In natura heart rate variability predicts subsequent alcohol use in individuals in early recovery from alcohol use disorder

Abstract

Impairment in autonomic self-regulatory functioning reflected by reduced heart rate variability (HRV) is a common feature of alcohol use disorder (AUD) and is believed to heighten AUD relapse risk. However, to date, no study has explored associations between in natura HRV and subsequent alcohol use among individuals seeking AUD recovery. In this study, 42 adults in the first year of a current AUD recovery attempt were monitored for four days using ambulatory electrocardiogram (ECG), followed by 90 days of alcohol use monitoring using timeline follow-back. HRV indices (independent variables) reflecting autonomic neurocardiac engagement were calculated from ECG recordings. Alcohol use (dependent variable) was calculated from timeline follow-back and expressed as percent days abstinent (PDA). The sample was 73.81% White/European American, 19.05% Black/African American, 4.76% Asian, and 2.38% Other race/Mixed race. As predicted, higher parasympathetically mediated HRV and lower heart rate (HR) were associated with greater PDA over 90-day follow-up. Additionally, interactions between these measures and baseline PDA indicated higher parasympathetically mediated HRV and lower HR mitigated the deleterious positive association between baseline and follow-up alcohol use. Including factors known to influence alcohol use and/or HRV in the models did not meaningfully alter their results. Findings are consistent with psychophysiological theories implicating autonomic self-regulatory functioning in AUD treatment outcomes and suggest that select HRV indices may have utility as indicants of risk for alcohol use lapse in individuals in early AUD recovery. Findings provide theoretical support for HRV Biofeedback for this population, which exercises the psychophysiological systems that support self-regulation.

INTRODUCTION

Alcohol use disorder (AUD) is one of the world’s most prevalent psychiatric conditions, annually accounting for the loss of approximately 99 million disability adjusted life years and 2.8 million deaths (1). This massive burden is partly due to the relapsing nature of substance use disorders. Given the enormous personal and public health burdens of AUD, helping individuals achieve AUD remission is of the highest importance.

A large body of research has sought to identify individual factors that heighten risk of alcohol use lapses (i.e., alcohol use episodes), which in turn may lead to AUD relapse (i.e., a return of AUD symptoms). While numerous psychological, social, and biological risk factors have been identified, negative affect has emerged as a prominent risk factor of alcohol use lapses (2). AUD is known to give rise to anxiety and stress, as well as other forms of negative affect, which may have a recursive relationship with alcohol use motivation (3). Individuals with AUD commonly report resorting to alcohol or other drugs to ameliorate these aversive affective states, with substance use behaviors then being negatively reinforced (4).

The brain areas implicated in the development and maintenance of substance use disorder have marked functional overlap with the structures of the central autonomic network (CAN), a group of neurological structures that modulate visceral functions and help maintain homeostasis via the autonomic nervous system in response to changing situational demands (5). The capacity to regulate affect and behavior is theorized to be, in large part, a function of these integrated brain–body systems (6). While the CAN influences affect and behavior through numerous physiological systems, its effects on the cardiovascular system are especially pronounced and quantifiable using measures of heart rate variability (HRV), the subtle changes in the time-intervals between heart beats (7). Numerous indices of HRV may be calculated from the electrocardiogram (ECG) waveform, with most established measures primarily reflecting parasympathetic effects on the heart’s rhythm.

Since HRV is a marker of affective regulation operating outside of conscious awareness, it may add important, objective information about affective and behavioral regulatory capacity to complement self-report measures in the study of AUD (8). Additionally, because HRV is an objective physiological measure not subject to the limitations of self-report (e.g., under-reporting; over-reporting; Hawthorne effect), and is easy and cost-effective to measure, HRV indices may have clinical utility for identifying patient risk. This is especially important in the context of addiction and addiction recovery, as difficulty self-identifying affective states is a hallmark of substance use disorders.

Well-functioning physiological systems are typically characterized by a high degree of complexity and flexibility, while illness and psychopathology are characterized by a loss of complexity and flexibility (9). This is particularly true regarding HRV. While the sympathetic nervous system tonically innervates the heart, cardiodynamic complexity and flexibility is primarily driven by the parasympathetic nervous system, which serves to slow down and dynamically modulate the heart’s rhythm. Compared to healthy controls, individuals with AUD typically exhibit lower parasympathetically mediated HRV (10).

Activation of stress neurocircuitry commonly associated with parasympathetic neurocardiac withdrawal—often referred to as vagal withdrawal—has also been well described in substance use disorders (11). Though parasympathetic withdrawal, with or without a concomitant increase in neurocardiac sympathetic activation, is usually adaptive as an acute response to a stressor or challenge, it is problematic when this state is sustained, such as in response to chronic stress (12).

