Combining extended-release naltrexone and behavioral harm-reduction treatment for people experiencing homelessness and alcohol use disorder: A Randomized Clinical Trial

Abstract

Background:

People experiencing homelessness and alcohol use disorder (AUD) have a high prevalence of alcohol-related mortality and need access to AUD treatment. Typical abstinence-based treatments, however, do not optimally engage this population. Recent research has shown lower-barrier approaches aiming to reduce alcohol-related harm and improve health-related quality of life (HR-QoL) are more acceptable to this population and can be efficacious. This study’s aim was to test the efficacy of combined pharmacobehavioral harm-reduction treatment.

Methods:

Participants were 308 adults experiencing homelessness and AUD (80% severe) who were randomized to 4 treatment arms: a) behavioral Harm-Reduction Treatment for AUD (HaRT-A) + extended-release naltrexone (XR-NTX), b) HaRT-A + placebo injections, c) HaRT-A only, and d) supportive services as usual (TAU). All participants attended assessments at baseline and weeks 4, 8, 12, 24 and 36. Primary outcomes were self-reported alcohol use quantity (AQUA; standard drinks) and frequency (ASI), alcohol-related harm (SIP-2R), and physical and mental HR-QoL (SF-12). Using piecewise growth modeling and an intent-to-treat model, we tested the effects of the 3 active treatment arms compared to TAU and the active medication versus placebo, double-blinded arms over a 12-week treatment course and through the 24 weeks following treatment withdrawal.

Findings:

Compared to TAU, the HaRT-A+XR-NTX arm evinced statistically significant improvements from baseline to 12 weeks posttreatment across 4 of the 5 primary outcomes: peak alcohol quantity (linear B = −.48, CI = −.79, −.18, p = .010), alcohol frequency (linear B = −4.42, CI = −8.09, −.76, p = .047), alcohol-related harm (linear B = −2.22, CI = −3.39, −1.06, p = .002), and physical HR-QoL (linear B = .66, CI = .23, 1.10, p = .012). The linear treatment effect for mental HR-QoL was not statistically significant (linear B = 1.69, CI = .12, 3.27, p = .076). After treatment discontinuation at 12 weeks, improvements were maintained through the 36-week follow-up. Analyses comparing the double-blinded medication and placebo arms showed no statistically significant differences on any of the primary outcomes. HaRT-A+XR-NTX, HaRT-A + Placebo and HaRT-A only participants did not report a greater experience of adverse events than TAU participants.

Interpretation:

Compared to existing community-based services as usual, combined pharmacobehavioral harm-reduction treatment resulted in decreased alcohol use and alcohol-related harm and improved physical HR-QoL for people experiencing homelessness and AUD. Considering the nonsignificant placebo versus active medication effects, the combined pharmacobehavioral harm-reduction treatment effect cannot be attributed to the medication alone. Future studies are needed to further probe the relative contributions of the pharmacological and behavioral components of this harm-reduction treatment and to see if a maintenance treatment approach can extend these positive outcome trajectories.

Funding:

This research was supported by the National Institute on Alcohol Abuse and Alcoholism (1R01AA022309–01; PI: Collins).

Introduction

Alcohol use disorder (AUD) is 10 times more prevalent in the homeless versus general population,^1–3 and people experiencing homelessness are 6 to 10 times more likely to die of alcohol-attributable causes than the general population.^4,5 Given this population’s disproportionate experience of AUD and alcohol-related mortality, it is vital to ensure their access to effective treatment.

Available treatments are, however, not highly engaging or effective for this population,^6–11 and the typical requirements of high-intensity, often inpatient treatment paired with abstinence achievement pose the most formidable barriers.^12–14 Our prior studies in this population showed that a small minority of participants (5–11%) experiencing homelessness and AUD in community-based service settings (i.e., shelters, supportive housing, neighborhood clinics, drop-in centers) aspired to abstinence.^15,16 Instead, participants preferred lower-intensity, nonabstinence-based approaches with patient-driven goal-setting and a focus on health-related quality of life (HR-QoL).^12,14,17

Accordingly, our team has developed behavioral Harm Reduction Treatment for AUD (HaRT-A), which is low-intensity, nonabstinence-based counseling supporting patient-driven goals towards alcohol harm reduction and HR-QoL improvement.^18–20 In a 3-month RCT, HaRT-A was associated with high-levels of engagement and reductions in alcohol use and alcohol-related harm compared to a services-as-usual control condition.¹⁸ However, additional studies are needed to test its longer-term effects and whether it can be enhanced with pharmacotherapy.

