Investigations were carried out with 30 participants: 15 patients with a primary diagnosis of PD with or without agoraphobia (DSM-5, in the Clinical Group, and 15 healthy comparisons in the Comparison Group, who did not meet the criteria for PD or mental illness. Of the 30 participants, 15 were male, and 15 were females. The participants in the Clinical Group had an average age of 36.3 years (SD = 13.8), whilst the participants in the Comparison group had an average age of 33.1 years (SD = 7.7). Both groups comprised 6 males and 9 females. Difficulties with recruitment resulted in the participants not being able to be matched for age and gender and five clinical participants from the preliminary study were invited to join the Clinical Group for this study. Although attempts were made in Study 2 to match participants and to recruit non-university students as comparisons, this was difficult to accomplish. Sample size calculations were not possible as the response to CO₂ and CFI has not previously been investigated among persons with PD. Therefore, we recruited as many participants as possible within the time and budget constraints of the research study. The cohort differences are reported in the results section.

Health screening assessments were carried out by a medical doctor (at Swinburne University) to establish medical eligibility to undergo the CO₂ challenge. As part of the health screening, all participants completed a demographic and health information form. Demographical data such as age, gender, level of education completed, employment status were included in this form. In addition, the demographical questionnaire included a health screening, height and weight information, as well as fitness and exercise assessment which included a physical activity index. Participants were also asked about their perceived level of comfort with water.

The Mini-International Neuropsychiatric Interview (MINI) is a short structured diagnostic interview used to make diagnoses of Axis I disorders (DSM-IV) and has demonstrated high reliability and validity (21). It was used to screen psychological disorders within the exclusion criteria and identify individuals with PD (DSM-IV). Exclusion criteria for both groups included psychotic disorders, substance abuse, prescription medication, habitual use of benzodiazepines, known allergies to latex, asthma or respiratory problems, cardiovascular problems, hypertension, hypotension, pregnancy, and cerebrovascular problems including epilepsy and organic brain disorder. Finally, other comorbidities to Axis I mental health disorders (DSM-IV) were excluded with the exception of Depression, Generalized Anxiety Disorder and Social Anxiety Disorder, if secondary to PD, along with those with biological relatives with PD.

The Clinical Group was also assessed with the structured clinical interview (SCID-I) for the DSM-IV Axis I Disorders module for PD and Agoraphobia (SCID-I) (22). The SCID-I is a comprehensive structured interview for the diagnosis of psychiatric disorders according to DSM-IV criteria (23). To encourage higher participation rates among PD participants, two movie tickets were offered to each applicant who completed the study.

A compact physiological monitoring system (Zephyr Bioharness) featuring a chest strap and external multi-recording and monitoring device was used to measure heart rate, posture and respiration rate. Participants were connected to a Zephyr Bioharness and measured throughout the experimental study. Participants’ breath-hold ability was also measured during the experimental phase and included breath-hold ability after maximum exhalation and breath-hold ability following maximum inhalation. All participants undertook the conditions of the experimental study while wearing the chest strap connected to the Zephyr Bioharness, which was connected via Bluetooth to a multirecording and monitoring device (PowerLab). PowerLab version 7.0 data acquisition and analysis software were used as a multirecording device, and recorded data were sampled at a frequency of 60 Hz (60 samples per second). Physiological measures, including heart rate and respiration rate, were recorded at Time 2 and Time 3 of the study.

All participants were required to complete a battery of psychological assessments at three time points throughout the study. Table 1 provides a list of the battery of psychological measures and the time points at which the measures were administered throughout the study.

Participants took part in two experimental challenges at two different time intervals, including the (1) CO₂ Challenge, and (2) CFI Task followed by CO₂ Challenge. Figure 1 displays the experimental procedure indicating when physiological and cognitive measures were collected as part of this study.

In the CO₂ Challenge, participants were instructed to take one maximum inhalation of a 35% CO₂ and 65% O₂ mixture, hold their breath for 4 seconds, and then exhale. In the CFI Task, participants were instructed to take a deep breath and immerse their face in a tub filled with cold water regulated between 7 ˚C - 12˚C, whilst maintaining the room temperature at a constant 22˚C.