Reduced parasympathetic neurocardiac control may reflect CAN dysregulation leading to a loss of capacity to flexibly and effectively respond to interoceptive and exteroceptive stressors (13). For individuals with AUD and other substance use disorders, this may increase reliance on alcohol and other drugs to cope, while also exacerbating challenges self-regulating appetitive drive and inhibiting substance use behaviors (5, 8). Though still debated, theories around somatic markers (14), interoception (15) and neural hemodynamics (16) also speculate how diminished parasympathetic neurocardiac control might contribute to problems with affective and behavioral regulation.

Previously, Quintana and colleagues (17) observed a negative association between laboratory-assessed, resting, parasympathetically mediated HRV and alcohol craving, while Garland et al. (18) found that greater parasympathetically mediated HRV reactivity to alcohol cues was associated with greater odds of alcohol use lapse at 6-month follow-up. However, to date, no study has tested associations between HRV and subsequent alcohol use among individuals seeking AUD recovery with in natura ECG monitoring methods. If prediction of alcohol lapses was possible using ambulatory HRV assessment, HRV could be an objective indicant of risk used to inform treatment planning, while also opening the way to real-time lapse risk detection using smartwatches and fitness monitors which already include the hardware to monitor HRV.

The present study sought to address this research gap using a four-day, ambulatory, ECG monitoring paradigm followed by 90-day monitoring of alcohol use using timeline follow-back with individuals in the first year of a current AUD recovery attempt. We hypothesized that greater parasympathetic neurocardiac engagement indicated by higher scores on HRV measures wholly or primarily reflecting parasympathetic effects on the heart, as well as lower HR would be associated with greater percent days abstinent (PDA) over follow-up (i.e., less alcohol use).

We also explored if parasympathetically mediated HRV and HR moderated an expected association between baseline and follow-up alcohol use. We hypothesized that greater baseline alcohol use combined with lower parasympathetically mediated HRV and higher HR would predict lower follow-up PDA (i.e., greater alcohol use), while greater parasympathetically mediated HRV and lower HR would buffer this vulnerability, resulting in greater PDA over follow-up (i.e., less alcohol use). Finally, we checked if factors known to influence alcohol use and/or HRV including age, AUD recovery motivation, drinking goal, medications, respiration, anxiety, depression, and AUD severity influenced results.

METHODS

Participants

Participants were recruited from the greater Boston area. Inclusion criteria included: 1) Being 18-65 years of age, 2) meeting past-year DSM-5 AUD criteria, 3) endorsing a current goal of alcohol abstinence, 4) being in the first year of a current AUD recovery attempt, and 5) participating in outpatient AUD treatment or an AUD mutual-help program (e.g., Alcoholics Anonymous). Since acute alcohol withdrawal can affect HRV, participants were required to have at least two weeks of alcohol and other drug abstinence before enrolling in the study to minimize the influence of physiological withdrawal symptoms. Exclusion criteria included: 1) having cardiac arrhythmias or serious medical conditions that may affect HRV (e.g., cardiovascular disease), 2) taking medications that directly influence HRV (e.g., beta-blockers), and 3) having an active substance use disorder related to a drug other than alcohol in the past year.

Procedure

Participants completed a study intake session in which baseline measures were administered, and they were fitted with an ECG monitor. As part of an ecological momentary assessment (EMA) monitoring protocol in the study (not utilized in the present analyses), an EMA application was also installed on participants’ smartphone (for EMA results, see 19, 20). Participants completed three surveys daily in which they reported their current emotional states on a scale of 0-10 (e.g., nervousness; sadness; craving; happiness). Each survey appeared at a random time within one of three, 3-hour blocks (between 10am-1pm, 2pm-5pm, & 7pm-10pm), Participants were also instructed to self-initiate a survey in moments when they felt high levels of stress, alcohol craving, or felt at risk for alcohol use.

Participants also completed four days of ECG monitoring, which included completing random and self-initiated EMA surveys several times per day. Ninety days following the end of the four-day EMA monitoring period, participants were remotely assessed for past 90-day alcohol use. This research was approved by the Mass General Brigham institutional review board (IRB# 2016P001178).

MEASURES

Percent days abstinent (PDA) at baseline

Timeline follow-back (21) was used to record the number of standard alcoholic drinks consumed over the past 30 days prior to study intake. Previous research has demonstrated that the timeline follow-back is a valid measure of alcohol consumption (21). Baseline PDA was calculated as the percentage of abstinent days in the past 30 days.

