Aberrant Cardiac Interoception in Psychosis Authors: Jayson Jeganathan, Megan E J Campbell, Nicolas Legrand, Micah Allen, Michael Breakspear Published: 2024 ABSTRACT AbstractBackground and HypothesisThere is mounting evidence that cardiac interoception, the perception of one’s heartbeat, is central to affective experiences. It has been proposed that symptoms of psychosis could arise from interoceptive dysfunction. Here we hypothesized that people with psychotic disorders would have a specific impairment in cardiac interoception, over and above broader perceptual deficits.Study Design43 adults with a history of psychosis (31 schizophrenia, 12 schizoaffective disorder) and 41 matched control participants completed a heart rate discrimination task. Participants responded to whether they perceived a sequence of auditory tones to be faster or slower than their heart rate. By trialing a range of auditory tone rates, we estimated a threshold for each participant, the difference between perceived heart rate and actual heart rate. To test whether differences were specific to interoception, participants completed an exteroceptive control condition, testing their discrimination of the rate of 2 sets of audible sounds instead of heart rate.Study ResultsParticipants with a history of psychosis had greater absolute differences between perceived and actual heart rate, indicating over- or under-estimation of heart rate compared to healthy controls. This difference was specific to the interoceptive condition, and not explained by group differences in exteroceptive perception.ConclusionsPsychotic disorders are associated with misestimation of heart rate. Further research may elucidate whether interoceptive abnormalities contribute to specific symptoms such as somatic delusions or affective features, and whether interoception could be a treatment target in psychotic disorders.Keywords: schizophrenia, psychosis, interoception, cardiac interoception, interoceptive Introduction Interoception is the perception of internal bodily states such as heart rate, gastric fullness, and respiration. There is substantial empirical evidence that cardiac interoception plays a role in subjective emotions.1,2 For instance, when participants are provided with false feedback that their heart rate is faster than their recorded heart rate, they tend to rate faces as more emotionally intense.3 Different emotions are associated with topographically distinct bodily sensations.4 Building on such findings, several theoretical formulations have argued that subjective feeling states are meta-phenomena arising from the processing of interoceptive sensations.5,6 Computational models have added further support for the close linkage of affective states and cardiac predictions.7,8 Many have suggested that interoceptive dysfunction could contribute to the expression of affective disturbances in mental illnesses.9 For example, there is a generally reduced capacity to accurately perceived bodily signals in anorexia nervosa.10 It has also been suggested that depression could arise from erroneous interpretation of interoceptive signals,11 learned helplessness about control over interosensations,12 or altered signaling in interoceptive circuitry.13 A review of empirical studies found converging evidence for increased cardiac interoceptive sensitivity in anxiety-related phenotypes.14 However, a recent meta-analysis did not find interoceptive accuracy to be different in depression or anxiety.15 Interoceptive abnormalities in psychotic disorders are of growing interest, with recent proposals that unexplained body sensations could lead to anxiety, disturbed body ownership, delusions of agency, somatic delusions, and paranoid delusions.6,16 Since bodily feedback influences subjective feeling states, we recently hypothesized that imprecise interoception might contribute to negative symptoms such as affective blunting and anhedonia.17 Uncertainty about the cause of bodily sensations could also contribute to a poor sense of self-efficacy.12 Cardiac interoception in schizophrenia has been previously investigated with the heartbeat counting task.18 In this task, participants are asked to silently count their heartbeats during intervals of different lengths, with accuracy measured as the concordance between reported and actual heartbeats. These 4 studies, with 23, 42, 18, and 19 participants with schizophrenia, respectively, found lower heartbeat counting accuracy in schizophrenia than healthy controls.19–22 One study found accuracy to be lower in schizophrenia than other mental illnesses.22 The presence of potential confounders, concerns about task validity, or smaller sample size mean that the evidence remains inconclusive. The heartbeat counting task cannot distinguish between inaccuracy in interoception from broader inaccuracies due to general perceptual dysfunction, working memory impairment, or temporal estimation biases.23 All of these factors are pertinent in psychotic illnesses. Second, these studies did not report whether the clinical cohort overestimated or underestimated their heart