Do Cardiorespiratory Variables Predict the Antinociceptive Effects of Deep and Slow Breathing?

Abstract

Abstract.

Deep and slow breathing (DSB) is a central part of behavioral exercises used for acute and chronic pain management. Its mechanisms of action are incompletely understood.

Objectives.

1) To test the effects of breathing frequency on experimental pain perception in a dose dependent fashion. 2) To test the effects of breathing frequency on cardiorespiratory variables hypothesized to mediate DSB analgesia. 3) To determine the potential of the cardiorespiratory variables to mediate antinociceptive DSB effects by regression analysis.

Design.

Single-blind, randomized, crossover trial.

Subjects.

Twenty healthy participants.

Interventions.

Visually paced breathing at 0.14 Hz, 0.10 Hz, 0.06 Hz, and resting frequency.

Outcome Measures.

Cardiorespiratory variables: RR-interval (= 60 seconds/heart rate), standard deviation of the RR-interval (SDRR), and respiratory CO₂. Experimental pain measures: heat pain thresholds, cold pain thresholds, pain intensity ratings, and pain unpleasantness ratings.

Results.

1) There was no effect of DSB frequency on experimental pain perception. 2) SDRR and respiratory CO₂ were significantly modulated by DSB frequency, while RR-interval was not. 3) Baseline-to-DSB and session-to-session differences in RR-interval significantly predicted pain perception within participants: Prolonged RR-intervals predicted lower pain ratings, while shortened RR-intervals predicted higher pain ratings. SDRR and respiratory CO₂ were not found to predict pain perception.

Conclusions.

The present study could not confirm hypotheses that the antinociceptive effects of DSB are related to changes in breathing frequency, heart rate variability, or hypoventilation/hyperventilation when applied as a short-term intervention. It could confirm the notion that increased cardiac parasympathetic activity is associated with reduced pain perception.

Pain, Breathing, Respiration, Heart Rate Variability, Biofeedback, Relaxation

Introduction

Deep and slow breathing (DSB) techniques are used in the multimodal therapy of chronic pain as a behavioral method of pain management . Most DSB techniques are characterized by a reduction of breathing frequency and a diaphragmal emphasis of breathing with a long exhalation phase. In clinical practice, biofeedback techniques often pace the frequency of DSB, i.e., the patient is instructed to inhale/exhale according to a moving bar shown on a screen . Despite the widespread popularity and some evidence of its effectiveness in painful conditions and experimental settings , the mechanisms of action of DSB are still not fully understood.

The majority of studies have suggested that the antinociceptive effects of DSB are mediated by psychological mechanisms, such as facilitation of emotional regulation , relaxation , or distraction . In addition, expectancy and placebo effects have to be considered . However, there is a body of evidence suggesting that DSB has elementary physiological consequences that may directly or indirectly moderate antinociceptive effects: Slow breathing frequencies have been reported to prolong RR-intervals (= 60[beats]/heart rate[beats/second] by definition) , and studies using heart rate biofeedback found that prolonged RR-intervals are associated with diminished pain perception . Further, DSB is likely to influence pulmonary gas exchange. Faster breathing frequencies may lead to hypocapnia, while very slow breathing frequencies may lead to hypercapnia. The latter has been shown to cause antinociceptive effects . Moreover, increases in heart rate variability (HRV) depend on breathing frequency. It is known that HRV reaches its maximum at a breathing frequency of approximately 0.10 Hz (6 breaths/minute), the so-called “resonant frequency” . Based on this observation, it has been hypothesized in the field of HRV biofeedback that DSB may modulate autonomic nervous system (ANS) activity and pain perception most efficiently at the resonant frequency .

In summary, depending on the breathing frequency, DSB might affect a number of cardiorespiratory variables, which might in turn mediate changes in pain perception. To our knowledge, the impact of DSB frequency on pain perception has not been studied so far in a dose-dependent fashion. It is further unclear which of the mentioned cardiorespiratory variables are influenced by DSB frequency and at the same time able to predict antinociceptive effects.

Therefore, the first objective of the present study was to test experimentally if pain perception was significantly affected by four different paced DSB frequencies. Pain perception was measured using thermal pain thresholds and by retrieving ratings of tonic heat stimuli. The second study objective was to test the effects of four paced breathing frequencies on the cardiorespiratory variables RR-interval, standard deviation of the RR-interval (SDRR), and respiratory CO₂. Finally, the third study objective was to determine if DSB-initiated changes in the measured cardiorespiratory variables could predict changes in nociception. Based on the studies mentioned earlier, it was expected that increases in RR-interval, SDRR, and respiratory CO₂ would predict antinociception.

