Interoception Wiki

← RADlab

Interoceptive awareness declines with age

11 min read

Study at a glance

  • Authors: Sahib S. Khalsa, David Rudrauf, Daniel Tranel
  • Year: 2009
  • Journal: Psychophysiology
  • DOI: 10.1111/j.1469-8986.2009.00859.x
  • N: 59
  • Design: Cross-sectional correlational study of resting cardiac interoception across adulthood. 59 neurologically/psychiatrically screened adults (23 M, 36 W), ages 22–63 (M=48, SD=11), performed a heartbeat discrimination (simultaneity-judgment) task on two visits up to 2 weeks apart, preceded by a pulse-detection control task screening for attention/decision-making capacity. Age, BMI, sex, and pulse-detection accuracy entered as predictors of normalized heartbeat-detection accuracy (A’, Grier 1971) in Pearson correlations, multiple regression, and logistic regression (good- vs poor-detector classification). Methods identical to Khalsa et al. (2008), the meditators study.

Key findings

  • Age inversely predicts objective heartbeat-detection accuracy, and it is the only demographic that does. Pearson r = −.49 (p<.0001) at Visit 1 and r = −.45 (p<.0005) at Visit 2; averaged across visits, age accounted for 30% of the variance in accuracy (R²=.30, β=−.53, t=−4.49, p<.003; Fig. 1). In multiple regression with BMI, sex, and pulse-detection accuracy also entered, age was the sole significant predictor at both visits (Visit 1 ΔR²=.27, β=−.46; Visit 2 ΔR²=.23, β=−.45).
  • BMI and sex were unrelated to accuracy (all ps>.1) — a null that contradicts prior undergraduate-sample findings (Rouse et al. 1988: BMI and sex predict heartbeat detection). The authors read the discrepancy as aging being disproportionately powerful in a wide-age sample, swamping the sex/BMI effects visible only within a narrow young range (or those effects being nonlinear across the lifespan).
  • The effect is reliable across sessions and survives categorically. Accuracy correlated across the two visits (single-measure ICC=.45, p<.0003, Cronbach’s α=.62). A logistic regression predicting good- vs poor-detector classification found age the only significant predictor (Visit 1 Wald=10.12, p<.002; Visit 2 Wald=5.80, p<.017). Good detectors were younger than poor detectors (Visit 1: 42.0 vs 53.5 yrs, t=4.51, p=.00002).
  • Sex does not modulate the age effect. The inverse age–accuracy relation held separately in men (Visit 1 r=−.46) and women (r=−.50); an ANOVA with sex as factor and age as covariate found a significant age effect (Visit 1 F(1,55)=16.3, p<.0003) and no sex main effect or sex×age interaction.
  • Mechanism is argued to be central and/or cutaneous, explicitly not cardiac afferents. The authors reason the decline is unlikely to arise in the heart’s own afferents (transplant patients before reafferentation detect normally; Barsky et al. 1998) and more likely reflects age-related cortical thinning in interoceptive regions (insula, primary somatosensory cortex) and/or decline of skin mechanoreceptors — Pacinian channels, whose vibrotactile sensitivity predicts heartbeat detection (Knapp, Ring & Brener 1997) and demonstrably declines with age — plus dampened sympathetic reactivity (a quieter heartbeat to detect). Attention was controlled for by the pulse-detection screen.

The paper the wiki’s aging pages had been queuing by name for three ingests, now read first-hand. Where MacCormack et al. (2021) and Volynets et al. (2020) show the felt or represented body’s role in emotion weakening across adulthood — and repeatedly hand the biological-mechanism question off to a citation — this is the citation. A brief report from the Iowa group (Sahib Khalsa, first author, with David Rudrauf and Daniel Tranel) measuring objective cardiac interoceptive accuracy across 22–63 years and finding it falls steeply: age alone accounts for 30% of the variance, with BMI and sex contributing nothing.

