Alcohol is the most misused substance in the United States [1]. The types of alcohol consumption behavior constitute a spectrum, from low-risk alcohol use to risky drinking and then to various levels of alcohol use disorder (AUD) [2]. Risky drinking includes binge drinking and heavy alcohol use, which in total affects more than 80 million Americans [1]. In addition, AUD affects more than 15 million Americans [1].

Traditionally, the interventions for alcohol misuse have been focused on the vigorous side of the spectrum, focusing on inpatient or outpatient rehabilitation programs [3], medication [4], and peer-led programs such as Alcoholics Anonymous [5]. Recently, cultural and lifestyle movements such as the Sober Curious Movement have increased the awareness of healthy alcohol use [6]. A large population who are on the less vigorous spectrum of unhealthy alcohol use started to be curious about treatments for alcohol use, increasing the attention on interventions for nondependent risky drinkers [7,8]. Thus, interventions that facilitate the reduction of alcohol use and the ease of craving benefit a large population.

Existing digital interventions for reducing alcohol use focused on long-term behavior change and coping. These interventions include digital psychotherapy [9-14], mindfulness practices [15,16], SMS text messages [8,17-20], and chatbots [18,21]. Despite the success of these interventions in clinical trials, they cannot provide immediate craving relief. As a result, individuals are vulnerable during acute craving episodes and are susceptible to high relapse rates [22,23]. Although the raising of mobile sensing enables just-in-time intervention systems that can detect when the users are at risk for alcohol consumption [24,25], the obtrusive nature (ie, the disruption of ongoing activities) of existing digital interventions is misaligned with the nature of at-risk contexts. As such, the potential of digital interventions and just-in-time systems can be further enabled by interventions that can achieve in-the-moment effects with minimal attention and effort.

Mindless intervention is an emerging type of intervention in the human-computer interaction community. Such intervention is delivered through mobile or wearable devices and can achieve in-the-moment effectiveness with minimal attention and engagement, hence mindless. Mindless interventions are primarily developed for regulating stress and anxiety [26-32]. The 2 main mechanisms that enable mindless interventions to regulate stress and anxiety are entrainment (eg, offset heart rate biofeedback [27,28]) and emotion regulation (eg, guided breathing [26,29-31] and affective touch [32]). Entrainment-based methods can influence the user’s physiological state by mitigating the autonomic nervous system (ANS), which is vital in regulating how the body reacts to external and internal stimuli [33].

The integration of mindless interventions into just-in-time intervention systems has the potential to mitigate cravings in-the-moment. However, careful translational evaluations are required. All existing mindless interventions were evaluated in a general population. However, the physiology of those who perform risky drinking or have AUD, regardless of severity, is different from the general, healthy population [34]. Prolonged and excessive alcohol use leads to neuroadaptations [35]. Such adaptations cause people with unhealthy alcohol use to have elevated parasympathetic nervous activity in response to alcohol cues, a phenomenon that’s not observed in the general population [34,36-40].

In this study, we aimed to characterize the translational effectiveness of an anxiety-mitigating mindless intervention on alcohol craving. We conducted this study as a formative step on people who performed risky drinking before experimenting with people with alcohol dependency and who needed clinical treatment. Among all mindless interventions, we chose to evaluate offset heart rate biofeedback. Offset heart rate biofeedback is delivered as a subtle vibration on the wrist [27]. Controlled laboratory evaluations have validated its ability to alter the activity of the parasympathetic branch of ANS and to mitigate anxiety [28]. We chose this intervention because (1) it best represents the merit of mindless interventions as it is nonobtrusive and requires little to no attention from the users and (2) it can be delivered by Apple Watch, allowing for scalability and low social stigma associated with use.

We hypothesized that risky drinkers who receive offset heart rate biofeedback during and after alcohol cues would experience a lower level of physiological response, a lower level of craving, and a lower level of anxiety compared with those receiving no intervention. We conducted an out-of-laboratory, controlled study with 26 participants. We found that offset heart rate biofeedback significantly mitigated both the physiological response and the subjective craving in response to alcohol cues.