Physiological Monitoring

Participants wore an eMotion Faros 180 ambulatory ECG monitor for four days, affixed using an elastic belt embedded with two ECG contact points worn around the chest (just under the pectoral muscles). Participants were encouraged to wear the device throughout the day without altering their behavior in any way. They were not asked to wear the device while asleep. Per accepted guidelines (22), ECG was sampled at 250Hz.

The eMotion Faros 180 ambulatory ECG monitor also includes a three-axial accelerometer used to assess movement. Accelerometry was sampled at 25Hz (23).

Drinking goal over follow-up

Because participants’ drinking goals could have changed from baseline (an abstinence goal at baseline was an inclusionary criterion for the study), at the 90-day follow-up, participants were asked to describe their alcohol use goal over the majority of the follow-up period. Participants chose from the following options: total abstinence, moderation, or no drinking goal, coded as 0, 1, 2 respectively.

Follow-up percent days abstinent

Timeline follow-back (21) was used to record the number of standard alcoholic drinks over the 90 days following the end of the EMA monitoring period, which ran for two extra days following the four-day ECG monitoring period. The 90-day follow-up was conducted using online forms and, if necessary, a concurrent phone call. Follow-up PDA was calculated as the percentage of days, in the past 90 days, in which no alcohol was consumed. In this paper, we define alcohol use lapses as episodes of alcohol consumption, which may or may not relate to AUD relapse (i.e., a return to active AUD symptomology; not measured in this study).

Control measures

Because factors that commonly covary with drinking and HRV might have been influencing study findings, after completing the initial analyses, we reran the study models including a number of additional variables as a check. These included: 1) Age; which is typically negatively associated with HRV. 2) Alcohol use disorder (AUD) recovery motivation; which has the potential to influence drinking outcomes among individuals in AUD treatment. If HRV could predict subsequent alcohol use, independent of recovery motivation, this would have important implications for HRV’s clinical utility to predict alcohol use risk. 3) Drinking goal; which may have changed over the follow-up period. 4) Medications (yes/no); although we excluded individuals from the study who were taking medications that directly affect HRV like beta-blockers, we allowed other medications. 5) Movement; this study utilized ambulatory ECG recordings. Although ambulatory recordings have the benefit of greater ecological validity, a consequent challenge is that ECG is being recorded under variable motor conditions, which has the potential to confound results. 6) Respiration; parasympathetic measures of HRV like pNN50, RMSSD, and HF HRV are influenced by respiration, although to varying degrees (24). Though we did not directly measure respiration in this study, respiration frequency can be inferred from the ECG waveform. 7) Anxiety, depression, and AUD severity; these measures tend to be negatively associated with parasympathetic neurocardiac engagement (25, 26) and positively associated with alcohol use among those with AUD (27).

Electrocardiogram recording processing

Since a specific aim of this research project was to explore moment-level associations between cardiac indices and self-reported affect, five-minute epochs of ECG recording aligning with EMA surveys were analyzed, with HRV indices calculated for each epoch. As noted above, participants were asked to complete three random EMA surveys per day and self-initiate surveys when they were feeling high levels of stress, alcohol craving, or at risk for alcohol use. Because participants experiencing more stress, alcohol craving, or risk for alcohol use would likely have completed more surveys, potentially biasing results, only HRV statistics aligning with random EMA epochs were included in the present analysis. After excluding epochs associated with self-initiated EMA surveys, on average, there were 8.13 (SD= 2.50; range= 2-12) epochs of ECG recording utilized per participant out of a total possible 12 random EMA surveys delivered during ECG monitoring.

ECG waveforms from the specified epochs were manually examined for irregularities such as missed, misplaced, or ectopic beats using Kubios (28). These irregularities were manually corrected, and if necessary, filters were applied.

In our analysis, we investigated a range of HRV indices. Though HRV indices are typically correlated with one another, multiple measures provide complementary information that can provide a more complete picture of the phenomena under investigation, while also supporting comparisons to other studies that collectively utilize an array of HRV indices.

Calculated time domain indices included heart rate (HR), the standard deviation of normal-normal intervals (SDNN), root mean squared of successive differences (RMSSD), percent of adjacent normal-normal intervals differing by greater than 50 milliseconds (pNN50), and the HRV triangular index, which is the integral of the beat-to-beat interval histogram divided by the height of the histogram. In the frequency domain, high frequency HRV (HF HRV; 0.15–0.4Hz) and low frequency HRV (LF HRV; 0.04–0.15Hz) indices were calculated using power spectral density analysis.

