Do Patients With Alcohol Dependence Respond to Placebo? Results From the COMBINE Study

Abstract

Objective:

The purpose of this study was to examine the nature of the effect of placebo medication plus accompanying medical management in the treatment of alcohol dependence.

Method:

The National Institute on Alcohol Abuse and Alcoholism COMBINE (Combining Medications and Behavioral Interventions) study, a randomized controlled double-blind trial of 1,383 alcohol-dependent patients, compared combinations of medications (acamprosate [Campral] and naltrexone [ReVia]) and behavioral therapy (medical management and specialist-delivered behavioral therapy) for alcohol dependence. This report focuses on a subset of that study population (n = 466) receiving (1) specialized behavioral therapy alone (without pills), (2) specialized behavioral therapy + placebo medication + medical management, or (3) placebo + medical management.

Results:

During 16 weeks of treatment, participants receiving behavioral therapy alone had a lower percentage of days abstinent (66.6%) than did the participants receiving placebo and medical management (73.1%) or those receiving specialized behavioral therapy + placebo + medical management (79.4%). The group receiving behavioral therapy alone relapsed to heavy drinking more often (79.0%) than those receiving behavioral therapy + placebo + medical management (71.2%). This report focuses on potential explanations for this finding. The two groups of participants receiving placebo + medical management were more likely to attend Alcoholics Anonymous meetings during treatment (32.7% and 32.0% vs 20.4%) and were less likely to withdraw from treatment (14.1% and 22.9% vs 29.3%).

Conclusions:

There appeared to be a significant “placebo effect” in the COMBINE Study, consisting of pill taking and seeing a health care professional. Contributing factors to the placebo response may have included pill taking itself, the benefits of meeting with a medical professional, repeated advice to attend Alcoholics Anonymous, and optimism about a medication effect.

The use of placebos in medication studies is common (Temple and Ellenberg, 2000), although controversial (Rothman and Michels, 1994). The purpose of placebo is to ensure that the superiority of an active treatment to that of no treatment can be attributed to the medication itself rather than the nonspecific effects of pill taking and medical attention. A positive response to placebo is common; for example, studies of major depressive disorder have reported a placebo response rate as high as 65% (Quitkin, 1999).

The power and universality of the “placebo effect” has been challenged, however. Hróbjartsson and Gøtzsche (2001) reviewed 114 studies of various disorders in which placebo was compared with no treatment. They found no effect for placebo in the 32 studies with binary outcomes. Placebo had a positive effect in studies with continuous outcomes but only in studies using subjective, not objective, outcomes. Thus, the placebo effect appears to be disease specific, not a uniform finding.

The nature of the placebo effect in alcohol dependence has previously been unknown. Understanding the placebo effect is particularly relevant in alcohol-dependent patients receiving concomitant behavioral therapy. Although the placebo effect is generally considered positive, the opposite effect could exist for patients receiving placebo medication in conjunction with behavioral therapy; one could hypothesize that some patients receiving medication may put less effort into their behavioral therapy because they expect the medication to change their drinking behavior. When such individuals receive placebo, behavioral therapy may be less effective than it would have been if delivered alone.

One opportunity to study the placebo effect in alcohol dependence was the multisite National Institute on Alcohol Abuse and Alcoholism (NIAAA) COMBINE (Combining Medications and Behavioral Interventions) study (Anton et al., 2006), in which we tested combinations of medications and behavioral treatments for alcohol-dependent participants. The medications were naltrexone (ReVia) and acamprosate (Campral). The behavioral treatments were medical management (MM; Pettinati et al., 2004), which focuses on medication adherence and support for abstinence, and the combined behavioral intervention (CBI; Miller, 2004), a moderate-intensity specialty behavioral treatment combining features of several evidence-based alcoholism treatments. The study primarily consisted of a 2 × 2 × 2 factorial eight-cell design, with participants receiving naltrexone or its placebo and acamprosate or its placebo. All participants in these eight cells received MM; half also received CBI. COMBINE investigators also added a ninth cell, which is the focus of this report: CBI alone, with no pills or MM. This afforded a unique opportunity to examine the placebo effect for these medications in alcohol dependence by comparing CBI alone versus CBI + placebo + MM. Because differences between these conditions could result from unequal amounts of treatment, we also compared CBI alone versus placebo + MM, with the latter group receiving fewer treatment hours than participants receiving CBI alone.

