Prazosin-Related Reductions in Daily Craving Mediate Treatment-Related Daily Alcohol Use Outcomes in Alcohol Use Disorder Authors: Stephanie Wemm, Rachel Hart, Gretchen Hermes, Zachary Harvanek, Rajita Sinha Published: 2026 ABSTRACT AbstractBackgroundPrazosin has been shown to reduce drinking, particularly in individuals with alcohol withdrawal symptoms, possibly due to reducing craving as shown in experimental studies. However, it is unclear whether prazosin decreases daily changes in alcohol craving and whether such changes drive prazosin-related changes in alcohol intake in the treatment of alcohol use disorder (AUD).MethodsUsing an intensive daily-level longitudinal design embedded within a 12-week, double-blind, placebo-controlled, randomized clinical trial of prazosin (16 mg/day) versus placebo (PBO) for the treatment of AUD, we examined treatment effects using daily phone surveys of alcohol craving and intake. Participants were treatment-seeking individuals (52 on prazosin, 45 on placebo, 97% of the treated sample) who met current criteria for AUD and who provided a total of 4289 daily surveys over the 12-week treatment period.ResultsPrazosin but not PBO significantly reduced daily craving (estimate [Est] = −0.902, 95% credibility interval [CI] [−1.296 to 0.475]), and this craving reduction predicted lower next-day drinking for participants taking prazosin (Est = −1.317, 95% CI [−2.196 to −0.599]) but not PBO (Est = −0.491, 95% CI [−1.147 to 0.087]), which represented a 66.7% versus 26.7% craving-mediated reduction in alcohol intake.ConclusionsUtilizing an intensive daily longitudinal design and random slope modeling, the current study examined daily craving change as a dynamic process associated with changes in daily alcohol intake during treatment. The findings support prazosin’s anticraving mechanism of action underlying prazosin-related reductions in alcohol intake.Keywords: Alcohol use disorder, Craving, Daily diary, Mechanism, Mediation, Prazosin Alcohol use disorder (AUD) is a chronic, relapsing condition contributing to substantial morbidity and mortality worldwide (1). Craving, a cornerstone symptom of AUD, serves as a key prognostic indicator associated with risk of relapse and treatment failure (2, 3, 4, 5, 6). Higher pretreatment craving prospectively predicted more heavy drinking days (≥4 drinks for women/≥5 drinks for men) and average drinks per drinking day while in outpatient behavioral treatment (3), underscoring the role of craving as a critical biobehavioral process influencing AUD treatment outcomes. Therefore, targeting craving in treatment development studies is essential for assessing a potential mechanism of change that supports AUD recovery. Craving is a multidimensional construct central to several addiction theories and can be broadly defined as wanting to drink alcohol or use drugs (7,8). The interplay between alcohol craving, arousal, and stress is intricately linked with disruptions in both central and peripheral stress physiology (9,10). Behaviorally, alcohol cues and stressors trigger dynamic increases in craving that predict higher alcohol consumption among individuals in treatment for AUD (5). Like cues, stress can provoke craving both through distress-relief motives and enhanced reward anticipation (11,12), creating a feed-forward cycle of increased drinking and sensitized wanting (10,13). Biologically, chronic drinking disrupts prefrontal-striatal reward and stress circuits and alters autonomic and hypothalamic-pituitary-adrenal axis function (10,14). These effects partially reflect alterations in noradrenergic and adrenergic pathways, key modulators of sympathetic arousal (15) and cue and stress reactivity (16). Animal studies have shown that alcohol both stimulates noradrenergic activity and increases drinking through changes in this system (17), leading to interest in adrenergic-targeted therapeutics. Several AUD medications appear to influence craving. Naltrexone, an opioid antagonist, reduces craving by mitigating endogenous opioid effects (18), which may forestall relapse by blunting the rewarding effects of drinking (19). Naltrexone also appears to be particularly effective for individuals with heavier drinking patterns (20) or those with elevated baseline alcohol craving levels (21,22). Gabapentin, although not U.S. Food and Drug Administration approved for AUD, reduces relapse shortly after detoxification through withdrawal-related mechanisms (23,24), although the effects are modest (25). Prazosin, an α1-adrenergic antagonist, has shown promise in reducing craving and alcohol consumption. Animal studies showed that prazosin attenuated excessive drinking by reducing adrenergic activity (26,27) and stress-potentiated alcohol approach behaviors (28). In human laboratory studies, patients treated with prazosin showed reduced stress- and cue-induced alcohol craving along with reduced anxiety and normalizing of the associated stress arousal response (29,30). A recent study in Europe found that individuals with higher levels of alcohol use derived more benefit from a combination treatment of prazosin and cyproheptadine (31). In studies examining the effect of prazosin treatment alone on alcohol consumption, prazosin was found to reduce the number of drinks consumed (32, 33, 34), heavy drinking days (32,35), and drinking days (33,35). Baseline