Measures of HRV known to reflect cardiac parasympathetic autonomic activation include RMSSD, pNN50, and HF HRV (29). SDNN and the triangular index reflect overall HRV (29) in that they reflect parasympathetic activation, but also some sympathetic influences. LF HRV is less well understood, with current theories positing the LF HRV reflects parasympathetic control with varying degrees of sympathetic influence (30) and baroreflex effects (31) depending on conditions. Greater scores on these measures, as well as lower HR, suggest greater neurocardiac, parasympathetic engagement.

ANALYSES

First, to identify possible multivariate outliers, Mahalanobis distances were calculated using the indices HR, HF HRV, and LF HRV. In instances where, taken together, HR, HF HRV, and LF HRV suggested a participant epoch was a statistical outlier, all HRV indices from this participant epoch were excluded from the analyses. SDNN, RMSSD, pNN50, HF HRV, and LF HRV were found to be excessively skewed and kurtotic and were therefore logarithmically transformed. Then, to calculate individual participants’ mean HRV measures across the four-day ECG monitoring period, an average score for each index was calculated for each participant.

To determine if these HRV measures can operate as stand-alone predictors of subsequent alcohol use, we first explored bivariate associations between 90-day follow-up PDA and each HRV index in general linear models using the GLM procedure in SAS 9.4. PDA over 90-day follow-up was the dependent variable, and HRV indices were the independent variables.

Then, to explore if HRV moderates a potential relationship between baseline alcohol use and PDA over 90-day follow-up, we added baseline PDA and the interaction term for baseline PDA and each HRV index into the general linear models.

Effect sizes were estimated using R-squared, with an effect size considered small if R²= 0.01, medium if R²= 0.09, and large if R²= 0.25.

Finally, to check if factors potentially co-related to follow-up PDA and HRV could be influencing PDA/HRV relationships, we added the following variables to our multivariate models to explore how these factors influenced results: 1) Age, 2) AUD recovery motivation (assessed with the Commitment to Sobriety Scale; 32), 3) drinking goal over follow-up (abstinence versus moderation), 4) medications (yes/no), 5) accelerometer-derived movement, 6) ECG-derived respiration, 7) anxiety (measured with the State-Trait Anxiety Inventory state section; 33), 8) depression (measured with the Beck Depression Inventory II; 34), and 9) AUD severity (measured with the Alcohol Dependence Scale; 35).

RESULTS

The sample was 61.90% male and 38.10% female (N= 42) and ranged from ages 18-65 (M= 41.59, SD= 12.60). The sample was 73.81% White/European American, 19.05% Black/African American, 4.76% Asian, and 2.38% Other race/Mixed race. Three participants were lost to follow-up and physiology data were missing for two participants (one due to a lost ECG device, and one due to ECG device failure) leaving a total of n= 37 included in the analysis. Multivariate outlier testing identified two participant epochs as outliers, which were omitted from the analyses.

Participants reported taking the following medications (n participants taking each medication in brackets): citalopram [1], escitalopram [1], sertraline [2], paroxetine [1], fluoxetine [1], duloxetine [1], venlafaxine [1], bupropion [4], naltrexone [6], buprenorphine / naloxone [1], acamprosate [2], lamotrigine [1], mirtazapine [1], oxcarbazepine [1], gabapentin [4], trazadone [1], topiramate [1], disulfiram [1], phenelzine, [1], hydroxyzine [1], loratadine [1], cetirizine [1], simvastatin [1], omeprazole [2], pregabalin [1], finasteride [1], lisinopril [1], hydrochlorothiazide [1], terbinafine [1], darunavir / cobicistat [1], emtricitabine / tenofovir [1], ustekinumab [1], and etonogestrel [1].

Descriptive statistics

Baseline measures indicated the sample, on average, had severe AUD, with a mean Alcohol Dependence Scale score (36) of 23.80 (SD= 8.49). The sample was highly motivated to resolve their AUD, with an average Commitment to Sobriety Scale (32) score of 26.35 (SD= 4.65). The sample had a mean Beck Depression Inventory II (34) score of 17.60 (SD= 12.08), suggesting mild depressive symptoms, and a mean State Anxiety Inventory (33) score of 32.78 (SD= 7.35), indicating mild symptoms of anxiety.

Mean and standard deviation HRV statistics for the ECG monitoring period are presented in Table 1.

Table 1.

Average percent days abstinent from alcohol over 90-day follow-up period, average heart rate variability (HRV) with standard deviations (left column), and unadjusted, general linear model results with percent days abstinent (PDA) from alcohol over 90-day follow-up as dependent variable, and HRV index as independent variable.

