Cognitive model of panic

The reason anyone measured heartbeat perception in a clinical population in the first place — and, on the evidence the wiki has, a theory that turns out not to need the measurement.

The model

Panic attacks arise when benign bodily sensations are catastrophically misinterpreted: a racing heart means a heart attack, breathlessness means suffocation, dizziness means collapse. The interpretation produces fear; the fear produces more arousal; the arousal supplies more sensations to misinterpret. The circle closes and the attack escalates.

Interoception enters because the model’s first term is a bodily sensation being noticed. Van der Does et al. (2000) give the standard reasoning, via Clark (1996): more accurate heart-rate perception in panic patients is “in line with the model, because the frequent focusing of attention on interoceptive sensations increases the probability for these sensations to be detected.”

That is the hypothesis the whole heartbeat-perception-in-panic literature was built to test. Ehlers & Breuer (1992) found the predicted difference and a decade of work followed.

The model in a proponent’s own words. Ehlers (1993) — now the wiki’s first-hand source — states the loop directly (citing Clark 1986, Barlow 1988, Ehlers Margraf & Roth 1988c): panic patients “suffer from interoceptive fears and… panic in response to the perception of benign physiological changes,” and “a positive feedback process between such sensations, their association with threat and the patient’s anxiety response is postulated.” She frames the model’s own limit honestly — it specifies the necessary conditions of panic (misinterpretation) but “we are not yet in the position to predict when exactly a panic attack will occur” — which is why her programme went looking for the situational and individual-difference factors (perception, attention, belief) that modulate the loop without being sufficient for it. And she reports the crucial negative control: panic patients are not generally more physiologically reactive (Roth et al. 1992), so the sensations being misinterpreted are ordinary ones, exactly as the model requires.

The negative control replicated in an analogue sample. Zoellner & Craske (1999) add an independent version of both halves of this. Infrequent panickers (selected by ASI ≥ 23) counted their heartbeats more accurately than controls — replicating Ehlers & Breuer (1992) from outside her lab — while showing no faster heart rate or any other physiological difference from controls, yet reporting far more anxiety and panic symptoms. The groups were discriminated by self-report, not physiology, which is the model’s own picture: ordinary bodies, extraordinary readings of them (see schema-guided-symptom-perception). Their caveat about heart rate is worth carrying but not over-reading — the analogue panickers, like Ehlers’s patients, did not have faster hearts, which removes the rate half of the cardiodynamic worry; stroke volume was not measured, so the amplitude half survives.

The twist: the model never needed it

The most useful thing this page records, and it comes from the model’s own camp.

Van der Does et al. — with Ehlers as third author — state plainly that accurate heart rate perception is not a necessary assumption of the cognitive model (Ehlers, Margraf & Roth 1988a), “so the results of this study have no consequences for the validity of the model.”

This is worth pausing on. A twenty-year empirical programme measuring cardiac perception in panic patients, and when the reanalysis complicates the finding, the theory it was testing shrugs. The model requires that sensations be misinterpreted, not that they be accurately detected — and misinterpretation of a sensation that is not there works at least as well as misinterpretation of one that is. Arguably better: a patient whose panic is not accompanied by any real heart-rate increase, but who reports one, is misinterpreting more floridly than a patient whose heart is genuinely racing.

So heartbeat perception was consistent with the model, never entailed by it, and the field spent two decades measuring a variable the theory was indifferent to. Recorded as a caution about auxiliary hypotheses: the measurement was adopted because it fit, and fitting is not the same as testing.

