1

Measuring Interoception: The Phase Adjustment

Task
Plans, D.1,2,3*, Ponzo, S.2*, Morelli, D.2,4, Cairo, M.2, Ring, C.5, Keating, C.T.6, Catmur,
C.7, Murphy, J.8, and Bird, G.3, 9

*These authors contributed equally.

1
INDEX Group, Department of Science, Innovation, Technology, and Entrepreneurship, University of
Exeter, United Kingdom
2
BioBeats Group Ltd, London, United Kingdom
3
Department of Experimental Psychology, University of Oxford, Oxford, United Kingdom
4
Department of Engineering Science, Institute of Biomedical Engineering, University of Oxford, Oxford,
United Kingdom
5
School of Sport, Exercise and Rehabilitation Sciences, University of Birmingham, Birmingham, United
Kingdom
6
School of Psychology, University of Birmingham, Birmingham, United Kingdom
7
Department of Psychology, Institute of Psychiatry, Psychology and Neuroscience, King’s College
London
8
Department of Psychology, Royal Holloway University of London, London, United Kingdom
9
Social, Genetic, and Developmental Psychiatry Centre, Institute of Psychiatry, Psychology and
Neuroscience, King’s College London

Correspondence should be addressed to:

David Plans, PhD
Department of Experimental Psychology
The University of Oxford
Woodstock Rd,
Oxford
OX2 6GG
United Kingdom

email: david.plans@psy.ox.ac.uk
phone: +44 (0) 7527016574
 2

Abstract

Interoception, the perception of one’s internal bodily state, has been repeatedly linked to mental
health, and atypical interoception proposed as a transdiagnostic risk factor. Despite the clinical
importance of interoception, existing measures are suboptimal as they are susceptible to
physiological and psychological confounds. Furthermore, existing measures are lab-based,
limiting both scalability and accessibility. Here we describe a novel measure, the Phase
Adjustment Task (PAT), administered via smartphone. During the task tones are presented at the
participant’s heartrate (recorded via smartphone camera), but out of phase with heartbeats.
Participants are required to adjust the phase relationship between tones and heartbeats until they
are synchronous. The design of the task excludes previous confounds, and data from 124
participants indicates both variance in performance across participants and that performance is
not affected by physiological or strategic confounds. Given its validity, accessibility and
scalability, the PAT task is suitable for large-scale clinical interoception research.
 3

Introduction

Interoception, the perception of the body’s internal state (1), has been linked to various
aspects of mental health. Interoception is thought to play a central role in emotion regulation
(2,3), social ability (4), craving and motivation (5,6), learning and decision making (7–13),
attachment (14), and monitoring of one’s own arousal, hunger/satiety, and pain (11,15,16). It is
therefore unsurprising that impaired or otherwise atypical interoception has been reported in
numerous psychiatric conditions, including anxiety, panic disorder, depression, schizophrenia,
alcohol and substance abuse, somatoform disorders, personality disorders, and eating disorders
(e.g. (9,17–22)). It has even been suggested that interoception may represent a common,
transdiagnostic, vulnerability factor for psychopathology; whereby individual differences in
interoception give rise to the ‘P-Factor’ - representing general vulnerability to develop and
maintain psychiatric symptoms (14,17,23–25).
Despite great interest in interoception and its relevance for mental health, measuring
interoception remains challenging. Most studies of interoception use one of two tasks; the
heartbeat counting task (HCT; (26,27) or a variant of the heartbeat detection task (HDT;
(28,29)). In the HCT, participants are asked to count the number of heartbeats they perceive in a
brief period, and accuracy is determined by the correspondence between their count and the
actual number of heartbeats. Despite widespread use, the HCT has been criticised on the basis
that good performance may be achieved simply by knowing one’s resting heart rate and using
this knowledge to estimate the number of heartbeats (30).
Concerns over the validity of the HCT have prompted increased focus on the HDT. In all
HDT variants, participants are required to detect whether an external stimulus (typically a tone)
is synchronous or asynchronous with their heartbeat (28,31). In the most common variant, the 2-
alternative-forced-choice (2AFC) HDT, participants are presented with a tone after a delay
following the heart’s R-wave of either 128ms (thought to be perceived as synchronous with the
heartbeat) or 384ms (asynchronous). Accuracy is determined by comparing the number of trials
correctly identified as synchronous or asynchronous (28). The advantage of the HDT over the
HCT is that knowledge of one’s heart rate is of no use when completing the HDT as both
synchronous and asynchronous tones are presented at the same frequency as the participant’s
heartbeat. The problematic feature of all 2AFC HDT variants, however, is that the use of
predefined delays for all participants relies on there being no individual differences in the delay
following cardiac contraction with which individuals perceive an external stimulus to be
synchronous with their heartbeat. This is problematic as large individual differences exist (31–
 4