Results of the COMBINE trial (Anton et al., 2006) revealed substantial improvement in drinking from baseline among all nine groups. Patients receiving (1) naltrexone + MM, (2) CBI + placebo + MM, or (3) naltrexone + MM + CBI all had a higher percentage of days abstinent (PDA) than those receiving placebo + MM. Naltrexone also produced a lower likelihood of heavy drinking than placebo.

The extensive collection of assessment data allows for further exploration of the possible reasons for any observable placebo response in COMBINE. This article explores potential reasons for the placebo response among this group of alcohol-dependent patients by focusing on the results of 466 participants receiving either (1) CBI alone, (2) CBI + placebo + MM, or (3) placebo + MM.

Method

Inclusion and exclusion criteria, assessments, recruitment, and randomization

The COMBINE Study was conducted at 11 clinical sites and included a coordinating center and collaboration with NIAAA. The study enrolled 1,383 participants recruited from advertisements and clinical referrals. Key inclusion criteria were (1) a diagnosis of alcohol dependence (Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition; American Psychiatric Association, 1994); (2) 4-21 days of abstinence immediately before randomization; and (3) an average of 21 or more (for men) or 14 or more (for women) drinks per week, with 2 or more heavy drinking days during a consecutive 30-day period within 90 days of intake. Key exclusion criteria were (1) a psychiatric disorder requiring pharmacotherapy, (2) medical illness precluding participation, or (3) another recent substance use disorder (except nicotine or cannabis). Assessments included substance use, medical and psychological status, laboratory data, and craving. The inclusion criteria and assessments are detailed elsewhere (COMBINE Study Research Group, 2003). Participants were randomized using a permuted block design and stratification by site. All participants gave informed consent.

Treatments

An overview of the COMBINE treatments is summarized elsewhere (COMBINE Study Research Group, 2003). Both behavioral therapies were manualized, and all sessions were audiotaped. Therapists received standardized training, and session tapes were monitored to ensure treatment fidelity (Miller et al., 2005).

Medications

This report focuses only on those participants receiving naltrexone and acamprosate placebos, all under double-blind conditions. Naltrexone placebo was administered as two pills daily, and acamprosate placebo was administered as two pills three times a day.

Medical management

During MM (Pettinati et al., 2004) visits, medical professionals (primarily physicians or nurses) reviewed side effects and other adverse events, discussed drinking, and encouraged medication adherence. The initial session lasted approximately 45 minutes; the eight subsequent sessions, held over 16 weeks, averaged 20 minutes each. MM supported abstinence by stressing medication adherence, discussing the benefits of abstinence, and encouraging attendance at support groups such as Alcoholics Anonymous (AA).

Combined behavioral intervention

CBI (Miller, 2004) combined approaches that have been found effective in treatment of alcohol dependence, including motivational enhancement therapy (Miller and Rollnick, 2002), cognitive-behavioral skills training (Kadden et al., 1992), encouragement of AA participation (Humphreys et al., 2004), and community reinforcement (Meyers and Smith, 1995). CBI used a flexible menu of options to address issues that the therapist and participant deemed important. Participants attended up to 20 hour-long sessions over 16 weeks, with the number and content of sessions determined by the therapist and participant together.

Statistical methods

Reported analyses used an intention-to-treat approach, including all randomized participants. The two prespecified, co-primary endpoints were PDA and time to relapse to the first day of heavy drinking (five or more standard drinks in a day for men, and four or more for women). Participants lost to follow-up were assumed to have relapsed to heavy drinking on the day after their last contact.

Differences between treatment groups (CBI alone, CBI + placebo + MM, and placebo + MM) at baseline were tested using one-way analysis of variance models, chi-square tests, or Fisher's exact test, according to the distributional characteristics of the parameter being evaluated. A mixed-effect general linear model was used for comparing differences among groups on PDA, which was computed monthly. The three treatment groups were analyzed as a fixed effect, and the repeated individual measures per patient over time were analyzed as a random effect. PDA within 30 days of baseline and site were covariates. AA attendance during treatment was also used as a covariate in secondary analyses. Tukey's honestly significant difference technique was used to control Type I error rates when making multiple pairwise comparisons. A proportional hazards model was used to determine time to relapse to heavy drinking and withdrawal from treatment. Logistic models with interaction terms adjustment were employed to test expectations for effectiveness of treatment upon withdrawal.

Results

Study sample

Baseline sociodemographic and clinical characteristics of the three cohorts of interest (CBI alone, CBI + placebo + MM, and placebo + MM) are listed in Table 1. None of these characteristics differed by group.

Table 1.