levels of alcohol withdrawal served as a significant moderator of prazosin’s benefit on alcohol use outcomes such that individuals with higher alcohol withdrawal at intake showed prazosin-related reductions in alcohol use outcomes over the 12 weeks of treatment as well as in craving assessed monthly relative to placebo (PBO)-treated individuals, but these effects were not observed in those with no or low levels of alcohol withdrawal symptoms (36). While this previous work identified an effect on both alcohol craving and intake, it did not test whether prazosin improves drinking outcomes by reducing dynamic day-by-day alcohol craving and craving-related drinking, i.e., through a mediated effect, and whether that effect is dependent on withdrawal status. If withdrawal is not a prognostic indicator of prazosin’s effects on craving-related drinking, craving would not be impacted by withdrawal, or alternatively, prazosin may reduce craving-related drinking only in people with higher withdrawal symptoms. Thus, further investigation into the potential efficacy of prazosin in reducing alcohol craving is warranted to ascertain how and why prazosin has effects on reducing drinking in patients with AUD. Daily intensive longitudinal assessment designs embedded within clinical trials offer a powerful approach to test whether medication-induced changes in craving mediate changes in drinking. Capturing fluctuations in craving via repeated assessments, particularly at the daily level, allows for high-resolution assessment of how a medication may influence daily alcohol intake during the trial. Furthermore, daily change in craving over time can be modeled to test whether the craving trajectory shifts following the introduction of medication or other intervention. Such modeled trajectories of a potential mediator or driver may then be used to predict shifts in the trajectory of alcohol intake, thus allowing for a formal test of mechanistic change as a mediator of medication-driven changes in drinking. The Current Study Based on the previous research presented above, an intensive longitudinal design of day-by-day dynamic craving and drinking was embedded within a 12-week clinical trial examining baseline alcohol withdrawal effects on prazosin’s efficacy on drinking outcomes in patients with AUD (36). Here, we present secondary analyses of daily survey data collected as part of this previously published randomized clinical trial (RCT) of prazosin versus PBO for AUD (36). While this study previously identified baseline alcohol withdrawal as a key prognostic indicator of prazosin’s efficacy on both monthly alcohol craving as measured by the Obsessive Compulsive Drinking Scale (OCDS) (37) and alcohol intake, whether prazosin treatment has an effect on dynamic daily changes in craving and drinking is not known. Accordingly, in the current study, we examined whether prazosin alters daily craving dynamics and the day-by-day relationship between craving and drinking. Using daily mobile survey data, we tested whether between-person differences in treatment condition (prazosin vs. placebo) predicted within-person changes in drinking via changes in craving over the course of the trial. Importantly, we used a random slope modeling approach to estimate individual trajectories of change in craving and drinking across time and to evaluate whether these within-person change processes differed across patients as a function of treatment condition. To our knowledge, few studies have applied trajectory-based mediation approaches to examine medication-related changes in dynamic craving-drinking processes in AUD pharmacotherapy trials. Based on previous laboratory work (29,30), we hypothesized that one mechanism by which prazosin was beneficial was by reducing craving-related drinking independent of baseline levels of alcohol withdrawal symptoms. As alcohol withdrawal at intake was found to moderate prazosin’s effect on drinking outcomes (36), we also examined whether the craving-drinking relationship was moderated by baseline level of alcohol withdrawal symptoms. Methods and Materials Participants Participants were 97 treatment-seeking adult men and women with AUD recruited from a medium-size New England urban community who provided daily phone survey assessments while enrolled in a 12-week RCT of prazosin versus PBO effects on alcohol use outcomes during the period from November 2012 to May 2017 (36) (Figure 1, CONSORT diagram). Participants were recruited via advertisements in newspapers, web-based and social media ads, or referrals for AUD treatment in the local area. Eligible participants were ≥18 years of age, met the current criteria for alcohol dependence in DSM-IV-TR (per the recruitment time frame from 2012–2017), were able to sign informed consent, and could read English and complete study evaluations. Participants were excluded if they met DSM-IV-TR criteria for abuse of or dependence on any other psychoactive substance except cannabis, nicotine, or caffeine; currently used psychoactive drugs, including anxiolytics, antidepressants (except selective serotonin reuptake inhibitors), naltrexone, or disulfiram; any psychotic disorder or current Axis I requiring specific attention; reported any significant underlying medical conditions such as cerebral, renal, thyroid, hepatic, or cardiac pathology that