Methods

Participants

Twenty healthy volunteers (50% male) were recruited by advertisement among university students. Absence of acute infections, obesity, internal, neurological or psychiatric disorders, acute or chronic pain, as well as absence of analgesic, psychoactive, cardiovascular, or antiallergic medication was ascertained in a structured clinical interview with a certified specialist. Written informed consent was obtained from all participants. The study was approved by the local ethics committee and is conform to the Declaration of Helsinki . Participants were informed that the study aim was to determine the effect of “several breathing exercises” on pain perception, but it was not explicated that the study focused on breathing frequency and that one session involved paced breathing at their own baseline frequency. Participants were not given any suggestions that a certain breathing condition may be more effective than the others.

Procedure

The present study was designed as a randomized, repeated-measures, single-blind trial with four sessions per participant. The experimental conditions were paced breathing at: 1) 0.14 Hz (8.4 breaths/minute), 2) 0.10 Hz (6 breaths/minute), 3) 0.06 Hz (3.6 breaths/minute), and 4) the individual's resting frequency. The breathing conditions were arranged in 20 different intervention sequences balanced for first session. Participants were randomly assigned to these intervention sequences using a free online tool . To get accustomed to the laboratory setting, all participants completed an additional training session before the first testing session, where paced breathing was exercised at different breathing frequencies and pain measurement procedures were practiced. Sessions were scheduled at least 7 days apart, to rule out carryover effects between the experimental conditions. Sessions were held always at the same time of day (±1 hour) to control for circadian effects. All testing was performed in a quiet room, with participants sitting in a comfortable reclining chair in a semisupine position. Participants were sitting for at least 10 minutes before any measurement was started. A full timeline of the procedures is shown in Table ⁰⁰⁰¹.

Table 1

Timeline of experimental procedures

For baseline resting period, participants were asked to relax, breathe normally, and sit quietly for 7 minutes. For the rest of the session, breathing was paced visually, with a vertically moving bar presented on-screen using EZ-Air Plus (BFE, retrieved from http://www.bfe.org/breathpacer.htm). The inspiration-to-expiration ratio was set to 1:2 for all frequencies . In each session, participants were instructed to pace their breathing according to a script adapted from Lehrer et al. focusing on a “constant diaphragmal breathing rhythm, while using pursed lips to control exhalation volume”, emphasizing that they should “breathe in a relaxed way” and “not try too hard.” They were further instructed to maintain paced breathing during and between all measurements. For the resting frequency condition, the breath pacer was set to the individual's mean resting frequency of that day's 5 minutes of rest.

R-R Intervals and HRV Measures

The “EXG” module of a “Biofeedback Expert 2000” system (Schuhfried, Mödling, Austria) was used to obtain a standard three lead setup electrocardiograms (ECGs). ECGs were recorded during 7 minutes of rest and 7 minutes of paced breathing at a sampling frequency of 1 kHz with instantaneous R-R interval detection. An ECG dataset was defined as the last 5 minutes of a given measuring period. Recording and data processing was performed following the appropriate guidelines .

SDRR was defined as the only outcome measure of HRV for the present study. Frequency domain measures of HRV, such as low- (LF) and high-frequency power (HF) were included for descriptive purposes, but excluded from further analysis, as they are only valid indices of ANS function, when breathing frequency is constant between experimental conditions . LF and HF represent the spectral power of HRV in frequency bands of 0.04–0.15 Hz and 0.15–40 Hz, respectively. Respiratory sinus arrhythmia (RSA) usually is the main source of HF power under resting conditions with a peak around 0.20 Hz. Paced breathing according to our experimental conditions switches RSA power below 0.15 Hz and therefore from HF to LF. The subsequent increase in LF and decrease in HF would therefore reflect the filter properties inherent to these measures instead of meaningful changes in vagal activity. The time domain measure root mean square successive difference (RMSSD) was dismissed from further analysis due to similar reasons .

All HRV variables were calculated using the software Kubios-HRV . All recordings were inspected visually and artifacts corrected with the software's built-in filter function. Recordings with more than three artifacts were to be excluded from analysis.

Respiratory Measures and CO₂

Respiratory amplitude was recorded using the “RESP” module of the biofeedback equipment. The module equals a strain gauge with a resolution of 0.2 mm and a sampling rate of 200 Hz. Recordings were visually checked and artifacts corrected. Analysis was performed using an adapted version of the open-source algorithm “peakdet” in combination with a moving-average low-pass filter with a window size of 0.5 seconds.

End-tidal partial pressure of respiratory CO₂ was measured at the end of the resting and breathing intervals using a tight-fitting anatomic mask, in combination with a CO₂ sensor connected to a Sirecust (SC) 9000XL (Siemens, Erlangen, Germany) health monitor. The CO₂ values of five subsequent breaths were recorded and averaged.

