Banellis et al. (2026) — Interoceptive ability is uncorrelated across respiratory and cardiac axes
The paper this wiki had been queuing by name for three ingests. micah-allen’s page, respiratory-interoception, interoceptive-control and interoceptive-taxonomy all cite it in advance as the study that would decide whether “interoceptive accuracy” has a person-level referent at all. Read first-hand, it mostly delivers — with one qualification the forecasts did not anticipate, and one positive result more interesting than the null.
What was done
241 participants from the Visceral Mind Project (Aarhus) completed two psychophysical tasks plus an exteroceptive control, all scored by the same hierarchical Bayesian machinery:
- HRDT — listen to your heart for five seconds, then judge whether a tone train is faster or slower than it; Psi-adaptive staircase in Δ-BPM units.
- RRST — two inhalations through a motorised circuit, one resistively loaded; judge which was harder.
- HRDT-Exteroception — the identical judgment made between two tone trains, no body involved.
Each yields four quantities: sensitivity (threshold), precision (slope), metacognitive bias (mean confidence) and metacognitive-efficiency (M-Ratio). The design is the taxonomy operationalised — four constructs, three modalities, one estimation pipeline — which is what makes the cross-correlation matrix worth more than the sum of the pairwise studies preceding it.
That uniformity is the paper’s real methodological argument. Earlier cross-modal work compared a counting score to a filter-detection score; a null there could always be blamed on the instruments being incommensurable. Here the two tasks differ in organ and stimulus and in nothing else that the analysis touches.
The null, and how good it is
| cross-modal pair | r | p_FDR | BF01 |
|---|---|---|---|
| cardiac ↔ respiratory sensitivity | −0.019 | 0.806 | 6.39 |
| cardiac ↔ respiratory precision | −0.022 | ns | 6.30 |
| cardiac ↔ respiratory metacognitive efficiency | — | 0.806 | 5.04 |
| cardiac ↔ respiratory confidence | 0.510 | <.001 | (evidence against null) |
The effect sizes are not small-but-present; they are at zero. And the study is powered to detect r ≥ 0.179, comfortably below the r = 0.360 metacognitive association Garfinkel et al. (2016) reported for these same two axes — the finding this literature’s domain-general reading mostly rests on. So this is a failed replication with the power to mean it, not an underpowered miss.
The honest brake: BF01 ≈ 5–7 is moderate evidence, and the authors say so. This licenses “no detectable common factor at rest in healthy young adults,” not “the channels are mechanistically independent.”
The positive result is the more interesting one
Confidence travelled everywhere. Respiratory↔auditory confidence at r = 0.642 is the largest correlation in the paper — larger than either interoceptive pair — and it links an interoceptive judgment to a judgment about tones.
Set the two halves side by side and the shape is clean:
- What you can do is organ-specific. Cardiac skill predicts nothing about respiratory skill.
- How sure you feel is a property of the person, and it does not care whether the signal came from inside the body or from a loudspeaker.
This is the interoceptive instance of a pattern well established in exteroceptive metacognition (Ais et al. 2016; Mazancieux et al. 2020; Rahnev 2025): bias generalises, efficiency does not. The wiki should record what it costs. Every questionnaire measure of sensibility — the maia, the Body Perception Questionnaire, the confidence ratings folded into “interoceptive awareness” — is a measure sitting on the domain-general side of this split. See metacognitive-efficiency.
Which supplies a mechanism for something interoceptive-sensitivity has recorded for four ingests without one: sensibility and accuracy keep failing to correlate, and the attachment-anxious profile scores high on noticing while the meditators score high on attention with no accuracy gain. If confidence is a trait about the person and performance is a fact about an organ, these are not anomalies to be explained one at a time. They are the same dissociation, met repeatedly.
What it does to the rest of the wiki
The channel-specificity worry is now first-hand and strong. interoceptive-taxonomy carries Ferentzi et al. (2018) — sensitivity is not a generalizable feature — at one remove, via Berntson & Khalsa’s citation. That paper did not include respiration; this one closes that gap in the axis pair the field cares most about, with better instruments. The consequence is stated on is-interoception-domain-general, created with this ingest.
