An efficacy trial of adaptive interventions for alcohol use disorder

Abstract

Background:

Adaptive interventions, sometimes referred to as “stepped care”, are those interventions in which the type or dosage of treatment offered to patients is tailored to baseline clinical presentation and then adjusted over time in response to patient progress or lack thereof. Currently, no adaptive brief interventions exist specifically for alcohol use disorder (AUD).

Method:

This study used a sequential multiple assignment randomized trial design with 160 individuals with AUD recruited both locally and nationally who had a goal to reduce or abstain from drinking. Participants received brief advice (BA) and then the study reassessed them three weeks later; the study randomized those who did not respond to BA, defined as reducing their drinking to low-risk guidelines, to two session of motivational interviewing (MI) or more BA. The study then reassessed participants at week 8. The study re-randomized nonresponders to receive either MI alone or MI plus behavioral self-control therapy (BSCT), also referred to as coping skills therapy, and evaluated participants at week 13.

Results:

Overall, participants receiving any BSCT made the greatest reductions in drinking. Participants who received MI at week 4 and BSCT at week 8 outperformed all other groups.

Conclusion:

Findings reveal that prolonged treatment, more sessions, and/or a specific combination of MI and BSCT provided optimal outcomes. Future research should determine whether such an algorithm holds across heterogenous groups of individuals with AUD.

1. Introduction

Alcohol use disorder (AUD) is a costly and heterogeneous disorder associated with numerous comorbid disorders and high rates of mortality (Degenhardt et al., 2018; Rehm et al., 2017). In 2018, about 5.8% of the U.S. adult population endorsed criteria for current AUD, 26.5% reported binge drinking (> 4 or 5 drinks for women and men, respectively), and 6.6% reported heavy drinking (binge drinking on > 5 days) in the past month (National Institute on Alcohol Abuse and Alcoholism, 2018). Despite widespread prevalence, fewer than 1 in 10 individuals with AUD report receiving treatment in the past year (Edlund et al., 2012). Expanding treatment options in mainstream health care for individuals with AUD, particularly among those who may not require specialty care, can greatly increase access to care. Specifically, delivery of brief interventions that provide goal-choice (i.e., moderation or abstinence) offered in mainstream health care settings, such as primary care, is likely to substantially increase uptake of treatment for AUD (Morgenstern et al., 2012; Willenbring, 2007). Research to improve brief AUD treatments for such settings is therefore a high priority.

1.1. The opportunity for adaptive interventions

One recommended approach for implementing brief interventions for AUD in primary care and other mainstream healthcare settings is to utilize adaptive interventions (AIs). AIs, also referred to as “stepped care”, are those interventions in which the type or dosage of treatment offered to patients is tailored to baseline clinical presentation and then adjusted over time in response to patients’ progress or lack thereof (Lei et al., 2012). AIs consist of 3 components: 1) a sequence of decision rules; 2) a set of possible treatments for each critical decision point; and 3) tailoring variables to monitor patients’ progress and identify nonresponse (Chakraborty, 2011; Lei et al., 2012). By dynamically responding to patient progress, AIs optimize intervention delivery, providing a cost effective and targeted response from providers prior to referral to specialty care, if needed.

AIs offer significant advantages over standard fixed treatments. First, AIs directly address a common problem of AUD treatment: treatment nonresponse. Studies show high levels of nonresponse even among patients receiving evidenced-based treatments (EBT) for AUD. Rates of nonresponse in major studies have ranged from 30% to 85% (Anton et al., 2006; Johnson et al., 2007; McKay, 2011; Project MATCH Research Group, 1997), with brief treatments for mild to moderate AUD showing upward of a 50% nonresponse rate (Kuerbis et al., 2012). While fixed treatments do not directly attend to the issue of nonresponse, a primary focus of AI research is how best to intervene with nonresponders in an effort to improve the overall efficacy of a sequence of treatments.

Second, validated AIs provide a set of decision support tools to select and adjust treatment for chronic diseases that balance efficacy and cost (Chakraborty, 2011); however, evidenced-based AIs are yet to be established for adults with AUD. Studies on EBTs provide information based on a static model of care (i.e., which treatment works best); however, in real world practice, clinicians typically adjust treatments over time based on patient response. This dynamic element of care is especially important in the treatment of chronic illnesses, where treatments often consist of a series of interventions tailored to individual patients’ characteristics, phase of illness (at-risk, acute, remission), and prior response to treatment. Traditional studies of EBTs do not provide guidance related to real-world implementation: what is the best and least restrictive treatment for a patient with these presenting problems, and if the patient fails to respond, what other treatment should be tried? AIs address these practice considerations using an empirical approach.