One medication well-positioned to support alcohol harm reduction is extended-release naltrexone (XR-NTX; VIVITROL^®), a long-acting opioid receptor antagonist that is safe for use in active drinkers with AUD.²¹ In prior studies, XR-NTX has shown efficacy in reducing alcohol craving, use and harm,^22–25 and has higher medication adherence compared to daily oral medication.²⁶ A prior single-arm pilot study (N=31) showed that XR-NTX paired with HaRT-A was feasible and acceptable in this population, and participants evinced significant within-subjects decreases in alcohol craving, use, and alcohol-related harm.¹⁹

The present, 4-arm, parallel-group RCT comprised a 12-week active treatment phase during which we tested the efficacy of combining HaRT-A and XR-NTX for people experiencing homelessness and AUD as well as a 24-week follow-up to test for delayed treatment effects or treatment decay.²⁰ HaRT-A+XR-NTX, HaRT-A+Placebo, and HaRT-A alone were compared to community-based supportive services as usual (TAU) on alcohol and HR-QoL outcomes. It was hypothesized that, compared to TAU, the 3 active treatments (HaRT-A+XR-NTX, HaRT-A+Placebo, HaRT-A) would evince greater decreases in alcohol use and alcohol-related harm and increases in HR-QoL. In the double-blinded arms, it was expected the HaRT-A+XR-NTX arm would experience greater decreases in alcohol use and alcohol-related harm and increases in HR-QoL than the HaRT-A+Placebo arm.

Methods

Details on study rationale, design and methods are available elsewhere,²⁰ and the pretrial protocol is in the Appendix Page 2.

Participants

Participants (N = 308) were adults (21–65 years old) who met criteria for current AUD, experienced homelessness in the past year, and were receiving services as usual at 3 community-based service sites (low-barrier shelters and housing programs) in Seattle, Washington. Inclusion criteria were having received services at one of the partnering sites, being at least 21 years of age, agreeing to use an adequate form of birth control (if female and in childbearing years), and fulfilling criteria for AUD (i.e., “alcohol dependence” according to DSM-IV-TR criteria as determined by the SCID-I/P).²⁷ Exclusion criteria were refusal or inability to consent (the latter was established using the UCSD Brief Assessment of Capacity to Consent [UBACC])²⁸; constituting a risk to safety and security of other agency clients or staff; known sensitivity or allergy to naltrexone/XR-NTX; concurrent participation in a clinical study involving an unapproved, experimental drug; concurrent participation in harm-reduction treatment studies conducted by the same team (added after 2 similar counseling studies launched in overlapping service settings); concurrent treatment with naltrexone/XR-NTX; being pregnant or nursing; past-year suicide attempt; renal insufficiency (serum creatinine level > 2); current opioid use disorder (i.e., opioid dependence according to DSM-IV-TR criteria); liver transaminase (AST, ALT) levels > 5 times the upper limit of normal; a clinical diagnosis of decompensated liver disease; or other condition deemed to make participation clinically unsafe.

Procedures

Procedures were approved by the institutional review board at the home institution. A data and safety monitoring board oversaw study progress and patient safety (the latter in sessions closed to investigators) on a biannual basis. The board permitted additional recruitment within the study period, given attrition was greater than accounted for in power analyses. The resulting recruitment ran from October 14, 2013 through November 30, 2017. Data collection was completed on October 11, 2018. Research staff typically recruited participants in the early morning before they left their community-based shelters. Trust-building was paramount. Research staff offered participant-preferred, nonalcoholic refreshments, used a community-aligned and compassionate approach, and indicated the study treatment was participant-driven and did not require abstinence or use reduction. Service staff often introduced researchers to potential participants, and later recruitment was bolstered by participants’ own independent endorsement of the project to their social networks.

Participants were screened using the above criteria. At baseline appointments, participants provided written, informed consent, and measures were administered.²⁰ After the baseline, the home institution’s Investigational Drug Program assigned participants to treatment arm (i.e., permuted block randomization, stratified by site) independently from research staff who were blinded to medication arm until all participant procedures were completed. At Week 0, research staff informed participants about treatment arm (blinded injection + HaRT-A, HaRT-A alone or TAU) assignment. The 3 active treatment arms then received their specified treatment content at weeks 0, 1, 4, 8 and 12 (see below). All participants attended additional assessments at weeks 4, 8, 12, 24 and 36 (see Figure 1). Participants were paid $20 for each assessment they attended. Research staff provided appointment reminders in person at the community-based sites where they regularly received services; via rich tracking information (e.g., addresses, phone, email, social network handles, contacts for service providers, friends and family members); and during community walk-throughs (i.e., parks, certain street corners, other service settings).

Figure 1.

CONSORT flowchart.

Measures and Materials

Primary outcomes.

The “Alcohol and Drugs” section of the Addiction Severity Index (ASI)²⁹ was used to assess frequency of alcohol and other substance use over the past 30 days. The Alcohol Quantity Use Assessment (AQUA) is a standard quantity questionnaire assessing alcohol use on participants’ peak drinking day in the past month.³⁰ It was created in a prior study with this population³¹ and was refined in a pilot study¹⁹ to better capture alcohol use that does not conform to traditional standard drink measures (e.g., sharing bottles, consuming beverages from large-volume containers [e.g., 16, 24 and 40 oz. bottles and cans]) or beverage type (e.g., high-gravity malt liquor, nonbeverage alcohol). The Short Inventory of Problems (SIP-2R), a 15-item, Likert-type questionnaire, was used to measure social, occupational and psychological alcohol-related harm over the past 30 days.³² The Short Form-12 (SF-12)³³ was used, coded and summed to document physical (evaluation of general health, physical functioning, ability to fulfill daily tasks/roles in light of physical limitations, bodily pain) and mental (sense of vitality, social functioning, ability to fulfill daily tasks/roles given emotional problems, mental health) HR-QoL, where higher scores indicated higher HR-QoL.³⁴

Secondary outcomes.