Testing of Pain Perception

Temperature stimuli were applied to the volar surface of the left lower arm with a 30 × 30 mm thermode connected to a Thermosensory Analyzer II (Medoc, Ramat Yishai, Israel). The thermode was kept in place by an elastic strap, while the arm comfortably rested on a tablet attached to the chair. The experimenter relocated the thermode once after the thresholding procedure and once after half of the tonic heat rating procedure. The three sites of stimulation were 3 cm, 9 cm, and 15 cm above the wrist, used in balanced sequence.

Heat and cold pain thresholds were determined using the method of limits according to established protocols . Nociceptive perception was measured using cold pain threshold (CPT) and heat pain threshold (HPT). In short, five stimuli were presented as decreasing (CPT) or increasing (HTP) temperatures starting from a baseline of 32°C. Participants were instructed to press a stop button at pain threshold defined according to a read-aloud script . Stimuli were separated by jittered interstimulus intervals of 11 ± 1 second. The first stimulus was defined as a trial stimulus and discarded from analysis. Thresholds were defined as the mean of the four other stimuli.

As a second measure of pain perception, a rating procedure was employed: 18 heat stimuli with temperature levels of 43, 44, 45, 46, 47, and 48°C were applied. Each level was presented three times. All stimuli were applied in pseudorandomized sequences, ensuring that no temperature occurred twice in a row. Sequences were randomly assigned across participants and conditions. All heat stimuli were applied for 20 seconds duration, oscillating ±1°C around the temperature levels at 0.5 Hz (speed: 2°C/second) to reduce adaption/sensitization effects . Thermode temperature ramp-up was set to 2°C/second, the interstimulus interval was 30 seconds. After nine stimuli, thermode location was switched. Before the procedure and after changing thermode location, two preconditioning stimuli of 43 and 47°C were applied to reduce adaption/sensitization effects. Participants could stop a stimulus if deemed unbearable at any time by pressing a button. In this case, the following interstimulus interval was automatically prolonged to keep testing duration constant. During the interstimulus intervals, participants were asked to enter ratings of pain intensity and unpleasantness on two visual analog scales (VASs) shown on-screen using a mouse. The VASs that ranged from “0” to “100” were ticked and labeled in steps of 10, and the cursor always started at “0.” Instructions were provided to the participant before the procedure by a read-aloud script. The distinction between pain intensity and pain unpleasantness was illustrated according to established protocols and resulted in two variables: pain intensity ratings and pain unpleasantness ratings. Both were calculated as the mean individual rating response to temperature steps 45, 46, 47, and 48°C. Ratings for 43 and 44°C were excluded from analysis as preliminary results indicated that ratings did not significantly differ from 0.

Statistics

SPSS 19.0.0.2 for Mac OS X (IBM, Armonk, New York) was used for statistical analysis. All statistical tests were performed using SPSS's GENLINMIXED procedure for generalized linear model (GLM) analysis with robust covariance estimation to correct for potential violations of the model assumptions. Before analysis, all variables were z-transformed to ensure grand mean centering and to standardize results. Further unstructured autoregressive covariance matrixes (no assumptions on repeated covariance structure are made) were used to model repeated covariance within participants across time and sessions for objectives one and three. For objective two, a first-order autoregressive covariance matrix (measures closer in time are expected to show stronger covariance) was used due to the limited degrees of freedom available. Within each study objective, Bonferroni correction was used to account for multiple comparisons.

For objective one, the effects of the factor “breathing frequency” on the dependent variables CPT, HPT, pain intensity ratings, and pain unpleasantness ratings were tested using four separate one-way analyses of variance (ANOVAs). For objective two, the effects of the fixed factors “time” (two levels: 5 minutes of baseline vs 5 minutes of paced breathing), “breathing frequency” (four levels: paced baseline frequency, 0.14 Hz, 0.10 Hz, and 0.06 Hz), and the interaction term “time × breathing frequency” on the dependent variables RR-interval, SDRR, and respiratory CO₂ were tested with three separate 2 × 4 ANOVAs. Significant effects were followed up using planned contrasts (Bonferroni corrected), testing all differences between baseline and paced breathing, as well as all differences between the four paced breathing conditions. For objective three, within-session changes from resting period to paced breathing (paced breathing minus resting period) and resting period measures of RR-interval, SDRR, and CO₂ were defined as predictive scalar factors for the dependent variables CPT, HPT, pain intensity ratings, and pain unpleasantness ratings in 12 separate GLMs.

Pearson's standardized residuals were examined for all models to guarantee robustness of the results. As a rule of thumb, 5% of cases would be expected to show an absolute value greater than 2.0 . So for the present sample size of 74 sessions, 3.7 sessions were expected to show an absolute value above this limit. Therefore, models with four values above 2.0 were to be reexamined, excluding these particular cases.