1.2. Research on AIs for substance use disorders (SUD)

Several rigorous AI studies exist for SUD populations treated in specialty care (Kranzler & McKay, 2012), providing evidence for the possibility of developing evidenced based AIs for AUD. These studies include high quality, programmatic research by McKay (2009) on continuing care with adults with a range of SUDs who received intensive outpatient care and by Kidorf et al. (2004) on adults receiving methadone treatment for opioid use disorder. For example, McKay showed that nonresponders to initial intensive outpatient (IOP) treatment with SUD fared better if they subsequently received intensive clinic-based care, whereas IOP responders with SUD fared better in telephone monitoring than intensive care (McKay, 2009). Studies by Kidorf et al. (2004) and Marlowe et al. (2008), which examined individuals involved in drug court, also demonstrated that stepped care interventions yield better outcomes for nonresponders than standard care.

1.3. Research on AIs for AUD

In contrast to those studies examining AIs for SUD, research on AIs for AUD is limited. A review identifying four studies reporting findings on AIs for AUD (Jaehne et al., 2012) found that the studies were methodologically limited or had inadequate differentiation in treatment contrasts. For example, Drummond et al. (2009) piloted an AI for men with AUD in primary care. Within that study participants were randomized to receive: 1) five minutes of advice and an educational pamphlet, or 2) the AI. Those assigned to the AI received a single session of counseling and were assessed one month later. Non-responders (those drinking >21 drinks in one week) were then assigned to 4 sessions of Motivational Enhancement Therapy (MET) and reassessed a month later. Non-responders at month two were referred to specialist treatment in the community. Although the study employed a strong control group for comparison to the AI, very few participants were non-responders to the initial session of counseling in the AI group. Of the original 54 participants in the AI group, only 17 received MET, and only 1 received a referral to specialty care. No significant group differences were found, which authors concluded was likely due to lack of power (only 18 individuals received stepped care). The review concluded that there was a need for more and higher quality studies on AI (Jaehne et al., 2012), such as using experimental procedures to adapt/assign the next stage of intervention.

1.4. The current study

To address gaps in knowledge about effective AIs for AUD, the current study aimed to test drinking outcome effects of stepped care treatment for individuals with AUD. We implemented a twelve week AI for individuals with AUD using a sequential multiple assignment randomized trial (Chakraborty, 2011; Lei et al., 2012) to test the efficacy of each stage of the assignment and overall intervention strategies. The AI included various combinations of brief advice (BA), motivational interviewing (MI), and behavioral self-control therapy (BSCT) over the course of twelve weeks, described further below.

2. Method

2.1. Study overview

We recruited 160 individuals with AUD (Registration no.: NCT02511808) to participate in a stepped care brief intervention to reduce heavy drinking. The study used a sequential multiple assignment randomized trial (SMART) to test “adaptive strategies” or sequences of interventions (Chakraborty, 2011; Lei et al., 2012). The study used brief forms of evidence-based interventions for AUD, which consisted of brief advice (BA), motivational interviewing (MI, Miller & Rollnick, 2013) and behavioral self-control training (BSCT, Hester, 1995). Briefly, SMART involves multiple intervention stages; each stage corresponds to one of the critical decisions involved in the AI. Participants move through the multiple stages, and at each stage the study randomly (re)assigns participant to one of several intervention options. SMART allows for an experimental comparison of different treatment options at each stage of the intervention (Lei et al., 2012), as well as pathways of treatment over the course of time. The Human Subjects Review Board at Northwell Health (protocol # 14-413) reviewed and approved the study procedures.

2.2. Participants

2.2.1. Recruitment.

The study used general advertising online and in local media to recruit participants, both nationally and locally (including within and outside of the health care system in which the study took place), seeking professional assistance to moderate their drinking. If a participant met initial eligibility criteria via phone screening, study staff scheduled the participant for an in-person or online (via telehealth) full screening assessment, depending on the participant’s preference. Sixty-six percent of participants chose telehealth as their preferred modality.

2.2.2. Study eligibility.

The study considered participants eligible if they (1) were between ages 18 and 75 and (2) had an estimated average weekly consumption for women: ≥15 standard drinks per week, or ≥12 standard drinks per week and >2 binge days (>4 standard drinks per sitting); for men: ≥24 standard drinks per week, or ≥14 drinks per week and >2 binge days (>5 standard drinks per sitting). The study excluded participants if they: (1) had an SUD (for any substance other than alcohol, marijuana, nicotine) or were regular (defined as greater than weekly use) nonalcohol drug users; (2) had a serious psychiatric disorder (e.g., bipolar disorder, schizophrenia); (3) exhibited current or lifetime physical alcohol withdrawal symptoms, as indicated by self-report of symptoms and/or having received inpatient AUD treatment (e.g., detoxification); and (4) were actively involved in another treatment for AUD (i.e., self-help groups, outpatient therapy) in the past 90 days. The exclusion criterion based on history of withdrawal was to ensure that an inappropriately brief treatment, implemented over time, would not put the participant at risk for withdrawal once again, or that the participant would not be mismatched with an inappropriate moderation-focused treatment.