34); in part because individuals perceive their heartbeat in different bodily locations (35), and
more peripheral locations give rise to later perception than central locations (31). Thus, some
individuals may perceive neither delay as synchronous, and falsely be judged to be non-
interoceptive (33).
Other HDT variants (the Method of Constant Stimuli (MCS) and 6-alternative-forced-
choice (6AFC) designs; (31–33,36)) do not present stimuli at delays presumed to be perceived as
synchronous and asynchronous. Instead, stimuli are presented at several delays and participants
indicate the delay they perceive to be synchronous. Accurate heartbeat perception is inferred
from the consistency of responses (i.e., the extent to which the same delay is selected as
synchronous); a non-interoceptive participant would select delays at random across trials.
These versions of the HDT are unaffected by heart rate knowledge and do not assume
that every participant perceives their heartbeat at a predetermined delay, but have some
limitations of their own, particularly for clinical studies. First, both versions typically involve a
large number of stimulus presentations in order to compute reliable consistency scores. This can
be problematic, particularly in populations characterized by low motivation, poor attention,
fatigue, or demand avoidance. Second, even the longest tasks employ only six delays covering
500ms after the onset of the R-wave, and it is possible that some participants can resolve
intervals smaller than 100ms. This is compounded when testing populations who require a high
degree of precision in their responses, such as those with Autism Spectrum Disorder, who may
become frustrated at the inability to select the exact delay appropriate for them. Third, they do
not account for the possibility that participants may perceive their heartbeat at different bodily
locations throughout the task, affecting the delay judged to be synchronous; if an interoceptive
participant perceives her heartbeat in several locations, and varies the location used to judge
synchronicity across trials, her consistency would be reduced and she would falsely be judged
non-interoceptive. Finally, these measures can only be administered under laboratory conditions
with specialised equipment (e.g. electrocardiograms), limiting testing to populations able to
travel to laboratories, and reducing the scope to test large samples.
In order to test theories linking interoception and mental health, and the utility of atypical
interoception as a biomarker for psychopathology, valid, reliable and accessible methods for
assessing interoception are needed. In this paper, we present a novel measure of interoception -
the Phase Adjustment Task (PAT) - designed to overcome the limitations of existing tests.
Importantly, in terms of accessibility and scalability, the task requires nothing more than a
smartphone, meaning participants can complete it anywhere. The smartphone camera is used to
detect heartbeats (37–39), and tones are presented at the same frequency as the participant’s
 5

heartbeat but out of phase with heartbeats. Participants use a dial to advance or delay the tones
until they perceive the tones to be synchronous (in-phase) with their heartbeat. Thus, the task
cannot be completed using knowledge of one’s own heart rate, and participants are able to adjust
the phase of the tones as they like, rather than having to select a predetermined delay. This
feature, plus a novel analysis method, means that multiple trials at multiple different delays are
not required, and so the task is much quicker than the MCS and 6AFC approaches. The speed of
the task further enhances the accessibility and inclusivity of this method, and increases its
suitability for clinical studies.
 6

Methods and Materials

Participants
182 healthy participants (excluding those pregnant, or with psychiatric or neurological
diagnoses, were recruited via Prolific (prolific.co). 58 participants failed to complete the task; the
final sample therefore comprised 124 participants (70 females, age range: 18-63, M=26.5,
SD=8.02). Ethical approval was obtained from the local ethics committee and informed consent
obtained from participants.