rate. Finally, the task cannot distinguish between participants with systematic misestimation of heart rate from those with imprecise estimates of heart rate; ie, it cannot dissociate interoceptive bias from accuracy. These limitations arise because the task does not sit within a clear psychometric framework that enables estimation of accuracy and precision. Here we used a recently validated interoceptive task24 which includes an exteroceptive control condition to exclude general perceptual dysfunction, and employs psychometric techniques that enable separable estimates of perceived heart rate and precision. We studied 84 participants (43 with psychosis and 41 controls) and hypothesized that those with a psychotic disorder systematically misperceive their heart rate as being either faster or slower than ground truth, while their perception of external stimuli would be relatively preserved. Methods Recruitment Clinical participants were recruited from outpatient mental health clinics in Hunter New England Local Health District in NSW, Australia. Case managers referred patients who fulfilled the DSM-5 criteria for broadly defined psychotic disorder (schizophrenia and schizoaffective disorder). These included some persons with ongoing positive symptoms but still able to fulfill task requirements. Referred patients were further screened for eligibility to participate. Control participants were recruited from the local community via advertisements on billboards, websites, and social media. Control participants disavowed a history of a mental health condition as defined by the Research Diagnostic Criteria (“never mentally ill”). Control participants additionally denied being previously hospitalized with any mental health condition. They also had no first-degree relative with severe mental illnesses such as schizophrenia or bipolar disorder. Exclusion criteria for both groups were a current substance use disorder, a neurological disorder, neurosurgery, or traumatic brain injury. Ethnicity was self-reported by the participants. The study was approved by the Hunter New England Human Research Ethics Committee (2020/ETH00531), and participants gave their informed consent to participate. A total of 43 adults aged 18–50 years with a history of diagnosed psychotic illness (clinical cohort) and 41 healthy controls matched for age, sex, and education were recruited for this study. Clinical interview Cognitive testing was completed with the Weschler Abbreviated Scale of Intelligence (WASI-II). Vocabulary and matrix reasoning tasks were used to score full-scale IQ. Clinical participants additionally completed a semi-structured clinical interview. Each participant’s clinical file was reviewed by one of the authors (J.J., a senior psychiatry registrar), including discharge summaries from hospitalizations, letters by outpatient psychiatrists, and recent progress notes by outpatient mental health clinicians. Information from this file review and the interview were used to verify the clinical diagnosis, and to complete the Positive and Negative Symptom Scale (PANSS), Hamilton Depression Rating Scale (HAM-D), Clinical Global Impression (CGI-S, a measure of symptoms severity), and the Social and Occupational Functioning Assessment Scale (SOFAS, a measure of functioning). Where the diagnosis remained in doubt, consensus was achieved by discussion between coauthors J.J. and M.B. (board registered psychiatrist). Heart Rate Discrimination Task The heart rate discrimination task has been fully described and validated elsewhere,24 but we provide a brief overview here. The task included interoceptive trials randomly interleaved with exteroceptive trials (figure 1a). Participants completed 80 trials over approximately 18 min. During each interoceptive trial, participants attended to their heart rate for 5 s while their heart rate was recorded with a pulse oximeter. Then they heard feedback auditory tones that were either faster or slower than their actual heart rate in the preceding 5 s. Participants were asked to indicate with a mouse click whether the tones were “Faster” or “Slower” than their actual heart rate. To estimate metacognitive ability, participants were then asked to provide a subjective confidence rating in their decision from 0 (uncertain) to 100 (certain). The rate difference between the actual recorded heart rate and the auditory feedback was generated by the established adaptive Bayesian “Psi” staircase method. For example, if the participant’s actual heart rate was 70 BPM, and the staircase generated a difference value of −15 BPM, the feedback tones were played at 55 BPM. In this scenario, the correct answer is for the participant to respond that the feedback tone was “slower” than their heart rate. The staircase procedure allows estimation of a threshold value (figure 1b). The threshold is the rate difference between feedback tones and actual heart rate at which the participant perceives them to have equal rate. Ten percent of trials were catch trials, where the feedback tones were presented at fixed intervals above and below the actual recorded heart