Interoception and Mental Health: A Roadmap

The field’s own map of itself. Where nearly every other study in this wiki is a single voice making a single argument, this is 24 named authors plus the roster of the Interoception Summit 2016 (Laureate Institute for Brain Research, November 2016, Warren-Foundation funded) trying to agree on what interoception is, how to measure it, how it goes wrong, and what to do next. It is the wiki’s first document whose purpose is to state the consensus rather than push a thesis — and that makes it the natural hub for the concept cluster the other papers each touch from one side.

Read the way the wiki reads everything else — for what it commits to and what it leaves undone — the roadmap’s most useful property is that it ratifies and organizes most of what is already here, while filling two or three gaps the wiki has been complaining about for several ingests. It contradicts nothing. It is, structurally, the review the other 30 studies were written into.

Organized around the meeting’s four themes.

Assessment: the field cannot agree on words

The paper opens where the wiki opened — on the taxonomy problem. “There is no generally agreed-on taxonomy for interoception science,” and both interoception and interoceptive awareness are, the authors argue, too broad to be useful: “interoceptive awareness” began life as a single EDI self-report subscale (Garner et al. 1983) and has since been stretched to cover any interoceptive feature reachable by report at all.

Their answer is Table 2, an eight-feature nomenclature for interoceptive awareness: attention, detection, magnitude, discrimination, accuracy (sensitivity), insight, sensibility, self-report. This is now the wiki’s third canonical taxonomy of interoceptive constructs, and the three do not line up cleanly — see interoceptive-taxonomy, which folds Khalsa’s version in alongside Farb et al.’s seven and Garfinkel et al.’s tripartite model and records where they collide. The load-bearing distinction the paper presses is the one all three share: sensation (the raw signal from bodily sensors) is not perception (what the system makes of it), and most of interoception happens outside conscious awareness entirely.

Interoception, the paper insists, is not one channel but all of them: cardiovascular, pulmonary, gastrointestinal, genitourinary, nociceptive, chemosensory, osmotic, thermoregulatory, immune, autonomic (Table 1). The field’s besetting sin is that it studies these siloed within single organ systems and single disciplines. The recommended correction is multilevel assessment — multiple systems and features in the same individual, top-down attention probes combined with bottom-up perturbations — and a repeated complaint that the efferent (visceromotor) limb is understudied: interoception is not just afferent read-out, it is the brain’s control loop over the body, and paradigms that measure visceromotor output barely exist.

Integration: one computation under the anatomy

The neural pathways section is a compact restatement of the wiki’s anatomy: afferent relays (NTS, parabrachial nucleus, PAG; thalamus, hypothalamus, hippocampus, amygdala; insula and somatosensory cortices) feeding a set of efferent visceromotor regions (anterior insula, ACC, subgenual and orbitofrontal, vmPFC, SMA, premotor). The paper notes without alarm that these regions “coincide closely” with the nociceptive and affective systems — the overlap Craig, Lindquist and Barrett each read differently.

The section that earns the paper its place in the theory cluster is Computational Theories of Interoception. Sensing alone cannot solve interoception because interosensations are noisy and ambiguous; the brain therefore runs a generative model and does Bayesian inference. Cast as hierarchical Bayesian models, interoception and homeostatic/allostatic control fall under one set of quantities — predictions, prediction errors, precisions — and surprise-minimization proceeds two ways: update beliefs (predictive-coding) or act to fulfill them (active-inference). Homeostatic (reactive) control sits at the bottom of the hierarchy; allostatic (prospective) control modulates homeostatic set-points from above. The striking mechanistic claim, carried in from Petzschner/Stephan: belief precision sets the force and pace of corrective action — “the tighter the expected range of bodily state, the more vigorous the elicited regulatory action” — which the authors offer as a novel account of psychosomatic phenomena and placebo. This is the seed of a computational psychosomatics, given its own page. The honest caveat is in the text: “the empirical evidence for hierarchical Bayesian principles of interoception… is indirect so far.”

The psychopathology figure (Fig 4) reproduces Barrett & Simmons’s Embodied Predictive Interoception Coding (EPIC) model directly — agranular visceromotor cortex issuing allostatic predictions, granular mid/posterior insula returning prediction error — and states the model’s mental-illness hypothesis: that interoceptive input becomes decoupled from interoceptive predictions, presenting as “noisy afferent interoceptive inputs.” See theory-of-constructed-emotion and insular-cortex.