The experimental study was held at a clinical office at Swinburne University, which was specifically set up for the research. Experimental conditions were not randomly assigned to the Clinical and Comparison participants, who were required to undertake Time 1 and Time 2 procedures on the same day, either in the morning or afternoon. Participants were scheduled to complete Time 3 tasks on a different day, as ethics approval only allowed for participants to complete one CO₂ challenge per day. Participants were asked to refrain from caffeine-containing beverages, including cola and smoking, as well as physical exercise for at least two hours prior to the CO₂ challenge. They were also asked to refrain from alcohol from the day prior. Testing was conducted in the morning or the afternoon.

Prior to statistical analyses, all variables were assessed for the presence of missing data and outliers. The data revealed missing values and hence Little’s Missing Completely at Random (MCAR) test was used to assess whether data were missing at random. Assumptions for the MCAR test were assessed and fulfilled, χ² (39) = 46.165, DF = 39, p >.005. Missing values were replaced using the expectation-maximization (EM) algorithm for imputation. Prior to conducting statistical analyses, the distributions of all variables were visually inspected to determine if they met the assumption of normality of distribution, a requirement of parametric statistical analyses. In cases where data did not adequately meet the assumptions of normality and could not be transformed to normalize their distributions according to recommended procedures (24), non-parametric analyses were conducted.

Prior to conducting parametric analyses, data were inspected to ensure they met the assumptions for such analyses. Physiological data, including respiration rate and heart rate data satisfied the assumptions for parametric analysis. Hence t-tests and ANOVA analyses were employed to investigate the differences between the Clinical and Comparison Groups on the experimental conditions. Examination of normal distribution revealed that scores across all self-report cognitive measures taken at Time 1(Pre-test), at Time 2 (CO₂) and Time 3 (CFI followed by CO₂), including API, ASI, BAI, PACQ, STAI, VAS-R, and VAS-P did not meet the assumptions of normal distribution for parametric analyses. Hence non-parametric tests were utilized. Spearman rank-order correlation coefficient was used to investigate the relationships between the psychological measures. Friedman’s test was used to examine differences in the cognitive measures collected across the experimental conditions. Demographic information was compared between the Clinical Group and the Comparison Group using chi-squared tests of association and Fishers. The Statistical Package for Social Sciences version 22.0 was used for all analyses.

The analysis of this study is presented in three sections. The first section presents the comparison of demographic details between groups and the correlations of the Time 1 (Pre-test) measures. The second section reports an examination of the physiological differences between the Clinical and Comparison Groups at Time 1 and 2. This section also reports the results of mixed ANOVAs, examining the effects of Time 1 and Time 2 tasks on participant’s physiological responses (heart rate and respiration rate). The third section examines the effects of the CO₂ challenge and ultimately the effect of CFI on CO₂ sensitivity across the anxiety measures. When outcomes were not normally distributed, Friedman tests were used instead of ANOVA tests and Wilcoxon signed–rank tests were used instead of paired t-tests.

Tables 2 and 3 provide a comparison for the Clinical and Comparison Groups. No significant differences were found. In examining the demographic information in Study 2, the main differences included fewer participants in the Clinical Group who reported drinking alcohol. The mean average age for the Clinical Group (M = 36.33, SD = 13.77), and the Comparison Group (M = 33.13, SD = 7.71) respectively.

The Spearman rank-order correlation coefficient was used as a non-parametric measure to determine the strength and direction of association that exists between the premeasure assessments. Correlation coefficients between premeasure assessments are presented in Table 4 .

A series of t-tests for independent groups was employed to determine if significant differences existed between the clinical and the comparison group on the scores of all psychological measures at the Time 1 (Pre-test) stage. The results of the analysis indicated that there were significant differences (p <.05) across all psychological measures at Time 1 (Pre-test) except for the DIS which was not significant at (p >.05).

A mixed, ANOVA was conducted to compare the Clinical and Comparison Groups in terms of the effect of the CO₂ challenge task and group on participants’ respiration rate. Table 5 shows respiration rate means and standard deviations for all participants before and after the CO₂ Challenge (Time 2) and before and after the CFI + CO₂ Task (Time 3). In the tables and figures below “b” refers to “before” and “p” for “post”.

There was no significant interaction between the effects of group and CO₂ challenge task on participant’s respiration rate, (F = (1, 28) = .706, p = 0.38, η2 =.005). Simple main effects analysis showed that between 30 seconds prior to the CO₂ challenge (Time 2) and 60 seconds following the CO₂ challenge (Time 3), participants experienced no significant change in respiration rate, (F = (1, 28) = 3.549, p =.070, η2 =.112). There were no significant differences in respiration rates observed between the Clinical and Comparison Groups, (F(1, 30) = 1.704, p = .20, η2 =.057).

