Published on 17 September 2026
Anorexia nervosa: a measurable gastric interoception deficit, an unproven prognosis
Verdonk C et al. JAMA Psychiatry, 2026, published online ahead of print 17 June 2026, e261301.
DOI 10.1001/jamapsychiatry.2026.1301
PMID 42307918
Scientific 75
Editorial 73
The essentials
Interoception, the perception of signals arising from the body, is described as impaired in anorexia nervosa, most often on the basis of questionnaires. This team measured it differently: a swallowed vibrating capsule delivers calibrated stimulations of the stomach, and the capacity to detect them is recorded. Sixty-two young women hospitalised for restrictive-type anorexia nervosa, weight restored, were compared with fifty-seven controls, then followed for six months. Detection of low-intensity stimulations is markedly worse in anorexia nervosa, with a Cohen’s d of -0.98, 95% confidence interval -1.51 to -0.44. No between-group difference, however, appears on brain response or on peripheral physiological measures. Of the fifty-four patients reassessed at six months, sixteen were in full relapse, and three session measures were associated with that status, with odds ratios whose confidence intervals are extremely wide. This is an exploratory, single-centre, mechanistic study conducted with a research device. None of it is available in clinical practice.
Context
Relapse is the central problem of anorexia nervosa, a disorder whose mortality remains among the highest in psychiatry. The authors recall relapse rates approaching 50% in the year following weight restoration. The usual predictors, discharge body mass index, illness duration, comorbidities, are robust but coarse, and they teach nothing about mechanism. Hence the interest of a measure that touches what makes the disorder what it is: the relationship to bodily sensations. That measure still needs to be something other than a self-report questionnaire, known to be influenced by anxiety and by the patient’s body representation. This group’s earlier work concerned mainly cardiac and respiratory signals. The gastric domain, though central to satiety and meal anticipation, had remained little explored, for lack of non-invasive tools.
Mechanism
What the capsule measures, and what it does not
The Vibrant capsule, once swallowed, delivers three-second mechanical vibrations in the stomach. The participant presses a button when she perceives a sensation she attributes to the capsule, and releases it when the sensation stops. Two blocks of roughly sixty stimulations, one at normal intensity, the other at enhanced intensity, are administered in randomised order. What is measured is a detection performance, expressed through a signal-detection-theory index, the normalised A-prime, complemented by a response bias, an omission rate and a response time. A Bayesian model then estimates latent parameters: initial prior beliefs about the presence of vibrations, interoceptive precision, and the speed of belief updating.
What is not measured is what happens outside the laboratory at mealtimes. The link between this detection performance and the subjective experience of satiety or gastric fullness in real-life situations cannot be inferred from this protocol. Translating a lower detection rate into a phrase such as the patient does not feel her stomach should therefore be avoided. A point that matters for interpretation: the between-group difference is seen only for low-intensity stimulations, and disappears when the stimulation is enhanced. The deficit concerns weak signals, not visceral sensitivity in general.
The study at a glance
| Population | |
| Women hospitalised for restrictive-type anorexia nervosa, weight restored, body mass index of at least 18.5, age 13 to 40, recruited from an eating-disorder treatment programme in Tulsa. Healthy controls matched on age and sex. 62 patients, mean age 18.9, standard deviation 4.5; 57 controls, mean age 20.7, standard deviation 5.3. Body mass index at the session: 19.9 versus 22.2. Comorbidities reported among patients: generalised anxiety disorder 43 of 62, major depressive disorder 23 of 62, obsessive-compulsive disorder 12 of 62. Ongoing psychotropic treatment in 57 of 62 patients. | |
| Intervention | |
| Mechanical stimulation of the stomach by an ingested vibrating capsule (Vibrant Ltd), two counterbalanced blocks of roughly 60 three-second stimulations, one at normal intensity, the other at enhanced intensity, after 30 minutes of rest. Simultaneous recording of electroencephalogram, electrogastrogram, electrocardiogram and skin conductance. | |
| Comparator | |
| Comparison between the two intensity blocks within each participant, and comparison between the anorexia nervosa group and the control group. No placebo capsule was administered: participants were informed they might receive an inert capsule, but all received an active vibrating capsule, which defines the single blind. | |
| Endpoints | |
| Experimental-session measures: perceptual accuracy, prior beliefs, interoceptive precision, learning rate, gastric evoked potentials, hunger. Six-month clinical endpoints: relapse status and symptom severity measured by the global score of the Eating Disorder Examination Questionnaire. | |
| Design | |
| Single-centre, single-blind crossover trial, with randomisation of stimulation-block order rather than of group assignment, followed by an observational longitudinal arm at one, three and six months in the anorexia nervosa group only. Data collected at the Laureate Institute for Brain Research. Registered as ClinicalTrials.gov NCT05111977, protocol published as an appendix, CONSORT compliance statement. |
Quality control
| Point checked | Judgement |
|---|---|
| Registration and protocol | Registered |
| Finding The trial is registered under NCT05111977 and the full protocol is appended to the publication, which remains rare and deserves credit. The protocol designated changes in gastrointestinal perception during stimulation periods as the primary endpoint, with prediction of clinical outcomes as a secondary objective. The publication presents both arms together, with no declared statistical hierarchy. | |