	Mean (SD)	Model			Follow-up PDA
		df	F	R2	β	SE
Follow-up PDA (DV)	89.92 (20.37)
Heart Rate (bpm)	87.19 (11.56)	1, 35	10.02**	0.22	−0.84**	0.26
SDNN (ms) a	55.63 (20.79)	1, 35	4.20*	0.11	16.10*	7.86
Triangular Index	11.27 (5.32)	1, 35	4.39*	0.11	1.87*	0.90
RMSSD (ms) a	23.63 (14.17)	1, 35	7.16*	0.17	15.28*	5.71
pNN50 (%) a	6.63 (8.60)	1, 35	5.33*	0.13	7.73*	3.35
HF HRV (ms^2) a	251.63 (289.83)	1, 35	7.10*	0.17	7.11*	2.67
LF HRV (ms^2) a	780.71 (664.00)	1, 35	13.29**	0.27	10.97***	3.01

Notes. Percent days abstinent (PDA) shown in table reflects PDA over 90-day follow-up. SD= standard deviation (in parentheses); DV= dependent variable; bpm= beats per minute; ms= milliseconds; SDNN= standard deviation of all normal-to-normal intervals; RMSSD= root of the mean squared differences of successive normal-to-normal intervals; pNN50= percent of normal-to-normal adjacent intervals greater than 50ms; HF HRV= high frequency heart rate variability; LF HRV= low frequency heart rate variability. ^a Logarithmically transformed data used for analyses; untransformed means are presented in first column. df= degrees of freedom; β= unstandardized beta coefficient; SE= standard error. * p< .05, ** p< .01.

Over 30-day timeline follow-back immediately prior to enrolling in the study, of participants endorsing any alcohol use in that period (57.50% of sample), the average drinks per drinking day was 2.00 (SD= 1.86; range= 0.03-7.53). The overall sample included in the present analyses had 87.92 (SD= 18.85) PDA at baseline.

Over 90-day timeline follow-back over the follow-up period, of participants endorsing any alcohol use (45.00% of sample), the average drinks per drinking day was 1.26 (SD= 1.24; range= 0.05-4.40). The total sample included in the present analysis had a mean of 89.92 (SD= 20.37) PDA over the 90-day follow-up period. On average, AUD recovery motivation remained high at 90-day follow-up with a mean Commitment to Sobriety Scale score of 23.92 (SD= 7.15).

On average, participants’ overall HRV expressed by SDNN was within normal limits relative to published averages for healthy adults, however, parasympathetically mediated HRV expressed by RMSSD and pNN50 was below previously reported averages for healthy adult populations (37, 38).

Bivariate model results

Findings from the bivariate general linear models are reported in Table 1. As hypothesized, greater parasympathetic neurocardiac engagement indicated by higher RMSSD, pNN50, HF HRV, LF HRV, SDNN, and triangular index scores, as well as lower HR predicted greater PDA (i.e., less alcohol use) over 90-day follow-up. Observed effect sizes for the bivariate associations were medium to large.

Interaction model results

Findings from the general linear models including the baseline PDA × HRV index terms are reported in Table 2. Adding baseline PDA and this interaction term to the models did not markedly change the initial bivariate results. All main effects of HRV and HR on PDA over 90-day follow-up remained statistically significant.

Table 2.

General linear model results with percent days abstinent (PDA) from alcohol over 90-day follow-up as dependent variable, and heart rate variability (HRV), baseline PDA, and HRV index × baseline PDA interactions as independent variables.

	Model			Follow-up PDA
	df	F	R2	β	SE
Heart Rate (bpm)	3, 33	28.57****	0.72	−3.81****	0.63
Baseline PDA				−2.85****	0.66
Baseline PDA × Heart Rate				0.04****	0.01
SDNN (ms) a	3, 33	24.98****	0.69	114.91****	20.82
Baseline PDA				5.21****	0.92
SDNN × Baseline PDA				−1.18****	0.24
Triangular Index	3, 33	18.86****	0.63	10.41***	2.36
Baseline PDA				1.93****	0.35
Triangular Index × Baseline PDA				−0.11***	0.03
RMSSD (ms) a	3, 33	24.51****	0.69	77.24****	14.09
Baseline PDA				2.87****	0.47
RMSSD × Baseline PDA				−0.79****	0.16
pNN50 (%) a	3, 33	24.68****	0.69	47.50****	8.56
Baseline PDA				1.25****	0.16
pNN50 × Baseline PDA				0.49****	0.10
HF HRV (ms^2) a	3, 33	25.48****	0.70	36.17****	6.45
Baseline PDA				2.23****	0.35
HF HRV × Baseline PDA				−0.37****	0.07
LF HRV (ms^2) a	3, 33	30.22****	0.73	39.01****	6.58
Baseline PDA				2.85****	0.48
LF HRV × Baseline PDA				−0.39****	0.08

Notes. PDA= percent days abstinent; bpm= beats per minute; ms= milliseconds; SDNN= standard deviation of all normal-to-normal intervals; RMSSD= root of the mean squared differences of successive normal-to-normal intervals; pNN50= percent of normal-to-normal adjacent intervals greater than 50ms; HF HRV= high frequency heart rate variability; LF HRV= low frequency heart rate variability. ^a Logarithmically transformed data used for analyses. df= degrees of freedom; β= unstandardized beta coefficient; SE= standard error. *** p< .001, **** p< .0001.