Baseline sociodemographic and clinical characteristics, by treatment^a

Characteristic	CBI alone (n = 157)		CBI + Placebo + MM Placebo + MM (n = 156)		Placebo + MM (n = 153)
	n	%	n	%	n	%
Sociodemographic
Age ≤ 44, sample median	81	51.6	87	55.8	79	51.6
Nonminority	121	77.1	114	73.1	120	78.4
Male	107	68.2	110	70.5	103	67.3
Married	65	41.4	78	50.0	68	44.4
Employed	109	69.4	112	71.8	122	79.7
Income ≥ $30,000	121	77.1	116	74.4	111	72.5
Education ≤ 12 years	44	28.0	47	30.1	45	29.4
	Mean (SD)		Mean (SD)		Mean (SD)
Clinical
Percentage days abstinent, past 30	23.5 (25.4)		24.3 (24.7)		24.3 (24.7)
Drinks/drinking day, past 30	11.8 (7.7)		12.6 (7.7)		12.6 (7.7)
Days heavy drinking, past 30	19.6 (8.8)		20.1( 8.5)		20.1 (8.5)
Alcohol Dependence Scaleb score	16.6 (7.0)		16.4 (7.3)		16.5 (7.2)
Drinking consequences, DrInCc	45.8 (20.3)		46.4 (20.1)		46.5 (20.2)
Obsessive Compulsive Drinking Scaled score	18.9 (10.0)		25.1 (7.6)		24.5 (7.6)

Notes: CBI = combined behavioral intervention; MM = medical management; DrInC = Drinker Inventory of Conse quences.

^aNone of the comparisons between groups are significantly different at p < .05;

^bmaximum possible score is 47;

^cmaximum possible score is 135;

^dmaximum possible score is 56.

Drinking outcomes

As reported previously (Anton et al., 2006), participants receiving CBI + placebo + MM and those receiving placebo + MM had a significantly higher PDA during treatment (79.4% and 73.1%, respectively) than did participants re-ceiving CBI alone (66.6%; see Table 2). The CBI + placebo + MM group also had more abstinent days than the placebo + MM group. Individuals receiving CBI alone relapsed to heavy drinking at a higher rate (79.0%) than did those receiving CBI + placebo + MM (71.2%; hazard ratio = 0.77, 95% confidence interval: 0.60-1.00, p = .049). There were no significant differences in relapse between participants receiving CBI alone and placebo + MM; 75.2% of the latter group relapsed.

Table 2.

Percentage of days abstinent at baseline (previous 90 days) and Week 16

Percentage days abstinent	Baseline		Week 16a		Mixed model p
	n	mean (SE)	n	mean, adj. (SE)
Placebo + MM	153	24.5 (2.0)	142	73.1 (2.2)	.0002
CBI + Placebo + MM	156	24.5 (2.0)	145	79.4 (2.2)
CBI alone	157	23.8 (2.0)	146	66.6 (2.2)

Notes: Adj. = adjusted. CBI = combined behavioral intervention; MM = medical management.

^aPairwise comparison tests using Tukey's honestly significant difference, corrected: 1 vs 2, p = .04; 1 vs 3, p = .034; 2 vs 3, p = .0001.

Understanding the placebo response

The main COMBINE outcome article (Anton et al., 2006) posited several potential explanations as to why participants receiving placebo + MM did better than those receiving CBI alone. These included (1) positive or negative expectations about one's treatment assignment, (2) daily pill taking as a reinforcer of motivation, (3) the nonspecific benefits of meeting with a medical professional regularly, and (4) the content of MM visits themselves. We have conducted additional analyses to try to understand further the nature of the placebo response in COMBINE.

Expectations

One common explanation for placebo response is that patients' expectations of a beneficial medication effect influence their response. To explore this, we examined results from the Treatment Experiences and Expectations questionnaire (D. Donovan, unpublished instrument), which asked participants at study entry, “How helpful do you feel the medications you might receive in this study will be in helping you change your drinking behavior?” The same question was posed regarding counseling and combined treatment. All three cohorts had nearly identical expectations of the effectiveness of medication (mean score of 3.8-3.9 on a 5-point scale, with 3 indicating “moderately effective” and 4 indicating “considerably effective”), counseling (3.8), and combined medication and counseling (4.2-4.3). Thus, a between-group difference in positive treatment expectations did not contribute to our findings.