would interfere with or be of potential harm during the study; or were hypotensive as indicated by a sitting blood pressure <90/60 mm Hg. The Human Investigation Committee of Yale University approved this study (ClinicalTrials.gov ID: NCT00585780), and all participants completed written informed consent prior to enrollment. CONSORT (Consolidated Standards of Reporting Trials) diagram for daily diary assessment study of alcohol craving and drinking outcomes during prazosin treatment trial. Procedures The intensive daily level longitudinal data were collected concurrently with the 12-week double-blind RCT (2-week dose titration, 10 weeks at target dose [16 mg/day] with a 5-day taper at the end) (36) to allow for a prospective day-by-day assessment of medication effects on daily craving and the craving-drinking relationship. After intake assessments were conducted to confirm eligibility, participants were enrolled in a 12-week randomized controlled trial and administered either placebo or 16 mg/day of prazosin. Target medication dosing was 3 times/day with 5 mg in the morning, 5 mg in the afternoon, and 6 mg at night. The titration and taper schedule are detailed in Supplemental Methods. Participants were not required to complete a minimum level of abstinence before enrolling in the study or have a certain level of alcohol withdrawal (Figure 2). Participants experiencing medically significant alcohol withdrawal and seeking medical alcohol detoxification were referred to either a local detoxification treatment facility or the affiliated inpatient unit. Participants who chose the latter completed an initial inpatient program for 3 to 4 weeks during which they began the study medication inpatient immediately following the medical alcohol detoxification protocol and then were released to an outpatient option at full dose. All participants completed a daily interactive voice response or smartphone application–based survey at 8 pm each night. Participants were reimbursed $2 per completed daily survey, with a $6 weekly bonus for full completion. Participants completed weekly visits to complete assessments and provided urine toxicology and breathalyzer results. Randomization and vitals collection are included in Supplemental Methods. Study design and data collection procedures. After intake and eligibility determination, participants with current alcohol use disorder were randomized to prazosin or placebo in a double-blind manner for 12 weeks (84 days) using a parallel group design. Participants completed daily diaries at 8 pm each day over 84 days. Participants were asked about their craving and the amount that they drank over the course of the day on surveys using an interactive voice recording or smartphone. At the bottom, sample data are shown from 2 representative participants. The dark blue, thin solid line represents subjective craving level on each day, and the orange bars indicate the amount of drinking on that day, as reported in the following day’s daily diary. The dashed blue and orange lines indicate each person’s between-subject craving and drinking, respectively. The solid orange line along the bottom of the figure indicates days with zero reported drinking. Gaps in the craving line and an absence of orange bars indicate missing daily observations rather than zero craving or zero drinking. t.i.d., 3 times/day. Assessments Sociodemographic Information A research associate collected information regarding the participants’ demographic characteristics and medical and psychiatric history via interviews and self-reports during the intake process. Baseline Alcohol Withdrawal Alcohol withdrawal was assessed at intake via the Clinical Institute Withdrawal Assessment for Alcohol-Revised (CIWA-Ar) (38) The CIWA-Ar is a validated 13-item assessment administered by a trained interviewer to assess for physiological signs (pulse, systolic and diastolic blood pressure), physical symptoms (headache, nausea/vomiting, sweating, tremor), and psychological symptoms (anxiety, agitation, and altered sensorium and perception). Each item is scored from 0 to 7, with higher scores indicating greater severity. Baseline Craving Alcohol craving was measured at intake using the Alcohol Urge Questionnaire (AUQ), which asks participants 8 questions about their craving and desire to drink alcohol on a 7-point Likert scale (1 = strongly disagree to 7 = strongly agree). Participants completed the AUQ at baseline to provide an initial measure of craving prior to beginning daily assessments. The internal consistency was Cronbach’s α = 0.88. Intensive Daily Assessment of Alcohol Craving and Intake Daily prompts via interactive voice response or a smartphone application (MetricWire) were completed nightly to track outcomes, as in previous work (5,36,39,40). Participants responded to these prompts at approximately 8 pm each day and reported their alcohol craving and the amount of alcohol consumed that day. The items assessing craving included “Right now, I could really use a drink” and “Right now, the idea of a drink is appealing,” to which participants indicated their agreement on a 5-point Likert scale (0 = definitely false to 4 = definitely true) (39,40). These 2 items were averaged to create a single index of daily craving. The 2-item measure provides a brief assessment of craving in the moment, which makes up one