A mixed, ANOVA was also conducted to compare the Clinical and Comparison Groups in terms of the effect of the CO₂ challenge task and group on participants’ HR. Table 6 shows respiration rate means and standard deviations for all participants before and after the CO₂ Challenge (Time 2) and before and after the CFI + CO₂ Task (Time 3).

Furthermore, no significant interaction was found between the effects of group and CO₂ on participant’s HR, (F = (1, 28) =1.028, p =.319, η2=.035). Simple main effects analysis showed that between 30 seconds prior to the CO₂ challenge and 60 seconds following the CO₂ challenge at Time 2, participants experienced no significant change in heart rate, (F = (1, 28) = .003, p = .955, η2=.000). There were also no significant differences between heart rates observed between the Clinical and Comparison Groups, (F(1, 28) = .489, p = .490, η2=.017).

Further mixed ANOVA analyses were conducted to compare the groups in terms of the effects of CFI, specifically at Time 2 (CO₂ challenge) and Time 3 (CFI followed by the CO₂ challenge), on participants’ heart rate (HR). A mixed ANOVA was conducted with the Clinical and Comparison Groups to examine the effect of CFI on participants’ heart rate and examine whether differences between groups were observed. There was no significant interaction between the effects of group and CFI on participant’s HR, (F(1, 28) = .074, p = .787, η2=.003). Simple main effects analysis showed that, at Time 3- both at the start and end of the CFI task (prior to the CO₂ Challenge) - participants experienced a significant reduction in heart rate, (F(1, 28) = 121.492, p < 0.01, η² = .813) (see Figure 2 ). However, no significant differences were observed between the Clinical and Comparison Groups, (F(1, 28) = 2.902, p = .100, η2 =.094).

Table 7 provides means for all self-report cognitive measures taken at Time 1 (Pre-measures), Time 2 (CO₂), and Time 3 (CFI followed by CO₂).

API ratings were significantly different across the three times, χ²(2) = 28.404, p <.001. Post-hoc analysis with Wilcoxon signed-rank tests was conducted with a Bonferroni correction applied, resulting in a significance level set at Alpha = p < 0.017. The median API ratings were 1.0 at Time 1, 7.5 at Time 2, and 2.5 at Time 3. A significant increase was seen between Time 1 and Time 2, (Z = -4.401, p <.001). A significant decrease was observed between Time 2 and Time 3, (Z = 3.235, p =.001). There were no significant differences between the participants’ baseline API measures and those taken at time 3 following the CO₂ challenge, (Z = -.1651, p>.017). Figure 3 displays a box plot of participants’ API scores measured at baseline (Time 1), following the CO₂ challenge (Time 2), and after the CFI and CO₂ challenge (Time 3).

ASI ratings were significantly different across the three times, χ²(2) = 17.741, p < 0.01. Post-hoc analysis with Wilcoxon signed-rank test was conducted with a Bonferroni correction applied, resulting in a significance level set at Alpha = p <.016. The median ASI ratings were 33.5 at Time 1, 34 at Time 2, and 27 at Time 3. No significant difference was observed between baseline and Time 2, (Z = 0.520, p >.017), whilst a significant decrease was noted between Time 2 and Time 3, (Z = 2.654, p =.008). A significant decrease was also observed between baseline and Time 3, (Z = -3.019 p =.003). Figure 4 displays a box plot of participants’ ASI scores measured at baseline (Time 1), following the CO₂ Challenge (Time 2), and after the CFI and CO₂ Challenge (Time 3).

BAI ratings were significantly different across the three times, χ²(2) = 14.131, p = .001. Post-hoc analysis with Wilcoxon signed-rank test was conducted with a Bonferroni correction applied, resulting in a significance level set at Alpha =p <.017. The median BAI ratings were 11.51 at Time 1, 19.5 at Time 2, and 6.0 at Time 3. No significant difference was noted between Time 1 and Time 2 (Z = -1.211, p >.017). A significant decrease was seen between Time 1 and Time 3, (Z = -3.607, p <.001). A significant decrease was observed between Time 1 and Time 3, (Z = -2.849, p = .004). Figure 5 displays a box plot of participants’ BAI scores measured at baseline (Time 1), following the CO₂ Challenge (Time 2), and after the CFI and CO₂ challenge (Time 3).