2.3. Procedures

This study used a SMART design with multiple points of randomization (see Fig 1). After a phone-based screening assessment, participants completed the consent form and quiz electronically via REDCap to ensure comprehension. The study conducted all follow-up assessments via telehealth (online video conference) or in-person, and the participant could choose which they preferred. Participants then completed a screening interview, during which they completed a battery of assessments to verify study eligibility. The study enrolled eligible participants and informed participants that they could receive up to 5 sessions of therapy. One week after screening, all participants completed their baseline assessment and (3) received an initial session of BA (20 min) from a research assistant (RA). The study modeled BA after general screening and brief intervention and involved providing brief personalized feedback about risk for AUD based on age, sex, drinking patterns and the participant’s score on the Alcohol Use Disorder Identification Test (AUDIT, Babor et al., 2001); providing guidelines for low-risk drinking; determining the person’s reasons for moderating; setting drinking goals; and discussing strategies and potential challenges for reducing drinking.

Figure 1.

CONSORT diagram of study procedures.

2.3.1. Randomization.

At week 4, the study reassessed participants. Study staff told those who had “responded” (defined as reducing their drinking to the level of the NIAAA safe drinking guidelines, i.e. for men under the age of 65: ≤ 14 standard drinks per week and < 5 standard drinks on one occasion; women and men over the age of 65 ≤ 7 standard drinks per week and < 4 standard drinks on one occasion) that they had reduced their drinking and thus would not be provided further treatment. Responders continued to participate in the remaining follow-up assessments. The study urn randomized those who were “nonresponders” to BA (based on responses to their AUDIT Alcohol Consumption Questions [AUDIT-C, Bush et al., 1998] , in other words the first three questions of the AUDIT, and gender) to either one 20 minute session of more BA (referred to here as BA Plus) or two, one hour sessions of MI. From study outset, study staff told participants to expect anywhere from 1 to 5 meetings with an RA and/or therapist for treatment. Study staff initially told participants why they would receive a specific amount of treatment to de-emphasize randomization as a point of success or failure on the part of the participant.

At week 8, the study reassessed participants again and urn randomized nonresponders a second time (using baseline AUDIT-C and gender) to additional treatment, consisting of one-hour long sessions. The study re-randomized those randomized at week 4 to BA Plus at week 8 to either (1) two sessions of MI or (2) one session of MI with four sessions of BSCT. Nonresponders previously randomized at week 4 to MI were re-randomized at week 8 to either (1) an additional session of MI or (2) 4 sessions of BSCT. Responders at week 8 did not receive further treatment but continued to participate in follow-up assessments. Participants then completed an end-of-treatment (week 13) and a post-treatment follow-up (week 24) assessment. During these visits, study staff offered these participants appropriate referrals for community treatment.

2.3.2. Timing of assessments, compensation, and retention.

Participants completed a battery of assessments at screen; baseline; weeks 4, 8, 13, and 24. Participants received escalating compensation ($20–$75) for each completed assessment. Retention was high, with rates for weeks 4, 8, 13, and 24 assessments at 92.0%, 84.7%, 77.9%, 74.2%, respectively. Participants also completed daily ecological momentary assessments (EMA) for the 12 weeks of the trial (see Kuerbis et al., 2020). This analysis did not use data from the EMA.

2.4. Study interventions

2.4.1. BA Plus.

BA Plus (20 min) consisted of eliciting the participants’ perspectives about the previous BA, providing feedback about current drinking, assessing current readiness for change using importance and confidence rulers, and then summarizing the participant’s statements in the session.

2.4.2. Motivational interviewing.

This study used an adapted version of MI from MET in Project MATCH (Miller et al., 1992; Project MATCH Research Group, 1993), utilized in our prior studies (Morgenstern et al., 2007; Morgenstern et al., 2012) and described here briefly. The study delivered structured personalized feedback (i.e., drinking behavior percentile ranks, AUD risk factors), a change plan, and other directive activities with the goal of eliciting self-motivational statements and strengthening commitment to change (Miller et al., 1992, pp. 13–32). Counseling emphasized either moderation or abstinence (depending on participants’ goals).

2.4.3. Behavioral self-control therapy.

BSCT is a manual-based cognitive behavioral therapy (CBT) for moderation of alcohol consumption that is empirically supported in the treatment of AUD (Morgenstern & McKay, 2007), which we have utilized in our previous studies (Morgenstern et al., 2007; Morgenstern et al., 2012). The current study shortened the BSCT from twelve to four sessions. Sessions focused on the development of skills to modify behavior patterns of excessive drinking, and were structured to include didactic skills training, an in-session performance exercise, and corresponding homework. We selected BSCT as the next step in the adaptive intervention with the assumption that MI would have addressed motivation and commitment to change yet participants may still lack the ability to implement their plan for change due to a lack of skills.