It does not overturn any wiki claim — it dissolves a class of them. No page here asserts cardiac–respiratory correlation, so nothing is contradicted. What is damaged is looser: the habit, running through this wiki because it runs through the field, of measuring interoception in the heart and then discussing interoception. Every study page that reports a cardiac finding and a general-sounding conclusion is now carrying an unstated modality restriction.
And it puts pressure on the wiki’s own newest construct. interoceptive-control was created two ingests ago holding open the question of whether control is a general capacity or a fact about breathing, explicitly pending “Banellis et al. 2026 on cardiac/respiratory decorrelation.” The answer is unhelpful in a specific way: this paper decorrelates perception across channels, which makes a general control construct less likely by analogy but tests nothing about control. The forecast was that this paper would settle it. It cannot.
The clinical implication is the authors’ own, and it is sharp. Interoceptive dysfunction is proposed as a transdiagnostic marker (interoceptive-psychopathology, khalsa-2018-interoception-roadmap) largely on cardiac evidence. If the axes are independent at rest, then “this disorder has an interoceptive deficit” measured cardiacally does not license expecting one respiratorily — and panic, the disorder with the strongest respiratory phenomenology, is one the field has mostly studied through the heart.
Where it leaves the escape routes
Two remain, and both are in the paper:
- Arousal. Coherence may exist under physiological challenge and vanish at rest — consistent with predictive-coding models on which interoceptive prediction errors matter most when the body is perturbed (Barrett & Simmons 2015). This is sahib-khalsa’s standing methodological position (isoproterenol) arriving as the leading defence of domain generality. Untested here.
- Task asymmetry. The HRDT is cardiac-plus-auditory; the RRST is respiratory alone. The authors propose pairing the HRDT with a tactile–auditory control, or adding an exteroceptive reference to the RRST, to make them comparably multisensory. Until then a residual criterion difference is not excluded.
A third the paper does not raise: a null correlation between two noisy measures is a weaker claim than it reads as. Cardiac M-Ratio required excluding 96 of 241 participants as psychophysically implausible; the RRST required 10. Whatever is producing that asymmetry is not modality-neutral, and the metacognitive null in particular is a comparison between a heavily-filtered variable and a lightly-filtered one.
The companion paper, ingested next
[[banellis-2026-body-wandering|Banellis et al. (2026, PNAS)]] — same first author, same senior author, same Visceral Mind Project cohort, different question — supplies the complement to the split above and should be read with it.
This paper says performance is organ-specific and confidence is domain-general. That one measures neither: it asks 536 people how much their resting thoughts were about the body, and finds the resulting body-wandering dimension is stable, trait-like (split-half ICCs 0.88–0.94), and predicts affect, autonomic arousal and symptom scores.
The join is that body-wandering sits squarely on the domain-general side of this paper’s split. It is a relationship to the body, not a reading of one — and it behaves exactly as this paper predicts such a quantity should: reliably person-level, while saying nothing about any perceptual ability. Two 2026 papers from one group, converging on the claim that what a person has interoceptively is a stance toward the body, and what they lack is any general skill at perceiving it.
The caution that comes with the pairing: the samples are drawn from the same cohort, so these are not independent replications of anything, and neither study measures both quantities in the same participants. The obvious next study — body-wandering scores against HRDT and RRST performance in the same people — is available to this group and has not been run.
Provenance notes
Peer reviewers named in the manuscript are Irena Arslanova and frederike-petzschner. Funding: Lundbeckfonden Fellowship and an ERC Starting Grant to micah-allen; no declared conflicts. Data and analysis code are public. First author Leah Banellis is given no researcher page here — single appearance in raw/, per the co-author convention; the position belongs to micah-allen, who is senior author and whose instrument programme this study both motivates and complicates.