What it measures, and why it is not the counting task

This is the wiki’s cleanest illustration of why “the heartbeat-detection task” is really two task families (see Heartbeat detection task). Khalsa et al. do not use Schandry mental-tracking/counting — the instrument nearly all the wiki’s interoception evidence rests on. They use a discrimination method in the Brener–Kluvitse/Brener–Liu–Ring lineage: tones triggered by the participant’s own R-waves, delivered either ~250–300 ms after the R-wave (the R-wave-to-pulse interval, perceived simultaneous with the heartbeat by accurate detectors) or ~650–700 ms after (perceived nonsimultaneous), with a forced-choice simultaneous/delayed judgment scored as A’ (a nonparametric d’ analog; Grier 1971). Because the tones carry the participant’s true rate and rhythm, a belief about one’s resting heart rate buys nothing — only actual perception of beat timing separates the intervals. This is the family built to remove counting’s guessing confound.

Two design features are worth carrying forward:

  • A pulse-detection control task. Before heartbeat detection, participants took their own wrist pulse and made the same simultaneity judgments against that (an exteroceptive signal). This screens for the ability to sustain task-focused attention and make simultaneity judgments at all; 3 participants failing it were excluded. When pulse-detection accuracy is entered as a regression predictor, it does not predict heartbeat accuracy and age still does — so the age effect is not merely age-related decline in attention or decision-making on a simultaneity task. This is a tighter attention control than most heartbeat studies run.
  • Two visits, so it is a trait estimate, not a one-shot. The cross-session ICC of .45 makes the age effect a property of a moderately reliable individual difference, not of a single noisy session — which matters given the lability the Van der Does reanalysis documents for the counting score.

The result, and the one demographic twist

Age inversely predicted accuracy at both visits (r ≈ −.45 to −.49) and as an averaged trait (R²=.30). In a model with BMI, sex, and pulse-detection accuracy competing, age was the only significant predictor — of both continuous accuracy and good/poor-detector classification.

The interesting null is BMI and sex. Prior undergraduate work (Rouse et al. 1988) had found both predict heartbeat detection — leaner and, in some studies, male participants detect better. Here, across a wide age range, neither did. Khalsa et al.’s reading is that aging is simply the dominant source of variance once the sample spans four decades, drowning out effects that are only visible within the narrow young-adult band most heartbeat studies use (or that sex/BMI act nonlinearly across the lifespan in a way this design cannot resolve). Either way it is a caution about the young-undergraduate base rate of this literature: predictors that look robust in 20-year-olds may be small relative to the aging gradient nobody was sampling.

Why the mechanism section matters to the wiki

Khalsa et al. do not just report the decline; they reason carefully about where it cannot be coming from, and the argument is directly useful to Heartbeat detection task and Maturational dualism:

  • Not the heart’s own afferents. Cardiac transplant patients, before their new heart is reinnervated, detect their heartbeats within the normal range (Barsky et al. 1998). So heartbeat “detection” does not primarily read cardiac afferent nerves — consistent with the long-standing view that the felt heartbeat is sensed partly through arterial pulsation transmitted to the skin (chest, neck, etc.).
  • Plausibly cutaneous. Fingertip vibrotactile sensitivity at 250 Hz — the Pacinian channel — predicts heartbeat-detection performance (Knapp, Ring & Brener 1997), and Pacinian sensitivity declines with age. If the felt heartbeat is partly a skin signal, an age-related skin-receptor decline is a candidate mechanism.
  • Plausibly central. Cortical thinning with age in the insula and primary somatosensory cortex — the regions implicated in cardiac awareness — occurs precisely in the middle-age epochs this sample over-represents (Salat et al. 2004; Sowell et al. 2003).
  • Plausibly cardiodynamic — which the paper does not flag but the wiki must. Age also dampens sympathetic reactivity, cardiac rate and contractility. A less forceful heart is a quieter signal to detect, and the wiki’s cardiodynamic confound (stroke volume predicts detection; Schandry et al. 1993) then predicts exactly this age decline with no loss of perceptual skill at all. Khalsa et al. mention reduced cardiac output as a mechanism but treat it as one route to genuinely reduced awareness; the wiki should hold open that the whole effect could be signal-strength rather than perception. Stroke volume was not measured here, so this study cannot separate the two — the same limit that dogs every result on the discrimination and counting pages alike.