Results

General

The mean age of the sample was 24.4 (range: 20.7–28.6) years. Two participants reported side effects: One participant experienced paresthesias in the upper limbs, as well as hiccups during three sessions. He was excluded from analysis as only one valid session remained. Another participant reported dizziness after the resting frequency condition. All data from this session were excluded from analysis except for the baseline measures. For another participant's session respiratory belt, RR-interval and SDRR data were lost due to equipment failure. Inspection of respiratory belt data indicated that all participants were able to maintain a stable breathing pattern throughout the HRV measurements, except for one participant who showed a respiratory rate of 0.89 Hz during the 0.06 Hz breathing frequency condition. Data from this session were excluded from analysis except for baseline measures. In total, the sample included 74 sessions within 19 participants, with one additional missing session for breathing frequency, RR, and SDRR, and no missing sessions for respiratory CO₂. Descriptive results for cardiorespiratory variables are shown in Table ⁰⁰⁰², and additional HRV results including frequency domain measures are shown in online supplement Table S1. Descriptive results for variables of pain perception are shown in Table ⁰⁰⁰³. Shapiro–Wilk's test for normality performed on within-subject centered variables indicated a nonnormal distribution of within-participant differences for breathing frequency (P < 0.001), HTP (P = 0.032), and RR-interval (P = 0.012). Intra-individual differences in CPT, pain intensity ratings, pain unpleasantness ratings, CO₂, and SDRR were normally distributed.

Table 2

Descriptives of the cardiorespiratory variables: Means ± standard deviation. N = 19 (74 sessions). “Baseline” data was recorded during 5 minutes at rest. Subsequently, 5 minutes of paced breathing according to the experimental conditions were recorded

Condition	Breathing Frequency (breaths/minute)		RR-Interval (ms)		SDRR (ms)		Respiratory CO2 (%vol)
	Baseline	Paced	Baseline	Paced	Baseline	Paced	Baseline	Paced
0.06 Hz = 3.6 breaths/minute	14.3 ± 4.2	3.7 ± 0.2	813.6 ± 124.4	819.5 ± 107.8	59.3 ± 21.9	98.0 ± 27.2	5.1 ± 0.4	5.4 ± 0.7
0.10 Hz = 6.0 breaths/minute	14.2 ± 3.4	6.0 ± 0.2	839.2 ± 122.4	853.2 ± 105.0	70.5 ± 22.7	109.2 ± 33.0	4.9 ± 0.4	4.9 ± 0.6
0.14 Hz = 8.4 breaths/minute	14.0 ± 3.5	8.5 ± 0.3	816.3 ± 132.4	829.7 ± 114.2	63.1 ± 21.8	84.8 ± 20.9	4.9 ± 0.5	4.9 ± 0.7
Paced resting frequency	14.0 ± 4.1	14.4 ± 4.0	847.7 ± 131.9	858.1 ± 120.6	62.7 ± 20.1	62.1 ± 20.7	4.9 ± 0.5	4.6 ± 0.7

SDRR = standard deviation of the RR-interval.

Table 3

Descriptives of the somatosensory variables: Means ± standard deviation, N = 19 (74 sessions)

Condition	Cold Pain Thresholds (°C)	Heat Pain Thresholds (°C)	Heat Intensity Ratings (Points)	Heat Unpleasantness Ratings (Points)
0.06 Hz (3.6 breaths/minute)	12.3 ± 9.5	46.8 ± 2.0	42.4 ± 17.4	41.4 ± 18.2
0.10 Hz (6.0 breaths/minute)	10.6 ± 9.1	46.5 ± 2.5	43.9 ± 19.5	43.3 ± 19.2
0.14 Hz (8.4breaths/minute)	12.3 ± 8.0	46.6 ± 1.8	42.1 ± 19.6	41.1 ± 19.6
Paced resting frequency	9.2 ± 8.0	46.9 ± 1.8	44.6 ± 19.5	41.5 ± 20.4

Somatosensory variables were recorded during constant paced breathing following the cardiorespiratory recordings. Intensity and unpleasantness ratings are shown as participant's mean ratings for stimuli with 45, 46, 47, and 48°C of thermode temperature on a 0–100 point visual analog scale.

Objective 1: Effects of Different Paced Breathing Frequencies on Pain Perception

No measure of pain perception was found significantly affected by the experimental condition. There were no significant differences in the effect of different slow-paced breathing frequencies on four variables on pain perception (see Table ⁰⁰⁰⁴).