What the reanalysis actually leaves for the model

Three findings that bear on panic, none of which is the one the literature was chasing:

• Accurate perception is not specific to panic disorder. No difference between panic patients and social phobia (25.0% accurate, numerically higher than panic’s 17.1%), GAD (20.0%), or specific phobia (11.1%). What the measure tracks is “continuous or frequent episodes of clinical anxiety,” not the diagnosis the model is about. (The comparison groups are small — n = 20, 15, 27 — so this is an underpowered null; see van-der-does-2000-heartbeat-perception-reanalysis.)
• The heart is rarely loud enough during real panic. Accuracy requires ~100+ bpm, and via McNally (1994), heart-rate increases during naturally occurring panic attacks are “rather modest (if they occur at all)” and largely confined to situational attacks. So during actual attacks most patients are in the range where nobody perceives accurately — the vicious circle, if it runs on perceived cardiac acceleration, is mostly running on something that is not being perceived.
• anxiety-sensitivity, not perception, is what distinguishes accurate perceivers — a belief about the harmfulness of bodily sensations, which is much closer to the “catastrophic misinterpretation” the model actually posits. The model’s real variable was in the data all along, wearing the perception measure as a disguise.

The clinical inversion

Because good heartbeat perception predicts worse treatment outcome (Ehlers 1995) and accurate perceivers are more anxiety-sensitive, interoceptive training looks contraindicated in panic — a rare and clean “more is worse” for is-more-interoceptive-awareness-better.

Van der Does et al. argue the opposite for one subgroup, and label the argument counter-intuitive and speculative themselves. Patients whose attacks come with no real HR increase but who report one are the schema-guided perceivers — reading a prior, not a heart. For them, interoceptive training or simply “being informed that what they feel is not their heart” might help, by reconnecting the report to the physiology or by discrediting the schema.

Note what this requires clinically: ambulatory HR monitoring to identify the subgroup, since the whole point is that the patient’s report is uninformative about their heart. The proposal is a stratification hypothesis, and the stratifying variable is one no clinician has by default. See interoceptive-training-clinical, interoceptive-exposure.

The field’s memory of why panic and interoception got married — and divorced

The Khalsa et al. (2018) roadmap supplies, in two sentences, the historical arc this page had been reconstructing from a single reanalysis. In the field’s own telling: “During the 1980s, biological psychiatry was inundated with observations of interoceptive disturbances in panic disorder… although the trend receded after it became clear that the etiological mechanism was broader than a single molecular receptor target.” The 1980s wave was the provocation literature — sodium lactate infusion (Gorman, Klein), CO₂ inhalation, isoproterenol — which reliably triggered panic in patients and made the disorder look like a discrete interoceptive-alarm malfunction with a molecular target. It receded when the mechanism proved broader than any single receptor.

Two things this adds to the page. First, it corroborates first-hand (from a consensus of the people involved) the same rise-and-fall shape the wiki inferred from Van der Does alone: panic was the reason interoception entered biological psychiatry, and the initial strong, specific story did not hold. Second, it reframes the heartbeat-perception programme this page is about as a second wave — after the pharmacological-provocation wave crested, cognitive-model-driven perception measurement carried the panic–interoception link into the 1990s and 2000s, and (per the twist above) measured a variable the theory did not require. The roadmap treats panic-disorder interoceptive dysfunction as settled enough to head its transdiagnostic table (see interoceptive-psychopathology) — so the link survived the divorce from a single-receptor account, just at the coarser grain of “interoceptive dysfunction is part of the disorder.”

Status in this wiki

Less thin than it was. Clark (1986, 1996) himself is still not in raw/, but the model is now stated first-hand by a proponent — Ehlers (1993) — rather than reconstructed solely from Van der Does et al.’s discussion (now supplemented by the Khalsa et al. consensus history above). The caution still stands in a weaker form: Ehlers is herself a partisan of the model, so the wiki’s account of what the model claims is inherited from people invested in it — but it is now inherited from a proponent stating her own view, not only from a reanalysis characterizing it. The page exists to hold the panic-disorder material the wiki had been accumulating on interoceptive-exposure with nowhere to put it, and to record the auxiliary-hypothesis lesson above.

The obvious gap: interoceptive-exposure is the treatment this model generates, is described on its own page as having “established efficacy in panic disorder,” and the wiki has read no source on that efficacy — nor on whether exposure works by changing interpretation (as the model says), by extinguishing conditioned fear, or by updating priors as Farb et al. would have it.