Multidimensional assessment of anticipated and experienced interoceptive states

Study at a glance

• Authors: Nicolas Wilzok, Emily M. Adamic, Sahib S. Khalsa, Ilona Croy
• Year: 2023
• Journal: Physiology & Behavior
• DOI: 10.1016/j.physbeh.2023.114265
• N: 52
• Design: Within-subject psychophysical paradigm validation; two novel cue-and-rate paradigms crossed with two interoceptive modalities (nociception, respiroception); linear mixed models; test-retest subsample (n=10, ~7 days)

Key findings

• Two new paradigms validated: the Accuracy of Interoceptive Anticipation paradigm (three graded cues, stimulus on 1/3 of trials, rate experience or — when no stimulus arrives — the anticipation retrospectively) and the Interoceptive Discrepancy paradigm (rate anticipation before every stimulus, then deliver a stimulus matching the cue or shifted ±1 level, then rate experience).
• Both are valid in both channels: cue level drove anticipation ratings (respiroception F[2,867]=705.9; nociception F[2,867]=257.9, both p<0.001) and stimulus strength drove experience ratings (F=533.6 and 107.2, p<0.001).
• The Discrepancy paradigm successfully induced the intended mismatch — the Increased condition produced underestimation and the Decreased condition overestimation, both differing from baseline (|d| 1.0–1.7 respiroceptive; 0.7–0.8 increased and 0.4–0.6 decreased nociceptive).
• Cross-modal correlation of discrepancy magnitude: r = 0.57, p < 0.001. People whose experience departed from their anticipation in breathing did the same in pain.
• Asymmetry: induced discrepancy was larger for underestimation (stimulus stronger than cued) than for overestimation, consistent with the pain literature that underpredicted pain hurts more (Rachman & Arntz 1991; Arntz 1996).
• An unpredicted baseline effect: at the highest cue, anticipation and experience diverged even when the stimulus matched the cue — participants overpredicted the strongest stimulus in both modalities (nociception t[761]=-7.3, p<0.001, d=-0.6).
• Test-retest over ~7 days: anticipation and experience ratings ICC 0.66–0.92; induced discrepancy ICC 0.78 (respiroceptive), 0.54–0.66 (nociceptive).
• Ratings were consistently highest on the sensory scale and lowest on the fear scale, in both modalities and both paradigms.

Limitations

• The two channels were not intensity-matched. Respiroceptive stimuli covered the full VAS range; nociceptive stimuli were ‘much weaker, clustered towards the lower end’. The headline cross-modal correlation is therefore between a well-sampled axis and a floor-compressed one.
• Three cue-stimulus levels cannot establish the psychophysical function. The models assume linearity over the tested range; exponential or plateau forms are untested, and the authors say so.
• Retest n = 10, drawn as the earliest-recruited participants rather than at random — the reliability claim, which is half the paper’s contribution, rests on ten people.
• A reporting error in the results. The respiroceptive discrepancy model reports ‘no effect of rating scale (F[4,1326] = 23.3; p < 0.001)’ — the same F and p already given for the interaction term, and a p < 0.001 described as a null. One of the two statements must be wrong and the paper does not resolve it.
• MAIA-2 and PHQ-9 were administered and never reported. Both appear in the methods; neither appears in the results. The obvious question — whether these laboratory anticipation measures relate to self-reported interoceptive sensibility — was collected and not answered.
• Stimuli were applied by hand by an unblinded experimenter (PinPricks behind a paravent, resistors behind a room divider), so cue-consistent application errors are not excluded by design.
• Healthy German students, 19–34, screened free of pain medication, chronic illness and mental disorder; conducted under COVID mask protocols. Cannot speak to cardiac or gastric channels, nor to thermal or chemical nociception.

Connections

• Methods: Accuracy of Interoceptive Anticipation and Interoceptive Discrepancy paradigms, Inspiratory Breathing Load (IBL) paradigm, Multidimensional Assessment of Interoceptive Awareness (MAIA)
• Concepts: Discrepancy awareness, Nociception, Respiratory interoception, Interoceptive inference, Predictive coding, Forecasting (internal-state prediction), Interoceptive precision, Insular cortex, Exteroception Boundary, Interoceptive psychopathology
• Debates: Is interoception domain-general or organ-specific?, inferential?
• Related studies: Harrison et al. (2021), Banellis et al. (2026), Khalsa et al. (2018), Petzschner et al. (2021), Haruki & Ogawa (2023)

A methods paper, and the wiki should read it as two things at once: a competent instrument-validation exercise that does what it says, and — unintentionally — the closest thing the wiki has to a test of the perturbation escape that Is interoception domain-general or organ-specific? ranks as its strongest unexamined defence.

What was built

Paulus & Stein’s thesis is that psychopathology lives in the anticipation of interoceptive states rather than in their sensing. The wiki has carried that claim since the roadmap ingest with a standing complaint attached (see Forecasting (internal-state prediction)): nobody had written down what the anticipation actually is, still less measured it. The inspiratory breathing load lineage — Paulus et al. (2012), Kruschwitz et al. (2019), Walter et al. (2020) — was the one attempt, and it had two defects Wilzok et al. set out to fix:

1. Anticipation was queried after the moment the stimulus would have arrived, so the rating is contaminated by the relief of its non-arrival.
2. One load level only, so the cue→sensation relation could not be traced.

The fixes give two paradigms, described in full on Accuracy of Interoceptive Anticipation and Interoceptive Discrepancy paradigms. Three graded cues (shades of grey) replace the single cue; the Interoceptive Discrepancy paradigm additionally takes the anticipation rating before every stimulus and then violates it by ±1 level on two-thirds of trials. Every trial is rated on three VAS — sensory, fear, affective — in both channels.