Table 4

General linear models indicate no effect of slow-paced breathing frequency on pain perception

Variable	dfdenominator	Model Term	F	P	pBonf
Cold pain threshold	70	Breathing frequency	2.599	0.059	0.236
Heat pain threshold			0.263	0.852	>1
Pain intensity ratings			1.242	0.301	>1
Pain unpleasantness ratings			0.568	0.638	>1

Breathing frequency conditions were: paced baseline frequency, 0.14, 0.10, and 0.06 Hz. N = 19 (74 sessions), df_numerator = 3.

p_Bonf = Bonferroni-corrected P value.

Objective 2: Effects of Different Paced Breathing Frequencies on Physiological Variables

RR-interval was not found significantly affected by factors time or breathing frequency, while significant time × breathing frequency interactions were found for SDRR and CO₂ (see Table ⁰⁰⁰⁵). For SDRR, planned comparisons revealed that all breathing exercises with the exception of paced resting frequency significantly increased mean SDRR compared with baseline. Further, mean SDRR significantly increased with decreasing breathing frequencies until the 0.1 Hz condition. The 3.6 b/minute condition did not significantly increase SDRR in comparison to the 0.1 Hz condition and the 0.14 Hz condition. Significant changes from baseline were only found for paced resting frequency, which decreased CO₂. However, respiratory CO₂ during the 0.06 Hz breathing condition was significantly elevated compared with paced resting frequency, and paced breathing at 0.10 and 0.14 Hz. Results for planned comparisons are shown in Table ⁰⁰⁰⁶.

Table 5

General linear models testing the effects of slow-paced breathing frequency on cardiorespiratory variables

Variable	dfdenominator	Model Term	F	P	pBonf
RR-interval	141	Time	0.019	0.889	>1
		Breathing Frequency	0.632	0.595	>1
		Time × Breathing Frequency	1.663	0.178	0.534
SDRR	141	Time	38.822	<0.001	<0.001
		Breathing frequency	22.223	<0.001	<0.001
		Time × breathing frequency	16.694	<0.001	<0.001
Respiratory CO2	142	Time	0.778	0.379	>1
		Breathing frequency	5.282	0.002	0.006
		Time × breathing frequency	5.006	0.002	0.006

N = 19 (74 sessions), df_numerator = 7 (df_Time = 1, df_Condition = 3, df_{Time*Condition} = 3). Significant effects are marked bold.

SDRR = standard deviation of the RR-interval; p_Bonf = Bonferroni-corrected P value.

Table 6

Planned paired contrasts following up significant effects of slow paced breathing frequency on cardiorespiratory variables N = 19 (74 sessions), Significant effects are marked bold

Planned Contrast			SDRR			Respiratory CO2
			t	P	pBonf	t	P	pBonf
0.06 Hz, RP	vs	0.06 Hz, PB	−8.371	<0.001	<0.001	−2.492	0.023	0.233
0.10 Hz, RP	vs	0.10 Hz, PB	−6.334	<0.001	<0.001	0.412	0.685	>1
0.14 Hz, RP	vs	0.14 Hz, PB	0.232	<0.001	<0.001	0.821	0.422	>1
RF, RP	vs	RF, PB	−2.136	0.820	>1	3.876	0.001	0.012
0.06 Hz, PB	vs	0.10 Hz, PB	1.880	0.048	0.475	3.896	0.001	0.012
0.06 Hz, PB	vs	0.14 Hz, PB	5.042	0.077	0.773	3.907	0.001	0.011
0.06 Hz, PB	vs	RF, PB	4.042	<0.001	0.001	4.109	0.001	0.008
0.10 Hz, PB	vs	0.14 Hz, PB	8.466	0.001	0.007	−0.008	0.993	>1
0.10 Hz, PB	vs	RF, PB	5.556	<0.001	<0.001	2.019	0.060	0.595
0.14 Hz, PB	vs	RF, PB	−8.371	<0.001	<0.001	2.020	0.059	0.594

SDRR = standard deviation of the RR-interval; p_Bonf = Bonferroni-corrected P value; RP = resting period; PB = paced breathing period; RF = resting frequency.