| Multiple comparisons | No correction |
| Finding A correction is applied where the number of tests is explicit: cluster-based permutation for the electroencephalogram, the Holm method for correlations, adjustment of post hoc contrasts. No correction, however, is reported for the prognostic regressions, where thirteen predictors are tested in a single table and two emerge significant, at p equal to 0.05 and 0.04. The article’s most interesting p-values are therefore the least protected ones. | |
| Sample size of the prognostic arm | Below plan |
| Finding The protocol targeted 130 evaluable participants, 65 per group, to compare relapse rates. The longitudinal analysis covers 54 patients, of whom 16 were in full relapse. An odds ratio estimated on sixteen events cannot be precise, and the published confidence intervals show it. | |
| Blinding and expectation effect | Single blind |
| Finding All participants received an active capsule, while being informed that an inert capsule was possible. This expectation manipulation is declared and consistent with the task, but it means there is no no-stimulation control condition allowing estimation of the false-alarm rate in the absence of real vibration. | |
| Nature of the prognostic arm | Observational |
| Finding Randomisation concerns only the order of stimulation blocks. The six-month follow-up is a cohort follow-up with no intervention. The association between a session measure and relapse says nothing about the direction of the relationship, nor about what effect acting on that measure would have. | |
| Availability of the paradigm | Research tool |
| Finding The Vibrant capsule holds US market authorisation for chronic idiopathic constipation, not for interoception assessment. The setup combines capsule, electroencephalography and computational modelling. It is neither an available test nor a treatment. | |
| Funding and conflicts of interest | Not verifiable here |
| Finding The appended protocol names the National Institute of Mental Health and the Laureate Institute for Brain Research as sponsors. The funding and conflict-of-interest disclosure sections of the published version do not appear in the documents consulted and could therefore not be verified. Their absence from these documents does not mean an absence of declared conflicts. | |
Results
| Result | Reported value |
|---|---|
| Perceptual accuracy, normal block | d = -0.98 [-1.51; -0.44], p = 0.001 |
| Reading A large effect size, but a wide interval. The omission rate points the same way, d = 1.02, interval 0.55 to 1.48, p < 0.001. The block-by-group interaction is significant for both measures, p = 0.007 and p = 0.002. | |
| Enhanced-intensity block | No difference, p = 0.87 and p = 0.79 |
| Reading A central result, often forgotten in secondary coverage: when the signal is strong, patients detect it as well as controls. The abnormality concerns weak signals, not a global insensitivity. An absence of difference is not proof of equivalence, but it constrains the interpretation. | |
| Computational parameters | Prior beliefs d = -0.31 [-0.67; 0.05], p = 0.05 |
| Reading The confidence interval crosses zero even as the p-value is reported as reaching the threshold. The between-block difference in interoceptive precision is clearer, d = 0.38, interval 0.02 to 0.75, p = 0.01. No between-group difference on overall interoceptive precision. | |
| Brain response and physiology | No between-group difference |
| Reading Gastric evoked potential amplitudes do not differ between groups, either at the onset or at the end of vibration. No difference either in heart-rate variability, electrogastrogram or skin conductance. Only resting heart rate is lower in patients. The described deficit is therefore behavioural and computational, not electrophysiological. | |
| Relapse at six months | 16 full relapses out of 54 patients reassessed |
| Reading Eight of 62 patients were lost to follow-up. The other 38 fell into an intermediate status, partial relapse, partial or full remission, partial recovery, grouped into a single category for analysis. This grouping is declared by the authors and motivated by the low variance available. | |
| Predictors of relapse | OR 3.82 [1.02; 15.91]; OR 5.37 [1.15; 32.04] |
| Reading Initial prior beliefs, p = 0.05, and response bias in the normal block, p = 0.04. The lower bounds barely clear 1 and the upper bounds reach 16 and 32. These estimates are compatible with a modest effect as much as with a considerable one. They were not corrected for the thirteen comparisons in the same table, nor validated on an independent sample. | |
| Gastric unpleasantness | OR 5.73 [1.38; 33.5], p = 0.03 |
| Reading This is a visual analogue scale completed during the task, not a composite score. This result comes from a second regression table with fourteen predictors, likewise uncorrected. Note an internal inconsistency in the source regarding the upper bound of this interval, discussed in the verification block. | |
| Symptom severity at six months | β = 5.84, R² = 0.16, p = 0.004 |
| Reading The strongest association in the longitudinal arm, between the between-block difference in interoceptive precision and the global score of the Eating Disorder Examination Questionnaire at six months. It explains 16% of the variance, leaving most of it unexplained. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Validity of the measure | Strength |
| Finding A psychophysical detection measure escapes the biases of interoception questionnaires, whose scores are known to be sensitive to anxiety and body image. The dissociation observed here confirms it: patients detect weak vibrations less well while reporting no gastric sensations different from those of controls. | |
| Robustness of the prognostic signal | Very wide intervals |
| Finding Sixteen events, thirteen to fourteen predictors tested per table, no correction, no external validation, no discriminative-performance measure reported. These are signals to explore, not established prognostic markers. The authors themselves speak of candidate biomarkers. | |