Our study provides the first empirical evidence demonstrating that a wearable device can unobtrusively mitigate alcohol cravings in real time. This initial evidence supports the effectiveness of offset heart rate biofeedback for risky drinkers. Based on our findings, we recommend future research replicate and build on this work to translate offset heart rate biofeedback into practice as wellness apps for promoting healthy drinking habits in risky drinkers or as digital prescriptions and just-in-time interventions for preventing relapse in individuals with AUD.

Participants are legally drinking adults (aged 21-65 years). Participants were recruited using institution-based networks, including Slack (Slack Technologies, LLC) channels, listserves, and physical flyers, as well as social media advertisements using Meta and X. Participants were also recruited from online communities such as Facebook (Meta Platforms, Inc) groups.

All participants were screened for at-risk alcohol use by the Alcohol Use Disorders Identification Test (AUDIT) [41]. Participants were included in the study if their AUDIT score was between 8 and 14 (hazardous or harmful alcohol consumption) at the time of screening. These are the chosen thresholds because this population exhibited a similar physiological response toward alcohol cues as those who have severe alcohol dependency [36,40], which was distinctive from healthy participants [39]. We did not include participants whose AUDIT score was 15 or above to avoid unnecessary harm to individuals with more severe alcohol misuse or individuals who were under clinical treatment.

Additional inclusion criteria were individuals need to (1) be aged between 21 and 65 years, (2) reside in the United States at the time of study, and (3) be iOS (Apple, Inc) users to install the intervention system. This age group was chosen because they are the typical adult age group who can legally consume alcohol in the United States. Adolescents and the older adult population have special needs in mitigating alcohol craving, which is beyond our interests. The exclusion criteria include individuals who were (1) abstaining from alcohol at the time of the study and (2) having additional physical or mental health conditions aside from hazardous alcohol consumption. These exclusion criteria were in place to avoid exposing at-risk individuals to alcohol triggers and to minimize confounding variables to the intervention.

The study used a between-subject design. All participants underwent the identical protocol. Participants were randomly assigned to the control or experimental group at the time of study. Sex, age, and ethnicity were matched among the 2 groups (Table 1). In total, 34 participants were recruited. Furthermore, 8 participants were not included in data analysis due to interruption during the study (n=3) and low physiological data quality (n=5). As a result, 26 participants (13 in each group) were included in the data analysis. Participants’ age ranged from 21 to 52 years (mean 29.31, SD 8.65 years for control group; mean 31.69, SD 9.87 years for experimental group). Participants’ race and ethnicity included African American (experimental: n=1), Asian (control: n=6, experiment: n=3), White (control: n=7, experiment: n=6), Latinx (experiment: n=2), and multiracial between White and Asian (experimental: n=1). All participants were cisgender and binary gendered (control: 7 female individuals and 6 male individuals; experimental: 8 female individuals and 5 male individuals).

The Institutional Review Board at Cornell University approved the study in March 2020 (IRB2009009824). We obtained informed consent from the participants once the participants were screened. Participants’ data were anonymized. Participants were compensated US $25 for completing the experiment and US $5 for returning the device.

We conducted a priori power analysis using G*Power [42] using the effect size determined by previous studies on offset heart rate biofeedback (d=1.07) [27]. We balanced the minimum allowed levels for type I and type II errors by choosing 1 – β = 0.8, following the clinical, preclinical, and laboratory study guidelines [43]. The power analysis yielded the sample size requirement for 1-tailed, 2-group comparisons as n=13 and N=26.

The experiments were conducted remotely between November 2020 and August 2022. We conducted an out-of-lab, between-group, controlled evaluation. All participants performed the identical experiment protocol during the study. Participants were randomly assigned into the experimental or control group, with considerations to match age, sex, and ethnicity.