Information
A smartphone application developed in swiftUI by BioBeats Ltd was used to administer
the PAT task. The application uses a camera-driven photoplethysmogram sensor that detects
heartbeats when participants place their finger over the camera (for details see (38,40)). This
method is comparable to lab-based methods (37). Participants completed a demographics
questionnaire (age, gender, height and weight) and other questionnaire measures, the results of
which will be reported elsewhere.

PAT Task: Design and Procedure

On each trial participants listen to a series of tones (duration = 156ms) at the same
frequency as their heart rate. They are asked to rotate a virtual dial on their smartphone to
advance or retard the tones until they perceive the tones to be synchronous with their heartbeats
(see Supplemental Information [SI] for full instructions). This is equivalent to allowing the
participant to continuously adjust the delay between their R-wave and the tone in order to
indicate the delay perceived as synchronous, but does not require the R-wave to be measured
(just the pulse at the finger using the smartphone camera; (38,39)). Interoceptive participants
(those able to perceive their heartbeats) are identified by consistency in the delay selected across
trials, as those who cannot perceive their heartbeat will choose delays at random.
Importantly, heartbeats were not used to trigger each tone (individuals may perceive their
heartbeat before it is recorded at the finger due to the delay between heart contraction and arrival
of the pulse at the finger). Instead, following a baseline pre-task recording period of 120 seconds,
 7

the heart rate was continuously estimated from beat-to-beat intervals every 3 seconds and this
estimate was used to predict when the next heartbeat would occur.
The tones were presented at the heartrate frequency but out of phase with heartbeats, so
that a single tone was played between each heartbeat and the next, and the delay between each
heartbeat and the tone was adjusted by the participant. The starting delay (or phase asynchrony
between heartbeats and tones) was randomly determined at the beginning of each trial.
Adjustments to the dial resulted in a change of delay that was customized for each participant
based on their heart rate - each turn of the dial advanced or delayed the tones by a set proportion
of the time between heartbeats (so that those with slower heart rates would not have to work
harder to make the tones synchronous with their perceived heartbeats). Participants were free to
make as many adjustments to the dial as they wished, and pressed a button to confirm when they
perceived the tones as synchronous with their heartbeat.
Simulated data were used to assess the impact of the number of trials on expected
consistency of selected delays. 20 trials constituted a good compromise between task duration
and the expected consistency profile, such that the risk of false negatives was minimised (see SI
for details). These 20 trials were preceded by 2 practice trials. After selecting their preferred
delay on each trial, participants rated their confidence in having successfully completed the trial
on a 10-point Visual-Analogue Scale (1=“Not at all confident”, 10=“Extremely confident”). In
addition, every 5 trials, participants indicated the location from which they felt their heartbeat on
the previous trial using a body map (Figure 1). In total, the task took approximately 15-20
minutes.

PAT Task: Scoring

Note that, as in the 6AFC and MCS tasks, there is not a ‘correct’ response in the PAT
task. Participants may perceive their heartbeat at a range of delays, and this may depend on the
location from which they perceive their heartbeat. The measure used to discriminate between
interoceptive and non-interoceptive participants is therefore the consistency of the selected phase
relationship or ‘delay’ across trials.
Given the periodic nature of heartbeats, analysis of the consistency of selected delays
should not be conducted using the standard deviation of selected delays. If, for example, a
participant completes two trials at 60bpm (interbeat interval 1000ms) and selects delays of -
 8

100ms and 900ms, a linear analysis of the mean and standard deviation would yield an average
of 500ms and high dispersion, with the participant classified as inconsistent (non-interoceptive).
However, -100ms and 900ms are the same value if the period is 1000ms (100ms before
heartbeats), indicating high consistency. This problem is solved if the selected delay (𝑑) is
expressed as a periodic function of the period (𝑝), so that a delay of 0ms is equivalent to a delay
which is equal to the period (𝑑 = 𝑝). This is implemented by expressing the delay as an arc of
length 2𝜋𝑑/𝑝 on a unitary circumference. Mapping each participant to a standardised
circumference allows individual variabilities in heart rate to be accounted for, as the space that
delays are mapped onto is not influenced by differences in period length. This formalisation
allows the similarity of angles to be used to define cross-trial consistency. In the previous
example, delays of -100ms and 900ms at 60bpm would result in the same point on the
circumference. In the PAT task, each movement of the dial corresponds to 𝛼 = 𝑑 + 2𝜋𝛼/𝑝,
and a measure of angular similarity can therefore be used to estimate the consistency of selected
!"#
delays. By expressing angles as complex numbers of modulus = 1 and argument = $
,