In all models, baseline PDA was positively associated with 90-day follow-up PDA, indicating greater alcohol use at baseline was associated with greater alcohol use over follow-up. Further, all models produced significant interaction terms with large effect sizes. Specifically, as hypothesized, greater baseline alcohol use in combination with lower RMSSD, pNN50, HF HRV, LF HRV, SDNN, and triangular index scores, as well as higher HR predicted lower PDA over follow-up (i.e., greater alcohol use; Figures 1 & 2). Conversely, when greater baseline alcohol use was coupled with higher RMSSD, pNN50, HF HRV, LF HRV, SDNN, and triangular index scores, as well as lower HR, the positive association between baseline and follow-up alcohol use was attenuated, such that these participants drank less over follow-up even though they may have been drinking in greater amounts at baseline.

Figure 1.

Interaction plots showing that parasympathetic neurocardiac engagement reflected by parasympathetic measures of heart rate variability (HRV) moderates the relationship between baseline percent days abstinent (PDA; independent variable) and 90-day follow-up PDA (dependent variable). Individuals demonstrating lower parasympathetic neurocardiac engagement characterized by lower RMSSD, pNN50, and HF HRV were more likely to continue drinking over follow-up in a pattern consistent with their baseline. Conversely, those exhibiting greater parasympathetic engagement (i.e., greater RMSSD, pNN50, and HF HRV scores) had greater PDA over follow-up (i.e., drank less), indicating that the positive relationship between baseline and follow-up alcohol use had been attenuated. Notes. RMSSD= Root of the mean squared differences of successive normal-to-normal intervals; pNN50= percent of normal-to-normal adjacent intervals greater than 50ms; HF HRV= high frequency heart rate variability

Figure 2.

Interaction plots showing parasympathetic engagement reflected by predominantly parasympathetic measures of heart rate variability (HRV) as well as heart rate (HR) moderates the relationship between baseline percent days abstinent (PDA; independent variable) and 90-day follow-up PDA (dependent variable). Individuals demonstrating lower parasympathetic neurocardiac engagement characterized by lower SDNN, triangular index, and LF HRV scores, as well as higher HR were more likely to continue drinking over follow-up in a pattern consistent with their baseline. Conversely, those exhibiting greater parasympathetic engagement (i.e., greater SDNN, triangular index, and LF HRV scores) and lower HR had greater PDA over follow-up (i.e., drank less), indicating that the positive relationship between baseline and follow-up alcohol use had been attenuated. Notes. SDNN= standard deviation of all normal-to-normal intervals; LF HRV= low frequency heart rate variability

Control measure checks

1) Adding age to our multivariate models did not substantively alter their results, nor was age associated with HR or HRV in our sample. 2) Neither baseline or follow-up Commitment to Sobriety Scale scores predicted PDA over follow-up (all p’s > .05), nor did having these measures in the model substantively change the initial results. 3) Notably, five participants changed from a goal of alcohol abstinence at baseline to alcohol moderation over the 90-day follow-up period, and had less PDA over follow-up, F(1, 35)= 42.47, p< .0001. While having a moderate drinking goal over follow-up was associated with less follow-up PDA in all models, adding this measure to the models did not meaningfully alter the initial results. 4) Adding medication to the models had a negligible impact on the associations observed. 5) Including accelerometer-measured movement in the models did not alter the initial findings in any meaningful way. 6) Including ECG-derived respiration in the models had negligible impact on the initial results. 7) Baseline State-Trait Anxiety Inventory (state section), Beck Depression Inventory II, and Alcohol Dependence Scale scores were not associated with follow-up PDA in our sample (all p’s > .05), nor did having these measures in the models markedly alter their initial results.

DISCUSSION

A growing body of research suggests that impairment in autonomic regulatory functioning reflected by lower parasympathetic neurocardiac engagement contributes to the persistence of substance use disorders and can impede addiction recovery (5, 10). While previous research has explored associations between parasympathetic neurocardiac engagement and precursors to alcohol use such as craving, there is a deficit of research testing direct associations between HRV parameters and subsequent substance use in individuals seeking recovery from substance use disorders.