Psychopathology: interoceptive dysfunction as transdiagnostic

The paper’s clinical core, and its most quotable consensus: interoceptive dysfunction is “increasingly recognized as an important component of different mental health conditions.” Table 3 lays out symptoms and signs across panic disorder, depression, eating disorders, somatic symptom disorders, substance use, PTSD, GAD, depersonalization/derealization, and autism. Mood and anxiety disorders are framed (via Paulus & Stein 2010) as failures to appropriately anticipate interoceptive states; eating disorders as abnormal caloric anticipation; addiction as abnormal responses to interoceptive cues on overlapping circuitry. The recommended lens is dimensional and transdiagnostic — RDoC — because several interoceptive processes never surface at the symptom-report level clinicians rely on. Five candidate dysfunction types are named: attentional bias (hypervigilance), distorted physiological sensitivity, cognitive bias (catastrophizing), abnormal sensibility, impaired insight. This whole strand gets its own page: interoceptive-psychopathology.

On biomarkers, the paper is sober: “there is currently limited evidence for interoceptive predictors of diagnostic, prognostic, or treatment status.” The most clinically sensitive measures, it speculates, will be perturbation probes that engage metacognitive beliefs under load — an interoceptive analogue of a cardiac stress test.

Two passages fill gaps the wiki has been flagging by name:

• The panic history, first-hand at last. The wiki’s cognitive-model-of-panic page exists but laments that Clark, Klein and the 1980s panic literature are “not in raw/” and known only secondhand through Van der Does. Khalsa et al. supply the field’s own memory of it: 1980s biological psychiatry was “inundated with observations of interoceptive disturbances in panic disorder” (lactate infusion, CO₂ inhalation, isoproterenol), and the wave “receded after it became clear that the etiological mechanism was broader than a single molecular receptor target.” That is the rise-and-fall arc the panic page reconstructs, now stated by the people who lived it.
• Interoceptive exposure, assessed by the field. The wiki’s interoceptive-exposure page has been asserting “established efficacy in panic disorder” for several ingests with no source read first-hand. This paper is not an efficacy trial, but it is an authoritative field statement: interoceptive exposure is “the most common application of interoceptive evaluation in current clinical practice” — and the manipulations “often fail to adequately reproduce the fear response, possibly because the patient retains full control over the stimulation… and the perturbation remains predictable,” few studies exist, long-term outcomes are unknown, and “none of these approaches has translated into clinical practice.” A candid downgrade of exactly the claim the exposure page was carrying on trust.

Current interoceptive treatments are catalogued (pharmacotherapy modulating interoceptive physiology; CBT with exposure/response prevention; capnometry-assisted respiratory training for panic; and MBSR, yoga and meditation/movement treatments aimed at metacognitive mind–body awareness — the last citing Farb et al. 2015, the wiki’s own author, as reference 127). On the horizon: Floatation-REST (attenuate exteroception to amplify interoception; Feinstein) and minimally-invasive perturbation tools (inspiratory loads, thermomodulation, transcutaneous vagus nerve stimulation).

Roadmap: what makes interoception special

The closing argument for why interoception deserves its own science rather than dissolution into “sensation”: it involves unusually high intra-brain connectivity (Kleckner et al. 2017), it is “tightly linked to the self and survival,” and it represents the present against the experienced past and anticipated future — so learning and malleability matter. Its conceptual frame overlaps with emotion and pain because all three serve homeostasis, and a key future task is deciding which neural systems for interoception, emotion, cognition and pain are “overlapping, interdigitating, or even possibly identical.” The document ends on its own limits: the improved models “remain theoretical and await further testing… the current document is best viewed as a work in progress.”

Why it matters to this wiki

Not for a new result — it has none — but for what it consolidates:

• It is the hub for the taxonomy strand (Farb, Garfinkel, now Khalsa), the computational strand (Seth, Seth & Friston, Barrett, now cast as clinical program), and the clinical strand (panic, exposure, MBCT/MBSR).
• It converts the wiki’s scattered clinical fragments — anxiety-sensitivity, cognitive-model-of-panic, interoceptive-training-clinical — into an explicit transdiagnostic framework (interoceptive-psychopathology) and a computational one (computational-psychiatry), each now a page.
• It ratifies the wiki’s own read that the field adopted the cardiac heartbeat tasks as standard practice without settling the validity question — the paper uses them as accuracy measures with no caveat, which is exactly the state is-the-heartbeat-counting-task-valid describes.

The thing to hold in view is that a consensus document is evidence about the field, not about the body. Where the roadmap agrees with the wiki’s harder-won readings that is reassuring; where it states a framework as settled (the Bayesian account) it also, to its credit, marks the evidence as indirect. Filed as the wiki’s orienting map, not as a finding.