Offset heart rate biofeedback relies on entrainment to be effective. Entrainment is the coupling and phase alignment between 2 oscillatory bodily systems [47]; it can be applied heart rate [48], brain waves [49], and movement [50]. Specifically, offset heart rate biofeedback uses a subtle vibration on the user’s wrist to entrain the user. The vibration is 30% slower or faster than the momentary heart rate [28]. In the initial controlled laboratory experiment, participants were deceived and informed that the offset heart rate represented their real heart rate [27]. In the follow-up, double-blind, controlled laboratory experiment, participants were informed that the offset heart rate feedback may or may not represent their real heart rate [28]. The results of the line of study showed that participants who received the slower heart rate feedback experienced lower self-reported anxiety and higher response in the parasympathetic branch of ANS compared with participants who received no intervention during and after stress stimulus.

We implemented the offset heart rate biofeedback on the Apple Watch following the engineering benchmark in the previous work [28]. The haptic cue is activated at a rate 30% slower than the momentary heart rate. For instance, if the individual’s heart rate is 90 beats per minute (bpm), then the Apple Watch would vibrate at 63 bpm. The upper bound of the tapping frequency is 65 bpm and the lower bound of the tapping frequency is 40 bpm [27,28]. These limits were applied to prevent the feedback from being too fast and arousal-inducing or too low such that it is natural. The tapping rate adapts to the user’s actual heart rate: if, in the previous example, the user’s heart rate drops to 72 bpm, the Apple Watch will follow along and reduce the tapping frequency to 50 bpm.

We developed a custom iOS and WatchOS app (Apple, Inc) to deliver the offset heart rate biofeedback. The app connects to an external heart rate monitor (Polar H10) through Bluetooth and streams the momentary heart rate to determine the frequency of offset heart rate biofeedback. We distributed the app through Apple’s TestFlight platform.

We collected three types of measures: (1) heart rate variability (HRV), (2) self-reported anxiety, and (3) self-reported craving.

We determined the normality of the metrics using the Shapiro-Wilk test and yielded nonnormal distributions. Thus, we treated all metrics as nonparametric variables. To assess the between-group differences in the negative baseline and positive baseline, we conducted the Mann-Whitney U Test.

All participants consumed alcohol within 7 days before the study. The median number of days those participants consumed alcohol before the study was 2 days for both groups. The median number of drinks participants consumed per day 7 days before the study was 2 drinks for both groups. As shown in Table 1, there were no significant differences between the control and experimental groups in AUDIT (P=.29), PHQ-9 (P=.15), and GAD-7 (P=.27). During the negative baseline, the 2 groups had no significant differences in RMSSD (P=.68), HF (P=.79), self-reported craving (P=.42), and self-reported anxiety (P=.62). During the positive baseline, the 2 groups had no significant differences in RMSSD (P=.98), HF (P=.87), perceived craving (P=.20), and perceived anxiety (P=.34).

We evaluated the effectiveness of offset heart rate biofeedback on HRV, self-reported craving, and self-reported anxiety in response to alcohol cues.

This paper evaluated the translational effectiveness of offset heart rate biofeedback (“the intervention”), a mindless intervention that had shown effect in mitigating anxiety, as an alcohol craving intervention. The results indicated that the intervention significantly lowered the physiological response to alcohol cues and reduced the self-reported craving during alcohol exposure among individuals who performed harmful and hazardous drinking (ie, risky drinkers).

The results from our study exhibited 3 main trends. First, participants who received the intervention experienced a steady level of parasympathetic activity (ie, HRV RMSSD and HRV HF) during alcohol exposure. In contrast, the participants who did not receive the intervention experienced elevated parasympathetic activity during alcohol exposure. Second, the self-reported craving followed the same trend as the physiological response. Third, the self-reported anxiety did not follow the same trend as the physiological response or the self-reported craving; neither alcohol exposure nor the intervention had a significant impact on self-reported anxiety.

The observations on the physiological response and self-reported craving are in alignment with the findings in previous studies—individuals who were moderately or heavily dependent on alcohol experienced elevated parasympathetic activity in response to alcohol cues; such elevation was higher than healthy drinkers [34,36-40]. A higher increase in cue-elicit parasympathetic activity is associated with a higher level of craving [37].