The present investigation is one of the first studies to explore associations between HRV and alcohol use in individuals seeking recovery from AUD, and the first to demonstrate the predictive utility of in natura, ECG-derived HRV on alcohol use in this population. Additionally, results highlight the value of ambulatory psychophysiological monitoring in substance use and mental health research.

As hypothesized, we found a medium effect size, positive association between parasympathetic neurocardiac engagement and alcohol use over 90-day follow-up, with RMSSD, pNN50, and HF HRV accounting for 11-17% of the variance in follow-up PDA. This extends a large body of empirical work indicating impaired parasympathetic neurocardiac control in individuals with AUD (39). While previous work has shown an association between parasympathetic neurocardiac deficits and affective states like craving (e.g., 17, 40), this study is among the first to show that these deficits are associated with actual alcohol use. Importantly, HRV was found to be predictive of subsequent alcohol use regardless of participants’ AUD recovery motivation or drinking goal over follow-up.

Notably, we also observed large effect size bivariate associations between HR and follow-up PDA (R²= 0.22), and LF HRV and follow-up PDA (R²= 0.27). Given HR and LF HRV are partially under sympathetic control, this finding suggests a relationship between sympathetic arousal and subsequent alcohol use in this population, however, studies using gold standard impedance cardiography will be needed to tease out these possible effects.

Although more work is also needed to explain the mechanisms behind these observed HRV/alcohol use associations, these findings generally fit with theories such as the neurovisceral integration model (41) that posit that autonomic dysregulation reflected by reductions in parasympathetic neurocardiac engagement leads to difficulties regulating affect and behaviors. Preliminary work also suggests neural prediction errors might contribute to exaggerated or blunted cardiovascular reactions that undermine adaptive responses to challenge (42). In theory, diminished parasympathetic neurocardiac engagement might also contribute to problems regulating emotions and behaviors from the bottom-up (afferent pathways) through disruption of cortical processing of cardioceptive signals that influence attention and somatosensory perception (reflected by heartbeat-evoked potentials; 14, 43), as well as through impaired regulation of neural hemodynamics (16).

Chronic stress has been shown to have a negative association with parasympathetically mediated HRV (44). It is possible that study participants exhibiting lower parasympathetic neurocardiac engagement may have been experiencing greater subjective stress – a known risk factor for alcohol use among individuals seeking AUD recovery (45). However, a post hoc data-check showed no significant bivariate associations between average subjective stress or craving captured in the parent study’s EMA data, and our HRV measures (p’s> .05).

This highlights how central autonomic mechanisms operating outside of conscious awareness might confer risk for this population and aligns with allostatic load theory (46), which is strongly implicated in contemporary theories of addiction etiology and maintenance (5, 11). Allostatic load theory posits that chronic disengagement of the parasympathetic nervous system and/or chronic engagement of the sympathetic nervous system can deplete an individual of physical and psychological resources and diminish their capacity to cope with stressors, potentially leading to disease and/or the perpetuation of disordered states like addiction. In individuals with AUD, it is also possible that repeated cycles of intoxication and alcohol withdrawal lead to a loss of regulatory flexibility in the cardiovascular system characterized by reduced HRV, as well as impairment in critical afferent, body-brain feedback loops (e.g., the baroreflex) that provide physiological context to the CAN, further undermining CAN regulatory control (5, 15).

Our models including the interaction term of baseline PDA × HRV reinforced findings from the bivariate models, with bivariate follow-up PDA and HRV associations remaining statistically significant after controlling for baseline PDA and the interaction term. As expected, baseline PDA was a robust predictor of 90-day follow-up PDA. Notably, statistically significant effects for the interaction term in all models revealed that, as hypothesized, HRV indices moderated the relationship between baseline and follow-up PDA.

Specifically, when greater baseline drinking co-occurred with lower parasympathetic neurocardiac engagement, greater subsequent alcohol use occurred, such that participants appeared to continue drinking over the follow-up period in a pattern similar to their baseline. However, when greater baseline drinking co-occurred with greater parasympathetic engagement, follow-up drinking was attenuated. These effects remained statistically significant after controlling for age, commitment to sobriety, drinking goal over follow-up, the use of medications, movement, ECG-inferred respiration, and symptoms of anxiety, depression, and AUD severity. Importantly, these findings suggest that greater autonomic self-regulatory capacity might be buffering individuals against alcohol use in early recovery from AUD.