rate. This ensured that once the staircase had converged, participant still received occasional trials that were faster or slower than their actual heart rate. Heart rate discrimination task. (a) Trial design. Participants were presented with interoceptive trials or exteroceptive in random, counterbalanced order. Participants attended to their heart rate or auditory tones. Then, participants decided whether a set of auditory tones was faster or slower than their heart rate (or reference tones). Finally, participants rated their subjective decision confidence. (b) Staircases for each condition from an example participant. Intensity values represent the difference between auditory tones and actual heart rate (or reference tones in the exteroceptive condition). Trials classified by participants as faster or slower are indicated with circles and squares, respectively. The shaded area indicates the 95% CI of the threshold posterior distribution. (c). The psychometric function estimated by the adaptive staircase, parameterized by the threshold value (horizontal lines) and slope, a measure of uncertainty around the threshold. The exteroceptive condition was used to control for general temporal estimation biases. In this condition, participants discriminated whether a series of feedback tones were faster or slower than an initial reference sequence of auditory tones. Tones for both the exteroceptive and interoceptive task were presented at equal volume. Visual stimuli were presented on a computer monitor of width 53 cm, resolution 1920 × 1080, placed at eye level, 50 cm from the participant’s eye, and auditory stimuli played over headphones at volume set comfortably for each participant. Photoplethysmography (PPG) was recorded at 100 Hz with a Nonin 8600FO pulse oximeter. Participants were asked to move the oximeter to a different finger until they could not feel pulsatile feedback. PPG signal analysis, including peak detection and artifact removal, was completed with Systole v0.1.3,25 as described previously.24 Statistical Analysis For each participant, we measured overall response quality using several metrics, to assess for group differences in random responses. These metrics measured whether the participant was more likely to make correct responses when the difference between actual heart rate and feedback tones was greater, whether correct responses were associated with faster button-press reaction time, and whether correct responses were associated with higher confidence ratings (see supplementary results 1). The heart rate discrimination task aims to estimate each participant’s psychometric function (figure 1b, right panel). The function describes how the difference between feedback tone rate and actual heart rate (or reference tone’s rate) predicts the probability that the participant responded “Faster.” The function is parameterized by threshold and slope. We primarily examined threshold, the rate difference (beats/min) between the feedback auditory tone and the preceding tone (or their actual heart rate) at which the participant is equally likely to judge the feedback tone to be faster or slower than the reference. Positive and negative thresholds, respectively, indicate systematic over- and under-estimation of heartrate. For instance, if a participant’s actual heart rate was 70 beats/min, and interoceptive threshold was −15, this corresponds to a perceived heart rate of 55 beats/min. We also tested for group differences in slope and metacognition. Greater slope indicates more uncertainty (less precision) around the threshold value. In addition, metacognitive measures were obtained from confidence ratings. Briefly, the measure meta-dʹ operationalizes metacognitive insight in analogy to discriminability (dʹ) in signal detection theory. Meta-dʹ captures the correspondence between confidence ratings and objective accuracy. As meta-dʹ is influenced by dʹ, we used the metacognitive efficiency (meta-dʹ/dʹ). Following the original experimental paradigm,24 for each participant, trials with outlier heart rate or heart rate variability values were excluded based on the absolute deviation around the median rule. Threshold and slope values were estimated for each subject using post hoc Bayesian modeling approach.24 Subjects whose interoceptive or exteroceptive thresholds were more than 3 standard deviations from the mean were excluded. Using this method, 3 outlier participants, all from the clinical cohort, were excluded. All 3 had highly deviant exteroceptive thresholds indicating either lack of understanding of the task, or gross perceptual dysfunction (supplementary table 1). The Jacques-Bera test was used to test for normality of measures. As all outcome measures were found to be normally distributed, the paired t-test was used to confirm prior work showing differences in threshold across conditions (interoceptive vs exteroceptive). We used the 2-sample Kolmogorov-Smirnov test to assess whether the clinical and control groups’ interoceptive thresholds were drawn from the same distribution. To investigate the main hypothesis that clinical