Objective 3: Predicting Pain Perception by DSB-Induced Cardiorespiratory Changes

Table ⁰⁰⁰⁷ shows the results of 12 regression analyses testing the relationships between the cardiorespiratory and nociceptive variables. Changes in RR-interval and baseline RR-interval both were highly significant predictors of pain unpleasantness ratings. For pain intensity ratings, change in RR-interval was a predictor of borderline significance, while RR-interval at baseline was a highly significant predictor. Including RR-interval change and RR-interval at baseline improved model fit for pain intensity ratings and pain unpleasantness ratings compared with an intercept-only model. Beta-coefficients indicated that in sessions where the RR-interval increased by 10 ms from baseline, pain unpleasantness ratings decreased by −0.23 ± 0.07 points. Beta-coefficients further indicated that in sessions where RR-interval was elevated by 10 ms at baseline, pain intensity ratings were −0.53 ± 0.11 points lower, and pain unpleasantness ratings −0.56 ± 0.12 points were lower. Change in RR-interval and baseline RR-interval could not explain significant proportions of variance for CPT or HPT and could not improve model fit. The same held true for changes in SDRR and respiratory CO₂. Examination of Pearson's residuals indicated that six cases were >2.0 for pain unpleasantness ratings (the largest case being −2.328). Reanalysis excluding the particular cases did not change the results in respect to model fit, direction of effect, and the criterion of significance. Further, a visual examination of the results indicated that one session could represent an outlier with undue influence on the linear results. Excluding this session did not change the results in respect to model fit, direction of effect, or the criterion of significance. All somatosensory variables were reexamined using one cardiorespiratory at a time for descriptive purposes and can be found in online supplement Table S2.

Table 7

General linear models testing if changes in cardiorespiratory variables predict measures of pain perception

Variable	Cofactor	Model Term	Effect			Coefficient		ΔAICC
			F	P	pBonf	Beta ± (SE)	t
Cold pain treshold	RR-interval	Intercept	2.115	0.128	>1	0.083 ± 0.204	0.408	5.710
		RR-interval change	0.018	0.894	>1	−0.008 ± 0.058	−0.134
		RR-interval baseline	3.970	0.050	0.600	−0.102 ± 0.051	−1.992
	SDRR	Intercept	1.349	0.266	>1	0.097 ± 0.205	0.475	5.343
		SDRR change	1.212	0.275	>1	0.078 ± 0.071	1.101
		SDRR baseline	1.272	0.263	>1	−0.062 ± 0.055	−1.128
	Respiratory CO2	Intercept	0.272	0.762	>1	0.105 ± 0.210	0.501	7.081
		CO2 change	0.483	0.489	>1	0.050 ± 0.073	0.695
		CO2 baseline	0.096	0.757	>1	0.022 ± 0.071	0.310
Heat pain threshold	RR-interval	Intercept	3.644	0.031	0.372	0.292 ± 0.123	2.365	2.919
		RR-interval change	0.795	0.376	>1	0.066 ± 0.074	0.892
		RR-interval baseline	4.743	0.033	0.396	0.150 ± 0.069	2.178
	SDRR	Intercept	0.346	0.709	>1	0.309 ± 0.133	2.317	5.522
		SDRR change	0.224	0.638	>1	0.027 ± 0.056	0.473
		SDRR baseline	0.561	0.456	>1	0.037 ± 0.049	0.749
	Respiratory CO2	Intercept	0.175	0.839	>1	0.281 ± 0.135	2.075	7.293
		CO2 change	0.090	0.765	>1	−0.018 ± 0.061	−0.300
		CO2 baseline	0.312	0.578	>1	0.036 ± 0.064	0.559
Intensity ratings	RR-interval	Intercept	11.189	<0.001	<0.001	0.211 ±0.204	1.034	−4.796
		RR-interval change	7.607	0.007	0.084	−0.120 ± 0.044	−2.758
		RR-interval baseline	22.063	<0.001	<0.001	−0.310±0.066	−4.697
	SDRR	Intercept	3.220	0.046	0.552	0.048 ± 0.206	0.233	0.598
		SDRR change	4.411	0.039	0.468	−0.074 ± 0.035	−2.100
		SDRR baseline	0.034	0.853	>1	−0.014–0.075	−0.186
	Respiratory CO2	Intercept	1.038	0.359	>1	−0.008 ± 0.187	−0.045	0.534
		CO2 change	2.021	0.160	>1	0.075 ± 0.053	1.422
		CO2 baseline	0.208	0.650	>1	0.020 ± 0.043	0.456
Unpleasantness rating	RR-interval	Intercept	12.737	<0.001	<0.001	0.128 ±0.183	0.701	−5.080
		RR-interval change	11.459	0.001	0.012	−0.134 ± 0.040	−3.385
		RR-interval baseline	23.513	<0.001	<0.001	−0.324 ± 0.067	−4.849
	SDRR	Intercept	0.315	0.731	>1	−0.019 ± 0.192	−0.102	−1.798
		SDRR change	0.603	0.440	>1	−0.039 ± 0.050	−0.777
		SDRR baseline	0.115	0.736	>1	0.026 ± 0.075	0.338
	Respiratory CO2	Intercept	0.755	0.474	>1	0.020 ± 0.181	0.112	−2.629
		CO2 change	0.010	0.919	>1	0.006 ± 0.063	0.102
		CO2 baseline	1.484	0.227	>1	0.072 ± 0.059	1.218

Cardiorespiratory variables at rest were included in analysis to account for session-to-session baseline differences. N = 19 (74 sessions), df_numerator = 4, df_denominator = 70. Significant effects are marked bold. ΔAICC = AICC_Model − AICC_{Intercept only} (a lower AICC indicates increased model fit).

