Anxiety sensitivity

The belief that arousal sensations are dangerous. Enters the wiki with Van der Does et al. (2000), where it turns out to be the only thing that distinguishes people who can count their heartbeats from people who cannot.

Why it belongs on the interoception side of the wiki

Most constructs here concern the signal (how strong, how well detected, how mapped). Anxiety sensitivity concerns what the signal is taken to mean — and it is dissociable from every other candidate. In the 2000 pool of 709 participants, accurate perceivers scored about half a standard deviation higher on the ASI (26.9 vs 21.0 for inaccurate perceivers, F = 3.8, p = 0.02), replicating Van der Does et al. (1997) in an independent sample.

They did not differ on:

• trait anxiety (STAI)
• state anxiety
• depressive symptoms
• somatosensory amplification (SSAS)
• age, body mass index, or actual number of heartbeats during the trials

That list is the interesting part. Accurate perceivers are not more anxious, not more depressed, and — strikingly — not more prone to amplify somatic sensations in general. They specifically believe bodily arousal is harmful.

In the interoceptive-taxonomy’s terms this sits closest to sensibility (beliefs and self-report about one’s own interoception) rather than accuracy, and the finding is that the wiki’s canonical accuracy measure is partly indexing a sensibility construct. See interoceptive-trait-prediction-error, where accuracy/sensibility mismatch is proposed as a trait-level prediction error and reported as correlated with anxiety.

The reframing it licenses

Ehlers (1995) found that good heartbeat perception predicts poor treatment outcome and recurrence of panic after remission — the clinical result that made cardiac interoception matter to clinicians in the first place, and one of the sharpest “more is worse” data points in this wiki (is-more-interoceptive-awareness-better).

Van der Does et al. propose that anxiety sensitivity is why. If the heartbeat score partly indexes the belief that bodily sensations are harmful, then “good perceivers do worse in treatment” is not a fact about perceptual ability being bad for you. It is a fact about catastrophic beliefs about the body predicting poor outcome, which is unsurprising and which the heartbeat task is picking up by proxy.

That is a deflationary reading of an alarming finding, and it generalizes: any clinical result of the form “high interoceptive accuracy predicts worse X” is a candidate for the same treatment.

The confound it creates, which runs the other way

Anxiety sensitivity does not merely correlate with the score — it supplies a mechanism for earning the score without perceiving anything.

Van der Does et al.’s artefact argument: nearly every participant undercounts. A participant who is nervous during the test, and who expects a racing heart because they believe arousal is dangerous, will count faster. Counting faster, in a population that systematically undercounts, moves you closer to the true value. You score as an accurate perceiver by being frightened of your heart.

So anxiety sensitivity is simultaneously:

• the one trait that distinguishes accurate perceivers (an empirical finding), and
• a route to scoring as an accurate perceiver without perceiving (a proposed artefact).

These are not independent — the second explains the first, and if it is right the ASI difference is not a discovery about accurate perceivers but a symptom of the task’s failure mode. Recorded on is-the-heartbeat-counting-task-valid.

A within-subject instance of the artefact, from a study that read it as attention. Zoellner & Craske (1999) found that within each participant, heartbeat-counting error fell as state anxiety rose (F(2,81)=10.04, p<.001), with no breakdown at high anxiety. That is the confound above happening within a person across trials: get more anxious, count more, score better. Two things to hold. First, it does not collide with the 2000 finding that accurate perceivers differ on the ASI but not on state anxiety — that is a between-person trait comparison at one time; this is a within-person state manipulation, and both can be true. Second, Zoellner & Craske read the effect as anxiety magnifying attention to the body (attentional-bias-to-threat), not as count inflation; their unsigned-error design cannot separate the two, so the study supplies the artefact’s mechanism and an innocent reading of it in the same result.

Distinct from trait anxiety, and the distinction does work

Worth stating because the two are easy to collapse and the wiki’s sources do not always keep them apart. Trait anxiety is a disposition to feel anxious. Anxiety sensitivity is a belief about what arousal means — you can be highly anxiety-sensitive without being especially anxious, and vice versa.

The 2000 data are a clean demonstration: accurate and inaccurate perceivers differ on the ASI and are indistinguishable on the STAI, in the same 709 people. Whatever the heartbeat task is tracking, it tracks the belief and not the affect.

A “more is better” datum, from the horizon

The Khalsa et al. (2018) roadmap adds the one place in the wiki where high anxiety sensitivity meets an intervention that reportedly helps it. Floatation-REST — floating in a dark, silent, skin-temperature tank to attenuate the outer world and amplify the inner — is reported (Feinstein et al. 2018) to elicit relaxation and interoceptive awareness specifically in individuals with high anxiety sensitivity, with short-term anxiolytic effect. If it holds (the data are single-session pilots), it is a rare “more interoceptive contact, better outcome” point in the population most defined by fearing bodily sensation — which is why it lands on is-more-interoceptive-awareness-better as evidence against the “more is worse” reading that the Ehlers/panic material supplies. The mechanism the roadmap proposes fits the ASI construct directly: the tank removes the predictability and control that let an anxiety-sensitive person avoid their own arousal, forcing tolerated contact — a subtraction-based cousin of interoceptive-exposure.