cohort systematically over- or under-estimate their heart rate, we tested for group differences in the absolute value of the threshold. A linear mixed model with random intercept was fitted to predict absolute threshold, with interoception-specific threshold differences in the clinical cohort identified by the interaction between group (clinical or control cohort) and condition (interoception or exteroception). A 2-way ANOVA without random intercepts was also fitted to investigate whether results depended on random subject-specific intercepts. To test the specificity of the main analysis, models were re-fitted with additional, potentially confounding covariates: heart rate, age, sex, IQ, and smoking status. Sensitivity to staircase convergence was tested by removing participants with inadequate staircase convergence. To clarify differences between diagnostic groups, the linear model was repeated with clinical subgroup (schizophrenia vs schizoaffective disorder) as the grouping variable. Exploratory t-tests were used to assess group differences in continuous task outcomes: interoceptive slope and metacognitive efficiency. T-tests were also used for group differences in continuous nonclinical variables such as age, IQ, and heartrate. Chi-squared tests were used to assess group difference in categorical variables such as sex and smoking status. In further exploratory analyses, Pearson’s R was used to test for correlations between task outcomes (interoceptive threshold, slope, and metacognitive efficiency) and clinical scores (PANSS subscales, HAM-D, CGI-S, and SOFAS). Statistical significance was set at alpha = 0.05. Results After excluding outliers, data were analyzed from 40 clinical participants (29 schizophrenia, 11 schizoaffective disorder) and 41 healthy controls. The clinical cohort was matched to the healthy controls in the distribution of age, years of education, IQ, and sex ratio. Smoking rates were significantly higher in the clinical cohort (table 1). In the healthy control group, 39 individuals identified as Caucasian and 2 as Asian. Within the clinical cohort, the composition was 35 Caucasian, 1 Asian, 1 African, and 3 Aboriginal or Torres Strait Islander. In the clinical cohort, means and standard deviations (SD) for clinical measures were as follows: PANSS positive symptom score was 11.7 (SD = 3.9), PANSS negative symptom score was 17.5 (SD = 5.8), PANSS general psychopathology score was 28.3 (SD = 5.8), HAM-D score was 3.5 (SD = 3.4) social and occupational functioning score (SOFAS) was 60.3 (SD = 16.0) indicating “some difficulty in functioning,” and Clinical Global Impression was 3.8 (SD = 1.4) where a score of 4 corresponds to “moderately ill” patients. All but 1 clinical participant were currently prescribed antipsychotic medication. Mean antipsychotic dose (chlorpromazine equivalent) was 578.4 mg (SD = 390.1). Table 1.Group Means for Demographic and Cognitive VariablesHealthy ControlsClinical CohortGroup Difference n 4140Age, mean years (SD)39.6 (8.4)37.0 (8.1) t = 1.41, P = .162Sex male, n (%)35 (85.4%)34 (85.0%)χ2 = 0.00, P = .998Education, mean years (SD)12.8 (3.0)12.5 (2.1) t = 0.488, P = .627IQ, mean (SD)109.4 (11.3)106.0 (10.9) t = 1.40; P = .166Smoking, n (%)6 (14.6%)22 (55.0%)χ2 = 12.9, P < .001 Group Means for Demographic and Cognitive Variables During the task, heart rate was significantly greater in the clinical cohort than the control group (t-test, t = 6.41, P < .001) (supplementary table 2). Heart rate was positively correlated with chlorpromazine equivalent antipsychotic dose (Pearson’s R +0.633, P < .001). There were no significant group differences in response quality to indicate random responding in the clinical cohort (supplementary results 1). First, we replicated the previous finding that healthy controls are biased toward underestimating heart rate. Interoceptive threshold was the difference between a participant’s perceived heart rate and actual heart rate, while exteroceptive threshold was the difference between a participant’s perception of the first and second set of auditory tones. Mean interoceptive threshold in controls was −5.84 beats/min. This underestimation bias was specific to interoception. A paired t-test within the control group revealed that threshold was significantly lower in the interoceptive condition than in the exteroceptive condition (meanIntero = −5.84, meanExtero = 0.33, t = −5.11, P < .001). A similar bias was also present in the clinical cohort (meanIntero = −7.40, meanExtero = 0.88, t = −4.70, P < .001). There was no group difference in this bias (supplementary table 2), nor any group by condition interaction (supplementary table 9). We then compared the distribution of interoceptive thresholds across the groups. Thresholds that deviate from zero correspond to systematic over- or under-estimation of heart rate. The distribution of interoceptive thresholds was significantly different between the clinical and control groups (Kolmogorov-Smirnov test D = 0.304, P = .040). The distribution had broader tails in the clinical group, indicating larger threshold deviations