We recruited participants for a laboratory study of behaviors related to drinking and driving using fliers placed around a university campus and nearby hospital. Interested individuals were screened by telephone. Inclusion criteria were: being 21–39 years old; having consumed ≥4 (women) or ≥5 (men) drinks on at least one occasion in the past year (history of binge drinking) without adverse effects (to minimize risk of adverse events during procedure); having averaged <4 drinking days and <12 drinks per week in the past 2 months (unlikely to be highly alcohol dependent); and being willing to use a ride share to get home after the study visit. Individuals were excluded if they desired alcohol treatment or received it in the past 6 months, evidenced alcohol use disorder per DSM-5 criteria, did not speak English, were pregnant, or had a medical condition that limits alcohol consumption. The telephone screen included the Columbia-Suicide Severity Rating Scale, and anyone experiencing suicidal ideation or behavior was excluded from the study and provided resources that addressed their suicidality (Posner et al., 2011).

Participants gave written informed consent, having been informed that they would be asked to drink three alcoholic beverages (vodka plus juice), at a dose equivalent to 4–6 drinks over the course of 2 hours, resulting in moderate levels of intoxication and blood alcohol. Study sessions were conducted with 1–3 participants in a clinical laboratory setting and lasted 5–6 hours. To prepare for the session, participants were instructed to eat a small meal before fasting for 2 hours, and to abstain from caffeine for 8 hours and alcohol for 24 hours. Midway through the session, participants received a small meal. Breath alcohol concentration (BrAC) was used as a proxy for BAC and measured with a BACtrack Mobile Pro device (BACtrack, San Francisco, CA). Prior research validated the accuracy of this device to measure BAC within ±0.001% of a police-grade breathalyzer and estimate BAC within ±0.01% of a blood test for the range of BACs under present investigation (Delgado et al., 2021).

Upon arrival, participants’ BAC was measured to confirm that it was 0.00%. A nurse trained in the protocol then administered a brief medical screening that included blood pressure and heart rate measurement, and a urine pregnancy test for female participants. Next, participants completed an intake survey that asked about basic demographic information and included the following measures: Driving History and Experience Questionnaire (Harrison & Fillmore, 2005); Drinking and Driving Questionnaire (McCarthy et al., 2012); 26 items of the Barratt Impulsiveness Scale-11 (Patton et al., 1995); the 4-item Brief Sensation Seeking Scale (Stephenson et al., 2003); and a 51-item temporal discounting task (Senecal et al., 2012).

Participants were given weight- and sex-based doses of 100-proof vodka (1.0 g per kg for men and 0.9 g per kg for women) mixed with orange juice, spaced 40 min apart (Delgado et al., 2021). These doses were intended to raise blood alcohol to the following levels: dose 1, 0.03–0.04%; dose 2, 0.07–0.08%; dose 3, 0.10% (Watson et al., 1981). Each dose was consumed within a standard 10-min window to minimize inter-participant variability in consumption rate. Twenty min after finishing each drink, participants were instructed to rinse with a non-alcoholic mouthwash before being administered the BACtrack Mobile Pro breathalyzer. After 3 initial readings, we continued to administer the breathalyzer every 15 min until the participant’s BAC reached 0.03% or less.

After screening, but prior to the laboratory session, eligible individuals were randomly assigned to either treatment (BAC Feedback) or control (No Feedback) conditions. After each BAC reading, individuals in the treatment group were shown their BAC while those in the control group were not shown their BAC. All participants then completed the dependent measures in silence, rating their level of intoxication, ability to drive a motor vehicle, and willingness to drive in that moment on 10-point visual analog scales (Harrison et al., 2007; Harrison & Fillmore, 2005; Van Dyke & Fillmore, 2014).

Participants were compensated with a $100 debit card. To ensure they could get home without driving, we provided a private rideshare trip from the clinical laboratory to their destination and waited with them until the driver arrived. The university Institutional Review Board approved the study, which was registered on ClinicalTrials.gov (NCT05028413).

The primary outcome in the analysis was self-reported willingness to drive while moderately intoxicated (0.04% ≤ BAC < 0.08%). Secondary outcomes were perceived intoxication and ability to drive. Outcome differences between treatment and control were analyzed using linear mixed effects models with a random intercept for each individual and fixed effects for BAC, treatment, and intoxication level. The treatment by intoxication level interaction was included to estimate the incremental impact of feedback on outcomes for each intoxication level. Thus, the sum of the treatment and interaction term for each intoxication level indicated its total treatment effect. For each outcome, models were fit both with and without sex, history of alcohol-impaired driving, and driving frequency to examine the sensitivity of results to confounding.

Additional linear mixed effects analyses were conducted to estimate the independent and interactive contributions of BAC limb (ascending vs. descending) and BAC feedback on self-reported outcomes while individuals were moderately intoxicated. The descending limb was defined idiosyncratically for each participant as all observations after peak measured BAC.

Data are available upon request for non-commercial research purposes.

