Effects of arousal modulation via resonance breathing on craving and affect in women with substance use disorder

Abstract

Background:

Craving for alcohol and other drugs is a complex in-the-moment experience that involves within-person changes in physiological arousal and affect. We evaluated the utility of a just-in-time, self-administered resonance breathing smartphone application (app) to reduce craving and improve affect in women during outpatient treatment for substance use disorders (SUD).

Methods:

Women (N = 57) receiving outpatient addiction treatment were randomized to practice either cardiovascular resonance breathing (0.1 Hz/6 breaths per minute) or a sham (~0.23 Hz/14 breaths per minute) in the face of urges over an 8-week intervention. Craving (Penn Alcohol Craving Scale) and affect (Positive and Negative Affect Scale) were collected weekly throughout the intervention. App data were uploaded weekly to assess frequency of use. Generalized Estimated Equations modeled craving and affect as a function of group randomization and app use frequency across the 8-week intervention.

Findings:

Higher levels of craving were associated with more frequent app use. The group X app use interaction was significant for craving. Frequent app use during the intervention phase was associated with lower craving levels in the resonance breathing group relative to the sham group over the 8-week intervention. There was no effect of app use frequency on affect measures.

Conclusions:

Women assigned to practice sham breathing who used the intervention frequently experienced elevations in craving that are commonly reported during outpatient SUD treatment. Women assigned to resonance breathing who used the intervention frequently did not experience such increases. Resonance breathing may be protective against triggers in outpatient treatment. Physiological mechanisms are discussed.

1. Introduction

Individuals with substance use disorder (SUD) commonly report craving, a complex, subjective, in-the-moment experience that is measurable with self-reports of increased negative affect (Kavanagh, May, & Andrade, 2009; Sinha et al., 2009) and physical discomfort and restlessness (McCaul, Hutton, Stephens, Xu, & Wand, 2017; Rosenberg, 2009; Sayette et al., 2000; Swift & Stout, 1992). Craving is also objectively measurable as heightened physiological arousal that is marked by increased skin conductance (Back et al., 2015; Kaplan et al., 1985; Tong, Bovbjerg, & Erblich, 2007), pulse (Back et al., 2014; Kennedy et al., 2015; Ooteman, Koeter, Vserheul, Schippers, & Brink, 2006; Sinha et al., 2003), blood pressure (Back et al., 2014; Sinha et al., 2003; Sinha, 2009), and stress hormone release (Back et al., 2014, 2015; Sinha et al., 2003), as well as shifts in heart rate variability (HRV) (Culbertson et al., 2010; Ingjaldsson, Laberg, & Thayer, 2003; Quintana, Guastella, McGregor, Hickie, & Kemp, 2013). These physiological adaptations are proposed to support behavioral inflexibility towards alcohol and other drugs through their role in attentional bias and executive functioning (Buckman, Vaschillo, Fonoberova, Mezíc, & Bates, 2018) and may promote the transition of subjective experiences of craving into active drug seeking and use.

Addressing craving in recovery from SUD requires learned techniques that prepare one for unexpected exposure to craving-inducing cues (Enkema & Bowen, 2017; Stalcup, Christian, Stalcup, Brown, & Galloway, 2006; Witkiewitz et al., 2011, 2013). Most current evidence-based treatments focus on effortful cognitive processes that aim to help one navigate past the craving state (Kober, Kross, Mischel, Hart, & Ochsner, 2010; Luehring-Jones, Louis, Dennis-Tiwary, & Erblich, 2017; Robinson et al., 2017; Stalcup et al., 2006; Tiffany, 1999; VerdejoGarcia, 2016; Wiers et al., 2015). However, it has long been established that cognition is altered during states of heightened arousal (McEwen & Sapolsky, 1995; Yerkes & Dodson, 1908); learned cognitive techniques thus may be less accessible after craving has begun. In order for cognitive approaches to work most effectively, the physiological systems that support hyperarousal may need to be dampened. Such physiological tools are currently available.