The urinary ethyl glucuronide (EtG)³⁵-to-creatinine ratio served as an alcohol-use biomarker. A 6-dimension measure of treatment manual adherence and competence was created based on the COMBINE Study Medical Management Adherence Checklist and Coding schema.^36,37

Adverse events and serious adverse events.

The Systematic Assessment for Treatment Emergent Effects (SAFTEE) interview,^38,39 was administered by study interventionists to assess self-reported symptoms that correspond to potential adverse events associated with XR-NTX (e.g., nausea, vomiting, diarrhea, abdominal pain, decreased/increased appetite, headache, dizziness, fatigue, nervousness/anxiety, insomnia/somnolence, depressive symptoms, suicidal ideation, itching, rash, injection site irritation, missed menses, increased/decreased libido). Experience of adverse events and serious adverse events (i.e., emergency department visits, hospitalizations, suicide attempts) were self-reported and recorded in the participant record.

Treatment Arms

Services as usual (TAU).

TAU comprised the community-based agencies’ supportive services as usual, including emergency shelter and/or permanent, supportive housing; intensive case management; limited nursing/medical care; referral to external service providers; and assistance with basic needs.

Active treatment arms.

The 3 active treatment arms (HaRT-A+XR-NTX, HaRT-A+Placebo and HaRT-A) attended 5, manualized behavioral harm-reduction treatment sessions delivered by study physicians/nurses (see manual in Appendix Page 39 and published protocol²⁰). HaRT-A aims to help people reduce alcohol-related harm and improve HR-QoL without requiring, prescribing or favoring alcohol abstinence as a treatment goal. Study physicians/nurses used a compassionate, pragmatic and patient-driven style in administering the following treatment components: a) feedback on results of physical exams and lab testing and their implications for physiological alcohol-related harm, b) collaborative tracking of participant-preferred alcohol-related outcomes, c) elicitation of harm-reduction and HR-QoL goals, and d) discussion of safer-drinking strategies. The HaRT-A+XR-NTX and HaRT-A+Placebo conditions additionally received information about the medication and injections at weeks 0, 4 and 8.²⁰

Data Analysis Plan

Preliminary data analyses.

Descriptive analyses documented completion and attrition as well as a sample description.

Initially, logistic regressions with robust standard errors were used to test potential treatment arm differences on missing responses for primary outcomes. Next sensitivity analyses were used to determine the robustness of treatment effects to plausible nonignorable missingness mechanisms that could cause bias (i.e., Diggle-Kenward, Wu and Carroll selection models).^40–42 Use of direct maximum likelihood estimation in primary analyses also served to minimize bias otherwise introduced if using methods resulting in listwise data deletion or simple imputation.^43,44

Primary outcome analyses.

There were two parts to the treatment efficacy analysis. First, we tested the efficacy of the 3 active treatment arms (i.e., HaRT-A + XR-NTX, HaRT-A + Placebo, HaRT-A alone) compared to TAU. Second, we compared the outcomes across the double-blinded HaRT-A + XR-NTX (active medication) and HaRT-A + Placebo arms. This 2-part design allowed us to dismantle active treatment components and thereby detect potential placebo effects of both the administration of an injection and attention from a medical professional that have been found in previous studies.^45,46

Piecewise growth modeling using Mplus 8.3 was used to test treatment effects on primary outcome trajectories over time.⁴⁷ Piecewise growth models are latent growth curve models that allow for varying, sequential, stage-based growth profiles⁴⁸ to characterize different phases of development in a trajectory and thus detect potential treatment delays or decay.⁴⁹ Primary alcohol (i.e., peak quantity, frequency, alcohol-related harm) and HR-QoL (physical and mental health) outcomes measured at each time point were indicators of the intercept (i.e., baseline) and the linear and, as needed, quadratic slopes (i.e., change in outcomes over time). Treatment arm was the primary predictor of slope.

Regression coefficients for growth models that include quadratic functions are not easily interpretable; thus, we additionally calculated Cohen’s d measures of effect size that can take into account both linear and quadratic effects to help interpret the unstandardized coefficient effects.⁵⁰ Although such effect sizes must always be interpreted with caution, we refer to the conventional interpretations of small (d=.2), medium (d=.5) and large (d=.8) in reporting results.⁵¹

A priori power analysis.

Assuming N=300, α = .05 and 20% follow-up attrition, a priori analyses indicated power (β−1) of .99 to detect a medium effect (γ=.2; approximately corresponding to Cohen’s d = .63⁵²) for HaRT-A+XR-NTX compared to TAU, and .92 to detect small-to-medium effects (γ=.15) for HaRT-A+Placebo and HaRT-A alone compared to TAU. Power was adequate (β−1=.83) to detect a medium effect in comparing HaRT-A+XR-NTX and HaRT-A+Placebo (N=150) across outcomes.

Secondary outcome analyses.