consistency can be computed as the modulus of the sum of the angles divided by the number of
! #
% )!"
considered angles, 𝑐𝑜𝑛𝑠𝑖𝑠𝑡𝑒𝑛𝑐𝑦(𝑑, 𝑝) = & 𝑚𝑜𝑑(∑&'(% 𝑒 $ #
). If angles are close to one

another, the modulus of their average will be close to 1; if they are randomly positioned the
modulus will approach 0. This consistency formula is equivalent to the Phase Locking Value
(PLV; (41)), a technique used to measure similarity between phases of periodic signals.
To account for the possibility that participants may perceive their heartbeat at different
bodily locations throughout the task, affecting which delay is judged to be synchronous, a variant
of PLV was used. The angles derived from each trial were analysed independently, and the
distance between each angle and its five nearest neighbours calculated. Consequently, a
participant declaring two clusters of delays, with delays close to each other but clusters far apart,
will still have high consistency. For each selected delay, the PLV from the 5 nearest neighbours
was computed, and the root mean square of those constituted the consistency measure.
 9

Results

Data were analyzed and plotted using the tydiverse , ggplot2 (42), RHRV (43), rstatix,
ggpubr and purrr packages for R. To test the null hypothesis that participants’ responses reflect
random behaviour, 5000 participants responding at random were simulated and the resultant
distribution compared to the distribution of real responses. Whilst the simulated data were
normally distributed (Figure 2A, red line), the distribution of real consistency values was
binomial (Figure 2A, black line) and only partially overlapped the simulated distribution. A
Wilcoxon-signed rank test revealed the simulated and real response distributions differed
significantly (Z=10.40, p < 0.0001, r=0.15), confirming that, as a group, real participants were
not responding randomly.
In order to classify interoceptive and non-interoceptive participants, a threshold value
based on the simulated distribution was defined, such that participants with consistency values
higher than 95% of the random distribution would be labelled as interoceptive at an alpha of
p=0.05. Using this threshold (blue dotted line, Figure 2A), 40/124 participants were interoceptive
(Figure 2B), consistent with previous findings that approximately ⅓ of healthy participants are
interoceptive (34). The suitability of this threshold was confirmed with an additional simulation
of 100,000 responders, which produced an identical threshold value.
The modal location of the four bodily location reports by each participant was used to
derive a summary statistic for each participant. The means and standard deviations of
consistency values in the 3 most commonly selected bodily locations did not differ (chest: N=
64, M=0.88, SD=0.035; neck: N=18, M=0.88, SD=0.034; fingers: N=17, M=0.89, SD=0.049).
Additionally, no difference was found in consistency values between participants declaring one
bodily location and those declaring multiple locations (t119.96=0.21704, p=0.8285, d=0.038).
Furthermore, the mode of the most selected location was used to check whether the average
delay chosen by interoceptive participants corresponded to previously reported delays between
tones and heartbeats. Mean delays were calculated using selected delays and heartbeat periods to
obtain a complex number signaling the mean delay (calculating the modulus as period/2π). The
mean delay for the most selected location by interoceptive participants (chest) was 97ms (median
66ms, SD 23ms), which, when added to the R-wave to finger pulse interval of 200ms at seated
 10

rest (44), fits with previous findings from the MCS that tones 200-300ms after the R-wave are
perceived as synchronous (31,33).
To examine the impact of participant response profiles on consistency, consistency scores
were correlated with the mean time spent on trials and the number of dial movements. Neither
correlation was significant in the total sample or in the interoceptive subgroup (all rs<.201, all
ps>0.155).
Physiological variables such as heart rate, Heart Rate Variability (HRV; SDNN, RMSSD
and pNN50; calculated from baseline period before testing) and BMI are reported in Table S1.
None of these variables differed significantly between interoceptive and non-interoceptive
participants after correction for multiple comparisons (see Table S1). No difference was
observed between males and females (t116.21=0.0047, p=0.9963, d=0.0008). There was also no
difference between confidence reported by interoceptive (M =5.79, SD=1.54) and non-
interoceptive (M=5.61, SD=1.87) participants (t91.536=0.562, p=0.575, d=0.101).
 11