Where it lands in the wiki

It de-stubs the aging thread’s biological leg. Age-related interoceptive decline, Maturational dualism, Interoceptive sensitivity and MacCormack all cite “Khalsa, Rudrauf & Tranel 2009” for the claim that objective interoceptive ability declines with age — the datum that lets those pages argue the felt/represented decline they measure has a bodily basis. That claim is now first-hand, and it is real: a 30%-of-variance age effect on an objective task, replicated across two sessions. This is the accuracy counterpart that the self-report sources’ designs could not supply.

But it does not close the gap it is invoked to close. The self-report pages want Khalsa to show that the same quantity they measure (felt intensity, interoceptive knowledge) declines because the body’s afferent signal declines. Khalsa shows an objective detection score declines — which is a third construct again (Interoceptive taxonomy (Farb et al. Box 1): accuracy, not sensibility or knowledge), and one whose age decline is itself confoundable with cardiodynamics rather than perception. So the aging thread now has first-hand evidence on all three constructs (knowledge, felt intensity, objective accuracy) declining together — which is the convergence Maturational dualism predicts — while none of the three establishes the mechanism, and the objective one is exactly the construct where “quieter signal” and “worse perceiver” are hardest to tell apart. The mechanism the whole thread hands off remains, even with this paper read, proposed rather than instrumented.

It quietly connects the aging thread to the decision thread. Khalsa et al.’s discussion is unusually forward-looking: they note the finding gives tools to test the Somatic marker hypothesis, asking whether age-related loss of conscious bodily feedback reduces the influence of emotion on decisions, and citing Denburg, Tranel & Bechara (2005) that some older adults’ real-life decision-making declines prematurely. That is precisely the “dark edge” MacCormack and Maturational dualism later develop — the same fading interoceptive input that eases emotion regulation should starve somatic-marker guidance — arriving here in 2009 as a research proposal rather than a finding. They even float interoceptive-feedback training or augmentation for aging (“similar to the use of exteroceptive aids such as eyeglasses and hearing aids”), a speculative bridge to Interoceptive training in clinical practice.

It is not a hard contradiction with anything. It corroborates the aging pages, adds an objective-accuracy leg the discrimination-task literature (Wiens, Nentjes) can host, and overturns no existing claim; its one surprising null (no BMI/sex effect) is flagged by the authors themselves against the undergraduate literature, not against the wiki.

Provenance note

First author Sahib Khalsa already has a page as the convener of the Interoception Summit and first author of the 2018 roadmap. This paper is nine years earlier and a different Khalsa: a hands-on lesion-and-psychophysiology empiricist in the Iowa (Damasio-lineage) tradition, before the Tulsa/LIBR clinical-consensus phase. The methods are shared with Khalsa et al. (2008), the meditators study the wiki already cites (on Interoceptive sensitivity and contemplative training increase objective interoceptive accuracy?) for the finding that experienced meditators do not show superior heartbeat-detection accuracy — the same instrument, same lab, turned on a different question. Co-authors David Rudrauf (imaging/modelling) and Daniel Tranel (the Iowa neuropsychology group, senior author) are held to the co-author convention here — named, no page — though Tranel is a candidate for one if the Iowa somatic-marker lineage recurs first-hand.

Graph View

  • Studies
  • Concepts
  • Methods
  • Debates
  • Gaps
  • Researchers, other
  • Studies
  • Concepts
  • Methods
  • Debates
  • Gaps
  • Researchers, other

Backlinks