2.4.4. Therapists and training.

Two master’s and four doctoral level therapists provided the psychotherapy interventions. The therapists were highly experienced providers of treatment for SUD (e.g., all but one had five or more years of experience providing MI and BSCT). Therapists participated in an initial, three-hour training on the protocol, and RAs, who delivered BA and BA Plus, attended a two-hour training on BA and multiple practice BAs with each other. Once weekly group supervisions of therapists and RAs (conducted separately) focused on ensuring fidelity to each protocol, and consisted of case discussion and review of session videotapes and refresher didactics. We initially assigned the one therapist who had fewer than five years of experience delivering MI and BSCT practice cases for training purposes.

2.4.5. Compliance with therapy.

Compliance with therapy was relatively high (see Table 1), with a majority of participants receiving the full dosage of their respective interventions.

Table 1.

Attendance to interventions by group.

	Possible # of Sessions	N	% Attended
Initial BA	1	160	100%
Week 4 Treatments
BA Plus	1	66	95%
MI	2	67	94% 1 session; 86% 2 sessions
Week 8 Treatments
Additional MI	1	28	82%
BSCT	4	24	95.8% 2 sessions; 83.3% 3 sessions; 66% 4 sessions
MI Only	2	30	97% 2 sessions
MI + BSCT	5	29	90% 5 sessions

2.5. Measures

2.5.1. Demographics.

A questionnaire collected data on age, gender, educational and occupational information, race and ethnicity, medical history, and history of substance use.

2.5.2. Alcohol use disorder.

The AUDIT-C, which comprises the first three items on the AUDIT, has adequate psychometric properties (Bush et al., 1998) and this study used it to determine preliminary eligibility for the study. The Composite International Diagnostic Instrument, Substance Abuse Module (CIDI-SAM; Cottler et al., 1989), a well-established diagnostic interview with excellent reliability and validity (Wittchen et al., 1991), assessed the number of AUD symptoms, according to the fifth Diagnostic and Statistical Manual (American Psychiatric Association, 2013).

2.5.3. Drinking outcomes.

The Timeline Followback Interview (TLFB, L. C. Sobell, Sobell, Leo, & Cancilla, 1988; M. B. Sobell et al., 1980) assessed frequency and intensity of alcohol use for 90 days prebaseline, during treatment (at weeks 5 and 8), and after treatment (weeks 13 and 24), each covering the time since the last assessment. The TLFB has demonstrated excellent psychometric properties (Carey et al., 2004; Dillon et al., 2005; Vinson et al., 2003). The study aggregated outcome data into two measures: the sum of standard drinks (SSD) and count of heavy drinking days (HDD) for the week prior to the affiliated assessment period (baseline, week 4, week 8, and week 13). We considered drinking days heavy if they exceeded the recommended guidelines of drinking >4 (women) or >5 (men) drinks on one occasion. To correspond to the outcome variable, the prebaseline TLFB data were aggregated into a summary variable for mean SSD and mean weekly HDD, to be used as covariates in the analyses.

2.6. Analytic plan

2.6.1. Initial tests of interventions.

Initial independent regression tests evaluated the main effects of week 4 intervention (BA Plus vs. MI) on week 8 outcomes and week 8 intervention (MI only vs. any BSCT) on week 13 outcomes. Outcomes were counts: sum of standard drinks (SSD) per week and count of heavy drinking days (HDD) per week. Log-linear regression models with a Poisson distribution accounted for overdispersion and provided the best fit for SSD and HSD count outcomes. Responders to initial BA, who the study never randomized at week 4, were not part of the model testing BA Plus vs. MI. Responders at week 8, who the study, therefore, did not randomize at week 8 were excluded from the model testing MI only vs. any BSCT. For outcomes at week 8, the study used week 4 drinking as a covariate. For outcomes at week 13, the study used week 8 as a covariate.

2.6.2. Testing the intervention strategies.

After testing the initial effects of the week 4 and week 8 interventions separately, we then tested the adaptive treatment strategies (i.e., combinations of treatment) on week 13 outcomes. This set of analyses included all responders after week 4. The current trial had four embedded strategies. The study randomized participants to two options at week 4. Participants who did not respond to the week-4 treatment were randomized a second time to two options at week 8. Following Nahum-Shani et al. (2015), we defined binary treatment variables A₁ (BA Plus vs. MI) and A₂ (MI only vs. any BSCT), corresponding to the randomization options at weeks 4 and 8, respectively. Specifically, we defined A₁= −1 for BA Plus and A₁=+1 for MI, and A₂= −1 for any BSCT and A₂=+1 for MI only.

In adaptive intervention trials, participants who respond to the first-stage treatment decision are not re-randomized to a second-stage treatment decision. Thus, initial responders must be accounted for in the analysis as though they were randomized to both of the two second-stage treatment strategies (Almirall et al., 2012; Lei et al., 2012). This replication process produces a sample where responders are overrepresented. Weighted Poisson regression models account for the resultant (weak) correlations within the data. Models utilized robust standard errors, using A₁ and A₂ and their interaction as explanatory variables, and the week 4 version of the response (SSD or HDD) as a covariate.