Yogic breathing, meditation, exercise, and HRV biofeedback (HRVbf) improve modulation of physiological arousal and are experienced subjectively as therapeutic (Blase & van Waning, 2019; Mason, Gerbarg, & Brown, 2021; Szulczewski, 2019). HRVbf typically involves multiple session, in-person training of a slowed respiratory technique (Lehrer & Gevirtz, 2014) and has garnered recent support as an effective body-focused intervention for many mental and physical health disorders (Goessl, Curtiss, & Hofmann, 2017; Lehrer & Gevirtz, 2014; Lehrer et al., 2020). Importantly, it decreases craving among inpatient and outpatient individuals with SUD (Alayan et al., 2018, 2019; Eddie et al., 2014, 2018; Penzlin, Siepmann, Illigens, Weidner, & Siepmann, 2015) and improves affect in non-SUD samples (Steffen, Austin, DeBarros, & Brown, 2017). There are, however, implementation challenges in practitioner-provided HRVbf as it is labor, resource, and time intensive (Lehrer & Gevirtz, 2014; Lehrer et al., 2000, 2013; Song & Lehrer, 2003).

Deconstructed to its most basic component, HRVbf is a slowed rhythmic breathing technique that paces breath with the natural cycle of the heart-brain feedback loop (i.e., the heart rate baroreflex), modulates blood pressure, and integrates cardiovascular signals in cognitive-emotional experience (Song & Lehrer, 2003; Vaschillo, Vaschillo, Buckman, Pandina, & Bates, 2011). Given that resonance in the baroreflex occurs between 0.075 and 0.12 Hz in humans (Vaschillo, Vaschillo, & Lehrer, 2006), voluntarily pacing one’s breath at 0.1 Hz or 6 breaths per minute, referred to as resonance breathing, provides a proxy for HRVbf. Both HRVbf and resonance breathing generate an immediate cardiovascular response observable as large, low frequency oscillations across the cardiovascular system and brain (Buckman et al., 2018; Pagaduan, Wu, Kameneva, & Lambert, 2019; Steffen et al., 2017; Vaschillo, Vaschillo, Buckman, Pandina, & Bates, 2012). Such oscillations are considered healthful because they expand an individual’s physiological capacity to adaptively respond to challenges (Alayan, Eller, Bates, & Carmody, 2018; Paccione & Jacobsen, 2019; Steffen et al., 2017). Resonance breathing has the advantage of being a body-focused intervention that can be self-administered on demand, which may be particularly useful for dampening arousal as the craving state arises, potentially also allowing one to access cognitively-mediated techniques learned in treatment.

This paper reports an 8-week self-administered resonance breathing app intervention in treatment-seeking women with SUD as an adjunct to treatment as usual. In contrast to the study of post-treatment outcomes, which is often limited to one or two follow-up time points, this work focused on weekly assessment of outcomes defined as clinically relevant states (craving, affect) during treatment. Generalized estimated equations (GEE) modeled the momentary processes of brief episodes of resonance breathing on levels of craving, and positive and negative affect, during the 8-week intervention as a function of randomized intervention group and frequency of app use. We hypothesized significantly lower levels of craving and negative affect as well as higher levels of positive affect among participants randomized to the resonance breathing group relative to a sham breathing group. More frequent app use was expected to further enhance these outcomes on an individual level.

2. Methods

2.1. RCT design

The Project IMPACT In-the-Moment Protection from Autonomic Capture by Trigger design used a parallel-assignment randomized clinical trial (NCT02579317) to test whether self-administered, in-the-moment, resonance breathing would improve outcomes for women receiving treatment for SUD. Urn randomization assigned participants to either the active or sham breathing intervention to maximize the probability of balanced groups with regard to important prognostic characteristics (age 8–30, >30 years; AUD diagnostic criteria met vs. not met) and to preserve unpredictability/allocation concealment. The protocol was approved by Rutgers Arts & Sciences Institutional Review Board for the Protection of Human Subjects Involved in Research. The full trial design details are presented in Supplemental Materials. Herein, we focus on trial elements that informed the current hypotheses.

2.2. Participants

Participants were recruited between November 2015 and March 2020 from a community outpatient substance use treatment facility that offers a continuum of care for women. This client-centered facility uses evidence-based treatment approaches that optimize clinical care for women and their children including seeking safety (Najavits, 2007), motivational interviewing (Miller & Rollnick, 1991), and child parent psychotherapy (CPP) (Lieberman, Van Horn, & Ippen, 2005). Consecutive admissions to the program were invited to take part in an 8-week paced breathing study with two arms: 6 breaths per minute [resonance breathing intervention group(RB)] or 14 breaths per minute (sham breathing control). Inclusion criteria included age between 18 and 65 years and not pregnant. Women who qualified as having a current or lifetime substance use disorder (alcohol included) but did not have recent use were not excluded from the study. Six women (10%) had achieved more than 30 days of abstinence. Women who exhibited severe mental health symptoms did not qualify for the intensive outpatient program; no further psychiatric criteria were applied. Due to the COVID-19 pandemic, three participants were discontinued prior to intervention; the trial was then terminated prior to full sample recruitment.