The “fear of fear” programme, first-hand

Ehlers (1993) — the wiki’s first first-hand Ehlers source — makes “rating bodily symptoms as dangerous” the third of the three characteristics her programme attributes to panic patients, alongside enhanced cardiac perception and attentional bias. Two things it adds to this page:

• The belief discriminates panic even after controlling for affect. Ehlers (1991): 110 panic patients, 81 infrequent panickers, 37 other-anxiety, 61 controls rated how probable, dangerous, and hard to cope with a list of symptoms were. Panic patients (and infrequent panickers) rated anxiety/panic symptoms as more dangerous than controls even when trait anxiety and depression were covaried out — the same ASI-is-not-STAI dissociation the 2000 pool would later show on the heartbeat task, reached from the belief side rather than the perception side. Converging measures (ASI, BSQ, ACQ, the ACQ Physical-Concerns factor specifically) all separate panic patients from other anxiety disorders.
• ASI predicts persistence prospectively. Ehlers’s preliminary prospective study of 46 infrequent panickers found the 17 who kept panicking over a year had higher baseline ASI than the 29 who did not. This is the fear-of-fear counterpart to the cardiac-perception-predicts-relapse finding, and it means the two prospective results the wiki leans on (perception → relapse; ASI → persistence) come from the same programme, giving the Van der Does reframing — that “good perceivers do worse” is really “anxiety-sensitive people do worse” — a first-hand basis it previously lacked.

A social-anxiety extension, from the constructionist side

Theriault, Young & Barrett (2021) reach this construct from an unexpected direction — social pressure. On their account the sense-of-should is an anticipatory aversion to violating others’ expectations, built by interpreting the interoceptive arousal that social prediction error generates. A footnote draws the clinical corollary directly: because the aversion is anticipatory, a person can be wrong about how disruptive their behaviour will be, and social anxiety may be exactly that error — “pathologically overestimat[ing] how severely their behavior will disrupt their social environment, or overweight[ing] aversive interoceptive experience” (they cite Khalsa et al. 2018).

That second clause is anxiety sensitivity in the wiki’s terms — over-weighting the meaning of aversive interoceptive signals — reached from a social rather than a cardiac-panic starting point. It suggests the ASI’s “beliefs about the danger of arousal” and the social-anxious over-estimation of social threat may be the same over-weighting operating on the same interoceptive read-out, one aimed at the body’s own signals and one at their social consequences. Recorded as a bridge rather than a result: Theriault et al. supply no measure, only the theoretical placement.

The ASI meets a non-cardiac task, and it does not behave the same way

This page’s first open item — that nothing here measures the ASI against a non-cardiac interoceptive task — is closed by Harrison et al. (2021), and the answer is not the one the cardiac material predicts.

Sixty adults split by trait anxiety completed the ASI-3 and the Filter Detection Task, a respiratory threshold-and-metacognition task. The moderate-anxiety group scored far higher on the ASI (18.4 vs 6.8, t = 6.9, p < 0.01) — the construct separates the groups as strongly as anything measured. What it correlated with, across the whole sample:

ASI-3 against	r	p
respiratory perceptual threshold	0.20	0.13 (ns)
metacognitive bias (mean confidence)	−0.30	0.02
metacognitive performance (M-Ratio)	−0.21	0.11 (ns)

So anxiety sensitivity tracks how confident people are about their breathing perception, and not measurably how well they perceive. Three consequences for this page:

The Van der Does deflation generalizes, and gets sharper. The reframing above says the ASI’s relationship to the cardiac counting score may be a belief masquerading as a perceptual ability. Here, in a channel where the score cannot be earned by believing (the load is physically present or absent; see is-the-heartbeat-counting-task-valid for why the cardiac task is different), the ASI attaches to the confidence rating and not to the threshold. That is what the deflationary reading predicts.

But the direction of the perceptual relationship reverses. In the cardiac literature, anxiety and panic go with better detection. On the FDT, the moderate-anxiety group was less respiratory-sensitive (3.0 vs 4.0 filters, Z = −2.4, p = 0.01), replicating Garfinkel et al. (2016a) and Tiller et al. (1987). Whether that is organ-specificity or an indictment of the cardiac instrument is unresolved — see is-interoception-domain-general and the study page.

“Anxiety sensitivity” and “anxiety” are not separable in this design. The ASI correlates 0.57–0.68 with state anxiety, GAD-7 and depression in this sample, and all four load together on the same principal component. The clean ASI-not-STAI dissociation the 2000 pool delivered on the heartbeat task has no counterpart here, because the groups were selected on trait anxiety in the first place. So the table above tells us where affect-in-general attaches, not specifically where the belief attaches.

Open

• Whether the ASI relationship is about interoception or about hearts specifically now has a partial answer (above) — but the discriminating study, both a cardiac and a respiratory task in one anxiety-selected sample, has still not been run.
• The construct’s relationship to the cognitive model of panic (catastrophic misinterpretation of bodily sensations is the model’s engine, and the ASI is close to a measure of the propensity to it) is now sourced first-hand via Ehlers (1993) — though the primary ASI paper (Reiss et al. 1986) and Clark’s model papers are still not in raw/.
• Its relationship to attachment-anxious interoceptive profiles — high noticing, low not-worrying (oldroyd-2019-attachment-interoception) — is unexplored and looks like the same phenomenon reached from developmental psychology.