| Prognostic inference | Association, not causality |
| Finding Lower detection could be a marker of severity, a sequela of undernutrition, an effect of the psychotropic medication taken by 92% of patients, or a factor contributing to relapse. The design does not separate these hypotheses. Adjustment is made for age and body mass index, not for treatments. | |
| External validity | Limited |
| Finding A single centre, a single inpatient unit, restrictive-type women only, 90% white participants, diagnoses made in routine care without a structured interview, purging or low-weight patients excluded. Transposition to purging presentations, to men, and to outpatient pathways remains entirely to be shown. The authors themselves list these limitations. | |
| Match between claim and evidence | Conclusion to temper |
| Finding The abstract concludes that these findings support the use of ingestible mechanosensory probes as tools for monitoring treatment response and preventing relapse. On sixteen events, without correction or replication, this wording outruns what the data establish. The body of the article is more measured and speaks of candidate biomarkers. | |
Level of evidence
Oxford level of evidence 2b. Confidence is fair on the existence of a deficit in detecting low-intensity gastric stimulations in these patients, an internal, controlled result consistent across two independent measures. It is low on the prognostic value of the identified markers, drawn from uncorrected exploratory analyses on sixteen events. It is nil on any individual clinical use, for lack of an available device, a validated threshold, or independent replication.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ Nothing changes on Monday morning, the tool does not exist in practice and the confidence intervals are too wide to draw anything individual from them. But the fact that the deficit concerns only weak signals, and disappears when you raise the intensity, tells you something useful about what you are trying to restore in a patient in remission. ”
What this means in practice: this study does not provide a test, it provides a working hypothesis about the relationship to bodily sensations after discharge. It justifies following future work on body-centred approaches, without anticipating their results.
What you can take from this
- Understand that the impairment in perceiving gastric signals in anorexia nervosa is measurable by means other than a questionnaire, and that it persists after weight restoration.
- Note that the observed deficit concerns weak stimulations and fades when the stimulation is enhanced, which is not the same as a global visceral insensitivity.
- Know how to answer a patient or family who may have heard of a relapse test: the device is experimental, neither available nor validated for individual use.
- Do not turn a group association into an individual prognosis. An odds ratio whose confidence interval runs from 1.02 to 15.91, estimated on sixteen relapses, does not translate into a probability for the patient in front of you.
- Continue basing relapse monitoring on established clinical elements, while bearing in mind that the relationship to bodily sensations is a legitimate focus of psychotherapeutic work.
Frequently asked questions
Could this device become a monitoring tool?
Nothing rules it out, but it would require, at minimum, independent replication, an evaluation of individual discriminative performance, correction for multiple comparisons, and a setup usable outside the laboratory. None of these steps has been taken.
Is the observed impairment a cause or a consequence?
The study cannot settle this. Prior undernutrition, illness severity, psychotropic medication present in 92% of patients, and a causal mechanism all remain equally compatible with the data.
Does the result hold for adolescents?
The included population ranges from 13 to 40 years old, with a mean age of 18.9, so it does include adolescents. But it is exclusively female, exclusively restrictive-type, and drawn from a single inpatient unit. No result is reported by age band.
Why does the brain arm show nothing?
Gastric evoked potential amplitudes do not differ between groups. The authors do, however, report stronger correlations, in the anorexia nervosa group, between these potentials and the behavioural measures. An absence of significant difference is not proof of absence of difference, particularly with this sample size.
Does this justify referral to body-centred therapies?
It makes the hypothesis plausible, it does not demonstrate it. No intervention was tested in this work, and the effectiveness of an approach cannot be inferred from the existence of a mechanism.
Annotated bibliography
Source study. Verdonk C, Mink K, Choquette E, Moseman SE, Mayeli A, Stewart JL, Paulus MP, Smith R, Khalsa SS. Gastrointestinal Interoception and Relapse in Anorexia Nervosa. JAMA Psychiatry. 2026 Jun 17:e261301. Published online ahead of print, no volume or pagination assigned to date. DOI 10.1001/jamapsychiatry.2026.1301. PMID 42307918. Registered as ClinicalTrials.gov NCT05111977. Three appendices accompany the article: trial protocol, online supplementary content, data-sharing statement.
Context reference, neurophysiological antecedent. Mayeli A, Al Zoubi O, White EJ, et al. Parieto-occipital ERP indicators of gut mechanosensation in humans. Nature Communications. 2023;14(1):3398. DOI 10.1038/s41467-023-39058-4. The work that established the electroencephalographic signature used here. PubMed identifier not established at the time of this review; the reference is reproduced as it appears in the source’s bibliography.
Context reference, computational antecedent. Smith R, Mayeli A, Taylor S, Al Zoubi O, Naegele J, Khalsa SS. Gut inference: a computational modelling approach. Biological Psychology. 2021;164:108152. DOI 10.1016/j.biopsycho.2021.108152. The Bayesian model from which the prior-belief and interoceptive-precision parameters are drawn. PubMed identifier not established at the time of this review.