Study allocation can be seen in the CONSORT diagram, Fig. 1. Of the 77 women who were randomized, daily app use data were missing for 20 participants who either failed to provide their phone for data uploads or dropped out of the study/treatment prior to app use data being collected. Partially missing substance use data, used as covariates, were imputed for five participants. The reportable sample included 57 women.

Fig. 1.

Consolidated Standards of Reporting Trials (CONSORT) flow diagram.

2.3. Recruitment

An exemplar timeline of study participation is presented in Fig. 2. At initial contact, all participants provided written informed consent. Participants then completed demographic and health screening information. During Week 2, a trained, graduate-level clinical researcher administered the alcohol and substance use disorder sections of the Structured Clinical Interview for DSM-5 [SCID-5 (First, Williams, Karg, & Spitzer, 2015) to verify these diagnoses and the Mini International Neuropsychiatric Interview (MINI) 7.0 (Sheehan, Lecrubier, Harnett-Sheehan, Amorim, Janvas, & Weiller, 1998) to assess psychiatric comorbidities. The following week, the researcher returned to conduct a 90-day TimeLine Follow Back in-person interview [TLFB (Sobell & Sobell, 1992) that assessed alcohol and other drug use (opiate, stimulant, nicotine, cannabis, hallucinogens) and craving [Penn Alcohol Craving Scale (PACS) (Flannery, Volpicelli, & Pettinati, 1999). An in-laboratory session was then scheduled.

Fig. 2.

Timeline of study involvement.

2.4. Pre-intervention laboratory phase

During the laboratory session, participants were given a light lunch and completed questionnaires including the PACS and the TLFB for the intervening time since last visit. Participants then completed a series of neurocognitive and physiological assessments that are not reported here (see Supplemental Materials).

2.5. Intervention

Upon completing the pre-intervention laboratory session, participants were given an iPhone programmed with CameraHRV ((c) Marco Altini, Amsterdam, Netherlands), an app that uses photoplethysmogram (PPG) in combination with a breathing pacer. Participants in both groups were instructed on how to open the app, enter the reason that prompted their app use, place their index finger over the camera lens to capture beat-to-beat cardiovascular data, and follow the app’s pacer (inhaling as the vertical breathing bar moved up and exhaling as it moved down). The pacer was preset at 6 breaths per minutes for the RB group and 14 breaths per minute for the sham group. Randomization was double blind; iPhone app programming (RB vs. sham) was conducted by one unblinded researcher. Participants were asked to use their app for 5 min any time they anticipated or experienced a trigger and/or any other reason that might encourage them to drink or use drugs (see supplement for flyer on trigger identification). In the event no such situations were encountered, participants were asked to use the app at the end of the day for 5 min.

Participants were engaged in the intervention phase of the study for ~ 8 weeks. Research personnel met with participants weekly to upload iPhone app use data and collect self-report measures. Participants completed the Positive And Negative Affect Scale (PANAS) (Watson, Clark, & Tellegen, 1988) to measure affect and were readministered the PACS and TLFB (since last visit).

The primary variables for the current study were collected during this phase. Because the majority of participants met criteria for AUD and at least one other substance, craving was calculated as the maximum PACS score between the alcohol and drug versions for each of the 8 weeks. Weekly positive and negative PANAS subscale scores were calculated for affect outcome variables. Number of app use episodes extracted from the iPhone app log quantified treatment dose.

2.6. Statistical analysis

Analyses were conducted using SAS 9.4 (Cary, NC: SAS Institute Inc). Demographic and substance use comparisons of the RB and sham groups were conducted using ANOVA and chi-square.

