An open-label, pragmatic, pilot randomized trial was conducted from January 2021 through December 2021. Each participant was randomly assigned in parallel to (1) IV ketamine (KET), (2) IM naltrexone (NTX), or (3) linkage alone (LA). All arms received enhanced linkage to our outpatient addiction clinic for continued AUD support. The primary clinical outcome was 30-day, all-cause, hospital readmission rate.

This study occurred at an urban, academic, safety-net hospital. Participants were recruited from the census of the addiction consultation service (ACS), a physician-led team that assists with evaluation, withdrawal management, pharmacotherapy, psychotherapy, care linkage, and harm reduction [26]. Enrollment, baseline data collection, and intervention were completed at bedside during participants’ index admission. Some follow-up data were collected at the outpatient addiction clinic visit if they presented. Follow-up attendance and readmission data were obtained through electronic health record (EHR) query.

This trial was approved by the Colorado Multiple Institutional Review Board.

To target those with high care utilization and ready access to our outpatient services, adult patients (age 18–65) with severe alcohol use disorder (six or more Diagnostic and Statistical Manual-V criteria [27]) were approached if they had one or more alcohol-related hospital admission(s) or emergency department (ED) visit(s) in the past year, had public (e.g., Medicaid) or private insurance, and were seen by the ACS. Exclusion criteria were: active COVID-19, being too medically ill for the interventions (AST/ALT > 5 × times upper-limit of normal, decompensated cirrhosis, glomerular filtration rate < 30 ml/minute, current/past acute coronary syndrome, cerebrovascular event, hypertensive crisis, cardiomyopathy, known elevated intracranial pressure, or platelets < 50/microliter), unresolved moderate/severe alcohol withdrawal, active delirium, active enrollment in another study, past-month receipt of IM naltrexone or IV ketamine, study medication intolerance, other substance use disorder (besides tobacco/cannabis), known/anticipated pregnancy or breast-feeding status, chronic/anticipated opioid use, unstable psychiatric illness (active psychosis or suicidality), and discharge to acute/residential treatment.

For appropriately identified patients, recruitment occurred by the principal investigator after their initial addiction consultation. Consented participants were immediately assigned to one of the three arms through a simple randomization feature in the EHR. Participants provided baseline data, completed their clinic intake including appointment scheduling (targeted within 7 days post-discharge), and received their assigned intervention at bedside before discharge. To better isolate the effects of the pharmacological interventions, inpatient teams were instructed not to prescribe oral naltrexone (or acamprosate or disulfiram) on discharge, although participants could receive any medication at follow-up if appropriate. Notably, prescribing discharge medications for AUD is rare; a recent systematic review [24] found just two studies implementing naltrexone prescribing for AUD on discharge. Their baseline prescribing rates were 0.0 and 1.6%. Additionally, evidence showing effectiveness of in-hospital prescribing of AUD medications is scarce [24]. Therefore we did not consider lack of oral AUD medication prescription on discharge to be an egregious departure from standard of care.

Acceptability data were collected 40 min after the start of pharmacologic intervention. Stipends were provided for enrolling ($10) and presenting to follow-up appointment ($20).

For the KET arm, the intravenous infusion (at a dose commonly used in depression studies [14, 28], 0.5 mg/kg over 40 min) was administered by a registered nurse with the principal investigator at bedside on or near the day of anticipated discharge. The participant was monitored with telemetry and continuous pulse oximetry during the infusion and until at least 2 h after completion (160 min post-initiation). Vital signs were recorded at 0-, 40-, and 160-min post-initiation. The participant was provided an eye covering and noise-cancelling headphones that played a standardized, relaxing soundtrack [29]. The participant received brief pre-infusion counseling on what to expect, with specific advice to (1) focus on deep breathing and (2) approach difficult internal experiences with curiosity. Physical symptoms were assessed immediately before initiation and upon infusion completion (at 40 min). Dissociative symptoms were also assessed upon infusion completion.