Discussion

The current study outlined the development of a novel measure of interoceptive ability -
the PAT task. Participants’ consistency scores differed significantly from random responses,
with 40/124 participants classified as interoceptive. Consistency scores did not differ according
to the bodily location from which the heartbeat was perceived, nor the number of locations.
Neither physiological characteristics of the participants, nor their response profiles, related to
interoceptive ability.
No relationship between performance on the PAT task and confidence was observed, but
it is worth noting that the confidence question was poorly specified; future versions of the PAT
task will employ separate questions concerning confidence in heartbeat perception and
confidence in task performance. Notably, previous studies of the relationship between self-
reported and objective measures of interoception produced mixed results (see (45)), and a lack of
a relationship is consistent with proposals that interoception is a multidimensional construct,
with dissociations between subjective and objective aspects (45,46).
Currently, the PAT task classifies participants as interoceptive or not; individual
differences in consistency scores are not treated as meaningful except as above or below the
threshold. The validity of a binary classification of participants as interoceptive or not should be
tested by further research, but is motivated by the suggestion that individual differences in
interoception should be considered not on task performance (where variance is likely to reflect
attentiveness or motivation) but as the percentage of time, or variety of circumstances, in which
one can perceive interoceptive signals.
In conclusion, the PAT task avoids problems with existing measures of interoceptive
ability. It is unaffected by knowledge of one’s own heart rate and makes use of, rather than falls
foul of, individual differences in the delay with which heartbeats are perceived. The task is also
far more scalable and accessible for clinical populations than currently available measures. We
therefore suggest that the PAT task represents a significant advance in methodology available for
clinical research on interoception.
 12

Acknowledgements

We are grateful to Leonardo Bartoloni for helpful discussion about analyses.

 13

Disclosures

DP, DM, and GB have a financial interest in BioBeats Ltd. MC and SP are employees of
BioBeats. CR, CTK, CC and JM have no conflicts of interest to declare.
 14

References

1. Craig AD (2002): How do you feel? Interoception: the sense of the physiological condition of the body.

Nat Rev Neurosci 3: 655–666.

2. Füstös J, Gramann K, Herbert BM, Pollatos O (2013): On the embodiment of emotion regulation:

interoceptive awareness facilitates reappraisal. Soc Cogn Affect Neurosci 8: 911–917.

3. Kever A, Pollatos O, Vermeulen N, Grynberg D (2015): Interoceptive sensitivity facilitates both

antecedent- and response-focused emotion regulation strategies. Personality and Individual

Differences 87: 20–23.

4. Stevens S, Gerlach AL, Cludius B, Silkens A, Craske MG, Hermann C (2011): Heartbeat perception in

social anxiety before and during speech anticipation. Behav Res Ther 49: 138–143.

5. Verdejo-Garcia A, Clark L, Dunn BD (2012): The role of interoception in addiction: A critical review.

Neurosci Biobehav Rev 36: 1857–1869.

6. Schmidt AF, Eulenbruch T, Langer C, Banger M (2013): Interoceptive awareness, tension reduction

expectancies and self-reported drinking behavior. Alcohol 48: 472–477.

7. Paulus MP, Tapert SF, Schulteis G (2009): The role of interoception and alliesthesia in addiction.

Pharmacol Biochem Behav 94: 1–7.

8. Paulus MP, Stewart JL (2014): Interoception and drug addiction. Neuropharmacology 76: 342–350.

9. Paulus MP, Stein MB (2010): Interoception in anxiety and depression. Brain Struct Funct 214: 451–

463.