Craving, positive affect, and negative affect were separately fitted with Poisson regression models using a Generalized Estimated Equation (GEE) with robust standard errors via the exchangeable working correlation matrix to account for correlation among observations within a subject over time (Zeger & Liang, 1986). GEE is a statistically powerful population-level test with modified variance estimators that preserve type 1 error with a recommended sample size of N ≥ 50 (Liang & Zeger, 1986; Wang, Kong, Li, & Zhang, 2016). Primary predictors were main and interactive effects of number of app use episodes per week (time-varying fixed effect) and intervention group (time invariant fixed effect; RB, sham). Correlations across the repeated measures of week over subjects were modeled. Week was not included as a predictor in the model as our primary interest was in the overall relationship between app use and outcome measures over the course of an 8-week intervention as opposed to week-to-week stability of the relationships. The exchangeable working correlations were 0.355, 0.454, and 0.441 for craving, positive affect, and negative affect, respectively. Demographic covariates included age and race. Number of substances used and daily cigarette use, both pre-study (90 days) and concurrent with the intervention were included as substance use covariates. The PACS craving score collected prior to the intervention (average between weeks 2–3 of the study) was a baseline covariate in the craving model. The PANAS was not collected prior to intervention; thus, no baseline affect measure was available.

For each of the three models, Cook’s d was calculated to identify possible influential data points (Cook, 1977; Preisser & Qaqish, 1996). One outlier present in the craving model was removed. No influential outliers were present in the affect models. Effects are described in-text with p-values and 95% confidence intervals (95 CI). Wald’s chi square, model fit p-values, and parameter estimates (unstandardized beta weights) can be found in Table 1.

Table 1.

GEE Model Statistics.

Model Term	Wald’s X2	X2	Beta	Beta
		p-value		p-value
DV: Craving
Group	3.02	0.082	0.297	0.049
App use frequency	6.26	0.012	0.068	<0.001
App use frequency X Group	5.34	0.021	−0.06	0.008
Covariates:
Baseline Craving	16.5	<0.001	0.089	<0.001
Age	0.68	0.409	0.009	0.356
Race	3.59	0.31	−0.033	0.658
Cigarettes/day pre-treatment	0.36	0.547	−0.009	0.005
Cigarettes/day treatment	1.29	0.256	0.0182	0.226
N substances (pre-treatment)	5.85	0.016	−0.246	0.005
N substances (treatment)	5.93	0.015	0.1721	0.003
DV: Positive Affect
Group	0.28	0.598	−0.034	0.592
App use frequency	0.56	0.453	−0.003	0.5
App use frequency X Group	2.63	0.105	0.012	0.101
Covariates:
Age	0.37	0.544	0.002	0.539
Race	0.78	0.679	−0.033	0.658
Cigarettes/day pre-treatment	2.09	0.149	0.011	0.097
Cigarettes/day treatment	2.72	0.099	−0.011	0.079
N substances (pre-treatment)	0.77	0.381	−0.03	0.387
N substances (treatment)	0.65	0.42	−0.03	0.414
DV: Negative Affect
Group	4.94	0.026	0.171	0.009
App use frequency	0.68	0.409	0.007	0.116
App use frequency X Group	1.01	0.315	−0.008	0.305
Covariates:
Age	0.73	0.392	−0.003	0.391
Race	6.43	0.04	−0.114	0.201
Cigarettes/day pre-treatment	0.09	0.769	−0.003	0.767
Cigarettes/day treatment	2.55	0.111	0.013	0.09
N substances (pre-treatment)	0.04	0.836	−0.011	0.838
N substances (treatment)	0.51	0.475	0.045	0.457

3. Results

3.1. Pre-intervention group differences

As shown in Table 2, there were no statistically significant differences in demographic, clinical, or substance use variables, or craving levels prior to intervention, between women randomized to the RB and sham groups. Table 3 shows demographic, clinical, and substance use variables for the final sample and the intent to treat sample.

Table 2.

Final Sample Characteristics by Randomized Group.

	RB	sham	t-test/X-square
	(n = 28)	(n = 29)
Demographics (N)
Age (mean years)	34.54 (8.01)	32.59 (9.00)	t(55) = 0.86
Race/Ethnicity**			X2(2) = 6.51
Black	2	10
White	24	17
Other	2	2
Hispanic	1	2
Education (mean years)	12.55 (1.74)	12.67 (1.95)	t(50) = −0.22
Substance Use (N)
Alcohol Use Disorder only	6	4	X2(1) = 1.46
Substance Use Disorder only	11	9	X2(1) = 1.15
Comorbid AUD/SUD	11	16	X2(2) = 2.79
Cigarettes/Day (90D pre-intervention mean)	13.09 (6.31)	9.59 (7.00)	t(33) = 1.45
Any substance use (90D pre-intervention mean)	26	25	X2(1)= 0.670
Cigarettes/Day (intervention)	7.77 (6.84)	8.07 (7.43)	t(41) = 0.01
Any substance use (intervention)	8	9	X2(1)= 0.087
Baseline Craving (mean)	10.89 (8.12)	8.85 (5.21)	t(47) = 0.94
Psychiatric Comorbidities (N)
Diagnosis	19	22	X2(1)= 0.391
Major Depressive Disorder	9	13
Bipolar Disorder (Type I and II)	3	0
Panic Disorder	5	7
Agoraphobia	3	3
Social Anxiety Disorder	3	3
Obsessive Compulsive Disorder	2	0
Posttraumatic Stress Disorder	4	7
Generalized Anxiety Disorder	4	1