Taken together, the present findings are noteworthy because they indicate that HRV indices may act as stand-alone predictors of alcohol use in individuals seeking AUD recovery. It is probable this association would extend to other substance use disorders, which are driven by similar neurological and psychophysiological processes. This should be tested in future studies. Moreover, HRV predicted subsequent alcohol use despite very high baseline and follow-up commitment to sobriety in the sample (mean scores of 26.35 and 23.92 respectively out of a total possible score of 30).

There are three, key clinical implications for these findings. Firstly, given HRV can be quickly and easily measured in the clinic or field using basic equipment and even smartphone apps, HRV has potential as an objective indicant of AUD relapse risk that could inform clinical decision-making. Secondly, results speak in a preliminary way to the exciting prospect of just-in-time relapse prevention interventions driven by real-time HRV monitoring using existing smartwatches and fitness monitors. Finally, these findings add support for the use of HRV Biofeedback and resonance paced breathing with individuals seeking addiction recovery – closely related treatments that use rhythmic breathing to stimulate the respiratory sinus arrhythmia and baroreflex to engage autonomic regulatory processes that support affective and behavioral regulation (47).

HRV Biofeedback and resonance paced breathing have previously been shown to reduce craving in individuals seeking addiction recovery (48). Notably, in a clinical trial of HRV Biofeedback for substance use disorder, we observed that lower resting baseline HRV was associated with increases in craving over three weeks of inpatient treatment; however, this trend was mitigated in participants receiving HRV Biofeedback (49). Thus, this intervention appeared to dissociate the relationship between initial psychophysiological vulnerability and craving changes during inpatient treatment. These findings suggest that HRV Biofeedback may be an especially useful adjunctive treatment for those with greater autonomic regulatory impairment. The present findings, as well as the emerging literature on HRV Biofeedback for craving, suggest this intervention could also mitigate risk for alcohol use lapses among individuals seeking recovery from AUD by helping restore autonomic homeostasis and executive control and/or reducing autonomic allostatic load during aversive affective states.

Limitations and future directions

Despite the strengths of the current study, there are a number of limitations that warrant discussion. Limitations include: 1) A small sample size. Future studies seeking to replicate these findings should include larger samples. 2) Alcohol use was not assessed during the four-day ECG monitoring period. 3) This study focused on alcohol use and individuals with AUD. Though participants were excluded from the study if they had an active substance use disorder related to a drug other than alcohol in the past year, it is possible some participants used drugs other than alcohol over the follow-up period. 4) We did not assess tobacco use so were unable to control for this factor, which may have influenced HRV measures in unknown ways. 5) The reliability of the HRV triangular index is optimal when calculated with recordings longer than 5 minutes; the fact we aggregated multiple epochs of ECG recording when calculating mean HRV indices may have partially mitigated this issue. 6) Individuals identifying as White (74%) formed the majority of our sample with participants identifying as Black being under-represented (19%). Future studies should recruit larger, more racially balanced samples to support exploration of how race might influence AUD/autonomic relationships. This is particularly important since prior research has found that individuals identifying as Black have on average higher resting parasympathetic HRV than their White counterparts, yet this difference does not appear to buffer this population against health disparities (50). 7) Though on average our sample had severe AUD, because we required a minimum of two weeks of alcohol abstinence before individuals could participate in the study, it is possible we ended up excluding some people with more severe AUD who were not able to achieve this period of abstinence. 8) Because HRV was calculated using epochs of ECG recording aligning with random EMA surveys, and participants did not necessarily respond to every random survey, it is possible that these epochs were systematically different in unknown ways. 9) 90-day timeline follow-back was completed online. Though participants were offered phone support to complete their follow-up drinking log, this is not the most ideal method for collecting timeline follow-back information. 10) Our analytic approach included multiple comparisons that increased the chance of type-II error. 11) This study used mean-level HRV metrics, which may not fully capture time-variant dynamics in HRV and their relationship to moment-level alcohol use. Future studies should explore these time-varying effects.

These limitations are tempered by the study’s strengths, including high ecological validity associated with the use of a naturalistic sample of individuals seeking AUD recovery and in natura ECG monitoring used in combination with EMA.

CONCLUSIONS

This is the first study to explore and demonstrate a relationship between ambulatory HRV and subsequent alcohol use in individuals seeking recovery from AUD. It extends a large body of theoretical and empirical work suggesting autonomic regulatory impairment in individuals with substance use disorders increases risk for lapses. Findings provide preliminary evidence suggesting HRV has utility as an indicant of alcohol use risk that could inform clinical decision making, and potentially drive just-in-time relapse prevention interventions. Findings also reinforce the potential utility of interventions like HRV Biofeedback and resonance frequency breathing, which are known to reduce negative affect in individuals seeking recovery from substance use disorders, potentially by exercising the psychophysiological systems underpinning affective and behavioral regulation.