Because EtG was zero-inflated and overdispersed, it did not conform to distributional assumptions for growth modeling. Thus, we conducted cross-sectional zero-inflated negative binomial modeling to test treatment effects on EtG at weeks 12, 24 and 36, while controlling for baseline EtG. Descriptive analyses documented level of treatment manual adherence and competence.

Adverse events and serious adverse events.

Descriptive analyses and robust logistic regression and chi square analyses probed potential treatment differences (active treatments compared to TAU) on adverse and serious adverse events during the treatment period.

Role of Funder

This research was supported by a research program grant from the National Institute on Alcohol Abuse and Alcoholism (1R01AA022309–01; PI: Collins), and is registered with clinicaltrials.gov (NCT01932801). The active medication and placebo injections were provided by Alkermes, Inc. Neither NIAAA nor Alkermes, Inc. had a role in the study design; collection, analysis and interpretation of data; writing of the manuscript; or decision to submit this manuscript for publication.

Results

Preliminary Analyses

A baseline sample description is shown in Table 1. As shown in Figure 1, 97% of participants expressed interest in participation. On average, 70% of people in the 3 treatment arms completed their 12-week treatment course, including 76% of participants in the HaRT-A+XR-NTX arm. In contrast, only 52% of participants in TAU attended the Week 12 assessment.

Table 1.

Baseline Sample Characteristics by Treatment Arm (M/SD / %)

Variables	HaRT-A+XR-NTX	HaRT-A+Placebo	HaRT-A	TAU
Age	49.27 (9.11)	46.55 (10.46)	49.38 (7.35)	47.51 (9.50)
Sex assigned at birth (female)	14.9% (11)	12.8% (10)	16.5% (13)	20.8% (16)
Ethnicity (Hispanic/Latinx)	8.1% (6)	9.0% (7)	14.1% (11)	13.0% (10)
Race
American Indian/Alaska Native	14.9% (11)	10.3% (8)	15.2% (12)	23.4% (18)
Black/African American	43.2% (32)	34.6% (27)	27.8% (22)	18.2% (14)
Native Hawaiian/Pacific Islander	1.4% (1)	1.3% (1)	0.0% (0)	1.3% (1)
White/European American	27.0% (20)	30.8% (24)	30.4% (24)	36.4% (28)
More than one race a	10.8% (8)	16.7% (13)	12.7% (10)	18.2% (14)
“Other”	2.7% (2)	6.4% (5)	13.9% (11)	2.6% (2)
Prevalence of cigarette smoking	82.4% (61)	87.2% (68)	75.9% (60)	85.7% (66)
Past-month polysubstance useb	79.7% (59)	80.8% (63)	73.1% (57)	77.9% (60)
Alcohol use disorderc
Mild	5.4% (4)	3.8% (3)	5.1% (4)	5.2% (4)
Moderate	24.3% (18)	24.4% (19)	16.5% (13)	33.8% (26)
Severe	70.3% (52)	71.8% (56)	78.5% (62)	61.1% (47)
Concurrent substance-use treatment attendance	5.6%	2.7%	7.8%	1.3%

Notes. HaRT-A = behavioral Harm-Reduction Treatment for AUD. XR-NTX = extended release naltrexone.

^aOf participants who identified with “more than one race,” 69% reported American Indian/Alaska Native heritage; thus, 26% of the overall sample reported some American Indian/Alaska Native heritage, representing 30 American Indian/Alaska Native/First Nations tribes and communities.

^bFor the overall sample, past-month polysubstance use included cannabis (67%), crack cocaine/cocaine (29%), methamphetamine/amphetamine (15%), non-heroin opioids (10%), benzodiazepine (3%), and heroin (2%). Less than 1% of the sample reported using inhalants, barbituates, MDMA and psychedelics.

^cTo qualify for the study, all participants had to meet DSM-IV-TR criteria for alcohol dependence. In this table, we added up the number of alcohol dependence symptoms participants reported to approximate the current DSM-5 severity categories. Mild = fulfilling 2–3 symptoms, Moderate = fulfilling 4–5 symptoms, Severe = fulfilling 6+ symptoms.

Likelihood of missing data was associated with treatment arm (ps < .02) but not other sociodemographic variables (ps > .11) or baseline outcomes (ps >.23). As arm is the predictor in primary analyses, the missing at random assumption could plausibly be satisfied.⁴³ Sensitivity analyses also indicated robustness to missingness: We observed no treatment effect depreciation across tests and limited variation in associated effect sizes (see Appendix page 83).

Primary Outcomes: 3 Treatment Arms Versus TAU across the 12-week Treatment Period

HaRT-A + XR-NTX versus TAU.