SDRR = standard deviation of the RR-interval; AICC = Akaike information criterion corrected for finite sample sizes.

Discussion

Overview

The present study tested effects of slow-paced breathing frequencies on experimental pain perception (objective 1), cardiorespiratory function (objective 2), and examined the relationship between changes in cardiorespiratory variables and experimental measures of pain perception (objective 3). Results indicated that breathing frequency had no significant effect on experimental pain perception and RR-interval. Only SDRR and respiratory pCO₂ were significantly affected by different breathing frequencies. Nevertheless, neither SDRR nor respiratory CO₂, could predict intra-individual changes in pain perception. In contrast, baseline-to-breathing as well as session-to-session differences in RR-interval were found to predict changes in pain perception. Prolonged RR-intervals predicted lower pain unpleasantness ratings, while shortened RR-intervals predicted higher ratings.

Breathing Frequency Was Not Found to Determine Antinociceptive Effects

The present study could not add evidence to the hypothesis that breathing frequency is the determining factor of DSB-induced hypoalgesia. This stands in contrast to two recent studies, which could show an effect of slow-paced breathing on experimental pain perception using paced breathing at normal frequency as a control condition . The best explanation for these conflicting results might be that the present study was the first to test the effects of slow breathing in a dose-dependent fashion, i.e., using several slow breathing frequencies. On the one hand, this approach might have diminished statistical power. On the other hand, the use of several slow breathing conditions might have reduced expectation effects. Finally, these results could point toward a third variable underlying DSB analgesia. Besides breathing frequency, the present study particularly aimed to follow up two other hypotheses on physiological phenomena accompanying DSB: modulation of cardiac autonomic function and modulation of respiratory gas concentrations.

RR-Interval Was Not Influenced by DSB, yet the Best Predictor of Pain Ratings

ANS activity is closely linked to nociception . RR-interval, heart rate, as well as HRV in general have been interpreted as measures of general ANS function in the past, “although inferences are clearly restricted to the level of the heart” . Nevertheless, these measures of cardiac autonomic function have been shown to be interrelated with pain perception in different contexts . Accordingly, we found that pain unpleasantness ratings significantly decreased when RR-interval increased during DSB. Although this effect did not reach significance for CPT, HPT, and pain intensity ratings, the same direction of effect could be observed (see Table ⁰⁰⁰⁷ and online supplement Table S2). Similarly, in sessions with elevated RR-interval at baseline, pain unpleasantness ratings and pain intensity ratings were decreased significantly. As increases in RR-interval are commonly interpreted as increases in parasympathetic activity or sympathetic withdrawal, our results are in accord with earlier animal and human studies linking increases in cardiac parasympathetic tone and sympathetic withdrawal to antinociceptive effects. However, it cannot be concluded that the DSB exercise was causally related to changes in cardiac ANS function and the associated effects on pain perception as breathing frequency was not found to significantly modulate RR-interval. Small but meaningful increases in RR-interval from baseline to DSB that remained undetected due to a lack of statistical power may have led to this outcome. The finding that paced DSB was insufficient to influence RR-interval stands in line with some earlier finding , but in contrast to others and indicates that DSB alone might not be efficient for manipulating heart rate. Biofeedback procedures designed to increase RR-interval might be more successful in doing so and should be considered as an intervention in future studies.

Hear Rate Variability during Slow Breathing Was Not Found to Predict Antinociception

SDRR is the most general measure of HRV. It has been used as an indirect measure of cardiac ANS activity and used to infer on general ANS function . Increases in SDRR are commonly interpreted to reflect parasympathetic activation and/or sympathetic withdrawal referring to experiments using cholinergic and beta-adrenergic receptor blockade . However, it is long known that slow breathing frequencies and increases in respiratory volume lead to dramatic increases in measures of HRV by mechanisms that obscure the association between HRV measures and the psychophysiologically relevant portion of ANS activity . Notwithstanding, HRV increases have been hypothesized to indicate resonance between the RSA mechanism and the baroceptor reflex , and proposed to be therapeutically beneficial when used to exercise parasympathetic function . Taking this hypotheses one step further, Chalaye et al. postulated that “slow deep breathing and HR biofeedback would produce the largest cardiac changes and the strongest analgesic responses” on experimental pain in a one-session experiment with healthy participants . Based on their finding of coinciding differences in pain tolerance thresholds and SDRR amplitude between a distraction and a slow breathing condition, they concluded: “modulation of HR and pain share a common neurophysiological pathway” . Our present experiment followed up this hypothesis by testing if DSB induced intra-individual SDRR amplitude changes can predict antinociceptive DSB effects in a dose-dependent fashion. Indeed, we could replicate that DSB at frequencies of around 0.1 Hz (“the resonant frequency”) causes large increases in SDRR and therefore general HRV. Consequently, SDRR should have been expected to predict a good degree of variation in experimental pain measures if a functional relationship between pain perception and general HRV existed. The fact that large DSB-induced changes in HRV did not predict any difference in pain perception is in accord with recent results by Martin et al. who could not find a relationship between breathing-induced changes in RMSSD and pain ratings . Together, these findings question if the “resonance” mechanism is relevant for DSB-induced antinociception, at least when applied without long-term training . It further underlines that HRV amplitude changes seen during slow breathing may reflect “mechanical effects of respiration” , rather than appropriate correlates of ANS function. The finding that RR-interval could predict differences in pain perception, while SDRR could not, supports the notion that RR-interval (or heart rate) at baseline is a better correlate of the psychophysiologically relevant ANS portion than HRV .