Hispanic responders = Mexican, Dominican, Honduran, Puerto Rican

^**The sham group consisted of a significantly greater proportion of Black participants than the RB group (p = 0.039). No other significant differences on characteristics were present between randomized group

Table 3.

Intent to Treat Sample Characteristics.

	Enrolled but not randomized (n = 30)	Randomized but Discontinued (n = 20)	Final sample (N = 57)	ANOVA/ X2
Demographics
Age (mean years)	31.07 (7.43)	29.78 (8.22)	33.54 (8.51)	F(2,101) = 1.81
Race (N)				X2(4) = 5.58
Black	11	3	12
White	17	14	41
Other	2	3	4
Hispanic	1	3	3
Education (mean years)	11.15 (3.58)	12.11 (2.65)	11.63 (3.28)	F(2,99) = 0.48
Substance Use (N)a
Alcohol Use Disorder only	2	3	10	X2(2) = 3.80
Substance Use Disorder only	8	7	20	X2(2) = 3.47
Comorbid AUD/ SUD	11	9	27	X2(4) = 0.834
Cigarettes/Day (pre-intervention mean)	9.94 (7.43)	8.35 (7.04)	11.21 (6.84)	F(2,72) = 0.93
Using Alc/Sub (pre-intervention mean)	28	20	50	X2(2) = 2.80
Cigarettes/Day (intervention)	-	6.06 (6.06)	7.92 (7.08)	F(1,66) = 0.71
Using Alc/Sub (intervention)	-	6	18	X2(1) = 0.0001
Psychiatric Comorbidities (N)
Diagnosis	14	11	41	X2(2) = 6.37**
Major Depressive Disorder	6	3	22
Bipolar Disorder (Type I and II)	2	1	3
Panic Disorder	6	3	12
Agoraphobia	4	3	6
Social Anxiety Disorder	4	3	6
Obsessive Compulsive Disorder	0	0	2
Posttraumatic Stress Disorder	2	4	11
Generalized Anxiety Disorder	2	0	5

Hispanic responders = Mexican, Dominican, Honduran, Puerto Rican

^**Women in the final sample were more likely to have comorbid mental health diagnoses than women who were enrolled but not randomized

^a10 women who were enrolled but not included in the final sample did not meet criteria as having either a current or lifetime AUD or SUD

3.2. App use and outcome measures

Women used their app an average of 3.7 times per week (SD = 2.37, range 0–21). There were no RB/sham group differences in weekly (t(55) = −0.28, p = 0.78) or total (t(55) = 0.25, p = 0.81) app use.

Mean craving levels in the full sample across the 8-week intervention can be seen in Fig. 3. Table 1 shows the GEE model fit statistics. Covariates including baseline craving and number of substances used prior to and during treatment were associated with craving levels during the intervention. A positive relationship between app use and craving was observed (95 CI: 0.037, 0.099, p < 0.001) indicating that participants who reported higher levels of craving also used their app more frequently. A significant group X app use interaction indicated that frequent app use during the intervention phase was associated with lower craving levels in the RB group relative to the sham group (95 CI: − 0.002, − 0.593, p = 0.008). Decomposition of the interaction revealed that frequent app use in the sham group was associated with higher levels of craving during the intervention (95 CI: 0.030, 0.092, p < 0.001), whereas frequent app use in the RB group was not (p = 0.38). The mesh surface plot in Fig. 4 illustrates the low stable craving levels (indicated by deep blue region) associated with more frequent app use in the RB group, while in the sham group craving levels remain elevated, even when the sham breathing app was used frequently.

Fig. 3.