10. Paulus MP (2007): Neural basis of reward and craving-a homeostatic point of view. Dialogues Clin

Neurosci 9: 379.

11. Werner NS, Jung K, Duschek S, Schandry R (2009): Enhanced cardiac perception is associated with

benefits in decision-making. Psychophysiology 46: 1123–1129.

12. Dunn BD, Galton HC, Morgan R, Evans D, Oliver C, Meyer M, Rhodri Cusack, Andrew D

Lawrence, Tim Dalgleish (2010): Listening to your heart. How interoception shapes emotion
 15

experience and intuitive decision making. Psychol Sci 21: 1835–1844.

13. Shah P, Catmur C, Bird G (2017): From heart to mind: Linking interoception, emotion, and theory of

mind. Cortex 93: 220–223.

14. Quattrocki E, Friston K (2014): Autism, oxytocin and interoception. Neurosci Biobehav Rev 47: 410–

430.

15. Pollatos O, Füstös J, Critchley HD (2012): On the generalised embodiment of pain: how interoceptive

sensitivity modulates cutaneous pain perception. Pain 153: 1680–1686.

16. Herbert BM, Herbert C, Pollatos O, Weimer K, Enck P, Sauer H, Zipfel S (2012): Effects of short-

term food deprivation on interoceptive awareness, feelings and autonomic cardiac activity. Biol

Psychol 89: 71–79.

17. Khalsa SS, Adolphs R, Cameron OG, Critchley HD, Davenport PW, Feinstein JS, et al. (2018):

Interoception and Mental Health: A Roadmap. Biol Psychiatry Cogn Neurosci Neuroimaging 3:

501–513.

18. Naqvi NH, Bechara A (2010): The insula and drug addiction: an interoceptive view of pleasure, urges,

and decision-making. Brain Struct Funct 214: 435–450.

19. Schaefer M, Egloff B, Witthöft M (2012): Is interoceptive awareness really altered in somatoform

disorders? Testing competing theories with two paradigms of heartbeat perception. J Abnorm

Psychol 121: 719–724.

20. Ardizzi M, Ambrosecchia M, Buratta L, Ferri F, Peciccia M, Donnari S, et al. (2016): Interoception

and Positive Symptoms in Schizophrenia. Front Hum Neurosci 10: 379.

21. Mussgay L, Klinkenberg N, Rüddel H (1999): Heart beat perception in patients with depressive,

somatoform, and personality disorders. J Psychophysiol 13: 27.

22. Barrett LF, Quigley KS, Hamilton P (2016): An active inference theory of allostasis and interoception

in depression. Philos Trans R Soc Lond B Biol Sci 371: 20160011.

23. Barrett LF, Simmons WK (2015): Interoceptive predictions in the brain. Nat Rev Neurosci 16: 419–

429.
 16

24. Murphy J, Brewer R, Catmur C, Bird G (2017): Interoception and psychopathology: A developmental

neuroscience perspective. Dev Cogn Neurosci 23: 45–56.

25. Caspi A, Houts RM, Belsky DW, Goldman-Mellor SJ, Harrington H, Israel S, et al. (2014): The p

Factor: One General Psychopathology Factor in the Structure of Psychiatric Disorders? Clin Psychol

Sci 2: 119–137.

26. Dale A, Anderson D (1978): Information variables in voluntary control and classical conditioning of

heart rate: field dependence and heart-rate perception. Percept Mot Skills 47: 79–85.

27. Schandry R (1981): Heart Beat Perception and Emotional Experience. Psychophysiology, pp 483–488.

28. Whitehead WE, Drescher VM, Heiman P, Blackwell B (1977): Relation of heart rate control to

heartbeat perception. Biofeedback Self Regul 2: 371–392.

29. Katkin ES, Reed SD, Deroo C (1983): A methodological analysis of 3 techniques for the assessment

of individual-differences in heartbeat detection. Psychophysiology 20: 452–452.

30. Windmann S, Schonecke OW, Fröhlig G, Maldener G (1999): Dissociating beliefs about heart rates

and actual heart rates in patients with cardiac pacemakers. Psychophysiology 36: 339–342.