The whole thing was run twice per participant, once with PinPricks on the forearm (32–512 mN) and once with inspiratory resistors (0–50 cmH₂O/LPS), order randomized.

The finding the paper leads with, and the one that matters here

The paper’s headline is that both paradigms are valid and reliable. They are, and the validation is unglamorous and sound: cues drive anticipation, stimuli drive experience, the discrepancy manipulation induces discrepancy, and it all holds up a week later.

The result with consequences for this wiki is a single correlation reported almost in passing:

r = 0.57 between the magnitude of induced discrepancy in respiroception and in nociception.

Two interoceptive channels, one sample, matched instrumentation by construction (literally the same paradigm, the same three scales, the same session), and a large positive cross-modal correlation. Every other cross-channel test in this wiki — Ferentzi, Crucianelli, Banellis — returns a null.

Why this is not (yet) a refutation of the decorrelation literature

It is tempting to file this as the first positive result in Is interoception domain-general or organ-specific? and let it fight Banellis. That would be wrong on three counts, and the page records all three.

It is a different quantity. Banellis decorrelated sensitivity, precision and metacognitive efficiency — first-order perceptual competence and its metacognitive readout. Wilzok correlated discrepancy, a signed difference between two self-reports made minutes apart on the same VAS. Nothing here is a psychophysical threshold. On the taxonomy the wiki uses, this sits nearer Discrepancy awareness and the metacognitive layer than the perceptual one — and Banellis’s one positive cross-modal result was also at that layer, in confidence.

Shared method variance is a complete alternative explanation, and the design maximizes it. The same person rates the same three scales in the same session in both channels. A participant who uses the extremes, or who anchors anticipation conservatively, produces correlated discrepancies with no shared interoceptive machinery whatever. This is exactly the caution the domain-general page already applies to Banellis’s confidence correlation — “mean confidence on any two VAS-rated forced-choice tasks will correlate partly through scale use” — and it applies here with more force, because Wilzok’s quantity is a difference between two VAS ratings rather than a single one, and because there is no exteroceptive control task to calibrate against. Banellis had an auditory comparison and it was decisive; Wilzok has none.

Nociception may not be a second interoceptive channel at all. See Nociception and Exteroception Boundary. The authors know this and address it in one paragraph, adopting the broad Craig-style inclusion on the grounds that all nociception serves homeostasis, against Ceunen, Vlaeyen & Van Diest’s narrower visceral-only proposal. If the narrow definition is right, this is a correlation between an interoceptive channel and a somatosensory one — which would be interesting for different reasons and would not bear on domain generality at all.

But it does put the perturbation escape on the board

Having said all that, the study does something no other cross-channel study in this wiki does, and it does it by accident.

Is interoception domain-general or organ-specific? ranks the escapes, and ranks first: all the null evidence is from resting healthy participants; a common central factor could exist and be invisible at rest. Sahib S. Khalsa’s isoproterenol programme is built on that premise — and Khalsa is an author here. The escape’s prediction is that cross-channel structure should appear under perturbation and not at rest.

Wilzok et al. perturb. Hard, and aversively, in both channels: pinprick pain and inspiratory loading are not resting states. And a large cross-modal correlation appears where the resting literature finds none.

That is a consistency, not a test, because the two literatures differ in quantity, instrument and channel set as well as in arousal. The discriminating experiment is still unrun and still the one the debate page names: the same battery, one sample, rest versus challenge. What changed is that the escape now has a datum pointing its way instead of being purely hypothetical, which is a real if modest movement on the wiki’s most consequential open question.

The authors’ own reading, and where it overreaches

Wilzok et al. conclude that discrepancy awareness “occurs at higher cognitive levels such as the insula, rather than being tied to primary sensory processing for one interoceptive modality per se,” citing the insula’s involvement in surprise (Loued-Khenissi et al. 2020) and in interoceptive prediction and prediction error.

The inference runs from a behavioural correlation to a neural locus with no imaging in the study, and the wiki should mark it as such. It is also in tension with Haruki & Ogawa (2023), where the insula is precisely where two interoceptive channels diverge — right dorsal anterior insula preferring cardiac attention, left dorsal middle insula classifying which organ is attended. “The insula integrates across channels” and “the insula distinguishes channels” are both live, and this paper asserts the first without testing it.

The overprediction result, which the paper undersells

Two findings about direction deserve more than they get:

• Induced discrepancy was larger when the stimulus exceeded the cue (underestimation) than when it fell short.
• At the highest cue, anticipation and experience diverged even on matched trials — participants systematically overpredicted the strongest stimulus.

Both are the classic pain asymmetry (Rachman & Arntz): people overpredict intense pain, and underpredicted pain hurts more than predicted pain. Wilzok et al. show it holds for inspiratory loading as well, which is the more interesting half and is stated in a single sentence. If the asymmetry is general across aversive interoceptive channels rather than a fact about pain, it is a constraint on any interoceptive inference account: the prior is not merely imprecise, it is biased upward for high-intensity states, and it stays biased after a familiarization phase that was sufficient to calibrate every other cue level.

Standing

A solid, modest instrument paper whose most valuable content is a correlation it does not foreground and a bias effect it mentions once. Its instruments are real additions — the first paradigms in this wiki that measure interoceptive anticipation as a graded, per-trial, pre-stimulus quantity in more than one channel — and its reliability data, thin as the n=10 makes them, are more than most interoceptive tasks have.

What it is not is the domain-generality result it could be mistaken for. The correlation is at the wrong layer, in a design that cannot separate shared machinery from shared scale use, across a channel pair one of which may not qualify. Recorded on Is interoception domain-general or organ-specific? as a perturbation-consistent datum with those three brakes attached, not as a field position that contests Banellis.