Sample mean for A) craving, and B) positive and negative affect across study involvement. Weeks 2 and 3 served as a baseline, intervention was ongoing weeks 4–11, and post-test was conducted during week 12. Both positive (dashed line) and negative (solid line) affect were within the normative range for a healthy population but wide variability is observed. Error bars are standard deviation. Note: PANAS was not collected at weeks 2, 3, or 12.

Fig. 4.

Mesh surface plots for craving levels across weeks 4–12 between the RB (left) and sham (right) groups. Low stable levels of craving are observed for frequent users of the app in the RB group. In contrast, frequent users of sham show elevated, volatile craving levels across all weeks.

Fig. 3 shows mean positive and negative affect over the 8-week intervention phase for the whole sample. Women in the RB group generally had higher negative affect (M_RB = 22.89, M_sham = 21.69, p = 0.009, 95 CI: 0.043, 0.299), but there was no association with frequency of app use. Negative affect is displayed as a mesh surface plot across week and frequency of app use (Fig. 5). The plot resembled that of the craving data in that there was a large amount of variability in the RB group with low stable levels of negative affect (deep blue region) among those who used the app frequently in the RB group, although overall group mean levels remained higher than the sham group. There were no main or interactive effects of app use or group on positive affect scores.

Fig. 5.

Mesh surface plot for negative affect across weeks 4–11 for RB (left panel) and sham (right panel). Similar to craving scores, low, stable levels are observed in the frequent users of the RB app, whereas frequent use of the sham app was associated with more volatile patterns. However, group means show that women in the RB group consistently had elevated negative affect scores relative to the sham.

3.3. Post-hoc analysis within race

Although this study was not powered to conduct within-race analyses, a post-hoc examination was performed to determine whether patterns were consistent across the available racial groups (White n = 41, Black n = 12, Hispanic n = 3, Middle Eastern n = 1). Too few Hispanic and Middle Eastern participants were enrolled to examine within and between effects of resonance breathing in these groups. Table 4 shows fit statistics for the two post-hoc GEE models conducted separately. Similar patterns of the group by app use interaction effect on craving within Black (95 CI: − 0.097, − 0.012, p = 0.013) and White participants (95 CI: − 0.115, − 0.014, p = 0.012) were observed.

Table 4.

Model Fit Statisitcs for Post-hoc Examination of Race-related Differences.

Model Term	Wald’s X2	X2 p-value	Beta	Beta p-value
DV: Craving (Black Women)
Group	1.55	0.21	1.62	0.04
App use frequency	0.46	0.5	0.041	0.072
*Group	2.82	0.03	−0.06	0.013
DV: Craving (White Women)
Group	4.6	0.03	0.617	0.007
App use frequency	6.53	0.01	0.084	<0.001
*Group	4.07	0.04	−0.07	0.012

4. Discussion

An escalation in craving during outpatient treatment for SUD is common as individuals are not removed from the triggers in their environment (Bordnick & Schmitz, 1998) that can unexpectedly instigate craving states (Buckner, Crosby, Silgado, Wonderlich, & Schmidt, 2012; Skinner & Aubin, 2010). The development of techniques that mitigate in-the-moment experiences of craving is therefore imperative for both short- and long-term recovery. This study is the first to show that self-administration of 5-minute resonance breathing episodes, often performed in the face of triggers, was protective against escalations in the craving typically experienced by women receiving outpatient SUD treatment. Specifically, frequent employment of the resonance breathing app was related to low, stable levels of craving during the 8-week intervention period, whereas frequent users of the sham app showed increases in craving.

The finding that episodic resonance breathing affects time-varying craving states extends prior studies that found craving reductions following multi-session, practitioner-administered courses of HRVbf (Alayan, Eddie, Eller, Bates, & Carmody, 2019; Eddie, Conway, Alayan, Buckman, & Bates, 2018; Leyro, Buckman, & Bates, 2019; Teeravisutkul, Chumchua, Saengcharnchai, & Leelahanaj, 2019). Craving involves the sudden onset of hyperarousal that interferes with emotional states and impairs cognition; such shifts may instigate automatic drug-seeking behavior. Mitigation of craving after a session of resonance breathing or HRVbf may be linked to the modulation of hyperarousal that occurs instantaneously and persists for a brief period afterward (Bates, Lesnewich, Uhouse, Gohel, & Buckman, 2019; Buckman et al., 2018; Prinsloo et al., 2011, 2013). As individuals dampen physiological manifestations of craving, they may be able to better regulate emotional states and access cognitive tools to avoid acting on craving states.