from zero (figure 2). To test the hypothesis that clinical cohort had larger misperceptions of their heart rate, we compared the absolute value of threshold across groups. A larger absolute threshold indicates a greater bias and corresponds to either over- or under-estimation of heart rate. A mixed linear model was fitted to predict absolute threshold, with fixed effects for condition (interoception or exteroception), group (clinical or control cohort), a group by condition interaction, and a random intercept for each participant. There was a significant group by condition interaction (P = .037). In the linear model, the interaction term had a coefficient of +3.80 (95% CI 0.23–7.37). That is, compared to the healthy controls, the clinical group had greater absolute interoceptive threshold by 3.80 beats/min (figure 2). There was also a significant main effect of condition, where absolute thresholds were greater in the interoceptive than the exteroceptive condition (see supplementary table 10 for full model results). A simpler test with 2-way ANOVA also supported a significant group by condition interaction effect (interaction, F = 4.09, P = .045, partial η2 = 0.025) (supplementary table 12). Thresholds (a) and absolute thresholds (b) in the clinical cohort and healthy controls. Each point represents a single participant’s threshold for either the interoceptive or exteroceptive condition. Half-violin plots represent the distribution of values in each group. Boxplots represent the quartiles of the distribution. Raw thresholds (a) correspond to the difference between perceived and actual heart rate. Absolute interoceptive thresholds were significantly greater in the clinical cohort, indicating greater degrees of heart rate over- or under-estimation. Exploratory follow-up analyses indicated that these results were robust to various potential confounds. Heart rate, age, sex, IQ, and smoking status were included as covariates to the mixed linear model. The group-by-condition interaction remained significant in this adjusted model (β = 3.80, P = .037) (supplementary table 11 for full model results, supplementary table 13 for ANOVA with the same covariates). We queried whether inadequate convergence of the adaptive staircase in the clinical cohort could account for these results. However, the interaction term remained significant (P = .021) even after removing 7 participants with inadequate staircase convergence (supplementary results 1). Results were similar when those with schizoaffective disorder (n = 11) was excluded from the clinical cohort, and the schizophrenia subcohort alone (n = 29) was compared to the control group (β = 4.29, P = .037). Compared to the schizoaffective disorder cohort (n = 11), the schizophrenia cohort (n = 29) had numerically greater absolute interoceptive thresholds (+2.87 beats/min). However, when these clinical subgroups were compared without the controls, there was no subgroup by condition interaction (β = 1.36, P = .745). We then compared group means for other outcomes: interoceptive slope and interoceptive metacognitive efficiency (supplementary table 2). Greater slope corresponds to imprecision, or more uncertainty around the threshold value. Interoceptive slope was not significantly different between groups (meancontrols = 9.41, meanclinical = 10.79, t-test, t = −1.50, P = .137). Interoceptive metacognitive efficiency also did not significantly differ between groups (meancontrols = 0.882, meanclinical = 0.861, t-test, t = −0.112, P = .911). We did not find any significant associations between absolute interoceptive threshold and PANSS positive symptom score (r = −0.079, P = .627), negative symptom score (r = 0.094, P = .565), or social and occupational functioning (r = −0.048, P = .769). Absolute interoceptive threshold did not significantly correlate with specific PANSS items (delusions, hallucinations, affective blunting, somatic concerns, and anxiety), depression (HAM-D), or antipsychotic dose (supplementary tables 3–5). Within the clinical group, absolute interoceptive threshold was not significantly affected by smoking status or sex (supplementary table 8). There were no significant differences in clinical measures between clinical participants with a positive interoceptive threshold (over-estimators) and those with a negative threshold (under-estimators) (supplementary table 17). Discussion These results demonstrate that people living with a chronic psychotic disorder have a perceptual deficit that is specific to interoception, and not explained by general perceptual disturbance or estimation biases. Unlike previous studies which could not dissociate bias from precision, our protocol explicitly decomposed interoceptive accuracy into 2 measures: bias in the threshold—which measures the difference between perceived and actual heart rate—and precision around the threshold. The clinical cohort were more likely to over- or under-estimate their heart rate, but did not have significant group differences in interoceptive precision. Although we did not observe associations between over- or under-estimation and