2.7. Hypotheses

We hypothesized that among nonresponders at week 4, those assigned to MI would significantly reduce drinking by week 8, when compared to those participants who received BA Plus. We also hypothesized that among nonresponders at week 8, those who were re-randomized to skills treatment (BSCT) would significantly reduce their drinking by week 13, when compared to individuals who only received MI. Finally, we hypothesized the adaptive regimen comprising nonresponder treatments of MI at week 4 and BSCT at week 8 would be associated with the largest reduction in drinking at week 13 of all the groups. We included covariates indicating whether the treatment was provided in-person or over video conferencing (telehealth), number of treatment sessions received, and gender in initial models but removed for nonsignificance.

3. Results

3.1. Demographics

Table 2 shows basic demographics of the participants. There were no significant condition differences at weeks 4 or 8 on demographics or AUD severity across condition.

Table 2.

Baseline characteristics of participants.

	Overall Sample (N = 160)
Variable	M or %	SD
Demographics
Age (years)	51.0	11.9
Female	69.4
Race/Ethnicity
Non-Hispanic, White/Caucasian	89.4
Hispanic/Latino, any race	6.3
Other	4.3
Education
High school diploma	5.7
Some college/Associates	21.3
Bachelor’s degree	27.5
Some graduate school or higher	45.6
Employment
Employed (full or part-time)	78
Unemployed/Looking for work	3.8
Not in labor force/not looking for work	18.2
Mean number AUD criteria met	6.8	2.2
Number of Criteria Endorsed for DSM-5
Alcohol Use Disorder (AUD)
2 – 3 criteria, mild AUD	7.5
4 – 5 criteria, moderate AUD	21.3
6 or more criteria, severe AUD	71.2

3.2. Response to initial BA

There were 18 responders to initial BA, leaving 133 individuals to be randomized at week 4 (Figure 1). Responders reduced their overall quantity of drinking (SSD), not the frequency of heavy drinking (HDD).

3.3. Intervention at week 4 predicting week 8 outcomes

3.3.1. SSD.

The regression adjusting for week 4 SSD yielded adjusted means of 23.49 (SE=1.42) and 21.10 (SE=1.42) for the BA Plus and MI groups at week 8, respectively, with no significant effect of condition (x²(1)=1.48, p=0.22). See Table 3 and Figure 3.

Table 3.

Drinking outcomes descriptives (mean (standard deviation)) by time point and group.

	All Participants (N=160)
Week 1 SSD	28.8 (15.3)
Week 1 HDD	3.5 (2.2)
	Responders at Week 4 (N=18)	Non-responders to Initial BA (N=133)
Week 4 SSD	6.2 (6.3)	26.0 (12.5)
Week 4 HDD	1.5 (.7)	3.8 (1.9)
	Responders at Week 8 (N=9)	Non-Responders at week 8, assigned to MI at Week 4 (N=52)		Non-Responders at week 8, assigned to BA Plus at Week 4 (N=59)
Week 8 SSD	17.0 (21.1)	23.5 (16.4)		23.7 (13.4)
Week 8 HDD	1.2 (2.0)	2.4 (2.3)		2.9 (2.4)
	All Pre-Week 8 Responders (N=27)	Non-Responders at Week 8 (N=111)
		MI	BSCT	MI	MBSCT
Week 13 SSD	11.1 (13.7)	22.8 (12.7)	17.8 (8.7)	21.4 (12.0)	18.0 (9.7)
Week 13 HDD	1.1 (1.6)	2.5 (2.0)	1.6 (1.6)	2.4 (2.2)	2.2 (2.1)

Note: SSD=sum of standard drinks; HDD=heavy drinking days; MI=Motivational Interviewing; BSCT=behavioral self-control therapy; MBSCT=Motivational Interviewing + BSCT

Figure 3.

Unadjusted means of SSD and HDD for week 4 to week 8 for those randomized to motivational interviewing (MI) or additional brief advice (BA Plus).

3.3.2. HDD.

The regression adjusting for week 4 HDD yielded adjusted means of 2.65 (SE=0.28) and 2.06 (SE=0.25) for the BA Plus and MI groups at week 8, respectively, with no significant effect of condition (x²(1)=2.73, p=0.10). See Table 3 and Figure 3.

3.4. Intervention at week 8 predicting week 13 outcomes

The study based comparisons on the 111 participants who were nonresponders to the week 4 conditions. See Table 3 and Figure 3.

3.4.1. SSD.

The regression adjusting for week 8 SSD yielded adjusted means of 20.80 (SE=1.23) and 17.44 (SE=1.16) for the MI only and any BSCT groups at week 13, respectively, with a significant effect of condition (x²(1)=4.04, p=0.045).

3.4.2. HDD.

The regression adjusting for week 8 HDD yielded adjusted means of 2.04 (SE=0.22) and 1.59 (SE=0.20) for the MI only and any BSCT groups at week 13, respectively, with no significant effect of condition (x²(1)=2.76, p=0.10).