31. Brener J, Kluvitse C (1988): Heartbeat detection: judgments of the simultaneity of external stimuli

and heartbeats. Psychophysiology 25: 554–561.

32. Clemens WJ (1984): Temporal arrangement of signals in heartbeat discrimination procedures.

Psychophysiology 21: 187–190.

33. Brener J, Liu X, Ring C (1993): A method of constant stimuli for examining heartbeat detection:

Comparison with the Brener-Kluvitse and Whitehead methods. Psychophysiology 30: 657–665.

34. Brener J, Ring C (2016): Towards a psychophysics of interoceptive processes: the measurement of

heartbeat detection. Philos Trans R Soc Lond B Biol Sci 371: 20160015.

35. Ring C, Brener J (1992): The temporal locations of heartbeat sensations. Psychophysiology 29: 535–

545.

36. Yates AJ, Jones KE, Marie GV, Hogben JH (1985): Detection of the heartbeat and events in the

cardiac cycle. Psychophysiology 22: 561–567.

 17

37. Murphy J, Brewer R, Coll M-P, Plans D, Hall M, Shiu SS, et al. (2019): I feel it in my finger:

Measurement device affects cardiac interoceptive accuracy. Biol Psychol 148: 107765.

38. Morelli D, Bartoloni L, Colombo M, Plans D, Clifton DA (2018): Profiling the propagation of error

from PPG to HRV features in a wearable physiological-monitoring device. Healthc Technol Lett 5:

59–64.

39. Bacciu D, Colombo M, Morelli D, Plans D (2018): Randomized neural networks for preference

learning with physiological data. Neurocomputing 298: 9-20.

40. Cropley M, Plans D, Morelli D, Sütterlin S, Inceoglu I, Thomas G, Chu C (2017): The Association

between Work-Related Rumination and Heart Rate Variability: A Field Study. Front Hum Neurosci

11: 27.

41. Lachaux JP, Rodriguez E, Martinerie J, Varela FJ (1999): Measuring phase synchrony in brain

signals. Hum Brain Mapp 8: 194–208.

42. Wickham H (2016). ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York. ISBN

978-3-319-24277-4, https://ggplot2.tidyverse.org.

43. Rodríguez-Liñares L, Vila X, Méndez AJ, Lado M, Olivieri D (2008): RHRV: An R-based software

package for heart rate variability analysis of ECG recordings. 3rd Iberian Conference in Systems and

Information Technologies (CISTI 2008) 565–574.

44. Obata Y, Ong QJ, Magruder JT, Grichkevitch H, Berkowitz DE, Nyhan D, et al. (2017): Noninvasive

Assessment of the Effect of Position and Exercise on Pulse Arrival to Peripheral Vascular Beds in

Healthy Volunteers. Front Physiol 8: 47.

45. Murphy J, Catmur C, Bird G (2019): Classifying individual differences in interoception: Implications

for the measurement of interoceptive awareness. Psychon Bull Rev 26: 1467–1471.

46. Garfinkel SN, Seth AK, Barrett AB, Suzuki K, Critchley HD (2015): Knowing your own heart:

distinguishing interoceptive accuracy from interoceptive awareness. Biol Psychol 104: 65–74.
 18

Figure Legends

Figure 1. Screenshots of the PAT task mobile application. A) Screenshot of one of the 20 trials
and B) body map screen.

Figure 2. A) Probability density function of data from simulated participants responding at

random (red line) and real participants’ data (black line). Blue line = 95% threshold value based
on simulated dataset, beyond which real participants are judged to be interoceptive; B) Mean
consistency values for interoceptive and non-interoceptive participants based on the 95%
threshold value; Solid line= median; Black dot= mean; Whiskers: upper whisker = min(max(x),
Q_1-1.5*IQR).
 19

Figure 1
 20

Figure 2
 21

Supplemental Information

Transcription of task instructions

For a full visualisation of the task, please refer to the task recording uploaded as part of the
submission. Please note that participants were free to move back and forth between the different
instruction screens.