clinical measures, it is conceivable that systematic misperception of heart rate may be experienced as anxiety or unexplained bodily sensations, especially in the context of relatively intact precision.16 These interoceptive abnormalities in schizophrenia depart from those seen in other mental illnesses.22 It remains unclear whether interoceptive differences are simply more severe in schizophrenia, or qualitatively different. The latter possibility is supported by the finding, compared to healthy controls, those with schizophrenia have opposite cardiac afferent effects on emotion processing.26 Further transdiagnostic studies utilizing the heart rate discrimination task may shed more light on differences across diagnoses. Although we measured and controlled for education, sex, IQ, and exteroceptive ability, there remain several potential factors that could affect group differences. While baseline heart rate was significantly higher in the clinical cohort and showed a trend-level relationship with absolute interoceptive threshold in the clinical cohort, group differences in interoception survived the addition of heart rate to the model. The other potential covariates can be divided into cognitive differences, psychiatric symptoms, and other demographic variables. Schizophrenia is associated with generalized cognitive impairment as well as specific deficits in verbal memory and processing speed. It is of substantial interest whether interoceptive deficits in schizophrenia are caused by cognitive impairment, or arise independently.27 The cognitive assessment used here (WASI-II) included vocabulary and matrix reasoning tests but not specific tests of working memory, attention, or processing speed. Working memory is less likely to be relevant, as it is similarly involved in both the interoceptive and exteroceptive conditions, and would likely impact equally on performance in both conditions. Other aspects of cognition may preferentially impact interoception. In particular, the interoceptive condition was intrinsically more difficult than the exteroceptive condition, requiring attentional shifts from external to internal stimuli, and multisensory processing rather than just auditory processing. Individuals with psychosis possess deficits in these areas,28 which may be more apparent when attending to subtle interoceptive signals than when attending to auditory tones. However, if cognitive differences caused interoceptive differences, we would have expected greater slope (more imprecision) in the clinical cohort rather than systematic deviations in threshold, as well as greater evidence of random responding (supplementary results 1). Disentangling cognitive deficits inherent to schizophrenia from interoceptive task performance is a challenge, and future studies may wish to add self-report measures of interoception,29 as well as specific assessments of attention and multisensory processing. Psychiatric symptoms could also differentially impact interoceptive performance. For example, hallucinations could interfere with attention, and somatic delusions could interfere with heartbeat perception. However, interoceptive differences did not covary with the corresponding PANSS items (supplementary table 4). While cardiovascular diseases and cardiac medications were not recorded, these could in principle influence interoceptive ability. However, heart rate did not significantly covary with interoceptive threshold in the control group (supplementary table 7). Body mass index and body composition could also impact interoception, particularly via the transmission and detection of pulsatile vascular impulses by participants. BMI is a potential confounder because antipsychotic medications frequently cause weight gain. Finally, due to the high proportion of males in our sample (85.4%), the generalizability of results to females remains unclear. Others have found that females have reduced interoceptive accuracy, but are able to notice bodily sensations more often.30,31 Participants were asked to place the pulse oximeter on a finger where they could not feel pulsatile feedback, but it is nevertheless possible that some participants completed the task using tactile feedback. Electrical measurements such as with an electrocardiogram would avoid this issue. Prior knowledge of one’s heart rate is a relevant factor in all noninvasive cardiac interoception tasks. If the control group had more accurate prior beliefs about their heart rate, this could explain their lower interoceptive threshold bias. Addressing this would require an interventional study where individuals’ heart rates are modulated away from their prior beliefs, eg, by exercise, chemical modulators, or electrical pacemakers. Notably, absolute interoceptive thresholds were not significantly associated with psychosis symptom scores. This could be related to variability in the task outcomes due to unmodelled sources of variance such as medications, cardiac conditions, or body mass index. While the PANSS has adequate inter-rater reliability,32 a more detailed quantification of delusions, hallucinations, and