The use of a sham breathing control provides evidence that resonance breathing does not function solely as a distractor. Paced breathing, even outside the resonance frequency, may divert focus from uncomfortable or anxiety-provoking states (Peretz & Gluck, 1999; Rezai, Goudarzian, Jafari-Koulaee, & Bagheri-Nesami, 2017). However, if efficacy of resonance breathing was attributable to distraction, then the sham breathing intervention should have conferred similar effects. The physiological mechanisms and the timescale by which resonance breathing and HRVbf impart clinical benefit in individuals with substance use and other mental health disorders are still not fully defined (Lehrer & Gevirtz, 2014), but are generally assumed to accrue cumulatively over time. Yet, the acute effects of resonance breathing and HRVbf on physiological and psychological processes are well established in laboratory-based studies. Cardiac muscle, vasculature, and afferent neural traffic respond instantaneously to a five-minute episode of resonance (Fonoberova et al., 2014). Brief episodes of resonance breathing and/or HRVbf immediately increase HRV and baroreflex sensitivity (Buckman et al., 2018; Lehrer & Gevirtz, 2014; Lehrer et al., 2020), and proximally decrease performance anxiety (Wells et al., 2012), improve cognitive performance (Laborde et al., 2021; Prinsloo et al., 2011; Wells et al., 2012); and alter neural reactivity to visual alcohol cues (Bates et al., 2019). The present study supports the potential utility of brief episodes of resonance breathing (and HRVbf) as a just-in-time intervention to dampen craving.

This is clinically significant in the context of the persisting need for greater health care accessibility. E-health smartphone apps are cost-effective and practical; a larger proportion of the United States population has access to cell phones than healthcare (Keisler-Starkey K, Bunch L. Health Insurance Coverage in the United States: 2019, 2019; Pew Research Center. Demographics of Mobile Device Ownership and Adoption in the United States [Internet]. Pew Research Center: Internet, Science, Tech. 2019, 2021) and smartphone penetration is growing worldwide. Early trials support health-based technology in modifying a variety of behaviors related to weight and stress management, and medication adherence (Balk-Møller, Poulsen, & Larsen, 2017; Harrer et al., 2018; Honary, Bell, Clinch, Vega, Kroll, & Sefi, 2020; Kazdin, 2015; Keyworth, Hart, Armitage, & Tully, 2018; Ramsey & Montgomery, 2014; Weber, Lorenz, & Hemmings, 2019). Delivering components of empirically based therapeutic approaches such as cognitive behavioral therapy, brief motivational interviewing, and social cognitive therapy via e-health apps has been shown to decrease substance use in various populations (Fowler, Holt, & Joshi, 2016; Gustafson et al., 2014; Marsch & Borodovsky, 2016). The results of this study extend this work and augment support for delivering app-based resonance breathing practice in the treatment of SUD.

Recent work has highlighted the inefficacy of certain psychiatric medications for racial minority groups (Price, Bruce, & Adinoff, 2021). The lack of improved treatment outcomes for commonly used SUD medications in Black and non-White Hispanic individuals (López, Barr, Reid-Quiñones, & de Arellano, 2017; Ray & Oslin, 2009) emphasizes the need to examine how our evidence-based treatments affect subgroups within our samples and to consider alternative biobehavioral interventions that may improve treatment outcomes for underrepresented individuals. Although power was limited, and replication in a larger, more diverse sample is necessary, the present post-hoc analyses support resonance breathing as a beneficial treatment adjunct for Black and White women.

Women in both the resonance breathing and sham groups exhibited moderate levels of negative affect through the intervention, with the resonance breathing group scores being slightly elevated. In the absence of a baseline assessment or an interaction with app use, it is unclear if this can be attributed to randomization or use of the intervention. The inability to detect effects may have been related to the restricted score range and the absence of baseline affect covariates. It may also indicate that brief episodes of resonance breathing during or in anticipation of substance use triggers may not instigate notable changes to fluctuating affective states, or that resonance breathing-induced changes in affect function through a different mechanism and require different modeling strategies. Each of these possibilities should be explored in future work.