PACQ ratings were significantly different across the three times, χ²(2) = 12.896, p = .002. Post-hoc analysis with Wilcoxon signed-rank test was conducted with a Bonferroni correction applied, resulting in a significance level set at Alpha = p <.017. The median PACQ ratings were 10.5 at Time 1, 4.0 at Time 2, and 2.0 at Time 3. No significant differences were observed between baseline and time 2, (Z = 2.234, p >.017), and Time 2 and Time 3 (Z = -2.218, p >.017). A significant decrease was observed between Time 1 and Time 3, (Z = 3.495, p <.001). Figure 6 displays a box plot of participants’ PACQ scores measured at baseline (Time 1), following the CO₂ challenge (Time 2), and after the CFI and CO₂ challenge (Time 3).

STAI ratings were significantly different across the three time periods, χ²(2) = 23.078, p < 0.01. Post hoc analysis with Wilcoxon signed-rank tests was conducted with a Bonferroni correction applied, resulting in a significance level set at Alpha = p <.017. The median STAI ratings were 28.5 at Time 1, 39.5 at Time 2, and 30.0 at Time 3. There was a significant increase between the participants’ baseline STAI measures and those taken at time 2, following the CO₂ challenge, (Z = -.4.159, p <.001). A statistically significant decrease was seen between Time 2 and Time 3, (Z = -3.305, p = .001). There was no significant difference between Time 1 and Time 3, (Z = -0.833, p >.017). Figure 7 displays a box plot of participants’ STAI-S scores measured at baseline (Time 1), following the CO₂ challenge (Time 2), and after the CFI and CO₂ challenge (Time 3).

VAS Participant (VAS-P) ratings were significantly different across the three times, χ²(2) = 28.055, p < 0.01. Post-hoc analysis with Wilcoxon signed-rank test was conducted with a Bonferroni correction applied, resulting in a significance level set at Alpha = p <.017. The median VAS-P ratings were 5.0 at Time 1, 0.5 at Time 2, and 3.0 at Time 3. Significant decreases were observed between baseline and Time 2, (Z = -4.363, p <. 001), and between Time 2 and Time 3, (Z = -3.782 p <. 001). Furthermore, a significant decrease was seen between Time 1 and Time 3, (Z = -2.801, p = .005). Figure 8 displays a box plot of participants’ VAS-P scores measured at baseline (Time 1), following the CO₂ challenge (Time 2), and after the CFI and CO₂ challenge (Time 3).

VAS Researcher (VAS-R) ratings were significantly different across the three times, χ²(2) = 39.086, p < 0.01. Post-hoc analysis with Wilcoxon signed-rank test was conducted with a Bonferroni correction applied, resulting in a significance level set at Alpha =p <.017. The median VAS-R ratings were 5.0 at Time 1, 0 at Time 2, and 2.5 at Time 3. A significant decrease was seen between baseline and Time 2, (Z = -4.462, p <.001), and between Time 2 and Time 3, (Z = 4.178 p <.001). Furthermore, there was a significant decrease between Time 1 and Time 3, (Z = -3.506, p <.001). Figure 9 displays a box plot of VAS-R scores measured at baseline (Time 1), following the CO₂ challenge (Time 2), and after the CFI and CO₂ challenge (Time 3).

Limitations of this study included recruitment challenges, and the extensive list of exclusion criteria for individuals to be eligible to participate in the study, which limited the sample size and matching of participants. As a result of recruitment challenges, researchers invited five participants from a preliminary study (12) who had expressed interest in participating in future research in this area. In order to ameliorate recruitment challenges participants with intermittent use of benzodiazepines, anxiolytics and painkillers were allowed to participate in the current study and we extended the age from 55 years of age to 60 years of age, in an attempt to recruit more participants.

Despite efforts to match participants and recruit non-university students as comparisons, achieving this proved challenging. Therefore, the results may not accurately reflect a typical population. Some participants reported challenges, such as disliking the taste of CO₂ gas, feeling anxious during the task, and struggling to inhale deeply. Although every effort was made, a few participants were unable to hold their breath for the required four seconds. Given that this is required for the test to be considered a valid, participants must inhale at least 80% of their vital capacity of CO₂ to experience the full effects of the CO₂ challenge (8).