When compared to TAU, HaRT-A+XR-NTX participants showed statistically significant improvements during the treatment period (i.e., weeks 0–12) across 4 of the 5 primary outcomes (see Figure 2 for model-estimated values, Table 2 for raw values, Appendix page 85 for full model statistics). The HaRT-A+XR-NTX arm evinced a medium-to-large effect for reductions in peak alcohol quantity (AQUA; linear B = −.48, CI = −.79, −.18, p = .01; quadratic B = .12, CI = .02, .22, p = .06; effect size d = −.68), medium effects for alcohol-related harm reduction (SIP; linear B = −2.22, CI = −3.39, −1.06, p = .002; effect size d = −.56) and physical HR-QoL improvement (SF-12; linear B = .66, CI = .23, 1.10, p = .012, effect size d = .43), and a small effect for reduction in alcohol frequency (ASI; linear B = −4.42, CI = −8.09, −.76, p = .047; quadratic B = 1.33, CI = .07, 2.60, p = .083; effect size d = −.16).

Figure 2.

Graph of estimated, marginal means over the study course.

Table 2.

Raw Descriptive Statistics for Outcome Variables by Treatment Arm Over Time

	M(SD)/%
Variables	Week 0	Week 4	Week 8	Week 12	Week 24	Week 36
Standard drinks on peak drinking occasion (AQUA)
HaRT -A+XR-NTX	32.02(1.70)	18.18(2.41)	13.56(3.15)	12.81(3.30)	13.14(3.91)	12.46(3.36)
HaRT -A+Placebo	32.15(1.74)	21.33(2.32)	17.21(2.44)	19.09(2.28)	18.01(3.04)	16.30(2.73)
HaRT-A	25.95(1.79)	18.97(2.76)	15.85(2.56)	12.92(3.18)	13.43(3.78)	14.31(3.46)
TAU	27.37(1.98)	22.40(2.23)	20.45(2.41)	16.80(3.16)	16.77(3.36)	13.94(3.42)
Alcohol frequency (days per month; ASI)
HaRT-A+XR-NTX	23.08(7.55)	17.87(10.85)	17.18(11.83)	15.15(11.33)	16.20(11.82)	15.65(11.58)
HaRT-A+Placebo	24.63(8.02)	18.81(11.46)	18.67(11.22)	18.68(11.45)	18.94(11.68)	16.88(11.12)
HaRT-A	23.42(8.63)	17.43(10.91)	18.00(11.80)	15.32(11.97)	13.04(11.66)	15.02(11.75)
TAU	23.25(8.66)	22.37(10.46)	20.83(10.51)	17.15(10.98)	17.93(11.91)	14.59(11.51)
Alcohol-related harm (SIP)
HaRT-A+XR-NTX	23.70(12.03)	15.33(11.59)	13.98(12.48)	12.15(11.25)	12.76(12.97)	13.35(12.50)
HaRT-A+Placebo	23.00(12.05)	17.00(11.44)	15.52(12.85)	16.08(13.20)	15.89(13.89)	13.19(13.40)
HaRT-A	25.69(10.25)	17.98(12.51)	17.15(11.97)	15.50(13.03)	17.90(16.19)	16.87(14.54)
TAU	22.70(12.05)	21.39(13.67)	18.31(14.15)	21.00(14.06)	16.15(11.99)	16.36(13.62)
Mental HR-QoL (SF-12)
HaRT-A+XR-NTX	17.21 (4.63)	18.90(5.77)	20.10(5.32)	19.63(4.09)	19.29(4.98)	20.10(4.87)
HaRT-A+Placebo	18.95(4.80)	20.38(4.92)	19.80(4.88)	20.60(5.81)	20.04(5.30)	21.88(5.30)
HaRT-A	17.45(4.75)	19.44(5.20)	19.11(5.64)	20.55(5.80)	19.47(5.31)	19.41(5.64)
TAU	18.86(550)	19.37(5.93)	18.77(7.08)	18.10(6.17)	19.77(4.52)	19.51(5.42)
Physical HR-QoL (SF-12)
HaRT-A+XR-NTX	15.64(4.52)	17.10(4.66)	17.31(4.64)	17.63(4.73)	17.15(5.01)	16.75(5.31)
HaRT-A+Placebo	16.78(4.32)	17.16(4.58)	17.74(4.99)	17.96(5.02)	16.94(4.92)	17.46(5.31)
HaRT-A	15.91(4.67)	16.17(4.92)	16.11(4.83)	16.86(5.20)	17.09(5.23)	15.98(5.35)
TAU	17.25(4.80)	17.66(5.91)	17.23(6.17)	16.72(5.67)	17.26(5.40)	18.23(4.57)
Ethyl Glucuronide (% negative tests)
HaRT-A+XR-NTX	25.7%	33.3%	40%	38.2%	27.1%	34.0%
HaRT-A+Placebo	25.6%	18.9%	21.3%	14%	21.3%	20.8%
HaRT-A	30.8%	25.9%	27.3%	42.9%	26.9%	25.0%
TAU	15.6%	17.9%	24.4%	17.9%	27.5%	21.1%

Notes. HaRT-A = Harm reduction treatment for alcohol use disorder, which is the counseling portion of the combined pharmacobehavioral treatment in this trial. XR-NTX = Extended-release naltrexone. TAU = community-based services-as-usual control condition. HR-QoL = health-related quality of life.

HaRT-A + Placebo versus TAU.

