be acknowledged. The central thesis is conceptual: its value lies in coherence and clinical utility, and it reframes the question but does not, by itself, supply outcome data. The phenotype-differential hypothesis is not established; it is grounded in BED-subtype and reward-eating research but has not been tested against GLP-1 RA outcomes, and the proposed phenotypes are idealized and overlapping. The historical analogy is illustrative, not mechanistic: the earlier agents and bariatric surgery differ from GLP-1 RAs in mechanism and risk profile, and the analogy concerns the pattern of late recognition, not equivalence of harm. As a critical narrative review rather than a systematic review, this synthesis is susceptible to selection and publication bias, and positive case material is likely over-represented. Finally, the direct evidence remains immature—low certainty in BED, essentially case-level in BN, and in AN it is restricted to harm-oriented case reports and commentary—and adequately powered trials with disorder-core endpoints and post-discontinuation follow-up are not yet available and may revise specific claims made here.

11. Conclusions

Whether GLP-1 RAs treat or mask eating disorders cannot be settled as a simple binary, because the question conflates three separable problems—measurement, phenotype and temporality—developed in Section 6, Section 7 and Section 8. In brief, the drug’s intended effects are scored as recovery by standard instruments; an agent acting on appetite and reward but not on the cognitive-affective core should not act uniformly across heterogeneous binge presentations; and the decisive outcome is the post-discontinuation trajectory, which depends on whether psychological treatment is delivered during the pharmacological window.

Mapped across the spectrum, the field resolves into a therapy–harm asymmetry: for binge-spectrum pathology, GLP-1 RAs are a candidate treatment of unproven efficacy and durability, whereas for restrictive pathology they are a candidate harm, mechanistically contraindicated and capable of concealing the disorder they reinforce. The most defensible current reading is that GLP-1 RAs can reduce the behavioral intensity of reward-driven binge eating and may thereby open a window in which psychological treatment becomes more tractable, but that they do not, alone, resolve an eating disorder.

12. Future Directions

Three lines of work would move the field forward. First, the factorial, phenotype-stratified trial outlined in Section 9.2—with disorder-core endpoints, mandatory drug-free follow-up, and an effect-modification analysis—is the single most informative next step, because it simultaneously addresses measurement, phenotype and temporality. Second, measurement science is needed: instruments validated specifically for use before and during GLP-1 RA therapy, able to separate pharmacological appetite modulation from genuine change in eating-disorder psychopathology, and incorporating clinician-rated and body-composition measures rather than self-report alone. Third, real-world pharmacovigilance and prospective cohorts—particularly in telehealth and compounded-product channels, and in adolescents and other under-studied groups—are required to detect emergent restrictive pathology and atypical AN that efficacy trials are not built to capture. Until these are in place, the precautionary, integrated approach set out in Section 9.3 and Table 3 remains the most defensible basis for clinical practice.

A fourth priority follows directly from the measurement problem and from the practical limits of phenotyping in non-specialist settings. If self-report instruments are partially invalidated during GLP-1 RA therapy, and if detailed psychological phenotyping is unavailable to most patients at the point of prescribing, the field will ultimately need objective, clinically accessible markers to support phenotype classification and treatment monitoring. Circulating endocrine signals are a natural place to look, because several of them link peripheral metabolic state to the central appetite and reward systems the framework implicates. Candidate markers include the incretin and appetite hormones already central to the mechanism—GIP, PYY and acylated ghrelin—whose treatment-induced trajectories might, in principle, index the degree of appetite and reward modulation an individual is experiencing. Of particular interest is neurotensin, measurable clinically as its stable precursor pro-neurotensin (pro-NT), which is a biologically plausible bridge between peripheral metabolic signaling and central reward regulation: neurotensin is an intestinal, fat-induced peptide that promotes lipid absorption and satiety signaling and is obligatory for high-fat-diet-induced obesity in animal models [42], while circulating pro-NT independently predicts incident diabetes, cardiovascular disease and obesity in humans [43,44]; centrally, neurotensin modulates dopaminergic transmission in the ventral tegmental area and nucleus accumbens and is implicated in reward-related behavior [45]. This convergence of intestinal fat signaling, satiety and dopaminergic reward makes pro-NT, and composite endocrine signatures more broadly, worth investigating as objective correlates of the reward- and restraint-linked phenotypes proposed here. The evidence base is still emerging, and no such marker is validated for eating-disorder phenotyping or for monitoring during GLP-1 RA therapy; establishing whether circulating pro-NT or multi-hormone panels track subjective craving, satiety and perceived loss of control—and whether they predict post-discontinuation trajectory—is a concrete and testable next step that could, if successful, make the framework implementable precisely where subjective assessment is weakest.