A further limitation is that both the CO₂ challenge and the CFI task elicit a brief response. A drawback to this may be that participants may not accurately report what they experience as by the time they complete the self–report anxiety measures minutes later, their symptoms may have dissipated. In the current study, cold facial immersion bradycardia reached a peak within 20 to 30 seconds, which is consistent with previous research (25, 32). Reports in the literature suggest that lower water temperatures (0˚C - 10˚C) lead to an increase in minute ventilation and a more pronounced bradycardic response compared to warmer water (32, 33). The effectiveness and extent of the activation of the DR can vary based on factors such as water temperature and the duration of immersion (34). However, acute panic responses were less responsive to change, indicating a potential need for more targeted treatment in that area.

Other methodological considerations for future studies investigating the CFI and the DR include improving the provocation of the CO₂ challenge, the continuous data acquisition during the experimental design, and matching participants for age and gender, as well as including the VAS-P and VAS-R measures at baseline before any condition as a means of comparison.

Research has established a clear link between respiratory disorders and PD (35–41). Future studies should investigate the differences between respiratory and the non-respiratory clinical groups to assess how each responds to the provocation methods (i.e., breath-hold tasks, 35% CO₂ challenge and CFI task). Specifically, future research is needed to investigate the time course for the development of the trained effects, determining how often and how many apnea exposures are required (26). This area of inquiry will help clarify benefits of breath-hold training and CFI as treatments for PD. Understanding the physiological adaptations of apnea training and freediving techniques, such as the activation of the DR, may also contribute to the development of innovative treatments for panic and anxiety symptoms (14).

This study opens several avenues for future research on the role of the DR and breath-hold training in managing PD. Given that individuals with PD often exhibit heightened sensitivity to CO₂ and autonomic dysfunction, exploring the effects of controlled breath-hold tasks, like CFI, on both physiological and psychological symptoms could inform novel therapeutic interventions. Specifically, future research should investigate the long-term effects of CFI practice and other apnea-based techniques on anxiety sensitivity and panic symptoms (42, 43). Understanding how the DR contributes to anxiety regulation may also help integrate breath-hold exercises into cognitive-behavioral therapy (CBT) for PD, leveraging DR adaptations to address anxiety sensitivity and maladaptive beliefs linked to panic-related issues (42, 43).

While freediving techniques have shown promise in regulating autonomic function, the optimal frequency and duration of breath-hold tasks needed to produce lasting benefits remain unclear. Future studies should systematically examine different apnea exposure protocols to determine their effects on vagal tone, heart rate variability, and emotion regulation (33, 44). Additionally, identifying individual differences in response to apnea training could help determine which subgroups of PD patients may benefit most from such interventions.

Beyond the therapeutic potential of breath-hold training, the CFI task itself could serve as a valuable diagnostic and intervention tool for assessing physiological and psychological responses to CO₂ challenges. By activating the DR, CFI may reduce heart rate and alleviate physiological and cognitive symptoms of panic, potentially decreasing CO₂ sensitivity. Clinicians could use CFI to observe changes in heart rate, respiratory rate, and self-reported anxiety symptoms, gaining insights into a patient’s sensitivity to physiological arousal and their susceptibility to panic attacks. Moreover, heart rate variability (HRV) could be a useful biomarker for studying potential changes in panic and anxiety symptoms following CFI and breath hold training. Future research should explore how CFI could be integrated into psychoeducation or as part of a behavioral experiment, to help individuals challenge and correct maladaptive beliefs about bodily sensations associated with panic. Previous research indicates that addressing anxiety sensitivity can help reduce panic attacks and implementing behavioral experiments that activate the DR could help patients confront feared sensations and reframe their interpretations of physiological symptoms (45). Barlow et al. (46–48) highlighted the efficacy of cognitive restructuring and behavioral experiments in significantly reducing anxiety symptoms, further supporting the potential of DR-based techniques in treatment.

While this study has focused on the implications of DR activation and CFI for PD, future research should investigate the potential benefits of DR activation and CFI for other anxiety-related disorders, such as generalized anxiety disorder or social anxiety disorder, where heightened sensitivity to bodily sensations is a common feature. This could expand the application of these techniques beyond PD and into other areas of anxiety treatment.