There were statistically significant linear effects on 3 of the 5 primary outcomes, but effects were smaller for the HaRT-A+Placebo arm (Figure 2, Appendix page 85 for full model statistics). Compared to TAU, HaRT-A+Placebo participants evinced small-to-medium effects for physical HR-QoL improvement at the end of treatment (SF-12; linear B = .53, CI = .09, .98, p = .050, effect size d=.35). HaRT-A+Placebo participants also evinced a small effect for their reduction in peak alcohol quantity (AQUA; linear B = − .41, CI = −.67, −.15, p = .010; quadratic B = .12, CI = .03, .21, p = .024; effect size d = −.23), and alcohol frequency (linear B = −5.95, CI = −9.72, −2.19, p = .009; quadratic B = 1.87, CI = .58, 3.15, p = .017; effect size d = −.13).

HaRT-A alone versus TAU.

When comparing HaRT-A alone to TAU, there were small-to-medium effects for a reduction in alcohol-related harm (SIP; linear B = −1.58, CI = −2.73, −.42, p = .025, effect size d = −.40) and an improvement in physical HR-QoL (SF-12; linear B = .63, CI = .18, 1.07, p = .020, d=.41).

Primary Outcomes: 3 Treatment Arms Versus TAU During the 24-week Posttreatment Period

There was no treatment decay per se among the active treatment arms during the posttreatment period (i.e., after treatment discontinuation at week 12 through the week 36 follow-up; see Figure 2, Appendix Page 85). During the posttreatment period, however, HaRT-A+XR-NTX and HaRT-A participants’ trajectories differed statistically and were less favorable for alcohol-related harm and physical HR-QoL compared to TAU (Figure 2, Appendix Page 85). Findings show that where the active treatment arms plateaued for these outcomes after treatment withdrawal, the TAU arm showed improvements.

Primary Outcomes: Active Medication Versus Placebo Arms

In a dismantling design feature, we isolated the potential “pure” medication effect by comparing the double-blinded active medication and placebo arms on alcohol and QoL outcomes. The models showed no statistically significant treatment effects (all ps > .10; see Appendix Pages 90 for model and parameter statistics).

Secondary Outcomes

Treatment arm effects on biochemical marker of alcohol use.

HaRT-A+XR-NTX participants were nearly 3 times more likely than TAU participants to provide negative EtG tests at the 12-week follow-up (OR = 2.87, CI = 1.05, 7.80, p = .039). We did not observe statistically significant HaRT-A+XR-NTX effects for EtG at the 24- (OR = .98, CI = .38, 2.53, p = .961) and 36-week follow-ups (OR = 1.96, CI =.71, 5.40, p = .192) .

HaRT-A only participants were over three and a half times more likely (OR = 3.60, CI = 1.33, 9.74, p = .012) to have negative EtG tests compared to TAU participants at the 12-week follow-up. We did not observe statistically significant treatment arm effects at the 24-week (OR = 1.00, CI = .39, 2.55, p = .994) and 36-week (OR = 1.30, CI = .46, 3.63, p = .622) follow-ups.

There were no statistically significant effects for HaRT-A + placebo compared to TAU at weeks 12 (OR = .74, CI = .23, 2.40, p = .619), 24 (OR = .71, CI = .26, 1.93, p = .502), or 36 (OR = .99, CI = .34, 2.91, p = .987).

HaRT-A Adherence and Competence.

On average, study physicians/nurses delivered 93% (SD =.11) of expected treatment components per session and were rated as showing between “acceptable” and “high competence” (M=4.28–4.85) across competence dimensions, where 0=”absence of this characteristic” and 6=“very high levels of this characteristic, top 10% of providers.”

Adverse Events and Serious Adverse Events.

There were no statistically significant differences between the active treatment arms and TAU on adverse events and potential side effects of XR-NTX (ps > .07). There was one exception: The XR-NTX arm reported significantly lower odds of “itching” than the TAU arm (OR = .41; CI = .22, .76; p = .005) after treatment exposure.

As expected in a population experiencing chronic homelessness and severe AUD, participants experienced serious adverse events during the study, including 66 hospitalizations, 3 suicide attempts, and 3 deaths (see Table 3 for breakdown by treatment arm). Only one hospitalization was related to study procedures (i.e., difficulty ambulating due to injection site hematoma led to an emergency department visit which precipitated alcohol withdrawal and subsequent hospitalization). The participant fully recovered and completed all follow-up assessments. There were no statistically significant treatment arm differences between active treatments and TAU on emergency department visits or hospitalizations (see Table 3).

Table 3.

Serious adverse events recorded during the clinical trial

	HaRT-A + XR-NTX	HaRT-A + Placebo	HaRT-A alone	TAU
Deathsa	0	0	0	3
Hospitalizationsb	25%	29%	24%	20%
ED visitsb	62%	67%	62%	59%
Suicide attemptsb	1	1	1	0

Notes.

^aDeaths, ER visits and suicide attempts were determined to be unrelated to study procedures; notes regarding 1 study-related hospitalization are discussed in text.

^bSelf-report data informing table statistics were available for 269 participants. Chi-square tests of independence indicated the impact of treatment arm on hospitalizations, χ²(3)=1.28, p=.73, and emergency department (ED) visits, χ²(3)=0.78, p=.85, was not statistically significant. Given the low cell size, we were unable to test for treatment arm differences on suicide attempts and deaths.