Paced Breathing at Baseline and Very Slow Frequencies Induced Hyperventilation/Hypoventilation

Hypercapnia-facilitated analgesia was the second candidate mechanism to be examined in the present study. Measures of respiratory CO₂ were used to detect potential consequences of DSB on respiratory gas exchange as hypercapnia had been shown to modulate acute pain perception . Indeed, significant differences in respiratory CO₂ could be found, which indicated that participants tended to develop hypercapnia during the 3.6 breaths/minute condition and hypocapnia during the paced resting frequency condition. These findings encourage the monitoring of hyperventilation/hypoventilation in the study of DSB exercises and replicate the finding that paced breathing at resting frequency is not just a simple “distraction” control condition, but has measurable consequences on respiratory gas exchange . Nevertheless, respiratory CO₂ could not explain relevant differences in any experimental measure of nociception. The magnitude of hyperventilation/hypoventilation might have been too small to lead to significant somatosensory consequences.

Limitations

The present findings have to be considered in the context of the following limitations: DSB effects were studied using experimental pain in healthy participants and may not generalize to clinical forms of pain and patient populations with ANS dysregulation. Caution must be exercised when comparing the results of our DSB protocol with other studies using breathing, relaxation, or distraction exercises: e.g., participants maintained visually paced breathing during sensory measurements at any time and only received a single training session. Therefore, the whole procedure might have been too demanding to have a relaxing effect. The simultaneous execution of paced DSB and thermal thresholding might also have diminished the sensitivity of the CPT and HPT measurements, which are affected by reaction time . We particularly emphasize that prolonged DSB training might yield entirely different results. It might enhance the effect of DSB by several mechanisms, but particularly because association effects were suggested to play a role in pain appraisal . Lastly, the decision to exclude participants reporting side effects from analysis might have led to an underestimation of hyperventilation/hypoventilation effects.

Conclusions

The present study supports the hypothesis that pain is as a homeostatic emotion depending on ANS function as inter-individual RR-interval differences at rest and changes in RR-interval were associated with measures of experimental pain perception. However, as the experimental conditions have neither influenced experimental pain perception directly, nor indirectly by altering RR-interval, we cannot conclude that breathing frequency is a determining factor of the proposed antinociceptive effects of DSB. Similarly, highly significant breathing-induced changes in HRV and respiratory gas concentration could not predict antinociceptive DSB effects.

The present results do not exclude the possibility that DSB therapies might be beneficial for chronic pain patients. DSB might have a different impact on patients with ANS dysfunction, and other mechanisms of action could account for its therapeutic efficacy. In this context, potential long-term training effects of DSB on cardiorespiratory function and pain perception might be of particular interest for future studies. As mentioned earlier, psychophysiological mechanisms like “facilitation of emotional regulation” , “relaxation” , or “distraction” pose relevant alternative mechanisms of action to be considered. Based on our results, we encourage further biofeedback studies to determine the relevant mechanisms of action of slow breathing exercises.

Note

An increase by 10 ms of RR-interval corresponds to an increase in heart rate by 0.9 beats/minute at mean breathing frequency. A decrease by 10 ms of RR-interval corresponds to a decrease in heart rate by 0.8 beats/minute at mean breathing frequency.

Financial disclosures/Conflicts of interest: Matthias Zunhammer is supported by a scholarship from the German National Academic Foundation. This research received no other specific grant from any funding agency in the public, commercial, or not-for-profit sectors. None of the authors has any conflict of interest to disclose.

Supporting Information

Author Webpage

Table S1. Descriptive HRV results: Means ± standard deviation. N = 19 (74 sessions). “Baseline” data was recorded during 5 minutes at rest. Subsequently, 5 minutes of paced breathing according to the experimental conditions were recorded

Table S2. Simple correlations between cardiorespiratory and somatosensory variables