Screen 1: “How well can you match a sound with your heartbeat? Let’s find out!”
Screen 2: “First, find a quiet place where you can sit comfortably upright with your earpho nes
on for around 10 minutes.”
Screen 3: “You will be asked to place your finger on the phone camera (on the back) so that the
app can read your heartbeat. Once your finger is in position, you will hear a series of sounds.
Each sound actually represents one of your own heartbeats!”
Screen 4: “It might seem like there is a delay between the sounds and the heartbeats you feel.
Play the video below to hear an example! [VIDEO 1]”
Screen 5: “In order to rectify the delay, you will be asked to move a dial until the sounds are in
sync with your heartbeats. Play the video below to hear an example! [VIDEO 2]”
Screen 6: “If you move the dial to the right, the delay between the heartbeat and the sound will
get longer; if you move it to the left, the delay will get shorter.”
Screen 7: “Want to know how this might look like? Press “continue” to watch a short tutorial.”
Screen 8: [VIDEO 3]
Screen 9: “After you have matched the sound with your heartbeat, you will be asked how sure
you are about the answer you gave. Press “confirm” then “continue” to start the following trial.
In this task, there will be 20 trials in total. [EXAMPLE OF THE CONFIDENCE SCALE]”
Screen 10: “You can feel your heartbeat in different places in your body, such as your chest or
yours fingers. You will be asked to indicate where you felt your heartbeat on a body map (like
the one below) once every 5 trials. You can choose any of the highlighted body parts or you can
select “nowhere” if you haven’t felt your heartbeat in any particular place. [EXAMPLE OF THE
BODY MAP]”
Screen 11: “For the duration of this task, please do not actively try to feel your pulse with your
hand; we are only interested in what you feel! When you are ready to start, please sit comfortably
upright with your earphones on and press “continue”.”
Screen 12: [HEART RATE BASELINE READING]
Screen 13: “You will now get a chance to do two practice trials. Focus on feeling your heartbeat
and try to match the sounds to your own heartbeat.”

The practice trials screens have the same instructions reported in Figure 1A for the main trials.
 22

Table S1. T-tests between interoceptive and non-interoceptive participants - time,

engagement, HR and HRV, BMI and confidence.

Variable T (df) p d

Time - 1.937 (105.94) 0.06 0.328

Movements - 1.589 (120.1) 0.115 0.25

Heart rate 2.351 (56.626) 0.0223 0.525

SDNN 0.313 (71.956) 0.755 0.062

RMSSD 0.939 (70.369) 0.3511 0.187

pNN50 1.058 (72.88) 0.2934 0.208

BMI 2.016 (44.326) 0.050 0.511

Movements: number of dial turns; time: mean time spent on each trial; heart rate
variability components: Standard Deviation of NN intervals (SDNN), Root Mean Square of
Standard Deviations (RMSSD), proportion of NN50 (the number of times successive heartbeat
intervals that exceed 50ms) divided by the total number of NN (R-R) intervals (pNN50). T(df); t
values and degrees of freedom. p; uncorrected p values, note none of these comparisons were
significant after multiple comparisons correction. d; Cohen’s d value.
 23

Figure S1. Simulation of consistency depending on trial numbers

Figure S1 (black line) shows the expected consistency of random delays as a function of the
number of trials (red lines correspond to the mean plus/minus 1 SD). The consistency measure
has some interesting properties. Consistency grows with the number of trials. This is expected
since the overall consistency is the average of the similarities of each delay considering only the
5 closest neighbours. As the number of trials grows, there will be more chances for each delay to
have similar delays close to it, producing high similarity, and consequently a high consistency
score. Alternatively, if the number of trials is low, to calculate the similarity of each delay it
becomes necessary to use most of the other delays that will be randomly spread across the circle,
therefore on average producing a lower consistency. Also, consistency is low for a small number
of trials. The other measures of consistency considered have the opposite trend, with high values
for a small number of trials and low consistency for a high number of trials. This makes this
measure a good choice for a small number of trials, such as in our case. If we designed an
experiment with a high number of trials, it would be best to use a different consistency metric,
since it would be difficult to distinguish between truly interoceptive participants and random
answers.