Of 25 individuals screened, 20 went on to participate (Figure S1): 13 individually, 4 as dyads, and 3 as a triad. Of these participants, 12 (60%) identified as male and 8 (40%) as female; 16 (80%) as white, 3 (15%) as Asian, and 1 (5%) as Black; none identified as being of Hispanic ethnicity (Table 1). Mean age was 27.1 (range 21–37). Mean peak BAC was 0.09% (range 0.06%–0.12%); Figure S2 shows BAC readings by group over time. Mean number of breath-testing rounds per participant, from first dose of alcohol until BAC declined to 0.03% or less was 12.6 (range 7–18). There were no adverse events during the study.

For a period of 105 min beginning 50 min after the final dose of alcohol, BAC Feedback participants were significantly more willing to drive than No Feedback participants (Figure S3).

At moderate levels of intoxication (0.04% ≤ BAC < 0.08%), BAC feedback was associated with less perceived intoxication (−2.0, CI: [−3.1, −0.8]), greater perceived ability to drive (2.4, CI: [1.0, 3.9]), and greater willingness to drive (3.9, CI: [2.5, 5.3]), all ps < .001 (Figure 1). These findings were robust to the inclusion of participant sex, alcohol-impaired driving history, and driving frequency in the models, all ps < .001. Tables S1–S3 provide full model statistics.

In the models that included BAC limb, treatment group, and their interaction term, BAC feedback was again associated with less perceived intoxication, p = .012, greater perceived ability to dive, p = .009, and greater willingness to drive, p < .001, while moderately intoxicated. Descending Limb was associated with less perceived intoxication, p = .047 and greater perceived ability to drive, p = .010, but not associated with willingness to drive, p = .21. The limb by treatment interaction was non-significant for perceived intoxication, p = .76, perceived ability to drive, p = .97, and willingness to drive, p = .078, and the results for treatment were directionally similar on both BAC limbs (Table S4, Figure S4).

At legally impaired levels (BAC ≥ 0.08%), BAC feedback did not impact perceived intoxication (−0.2, CI: [−1.6, 1.3]), p = .83, or ability to drive (0.0, CI: [−1.9, 1.9]), p = 1, but individuals who received feedback were more willing to drive (2.3, CI: [0.2, 4.3]), p = .029. With covariates included, there were no significant differences between treatment and control groups on any of these outcomes, all ps > .09.

In this laboratory-based, randomized controlled trial, we examined how receiving feedback about one’s BAC, up to a BAC of about 0.10%, impacted self-reported willingness to drive. When individuals were informed that their blood alcohol level was in the moderately intoxicated range (0.04% ≤ BAC < 0.08%), they reported feeling less intoxicated, better able to drive, and more willing to drive than an uninformed control group. When their BAC was at or above the legal limit (≥ 0.08%), after controlling for potential confounders, there were no differences between informed and uninformed groups on these measures. This pattern of results is consistent with the hypothesis that knowing that one’s BAC is below the legal limit could lead to a greater willingness to drive despite being objectively impaired.

This unintended consequence of providing breathalyzer feedback is concerning, given that the relative risk of a crash at a “legal” BAC of 0.07% is twice that of a BAC of 0.00% (Compton et al., 2002). To mitigate this unintended consequence, we recommend that breathalyzer apps provide clear, salient feedback about potential driving impairments alongside BAC readings (Haney et al., 2024). We also urge greater scrutiny of the consumer breathalyzer market. Many publicly available devices significantly underestimate BAC—misinforming some users that they are not legally impaired, thereby compounding the problem of impaired drivers who feel safe to drive after a <0.08% reading (Delgado et al., 2021). We note that this problem could be addressed more systemically by adopting recommendations to lower the federal BAC limit to 0.05%, in line with most countries that set a limit (National Academies of Sciences, 2018).

The present study had limitations. The sample was small and non-representative. Self-reported willingness to drive was used as a proxy for the decision to drive. Peak BAC averaged 0.09% and maxed out at 0.12%, so we cannot draw conclusions about greater levels of intoxication. Because participants were given only their BAC reading, further research is needed to evaluate how app messaging affects willingness to drive and what messages might best inform individuals about the potential consequences of their measured BAC. To mimic a naturalistic, bar-like atmosphere, drinks were spread out over time, and individuals participating in the same session were free to chat when measurements were not being taken. Nevertheless, this was a laboratory experiment, and results may not generalize to real-world contexts.

If the potential downsides of smartphone-paired breathalyzers can be minimized, they could be used in a variety of ways to reduce drinking and driving (Luczak & Ramchandani, 2019; Quanbeck et al., 2014; Tofighi et al., 2019). First, with repeated breathalyzer use, individuals may become more accurate at estimating their level of impairment. Second, integrating breathalyzers with rideshare apps could improve the accessibility of safe alternatives to driving. Third, individuals’ sharing their BAC readings with others could add a measure of accountability. Fourth, integrating breathalyzers with driving monitoring apps could enable delivery of monetary and other rewards for abstaining from drinking and driving (Alessi & Petry, 2013; Ebert et al., 2025; Petry et al., 2000; Sloan et al., 2014). As interventions are built around smartphone breathalyzer technology, we must remain vigilant to potential unintended consequences.