3.5. Testing the adaptive strategies

3.5.1. SSD at week 13.

Weighted Poisson regression adjusting for week 4 SSD did not support the significance of an A₁ by A₂ interaction between the week 4 and week 8 treatments (x²(1) = 0.07, p = 0.80), with the interaction model showing no significant pairwise differences among the four strategies (p>0.05 for all comparisons). See Figure 4. Dropping the interaction, a main-effects model showed no significant effect of week 4 treatment (x²(1) = 0.24, p = 0.63), with an adjusted mean of 17.81 (SE=1.31) for the participants receiving MI at week 4 compared to 18.61 (SE=0.96) for those receiving BA Plus. There was a significant effect of week 8 treatment (x²(1) = 4.92, p = 0.03), with participants who received MI only in their week 8 treatment having an adjusted mean SSD of 19.82 (SE=1.08), compared to 16.72 (SE=1.04) for those receiving any BSCT.

Figure 4.

Unadjusted means for SSD and HDD for week 8 to week 13 for those who were re-randomized at week 8 (non-responders).

MI=Motivational Interviewing; BSCT=behavioral self-control therapy; BA= Brief Advice Plus.

3.5.2. HDD at week 13.

The A₁ by A₂ interaction between the week 4 and week 8 treatments approached significance (x²(1) = 3.27, p = 0.07), with adjusted means of 2.02 (SE=0.21) for BA Plus + MBSCT, 1.28 (SE=0.24) for MI + BSCT, 2.00 (SE=0.28) for BA Plus + MI and 2.10 (SE=0.31) for MI + MI. Uncorrected pairwise comparisons suggested significant adaptive-strategy differences between the BA Plus + MBSCT and MI + BSCT (p=0.03); BA Plus + MI and MI + BSCT (p=0.05); and MI + BSCT and MI + MI (p=0.03) strategies. See Figure 4.

The main effects model showed no significant difference between week 4 treatments (x² (1)=1.11, p=0.29), with adjusted mean of 1.71 (SE=0.21) for MI and 2.00 (SE=0.19) for BA Plus, or for week 8 treatments (x²(1)=2.22, p=0.14), with an adjusted mean of 2.04 (SE=0.22) for only MI and 1.67 (SE=0.17) for any BSCT.

4. Discussion

This study attempted to identify effective adaptive brief interventions for AUD. We aimed to identify specific strategies that better address nonresponse to successive brief AUD treatments provided at the frontlines of behavioral health care. To do this, we tested both the main effects of two separate brief interventions, MI vs. BA Plus and MI vs. BSCT, and their combination (i.e., adaptive strategies), over the course of 12 weeks.

4.1. Main effects of brief treatments among nonresponders only

For the main effects, we first hypothesized that, among nonresponders to initial BA, week 4 MI would outperform week 4 BA Plus. This hypothesis was not supported. Among the nonresponders to initial BA who we randomized at week 4, there were null findings for MI vs. BA Plus at week 8 across both drinking outcomes. Potential explanations for this null finding are discussed further below and in the limitations section.

Among nonresponders at week 8, we hypothesized that BSCT would outperform MI alone, and findings partially supported this. Among the week 8 nonresponders, those who received any BSCT demonstrated a lower quantity of drinking overall, but no difference in heavy drinking days, at week 13, compared to those who received only MI. It may be that those who struggle to change their drinking (i.e., nonresponders) are most helped by a brief skills-based approach. Participants for this study responded to advertisements calling for volunteers who were interested in reducing their drinking. This suggests some level of existing motivation to change one’s drinking at study initiation, and it may be one explanation for why MI did not provide the dramatic impact that this study hypothesized. In this context, learning skills on how to implement change may be the most critical information in the context of being already motivated enough for change to pursue a brief treatment for AUD.

4.2. Effects of adaptive strategies

For the adaptive strategies, we expected that, when accounting for responders in the analyses, strategies designed to adapt to treatment nonresponse by adding an additional, distinct treatment would yield the biggest reductions in quantity of drinking and frequency of heavy drinking. Specifically, we expected that receiving MI at week 4 and BSCT at week 8 would yield the best outcomes on both SSD and HDD. Our hypotheses, in this regard, were supported partially for SSD and more fully for HDD, and findings revealed a somewhat hierarchical set of outcomes. When accounting for responders, using any BSCT to adapt to consistent non-response later in treatment was superior to providing only MI for lowering quantity of drinking overall at the end of treatment. However, there was no treatment interaction, suggesting the addition of BSCT, regardless of what was previously received, was important for clinically meaningful change. Interestingly, counter to hypotheses, adding MI alone at week 8 to adapt to nonresponse to BA Plus delivered at week 4 did not significantly improve outcomes for this group. Again, this may be due to participants already having significant motivation for change when they entered the study, but potentially no understanding of how to implement changes behaviorally.