The trend in self-reported anxiety was not in alignment with previous studies. In the previous work that implemented offset heart rate biofeedback as an anxiety intervention [27,28,30], the intervention led to increased parasympathetic activity and lower self-reported anxiety. In contrast, the intervention implemented in our study led to a stable parasympathetic activity while having no effect on self-reported anxiety. This observation, although not aligned with previous studies, was sensible because this study did not manipulate alcohol craving through anxiety. It is possible that alcohol craving would lead to elevated anxiety [63]. However, such response was not observed in this population, possibly because their dependency on alcohol was moderate.

Based on the 3 observed trends, we hypothesized that the intervention did not mitigate craving by reducing the participants’ anxiety but by stabilizing the physiological response elicited by alcohol cues. The reduced change in parasympathetic activity could subsequently lead to a stabilized subjective experience of craving (as denoted by self-reported craving). Following this hypothesis, the previously observed effect on stress and anxiety could be that the entrainment signal stabilizes the stress response (ie, sympathetic activation in response to stress manipulation), as shown in higher parasympathetic activity in those who received the intervention [28] during stress manipulation. The little increase in self-reported anxiety levels during and after stress manipulation could be an effect of the stabilized physiological response, similar to what was observed in this study [27,28]. To verify this speculation on the mechanism behind offset heart rate biofeedback, future studies should replicate the study procedure with anxiety-elicited craving and compare the physiological effect with this study.

As a formative study, this study aimed to characterize the effect of offset heart rate biofeedback on cue-elicit alcohol craving in a controlled setting. The promising trends in reducing both physiological and self-reported cravings suggest new possibilities for digital and wearable interventions targeting alcohol misuse. Unlike existing digital interventions, offset heart rate biofeedback has the potential to directly reduce craving unobtrusively. Compared with digital psychotherapy, push messages, or chatbots, this intervention provides a more direct and immediate influence on craving. Compared with mindfulness exercises, this intervention requires much less engagement and attention from the user. As such, offset heart rate biofeedback is particularly impactful in day-to-day living and outpatient settings. For risky drinkers, offset heart rate biofeedback can be integrated into wellness apps to help control the urge to drink and promote healthier alcohol use. For individuals with AUD, offset heart rate biofeedback can serve as a digital prescription in outpatient settings, either as a standalone tool for immediate craving relief or integrated into just-in-time sensing systems to prevent relapse during high-risk situations.

This study raised new questions for future research to validate its generalizability and further investigate the mechanism of offset heart rate biofeedback. First, for risky drinkers, the results of this study served as the initial evidence of offset heart rate biofeedback in craving mitigation. Future studies should replicate its results using alternative cue-elicitation methods and test its effectiveness in naturalistic settings. Second, this study deliberately evaluated participants with risky drinking behaviors (ie, those who were not under the clinical definition of AUD). This limits the generalizability of the results in this study to the clinical population. Thus, future studies should replicate this study in the clinical population to understand whether this intervention is effective among individuals with AUD. Finally, replicating this study with anxiety-elicited craving can deepen the understanding of the mechanisms behind offset heart rate biofeedback. A more profound theoretical framework will bolster confidence in evaluating this intervention in a population with severe alcohol dependency.

This paper presents the translational evaluation of offset heart rate biofeedback, an iconic type of mindless intervention, extending its application from anxiety mitigation to alcohol craving mitigation. Through an out-of-laboratory, controlled, between-group evaluation with participants who practiced harmful or hazardous drinking, we discovered that offset heart rate biofeedback significantly reduced both physiological responses to alcohol cues and self-reported cravings during alcohol exposure. This paper marks the first empirical evidence that a wearable device can unobtrusively and immediately mitigate alcohol cravings, creating new opportunities for digital interventions to support a broad range of individuals—from risky drinkers to those with AUD. Following this formative work, future studies should investigate the generalizability of this study’s results by conducting evaluations in more naturalistic settings and among populations with more severe alcohol misuse.