aspect of the multidimensional craving process and captures the core motivational component of in-the-moment alcohol desire (urge and appeal) that may be closely linked to acute stress- or arousal-related noradrenergic activation. Daily craving was assessed only via the daily interactive voice recording (IVR)/smartphone reports. As with daily craving, drinking behavior was also assessed during the daily phone assessments. When reporting daily drinking behavior on the phone, participants were asked how many glasses of beer, wine, and/or mixed drinks they consumed per day. The quantity of each type of drink was summed for each day to create an index of total drinks consumed per day. Thus, drinking data used in the current analyses were derived exclusively from the daily IVR/smartphone reports. Data Analysis Plan The power analysis is described in the Supplement. Descriptive statistics (psych packages) (41,42) and graphs (ggplot2 and ggeffects) were generated in R 4.2.1. We calculated the intraclass correlation coefficient for craving in Mplus to determine the percentage of variance in daily craving and drinking that was accounted for across participants at the between-subjects level. Day in study was weighted by dividing by 84 so that the estimates of change would reflect the change from day 1 in the study to day 84 to represent the 12-week period. Two sets of main analyses were conducted using multilevel structural equation modeling (MSEM) with random slopes and Bayesian estimation in Mplus version 8.7 with noninformative priors and mean-point estimators (see Figure S1 for a conceptual model). MSEM allows for decomposition of within-person and between-person variance in the variables of interest. Statistical significance was evaluated using 95% credibility intervals (CIs) and posterior predictive p values, which are interpreted differently than frequentist p values and represent the posterior-averaged probability of observing data as extreme as or more extreme than that observed under the model. A key feature of the current analytic approach is the use of random slopes to capture within-person dynamic change processes daily over time and evaluate these processes at the between-person level. Specifically, individual trajectories of change in craving and drinking across the study period were estimated for each participant, and these person-specific slopes were then used to examine whether treatment condition predicted differences in these dynamic change processes. This approach allows for testing how within-person associations unfolding over time vary as a function of treatment group. In the first model, we assessed whether the effect of prazosin versus PBO on changes in daily drinking was mediated by changes in daily craving, controlling for pretreatment alcohol withdrawal scores. A prospective (lagged) structure was specified such that craving reported on the previous evening predicted next-day drinking. At level 1, repeated measures of craving and drinking were used to estimate individual slopes of change over time. At level 2, these slopes were entered into a mediation model in which treatment condition predicted change in craving (a path), change in craving predicted change in drinking (b path), and treatment condition predicted change in drinking (c′ path). Model constraints were used to estimate condition-specific indirect effects for prazosin and PBO. Thus, the a path and b path reflect both the direction of change (the positive or negative sign) and the magnitude of change (the estimate) in the slopes of craving and drinking. In the second model, we examined moderated mediation to assess whether alcohol withdrawal at baseline moderated any of the paths in the mediation model. Missing daily data reflected nonresponse to the IVR/smartphone surveys. Days during an inpatient stay were treated as missing. An additional sensitivity analysis was used to assess whether controlling for individuals who initiated the study as inpatients changed the analyses. Results Demographic and Survey Response Rate Preliminary Analyses Participants’ demographic characteristics are summarized in Table 1. Participants randomized to PBO and those randomized to prazosin did not differ significantly on demographic or clinical variables of interest with the exception of daily craving, which was higher in the prazosin group on average. Of the 100 participants in the original RCT, 97 participants completed daily diary data, providing a total of 4289 daily survey entries. Compliance was calculated as the proportion of days completed of the total number of days each participant remained enrolled in the study; using this denominator, average compliance was 62.4% (∼4.4 days of 7 for each week), consistent with past meta-analyses suggesting that individuals with substance dependence complete 69.8% (95% CI [60.97 to 77.95]) of surveys on average (43). The baseline characteristics and daily data completion rates were comparable across the prazosin/PBO treatment groups (Table 1). Table 1Participant Characteristics and Ecological Momentary Assessment CompletionPrazosin, n = 52Placebo, n = 45p ValueDemographic and Clinical HistorySex, Female32 (62%)31 (69%).587Age, Years41.24 (10.99)41.44 (10.49).926Race Black, non-Hispanic24 (46.15%)25 (55.56%).518 Other or Hispanic4 (7.69%)4 (8.89%) White, non-Hispanic24 (46.15%)16 (35.56%)Education, Years13.56 (2.87)13.27 (1.74).555Inpatient7 (13.46%)4 (8.89%).748Completers33 (63.46%)29 (64.44%)>.999Baseline Alcohol Use PatternsCIWA-Ar Score4.15 (5.81)3.76 (4.27).705Baseline Craving, AUQ25.68 (12.48)23.36 (10.38).324Baseline Past Month Alcohol Used20.19 (9.25)20.76 (8.19).757Baseline Average Drinks/Drinking Day10.96 (14.31)9.19 (6.37).512Years of Alcohol Use16.88 (9.52)18.29 (11.07).518Intensive Average Daily Smartphone Diary Reports During 12-Week TreatmentDaily Entries2223 (51.83%)2066 (48.17%).589Average Daily Craving1.54 (0.92)0.97 (0.76).002Average Drinks on Drinking Day5.39 (8.30)3.55 (2.35).190Average % Drinking Days70.1% (30.6%)60.7% (32.8%).153Average % Heavy Drinking Days24.5% (36.8%)13.6% (27.6%).112Change in Cravinga−0.45 (1.05)−0.18 (0.74).267Change in Drinks per Daya−0.47 (2.91)−0.32 (1.37).805Change in % Heavy Drinking Daysa−12.14 (30.02)−1.56 (21.19).135Change in % Drinking Daysa−3.82 (41.25)0.15 (29.09).679Values are presented as n (%) or mean (SD). Self-report daily surveys were obtained daily over the 84-day trial. Baseline drinking was obtained via timeline follow-back interview.AUQ, Alcohol Urge Questionnaire; CIWA-Ar, Clinical Institute Withdrawal Assessment-Alcohol Revised.aChange refers to the difference between the last 2 weeks of each participant’s participation in the study and the first 2 weeks of the study (titration period). Participant Characteristics and Ecological Momentary Assessment Completion Values are presented as n (%) or mean (SD). Self-report daily surveys were obtained daily over the 84-day trial. Baseline drinking was obtained via timeline follow-back interview. AUQ, Alcohol Urge Questionnaire; CIWA-Ar, Clinical Institute Withdrawal Assessment-Alcohol Revised. Change refers to the difference between the last 2 weeks of each participant’s participation in the study and the first 2 weeks of the study (titration period). First Model: Mediation Model The results from the mediation model are included in Table 2 and depicted in Figure 3 (full model results can be found in Table S1). Weekly averaged week 1–corrected change in daily craving and change in next-day drinking are depicted in Figure S2A, B. Individuals randomized to prazosin showed significantly greater reductions in daily craving across the study compared with those randomized to PBO (a path). Participants receiving prazosin reduced their daily craving by nearly 1 point on the 5-point rating scale from the beginning to the end of the study (estimate [Est] = −0.902, 95% CI [−1.296 to −0.475], posterior p value < .001), whereas those randomized to PBO showed a smaller reduction (Est = −0.338, 95% CI [−0.711 to 0.047], posterior p value = .038)a (Table 2 and Figure 3B). Relative to week 1 craving levels, this corresponded to a 46.2% reduction in craving in individuals receiving prazosin (week 1 mean = 1.95, SE = 0.09 points) compared with a 27.2% reduction in those receiving PBO (week 1 craving average mean = 1.24, SE = 0.09 points). Table 2Final Focal Mediation Model Investigating the Indirect Effect of Prazosin on Daily Drinking Via Its Reduction on Daily CravingParameterEstimatePosterior SDp95% CIBetweenaΔmx: Prazosin→Change in Craving−0.5640.287.020∗−1.091 to −0.029bΔyΔm: Change in Craving→Change in Drinking1.4540.261<.001∗0.946 to 2.048cΔyx: Prazosin→Change in Drinking0.4000.460.170−0.409 to 1.341Withinbym: Daily Craving→Next-Day Drinking0.2680.050<.001∗0.170 to 0.376Standardized Change in Craving−0.1990.028<.001∗−0.257 to −0.140Standardized Change in Drinking−0.0660.025.003∗−0.112 to −0.017Mediation ResultsPrazosin Indirect: aΔmx × bΔyΔm−1.3170.406<.001∗−2.196 to −0.599Prazosin Total: aΔmx × bΔyΔm + cyx−0.9170.519.036−1.952 to 0.119Placebo Indirect: aΔmx × bΔyΔm−0.4910.304.038−1.147 to 0.087Placebo Total: aΔmx × bΔyΔm + cyx−0.0910.554.394−1.058 to 1.117p is the Bayesian 1-tailed p value or the proportion of the posterior distribution that overlaps 0 (for positive estimates, proportion below 0; for negative estimates, proportion above 0). aΔmx is the effect of prazosin on change in craving from the beginning of the study to the end of the study; bΔyΔm is the effect of change in craving from the beginning of the study to the end of the study on change in drinking; cΔyx and c′Δyx (the latter not shown in the table) refers to the direct effect of prazosin on change in drinking from the beginning of the study to the end of the study. All change estimates reflect the change from day 1 to day 84 in the study. The Bayesian CI encompasses the lower 2.5% and 97.5% in the posterior distribution. All estimates are unstandardized, apart from the slope change in craving and slope change in drinking, as indicated in the table.