somatic hallucinations may assist in identifying links between specific symptoms and interoception. Alternatively, interoceptive dysfunction could be a stable trait feature, unlike psychotic symptoms that can fluctuate over time. Longitudinal assessments could clarify whether interoceptive deficits covary with symptoms across an individual’s illness trajectory or are a core, enduring effect. Our result raises questions about the biological basis for interoceptive differences in psychotic disorders. Converging evidence from primate fiber tracking and human functional MRI suggests that the insula facilitates the effect of bodily signals on emotion.33,34 In mice, inducing tachycardia using an optogenetic cardiac pacemaker increased anxiety-like behaviors in risky contexts, while optogenetic inhibition of the posterior insula attenuated this effect.35 The biological basis of the distinction between interoceptive bias and precision, which we operationalized as interoceptive threshold and slope, remains unknown. Schizophrenia is characterized by functional and effective connectivity abnormalities in the insular cortex, the region crucial for processing interoceptive signals.36,37 Functional imaging studies of the insula during interoceptive tasks could contribute to our understanding of shifts in interoceptive signaling in schizophrenia. Future research could also clarify whether interoceptive biases are stable across time or vary with psychiatric treatment. Another unresolved issue for future investigation is whether interoceptive dysfunction in psychosis extends beyond cardiac interoception to other modalities, and what role these play in psychiatric phenomenology. Visceral pain often manifests as referred pain on superficial body structures due to the convergence of somatic and visceral afferents in the spinal cord.38 Could dysfunctional interoception in specific visceral structures present with tactile hallucinations at corresponding superficial structures? These questions could be explored using novel tasks like the water load test for gastric interoception39 or the respiratory resistance sensitivity task.40 While there has been burgeoning interest into the science of interoception and its relevance in psychiatric disorders, theoretical proposals need to be empirically supported by well-controlled studies. Here we built upon theoretical proposals implicating interoceptive dysfunction in psychosis, to empirically test this proposal. We found that individuals with psychosis are more likely to over- or under-estimate their heart rate. Further research is needed to understand the biological basis of interoceptive differences in psychosis, and the relationship between interoception and psychiatric symptomatology. Supplementary Material Supplementary material is available at https://academic.oup.com/schizophreniabulletin/. Acknowledgments The authors have declared that there are no conflicts of interest in relation to the subject of this study. Contributor Information Jayson Jeganathan, School of Psychology, College of Engineering, Science and the Environment, University of Newcastle, Newcastle, NSW, Australia; Hunter Medical Research Institute, Newcastle, NSW, Australia. Megan E J Campbell, School of Psychology, College of Engineering, Science and the Environment, University of Newcastle, Newcastle, NSW, Australia; Hunter Medical Research Institute, Newcastle, NSW, Australia. Nicolas Legrand, School of Culture and Society, Interacting Minds Centre, Aarhus University, Aarhus, Denmark. Micah Allen, Center of Functionally Integrative Neuroscience, Aarhus University, Aarhus, Denmark; Department of Psychiatry, University of Cambridge, Cambridge, UK. Michael Breakspear, School of Psychology, College of Engineering, Science and the Environment, University of Newcastle, Newcastle, NSW, Australia; Hunter Medical Research Institute, Newcastle, NSW, Australia; School of Medicine and Public Health, College of Medicine, Health and Wellbeing, University of Newcastle, Newcastle, NSW, Australia. Funding This work was supported by the Royal Australia New Zealand College of Psychiatrists (Beverley Raphael scholarship to J.J.), the National Health and Medical Research Council (APP2013829 to J.J., APP2008612 to M.B.), and the Rainbow Foundation (J.J.). Author Contributions J.J. recruited participants, completed the task, completed clinical interview, and drafted the manuscript. M.C. provided supervision, assisted with task implementation, and contributed to the manuscript. N.L. and M.A. assisted with task implementation contributed to the manuscript. M.B. provided supervision and contributed to the manuscript. J.J., M.A., and M.B. were all involved in designing the experiment. Data Availability Code for the experiment is available in the original Cardioception Python Package (https://github.com/embodied-computation-group/Cardioception References Associated Data Supplementary Materials Data Availability Statement Code for the experiment is available in the original Cardioception Python Package (https://github.com/embodied-computation-group/Cardioception