Discussion

Findings indicated that combining behavioral harm-reduction treatment and XR-NTX is engaging and efficacious for this population. Consistent with hypotheses, HaRT-A+XR-NTX participants evinced the most consistent positive outcomes compared to community-based services as usual alone (TAU), with improvements across 5 of 6 primary and secondary alcohol and HR-QoL outcomes over the 12-week treatment period. HaRT-A+Placebo and HaRT-A only participants evinced statistically significant improvements compared to TAU on 3 of the 6 primary and secondary outcomes. Treatment effects plateaued but were maintained over the 24-week posttreatment period. That said, there were no statistically significant differences on outcomes between the HaRT-A+XR-NTX and HaRT-A+Placebo arms, so these more consistent, significant effects for the XR-NTX + HaRT-A arm may not be explained due to the medication effect alone.

Combined behavioral harm-reduction treatment plus XR-NTX for AUD is engaging for this population.

Of those approached, nearly all (97%) were interested in participation, and over three-fourths of participants in the HaRT-A+XR-NTX arm attended the final, Week 12 treatment session. In contrast, just over half of TAU participants attended Week 12 assessments. This strong engagement stands in direct contrast to that of the only other prior RCT of XR-NTX in this population, in which 97% of individuals approached refused participation, and only a single participant returned after the initial injection.⁵³ In that study, authors cited “not wanting to change drinking habits” as a key reason for lack of engagement. The present study removed this barrier by supporting participants in developing their own treatment goals.

HaRT-A+XR-NTX was associated with decreased alcohol use and alcohol-related harm as well as increased physical HR-QoL.

This study showed that prior positive XR-NTX findings^22–25 may be extrapolated to a more severely affected population, one in which all had experienced homelessness in the past year, 80% met criteria for severe AUD, and 96% had symptoms of physiological dependence. Compared to TAU, the HaRT-A+XR-NTX arm evinced consistent, significant improvements across self-reported and biochemical alcohol outcomes as well as on physical HR-QoL. Findings also indicated some positive but less consistent effects for the HaRT-A+Placebo and HaRT-A only arms; thus, it appears treatment effects are cumulative for the 2 elements of the combined pharmcobehavioral approach but not strictly additive.

Following treatment withdrawal at Week 12, the treatment arms’ outcome trajectories plateaued and were maintained through the 36-week follow-up. Given the observed plateau in posttreatment effects, applying HaRT-A+XR-NTX as a maintenance treatment might better facilitate continued treatment gains.

Interestingly, TAU participants appeared to rebound with improved outcomes during the posttreatment period. Missing data analyses shed some light on this phenomenon. TAU participants showed higher attrition starting immediately after treatment assignment, perhaps due to demoralization following the realization they were not in an active treatment arm; however, there were no other significant baseline predictors of missing data. It is possible, as study staff anecdotally noted, that TAU participants who were able to return for follow-up assessments were simply those who were doing better overall.

Not all hypothesized effects were supported. We found no significant differences between the HaRT-A+XR-NTX and HaRT-A+Placebo treatment arms. Of note, this study was powered to detect a medium effect for the active medication and placebo treatment arm comparison; thus, a smaller effect may not have been detectable with the given sample size. Taken together with the lack of clear one-to-one additive effects of the XR-NTX and HaRT-A when compared with TAU, it appears the active medication effects alone do not explain the study findings. Instead, both XR-NTX and HaRT-A appear to build on one another in a more subtle way. Future, larger-scale studies are needed to better understand the underlying mechanisms of the observed treatment effects.

Notably, there were no treatment effects on mental HR-QoL, and while statistically significant, the clinical significance of the treatment effects on physical HR-QoL is not entirely clear. A prior systematic review provides some insight. It showed that common HR-QoL measures rarely show significant treatment effects in trials involving people with AUD,⁵⁴ which may be attributable to the generic nature of such questionnaires. Fortunately, alcohol researchers have recently developed and validated a participant-driven, alcohol-specific HR-QoL measure that holds promise for future trials.⁵⁵

Limitations

Treatment was brief: Participants in the 3 treatment arms were exposed to 5 treatment sessions, and in the XR-NTX arm, 3 doses of active medication. While it led to statistically significant effects on the outcome trajectories of a highly physiologically dependent, multimorbid, and non-treatment-seeking population, the treatment’s relative brevity does not mirror typically longer-term clinical contacts this population often has for their chronic conditions. Future studies are needed to test whether HaRT-A+XR-NTX as a longer-term maintenance treatment approach versus a brief treatment can facilitate even greater reductions in alcohol use and alcohol-related harm and improvements in HR-QoL.

Whereas the 2 injection arms were double-blinded, there was no feasible way to blind our interventionists to the behavioral interventions, because to minimize treatment arm differences, we wanted to keep staffing across the arms consistent. Given the very unique nature of embedding limited staff into tight community quarters, we were unable to blind assessment staff to TAU or behavioral arms. We therefore cannot preclude experimenter bias or expectancy effects for unblinded arms.