∗Significant result, p < .05.CI, credibility interval. Final Focal Mediation Model Investigating the Indirect Effect of Prazosin on Daily Drinking Via Its Reduction on Daily Craving p is the Bayesian 1-tailed p value or the proportion of the posterior distribution that overlaps 0 (for positive estimates, proportion below 0; for negative estimates, proportion above 0). aΔmx is the effect of prazosin on change in craving from the beginning of the study to the end of the study; bΔyΔm is the effect of change in craving from the beginning of the study to the end of the study on change in drinking; cΔyx and c′Δyx (the latter not shown in the table) refers to the direct effect of prazosin on change in drinking from the beginning of the study to the end of the study. All change estimates reflect the change from day 1 to day 84 in the study. The Bayesian CI encompasses the lower 2.5% and 97.5% in the posterior distribution. All estimates are unstandardized, apart from the slope change in craving and slope change in drinking, as indicated in the table. ∗Significant result, p < .05. CI, credibility interval. Effects of prazosin vs. placebo on longitudinal changes in craving and drinking and their mediated relationship over the 12-week trial. (A) Conceptual mediation model depicting the effect of treatment condition (prazosin vs. placebo) on change in daily craving (s1) and subsequent change in next-day drinks (s2). (B,C) Estimated daily craving reductions from day 1 to day 84 (a path) (B) and estimated reduction in daily drinks (total effect) (C) by treatment condition. Bars represent model-estimated change with corresponding uncertainty intervals. (D) Model-implied trajectories showing the association between percentage change in craving and corresponding percentage reduction in drinks/day across the study period by treatment condition. Shaded regions represent the posterior SDs. Consistent with the craving mediation hypothesis, prazosin treatment significantly reduced drinking through its effect on craving reductions (Est = −1.317, 95% CI [−2.196 to −0.599], posterior p value < .001) (Figure 3D). Prazosin-related reductions in craving were associated with a predicted daily 1.317-drink reduction across the treatment period, corresponding to a 66.9% reduction relative to averaged week 1 drinking levels (Figure 3D). Importantly, this estimate reflects the portion of the treatment effect operating through craving reductions rather than the total effect of treatment on drinking outcomes. In contrast, the indirect effect for PBO was not significant, as the 95% CI included 0 (Est = −0.491, 95% CI [−1.147 to 0.087], posterior p value = .038), corresponding to a smaller, nonsignificant 26.8% reduction in drinking relative to week 1 drinking levels (Table 2 and Figure 3D). In sensitivity analyses, controlling whether a participant initiated the study during an inpatient stay did not change the overall pattern of results. Second Model: Moderated Mediation of Alcohol Withdrawal on Craving-Related Drinking Given previous findings that pretreatment levels of withdrawal moderated the effect of prazosin on monthly OCDS alcohol craving and drinking, we conducted a moderated mediation analysis of the craving mediation model using pretreatment withdrawal as a moderator. Pretreatment withdrawal did not moderate any path related to change in daily craving and drinking, i.e., there were no differences in treatment effect on daily craving-drinking relationship by alcohol withdrawal scores. As expected based on previously reported results (36), pretreatment withdrawal moderated the association between prazosin treatment and average drinking in the study (see Table S1 for full model results), such that patients taking prazosin had lower drinking on average if they also had higher withdrawal (Est = −0.214, 95% CI [−0.380 to −0.048], posterior p value = .005), without impacting the daily craving-drinking relationship (Figure S3). Discussion In the current investigation, we utilized an intensive daily-level longitudinal design embedded within a 12-week RCT of prazosin/placebo for AUD to examine whether daily changes in craving served as a mediator of prazosin’s treatment effects on daily alcohol consumption. Over the course of the 12-week double-blind RCT, individuals on prazosin, but not those on placebo, exhibited a significant reduction in daily craving, and that daily craving reduction in turn predicted decreases in reported daily alcohol drinks consumed. Notably, these findings remained significant even after accounting for both the main effect and the moderating effect of pretreatment withdrawal, a prognostic marker of who benefited the most from prazosin treatment as reported previously (36). In the context of a powerful daily longitudinal design, the current findings provide evidence that prazosin-related reductions in daily craving may represent an important mechanism associated with reductions in daily drinks consumed during treatment. From a clinical perspective, the observed craving-mediated reductions in daily drinks (∼1.3 drinks per day) may translate into clinically meaningful shifts, such as a reduction in severity from high to medium risk in World Health Organization risk drinking levels (44), highlighting craving as a targetable symptom that directly results in clinically relevant change in drinking outcome during treatment. Short-term treatments, such as medically managed withdrawal, may alleviate acute withdrawal symptoms, but additional treatment to prevent relapse and sustained drinking reductions are needed (45). This need suggests that targeted treatments that address specific dynamic daily processes such as craving could be of benefit for sustained positive treatment response to support recovery from AUD. Consistent with prior findings (5,46), the current results pointed to craving as an immediate, proximal driver of drinking behavior during the trial. Furthermore, the current findings showing that prazosin can effectively target