For heavy drinking days, the effect was slightly more complex. While, again, receiving BSCT was critical for participants’ reducing their number of heavy drinking days, those who received MI starting in week 4 and then received BSCT at week 8 showed the greatest level of reduction. In fact, pairwise comparisons demonstrated a significant difference in week 13 HDD between the two strategies that received both MI and BSCT. The difference in the strategies lay in the timing and amount of MI (2 sessions starting at week 4 vs. 1 session starting at week 8). Those who received two sessions of MI followed by four sessions of BSCT had the lowest HDD of any group.

There are several potential explanations for the unique performance of MI + BSCT. First, the MI + BSCT group had the most possible sessions (6) over the course of 8 weeks, and, thus, it received the largest dosage of treatment over the longest period of time. While we accounted for number of sessions received as a covariate and discovered it was nonsignificant, the potential impact of more sessions may be particularly important and impactful given that more than two-thirds of the sample was female. Women are known to respond to greater dosage of treatment for SUD over longer periods compared to men (Greenfield et al., 2007); though research has not previously established this for brief treatments. Second, MI at week 4 may have primed participants in such a way as to provide an optimal foundation for skills-based treatment, perhaps because there was more time to explore one’s ambivalence and motivation and to prepare for change between weeks 4 to 8 (i.e., 2 sessions attending to motivation over 4 weeks compared to only 1 session a week prior to starting skills-based work). MI may provide a forum for optimal formation of therapeutic alliance, thus providing a trusting and established rapport for initiating BSCT. MI at week 4 may also have primed greater change to occur in BSCT at week 8 because during weeks 4–8 participants may have paid more overt attention to self-monitoring compared to those who received additional BA. Third, clients who received more sessions may have been attempting to change during a time in which they received active support over a longer period of time than those who received BA Plus at week 4 and then 1 session of MI + 4 sessions of BSCT at week 8. If that were true, if we followed the BA Plus + MBSCT group four weeks later, there might be an equivalent level of reduction to the MI + BSCT groups.

4.3. Potential factors in the lack of effect of MI

While the null findings for MI vs. BA at week 8 are puzzling, there may be a few reasons why drinking outcomes were equivalent for those who received MI between weeks 4 and 8 and those who received only BA. Our attempt to identify when to assess and initiate an additional intervention may have been suboptimal. Four weeks may be too long to wait to assess for an intervention effect of MI and then potentially alter the intervention. In a previous study with a similar sample, MI had significantly greater gains over comparison groups after one session (Morgenstern et al., 2012); yet by the second week, the other groups had caught up to the gains that the MI group made.

Conversely, a longer assessment period may have been required to see delayed condition differences or a sleeper effect at week 8. Given the numerous efficacy trials on MI, however, it is important to note that BA, which was modeled after SBIRT in an attempt to capture usual care in a hospital and/or primary care setting, potentially contained some of the active ingredients of MI, such as having clients delineate the reasons for change. Additionally, while MI alone underperformed among this sample, it still played a critical role in optimizing the impact of BSCT.

4.4. Clinical implications

While participants were not exclusively recruited from primary care (recruitment occurred both within and outside of the local healthcare system and nationally) and the interventionists were not primary care providers, this study attempted to provide a sense of how behavioral health interventions might work outside specialty care and within mainstream healthcare settings. Thus, this study provides important guidance for implementing adaptive interventions at the frontlines of the health care system—particularly in primary care, which is experiencing an expansion in its role of addressing behavioral health. For example, the lab in which this study took place is a part of a large health care system that utilizes “health coaches” to initially assess and treat behavioral health problems. Health coaches are educated at both the bachelor’s and master’s levels. This study trained RAs using the same curriculum that is used for the health coaches, and they were meant to represent a minimal intervention based on psychoeducation—a low-cost intervention that could be widely adapted in a health care facility.

A simple but important finding is that in the context of nonresponse, providing more of the same intervention is not helpful to further reduce drinking. There are three additional potential takeaway points for clinical care: (1) change is possible after initial nonresponse to a brief intervention for AUD without referral to specialty care; (2) adapting the approach to include skills-based training among those who struggle to change provides the opportunity for the greatest therapeutic effect; and 3) the optimal algorithm of treatment may be that if individuals do not respond to initial brief advice, then prescribing and/or offering both MI (at least two sessions) followed by BSCT (4 sessions) across 8 weeks is critical to achieve optimal outcomes for individuals with AUD.

While MI has been disseminated and proliferated across disciplines and professions, behavioral skills training has not. Considering response to treatment improved with continued adaptation of treatment, particularly the addition of BSCT, underscores the necessity of ongoing assessment and intervention implemented by behavioral health care specialists across health care settings. While such an algorithm may increase burden on health care providers, such as those in primary care, it will likely reduce harm among patients seen in these settings, and, subsequently, reduce health care and societal costs overall. Importantly, providers need to be aware that the process of change in drinking may be slower than they might otherwise expect; such change could happen over the course of 12 weeks or longer. Even if progress is slow, providers must communicate their belief in their patients’ ability to change. Those patients who do not respond to initial BA may take another two months to make clinically meaningful change, regardless of the intervention. Such patients may not necessarily require referral to specialty care; their response may indicate that the initial interventions we offer individuals with AUD who do not respond to BA need to be slightly more intensive and last for more than just a session or two for optimal reduction in harm.