For the NTX arm, the gluteal injection (380 mg) was administered by a registered nurse at bedside on or near the day of anticipated discharge. Nurse educators were trained in correct administration, and they subsequently supervised other nurses. Physical symptoms were assessed immediately before and immediately after injection. Vital signs were recorded at 0- and 40-min post-injection.

For the LA arm, no pharmacologic intervention was given as part of the study, but participants still received outpatient addiction clinic linkage and the research stipends. For comparison with other arms, baseline physical symptoms and vital signs were assessed on or near the day of anticipated discharge.

The primary clinical outcome was all-cause, 30-day hospital readmission rate, assessed through EHR query that detects admissions to at least 34 other hospitals in the state (as of the start of the trial). This outcome was selected because: (1) it is highly relevant to clinicians [30], policymakers [31], and patients who often experience disease-related stigma in hospital settings [32, 33], and (2) it is not dependent on patient follow-up or recall, thus ensuring near-complete data collection among this inherently unstable population.

Primary feasibility outcomes included study recruitment rate, patient acceptability (10-point Likert scale), and adverse events. Main secondary clinical outcomes were 30-day all-cause emergency department (inclusive of our institution’s withdrawal management facility) visit rate and 14-day addiction clinic encounter (in-person or telehealth), ascertained through EHR query.

At baseline, self-reported daily drinking was recorded using a modified Timeline Follow Back (TLFB) [34] method (recording 7 days preceding admission). Adverse childhood events (ACE) were recorded using the 10-item ACE questionnaire [35]. Recent depressive symptoms were recorded using the Patient Health Questionnaire 9 (PHQ9) [36], a 9-item instrument commonly used for outpatient screening.

We assessed anticipated and perceived effectiveness of intervention in terms of reducing alcohol intake (10-point Likert scale) immediately following pharmacologic intervention and at follow-up, respectively.

We recorded immediate, treatment-emergent symptoms through a standardized, open-ended symptom inquiry (“Pay attention to your body. What symptoms are you noticing?”) assessed immediately pre- and post-pharmacologic intervention. Immediate dissociative symptoms from ketamine were recorded using the Clinician-Administered Dissociative States Scale (CADSS) [37], a 23-item questionnaire developed to measure dissociation due to acute stressors. For those that presented to their follow-up visit, adverse events since hospital discharge were recorded using the Patient Rated Inventory of Side Effects (PRISE) [38], a self-report, system-based (e.g. “cardiovascular”) questionnaire that assesses the presence and tolerability of symptoms.

Data were analyzed between the LA arm and either KET or NTX arms. Descriptive statistics were used to analyze demographic variables, study feasibility, and acceptability. Hypothesis tests resulting in p-values for statistical significance were conducted judiciously; Table 1 (baseline characteristics, by arm) excludes p-values entirely because there is inherently no uncertainty that observed differences are based on random chance [39, 40]. When tested, continuous numerical data (e.g., vital signs) were compared via an appropriate independent or paired samples t-test. Proportional or count data (e.g., primary/secondary binary outcomes) were compared with a chi-squared test of independence, with effect sizes reported as relative risks. All reported p-values were considered significant at an alpha of 5%. In Table 1, missing values were included in “other” or “unknown” for all self-reported demographics, except for family history (considered negative if missing). Primary (readmission) and secondary clinical outcomes (ED visit, clinic attendance) did not require participant responses, so all participants were included in intent-to-treat analysis. Post-hoc, relative risks for readmissions using a per-protocol analysis and using alcohol-related readmissions were explored. Alcohol-relatedness was defined as alcohol consumption likely contributing to the indication for admission, as documented by the primary hospital team’s discharge summary. When readmission encounter details were unavailable, the readmission was considered alcohol-related by default. Due to small sample size, when available, partial data were included in the appendices (acceptability, adverse events, etc.). Data were analyzed in Python (version 3.7.6).