this daily craving-drinking relationship are consistent with previous experimental research showing that prazosin reduced stress-induced and cue-induced craving in the laboratory (30) and at monthly and weekly assessments during the trial (36). Our previous findings also suggest that withdrawal is a critical prognostic indicator of alcohol consumption and of response to prazosin treatment. However, because baseline withdrawal did not moderate the mediated pathway, the current findings do not fully explain why prazosin appears to confer greater overall drinking benefit in individuals with higher withdrawal severity. Nonetheless, these findings suggest that daily craving reduction may represent a pathway through which prazosin reduces daily drinking that is independent of withdrawal severity. Identifying both who would benefit most from prazosin prognostically and how prazosin may be efficacious in reducing daily alcohol intake during treatment is critical to advancing treatment development research and for understanding mechanisms of alcohol behavior change (47). Such investigations promise to precisely identify the patient population and the mechanism by which pharmacological treatments may have their effects in AUD. Intensive daily-level longitudinal designs embedded within an RCT are useful because they permit granular assessment of symptom change and treatment-related processes across the study period. The current study provides a framework for leveraging such daily subjective reports during a clinical trial of an investigational drug to assess underlying mechanisms of a drug effect using an ecologically valid design. This approach allows researchers to examine both between-person differences in the extent of dynamic daily changes over treatment and individual differences in within-person day-to-day change processes summarized through random slopes. Notably, although individuals in the prazosin group appeared to have higher average levels of craving and drinking early in treatment, participants reached full medication dose by end of week 2, after which daily craving and drinks/day remained relatively stable in the placebo group while individuals receiving prazosin continued to show reductions. This pattern underscores the importance of distinguishing between between-person differences in average group level and within-person changes over time. Using intensive longitudinal data, such as daily diary or ecological momentary assessment, allows for modeling day-by-day changes in symptoms as they shift over time. These frequent measurements can reveal patterns that might be overlooked with less frequent or aggregate mean level changes, which is particularly important given that craving and drinking vary across contexts and days. Collectively, our findings from this granular data, along with previous experimental research (29,30), suggest that prazosin significantly affected day-to-day changes in craving that in turn impacted next-day drinks/day. The findings have important implications for pharmacotherapy development, indicating that investigational drugs may influence dynamic psychiatric processes that can be captured using intensive longitudinal methods. Chronic alcohol use alters neural circuits underlying craving and motivational drive, partially distinct from those associated with withdrawal (48,49). Noradrenergic sensitization in prefrontal-striatal pathways and extended amygdala regions, including the bed nucleus of the stria terminalis, contributes to heightened cue-driven motivation and craving (50,51). Converging evidence suggests that these craving-related circuit disruptions predict future heavy drinking (50). Thus, prazosin’s effects on these disrupted noradrenergic circuits may underlie the dynamic neural processes driving craving, alongside broader alcohol-related neuroadaptations that sustain compulsive drinking. Furthermore, the brief craving assessment included in this study was sensitive enough to detect prazosin-driven shifts in alcohol craving, suggesting that even a brief self-report measure can capture noradrenergic modulation of craving in daily life. The current study underscored the value of focusing on mechanisms of behavior change, such as craving, to understand how treatments exert their effects. Craving is a dynamic, proximal mechanism that drives alcohol use (5,46), and our findings captured these dynamic changes in the craving-drinking relationship and showed that prazosin reduced daily craving, and such reductions in daily craving mediated prazosin’s impact on daily drinking behavior. By leveraging high-resolution data, we were able to model lagged associations, showing that craving on one day predicted subsequent drinking on the next day. These findings extend the state of pharmacological mechanistic research because rather than examining dynamic changes only on the time scale of weeks or months, we can detect processes unfolding on a daily basis to understand how and why specific interventions work (52). Interventions, such as naltrexone and topiramate, have been shown to reduce overall craving measured at the daily level and have shown that craving mediated the association between the intervention and drinking outcomes (53,54). Notably, naltrexone’s anticraving mechanism of action in AUD has been primarily based on preclinical mechanisms of action