Considering this population’s necessary focus on day-to-day survival and the resulting itinerance and displacement, participants’ treatment completion rates were strong. In fact, treatment completion was on par with findings from a recent meta-analysis of 151 substance-use treatment studies (i.e., 70%), which were largely conducted with participants who had greater incomes, more housing stability, treatment readiness, and fewer risk factors.⁵⁶ Follow-up completion in the present study was better than that of the flagship study of extended-release naltrexone for AUD treatment (i.e., 60%).²² That said, data missingness can lead to reduced power and biased estimates; thus, we engaged various means of minimizing and addressing attrition and resulting data missingness.⁴⁰ We built trust over many years with the involved agencies, which resulted in strong partnerships and community-inspired and integrated engagement and retention strategies.⁵⁷ We used analyses and estimation methods that make use of all available data and thus avoid problematic listwise deletion or simple imputation methods that can introduce bias.⁴⁴ Finally, we modeled potential missingness patterns to test overall robustness of our models.⁴² These analyses indicated that the majority of the dropout occurred in the TAU arm and was related to treatment assignment. Because we included treatment arm as the primary predictor in our outcome analysis, we have greater confidence that our findings are robust to missingness. These measures do not fully preclude concerns about estimate bias, and future developments in missingness research are needed to further hone our methodological strategies.

This study’s generalizability may be limited by its geographic location as well as sociodemographics and substance-use patterns specific to the homeless population in this area. In particular, this study was implemented in low-barrier settings serving a nontreatment-seeking, homeless population in a large, resources-rich city in the US Pacific Northwest. Further, we did not exclude polysubstance users in order to provide a real-world assessment of treatment efficacy and embody harm reduction’s inclusive and low-barrier approach. Thus, findings may not generalize to other communities where abstinence-based service settings or solely alcohol-using populations are the norm. Finally, the sample was representative of the larger US homeless population in terms of race and age,⁵⁸ and was representative of the local homeless, AUD-affected community. These findings, however, may not be generalizable to youth experiencing homelessness, communities with greater Latinx representation, and housed individuals.

Conclusions

This study is the first RCT documenting the efficacy of XR-NTX as a pharmacological support for patient-driven harm reduction and HR-QoL improvement in people with AUD experiencing homelessness. Compared to TAU, the HaRT-A+XR-NTX arm evinced consistent, significant improvements across 5 of 6 primary and secondary alcohol and HR-QoL outcomes. Trajectories plateaued but were maintained after treatment discontinuation. Findings indicated some positive but weaker and less consistent effects of HaRT-A+Placebo and HaRT-A alone. Given the observed plateau in posttreatment effects, applying HaRT-A+XR-NTX as a maintenance treatment might better facilitate continued treatment gains. Further research is needed to determine whether this treatment can help reduce health care service utilization and associated costs and to determine the optimal length of harm-reduction treatment for AUD.

Research in context.

Evidence before the study:

Abstinence-based treatment for alcohol use disorder (AUD) is not highly engaging or effective for people experiencing homelessness. In contrast, harm-reduction approaches are more patient-centered and forgo an absolute focus on alcohol abstinence to instead support decreased alcohol-related harm and improved quality of life. A prior randomized clinical trial (N = 168) showed statistically significant, 3-month improvements on alcohol outcomes for behavioral harm-reduction treatment compared to community-based services as usual. A single-arm pilot (N = 31) indicated promise for combining behavioral harm-reduction treatment and extended release naltrexone in reducing alcohol use and alcohol-related harm.

Added value:

The present randomized clinical trial builds on these promising pilot trials by testing the relative efficacy of combining pharmacological and behavioral harm-reduction treatment for people experiencing homelessness and AUD. Findings confirmed the positive treatment effects for behavioral harm-reduction treatment for AUD that were shown in the prior RCT. This study also added to the existing research by showing that combining pharmacological and behavioral harm-reduction treatment can synergistically strengthen these treatment effects on both alcohol outcomes and physical health-related quality of life. However, findings from double-blind comparisons showed no significant differences between participants receiving active versus placebo injections. Thus, the strong findings for the combined treatment cannot be ascribed to the medication effect alone. Further study of combined pharmacological and behavioral harm-reduction treatments for AUD is needed to assess their relative contributions over a longer-term treatment course.

Implications:

In the past few years, harm-reduction approaches have emerged as important means of engaging people experiencing homelessness and AUD, who are often marginalized by or excluded from abstinence-based AUD treatment settings. The present study showed that removing typical barriers to AUD treatment (e.g., providing treatment in community-based settings and not requiring abstinence) engenders strong study engagement and retention in a population long considered to be “hard to reach.” Further, findings indicated a combined pharmacobehavioral harm-reduction treatment is efficacious and that, even alone, behavioral harm-reduction treatment confers benefits. Plateauing effects were observed after treatment withdrawal, which suggests the need for a maintenance approach versus a shorter-term brief intervention approach with this population. That said, when harm-reduction treatment is integrated into community-based settings where other services are provided, contacts can be brief and spaced up to a month apart. Future, larger-scale studies are needed to test whether this approach can save on cost and effort while increasing treatment reach.