4.5. Limitations

There are several limitations to this study that indicate our findings should be interpreted with appropriate caution. The study had fewer responders than anticipated. The threshold for “response” may have been too strict—to drink at or under NIAAA low risk guidelines. Using such a threshold precludes detecting potentially large responses to treatment (e.g., reducing drinking by half) that still exceed low-risk guidelines. Despite this unexpected lack of response, this study can still provide guidance for nonresponders to brief interventions for AUD using very high standards. Perhaps quantity and intensity of drinking are only partial indicators of the harms of alcohol (Pearson et al., 2017), but other indicators of harm (e.g., alcohol consequences, endorsement of AUD criteria, etc.) were beyond the scope of this analysis.

Our sample also had limited ethnic and socioeconomic diversity (mostly white, well-educated, and employed), which reflects the local demography. While we attempted to recruit a diverse sample (i.e., through targeted advertisement campaigns, expanding to telehealth), several barriers likely precluded achieving greater diversity. Such barriers include a homogenously populated geographic location, the socioeconomic barriers that accompany current telehealth interventions (e.g., access to broadband Internet) (Weigel et al., 2020), and the widespread availability of local clinical services for SUDs. Thus, the generalizability of the study findings may be constrained to those individuals with AUD who match our sample.

In addition, there are limitations in our ability to conclude the nature of change within each of the adaptive strategies. This set of analyses do not explore week-to-week change that could illuminate further group differences and how and when they changed. We do not know how participants reacted to ending or continuing treatment at any particular point during the study. Each point of randomization is a potential opportunity to reinforce a participant’s motivation and opportunity for change. This study did not assess such detailed nuances of the adaptive intervention process.

An important component of any study is discrimination between conditions. It could be suggested that week 4 MI and BA were poorly discriminated thus yielding equivalent outcomes. While the two conditions shared features (i.e., providing feedback, having participants list reasons for change), there were at least three concrete and important distinguishing features that alone should have contributed to condition differences. One, there was a difference in provider: master’s or doctoral level therapists delivered MI at week 4, while research assistants provided BA. Two, the interventions had a difference in dosage: MI at week 4 consisted of two 1-hour sessions, while BA Plus was only 1 20-minute session. Three, only the MI condition implemented change plans. Self-evaluation reports from therapists, weekly supervision of both therapists and research assistants, and careful review of audiotapes ensured fidelity to each condition. Interestingly, despite these equivalent outcomes, the difference between the two week 4 conditions emerged within one of the adaptive strategies: receiving two sessions of MI between weeks 4 and 8 was instrumental for optimal change upon receipt of BSCT. A cumulative effect of treatments, therefore, may be greater than the independent parts.

Finally, it is possible that a stronger algorithm for an adaptive intervention for adults with AUD could be formulated and tested if there was a different approach to assessment and determining the initial intervention. For example, one could test motivation and skills capability from the beginning and based on the results initiate a specific intervention. Relatedly, it is possible that lack of response to MI does not necessarily mean that continued drinking is due to participants lacking the skills to change. It could be that inadequate planning occurred during the MI phase of treatment. Interestingly, this study addressed this by having participants re-randomized to MI vs. MI + CBT—providing an opportunity to re-plan. Still, receiving any BSCT outperformed MI alone.

4.6. Future research

Future studies should explore more combinations of brief treatments, at a variety of timings, to better hone algorithms for individuals with AUD presenting to different sectors of the health care and social service systems. Studies should include a larger sample, with greater racial, ethnic, and other demographic heterogeneity to ensure algorithms hold across distinct groups. Differences across gender and age also need to be explored.

4.7. Conclusion

This study is an important first step in testing and honing an adaptive intervention algorithm specifically for the treatment of AUD. While there is a proliferation of SBIRT interventions, heretofore there have been no clear directives for optimizing these interventions in a systematic way. Most providers make their decisions in isolation, within contexts for which clinical trials can rarely account. This study attempted to mimic a real-world context for providing increasingly intensive interventions in the context of nonresponse to provide more information about which strategies will yield the greatest reduction in harm for individuals with AUD. While more detailed and nuanced information is needed, this study provides concrete information for providers interested in optimizing their frontline interventions.

Figure 2.

Unadjusted means of SSD and HDD at weeks 1 and 4 for responders (N=18) vs. nonresponders (N=133).

Highlights.

• Alcohol use disorder (AUD) is often treated initially using brief interventions.

• Optimal timing, dosage, order, and combination of brief treatments is unknown.

• A 12-week Sequential Multiple Assignment Randomized Trial was implemented.

• A combination of motivational interviewing and/or coping skills therapy were tested.

• Participants who had any skills-based therapy made the greatest change.

• The strategy of receiving 2 MI sessions followed by skills therapy performed best.