5. Strengths and limitations

This work builds upon the broader mechanisms of behavioral change literature, which emphasizes the dynamic shifts that occur across many aspects of recovery (Buckman et al., 2018; Chung et al., 2016; Kazdin & Nock, 2003; Kelly, Hoeppner, Stout, & Pagano, 2012). Craving is a time-varying experience that is highly influenced by environment (Drummond, Litten, Lowman, & Hunt, 2000; Skinner & Aubin, 2010; Tiffany & Wray, 2012) and requires a longitudinal perspective to detect notable within-subject variability in the natural environment. The intervention and analyses were both structured to provide the resolution needed to capture craving states within-subject, using up to 9 time points per participant and an e-health app that allowed us to track outpatients’ subjective internal craving states and their use of a tool to navigate through it. Our weekly in-lab assessment of craving provides greater temporal context than typical cross-sectional reports of craving, but could be improved upon through the use of ecological momentary assessment (EMA) data collection or app-administered assessments at the time of breathing practice; both of which would provide a more temporally precise relationship between app use and modulation of craving (Cleveland et al., 2021; Hartwell & Ray, 2018; Serre, Fatseas, Swendsen, & Auriacombe, 2015).

In considering the relative merits of resonance breathing versus HRVbf to mitigate momentary craving states, we note that individual variability exists in the resonance frequency [typically 0.075–0.12 Hz, (Vaschillo et al., 2006)]. This is accommodated to some extent during HRVbf but not resonance breathing, which instead trains breathing to the average resonance frequency of 0.1 Hz. The results of the present study align with other evidence that breathing at 6 breaths per minute engages the baroreflex, induces large cardiac oscillations, and promotes adaptive physiology and emotional functioning (Russo, Santarelli, & O’Rourke, 2017; Song & Lehrer, 2003; Steffen et al., 2017, 2021). Resonance breathing was chosen for the current study due to its ease of use and simplicity for implementation in clinical practice.

The recruited sample was from an understudied population of treatment seeking women who were parenting young children and often had co-occurring SUDs and mild-to-moderate mental health comorbidities. Some had achieved abstinence more than 30 days prior to study initiation. It is not known if an app-based intervention would be effective for those earlier in their recovery or with different symptom severities. Future work should examine who this technique most benefits and how its efficacy translates to those requiring more intensive treatment paradigms. The present results also suggest that those who report relatively higher craving levels at treatment entry may be most likely to use this e-health intervention and to benefit from its use.

The current trial was terminated prematurely due to COVID-19 and the target enrollment was not achieved. The current analytic plan for longitudinal data was selected, in part, for its flexibility to estimate effects with relatively smaller samples. This analysis plan provided some protection against inadequate power and provides insight into the mobility of craving between pre-post measures, although smaller effect sizes may not have been detectable with the current sample and replication is necessary. Due to the relatively small sample paired with short follow-up periods (1 and 3 months), the study was not sufficiently powered to analyze relapse rates, however, longitudinal improvements in craving, independent of changes in substance use behavior, may contribute to improved recovery (Laudet, 2011; Witkiewitz et al., 2017).

The a priori hypotheses of the current work were to evaluate the efficacy of an app-based resonance breathing intervention over a course of 8 weeks. Our analytic approach was selected to capture general changes in craving due to app use as opposed to how the relationship between app use and craving varies from week to week. The GEE model with a marginal covariance structure for longitudinal data takes into account the temporal effects of the repeated measures by way of modeling the individual variances of the outcome variable at each time point and the covariances/correlations among those time points, without including fixed effects of week. This approach allows us to leverage the full robustness of the data while not violating the assumption of independence across repeated time points.

Finally, while we speculate that the resonance breathing app may improve access to cognitively mediated tools learned in treatment by dampening physiological arousal, we acknowledge that some cognitive effort is needed to identify a craving state, open the app, and practice resonance breathing. Continued development of wearable biosensors to identify moments of physiological arousal and automatically alert the wearer could further remove the cognitive load needed to navigate the craving state.

6. Conclusion

Historically, the instantaneous, dynamic, and multi-system features of clinically relevant internal states such as craving have created challenges for standard treatment approaches. These same aspects make just-in-time interventions, like app-based resonance breathing, particularly suitable for modulating craving in daily life. We found that brief, self-administered episodes of resonance breathing inhibited craving escalations that often occur during outpatient treatment for SUD. The use of a resonance breathing app during hyperarousal may provide people with a readily available tool to “reset” their physiological state in the face of triggers and thus expand the window of time in which treatment-based techniques are accessible before the craving state progresses into drug-seeking behavior.