and human experimental studies without direct testing of whether naltrexone intervention affects daily craving and presumed craving-mediated alcohol intake in the real world during the pharmacotherapy period. The current demonstration that prazosin reduced drinking via its effect on daily craving and that day-to-day changes in craving predicted later drinks the next day addresses an important gap by explicitly incorporating a temporal component at the daily level and over time, a significant gap in much of the current mechanistic work in pharmacological research (55). The current study has several strengths, including the use of an ecologically valid, intensive longitudinal daily data collection method that enabled the disaggregation of proximal drivers of drinking, such as craving, from prognostic markers of prazosin treatment response, such as withdrawal. However, certain limitations should be considered. First, prazosin may have a more significant effect on dynamic ecological momentary craving states captured multiple times per day. Therefore, the design captures day-level fluctuations but may not detect shorter within-day momentary time scale neurobehavioral processes, including stress-reactive or cue-elicited changes in craving and arousal. Although only retrospective over a 12- to 24-hour span, this daily report may still inadvertently alter how patients perceive their cravings. Despite the relatively short recall period, the act of retrospection could influence patients’ perceptions of their craving levels. To avoid the effect of retrospection, only drinking captured during the evening survey was used in the current study. However, this poses a potential limitation as drinking occurring after the approximately 8 pm survey window was not separately assessed and therefore is not represented in same-day reports. Compliance with daily reporting was moderate, with a mean of 4.4 of 7 daily reports weekly during the trial, although consistent with past research (43). Missingness might not have been random, such as being related to heavier drinking or other clinical features; however, all studies examining these processes in AUD could be subject to this bias. Nonetheless, it is important to acknowledge that such missingness could bias estimation of within-person associations and indirect effects. A final limitation is that we could not fully explore the role of alcohol withdrawal in the daily dynamics between craving and drinking. To do so, future studies should use daily measures of alcohol withdrawal to determine whether these symptoms also serve as a mechanism by which prazosin exerts its effects on drinking. Conclusions The current study used intensive day-level longitudinal data and random slope modeling to show that prazosin-related reductions in daily craving were associated with subsequent reductions in daily drinks each day during treatment. These findings extend prior laboratory and clinical work by supporting craving reduction as an important symptom of real-world daily life to be targeted in treatment. This methodology not only replicated laboratory findings but also provided ecologically valid support for prazosin’s efficacy as an anticraving medication. Additionally, utilizing the intensive day-level design, the current findings revealed that prazosin significantly affects dynamic changes in craving: Prazosin-related decreases in daily craving mediated predicted reductions in daily alcohol intake during treatment independent of alcohol withdrawal, thereby demonstrating that craving may serve as a targetable mechanistic marker of how prazosin may be of benefit in AUD treatment. The current study validates a daily longitudinal design for understanding mechanisms of behavior change in the treatment of AUD to understand intervention effects on day-to-day drinking reductions. Acknowledgments and Disclosures This work was supported by the U.S. National Institutes of Health (Grant Nos. R01AA020504 and R01AA029113 [to RS]) and the Robert E. Leet and Clara Guthrie Patterson Trust Mentored Research Award (to SW). RS designed and supervised all aspects of the study. RH and GH acquired data and conducted clinical assessments. SW and RS analyzed and interpreted the data and prepared the article. ZH and GH edited and approved the article. We thank Drs. Verica Milivojevic and Nia Fogelman of the Yale Stress Center for their assistance. We also extend our gratitude to all staff at the Yale Stress Center for their contribution to data collection and management. Data are available upon reasonable request. RS is a scientific consultant for Imbrium Therapeutics and has received research support from Aelis Farma, CT Pharma, and Tenacia Biotechnologies. All other authors report no biomedical financial interests or potential conflicts of interest. ClinicalTrials.gov: Prazosin to Reduce Stress-Induced Alcohol/Drug Craving and Relapse; https://clinicaltrials.gov/study/NCT00585780?tab=results; NCT00585780. Supplementary material cited in this article is available online at https://doi.org/10.1016/j.bpsgos.2026.100772. Denotes not significant as the 95% CI crosses 0. Supplementary Material Supplemental Methods, Results, Figures S1–S3, and Table S1 Key Resources Table References Associated Data Supplementary Materials Supplemental Methods, Results, Figures S1–S3, and Table S1 Key Resources Table