As an indirect measure of negative expectations, we examined early withdrawal from treatment. We found that 29.3% of participants receiving CBI alone withdrew from treatment (defined as three or more consecutive missed MM and/or CBI visits, or no visits in a given month), compared with 14.1% of participants receiving CBI + placebo + MM, and 22.9% of participants receiving placebo + MM (χ² = 10.57, 2 df, p = .005). Figure 1 illustrates the time course of treatment retention among these three groups, showing the greatest likelihood of early withdrawal in the cohort of CBI alone (log rank test, p = .037).

Figure 1.

Treatment retention among COMBINE participants receiving no active medication. MM = medical management; CBI = combined behavioral intervention.

We then examined whether participants with higher expectations of medication efficacy were more likely to withdraw from treatment when assigned to CBI alone. We performed a logistic regression analysis to examine potential moderating effects of expectations of medication efficacy or CBI on treatment assignment to predict probability of treatment withdrawal. Higher expectation of medication efficacy was a weak but nonsignificant (p = .07) independent predictor of greater treatment withdrawal. Level of expectation of CBI efficacy was neither an independent predictor of treatment withdrawal nor did it have moderating effects on treatment group as a predictor of probability of withdrawal. We thus concluded that participants' expectations of treatment did not work as a moderator with treatment group to affect probability of treatment withdrawal.

Treatment attendance

We found that 71.4% of participants receiving CBI + placebo + MM attended 9 or more CBI sessions (the number designated by COMBINE investigators as “adequate”), as opposed to 63.5% of participants receiving CBI alone; the mean (SD) number of CBI sessions attended was 10.4 (3.7) for the CBI + placebo + MM patients, and 10.0 (3.7) for the patients receiving CBI alone. Neither difference was statistically significant. Placebo + MM participants attended a mean of 7.7 (2.2) MM sessions out of a maximum of 9; those receiving CBI + placebo + MM attended a mean of 7.8 (2.2) sessions. To determine the possible differential effects of attention on outcome, the groups were compared on total hours of treatment. CBI-alone participants averaged 8.8 (4.4) treatment hours; CBI + placebo + MM participants averaged 12.3 (4.9) hours, and placebo + MM participants averaged 3.2 (1.1) hours (F = 213.61, 2/461 df, p < .0001).

Content of medical management visits

Placebo can never be given by itself in a study but is always delivered in conjunction with some form of MM by a clinician who does not know that the participant is receiving a placebo. The MM treatment in the COMBINE Study, therefore, may itself have had beneficial effects. MM clinicians primarily stressed medication adherence but also regularly recommended abstinence. The latter feature of MM, however, did not differentiate MM from CBI.

Because another core feature of MM was encouragement to attend AA, and because AA attendance has previously been associated with good drinking outcomes (Donovan, 1999; Moos and Moos, 2004; Morgenstern et al., 1997), we used the Form 90 (Miller, 1996) to examine AA attendance at baseline, Week 8, and Week 16 in the group receiving CBI alone and the two groups receiving placebo + MM (see Table 3). Although a nearly identical percentage of participants in the three groups attended AA during the week before intake, a significantly higher percentage of participants in the two groups receiving placebo + MM (32.7% and 32.0%) attended AA during treatment than did those receiving CBI alone (20.4%).

Table 3.

Percentage of patients attending Alcoholics Anonymous (AA) in the week before baseline and during treatment

Patients attending AA	Baseline		During treatment
	n	%	n	%
Placebo + MM	35	22.9	49	32.0
CBI + Placebo + MM	34	21.8	51	32.7
CBI alone	35	22.3	32	20.4

Notes: MM = medical management; CBI = combined behavioral intervention.

χ² = 7.38, 2 df, p = .025.

AA attendance during treatment was a significant predictor in the mixed model of PDA during treatment (F = 4.54, 2/435 df, p = .03). However, when our overall analysis of PDA for the three groups was adjusted for AA attendance during treatment, the differences among the groups remained significant in the same direction (p < .0001), suggesting that differential AA attendance could not account for the between-groups differences.

Discussion

The NIAAA COMBINE Study offered an opportunity to examine the effect of placebo medication in participants also receiving alcohol-specific behavioral therapy. We found that participants who received behavioral therapy plus placebo and MM visits had significantly better drinking outcomes than did participants who received the same behavioral therapy without taking pills and attending medical visits. Moreover, participants receiving only placebo and MM had more abstinent days than did participants receiving behavioral therapy alone. Thus, there appeared to be a significant placebo effect in this population.

A number of explanations have been proposed for the placebo effect. One explanation involves the increased attention that participants receive when taking placebo (Brown, 1994). This appears plausible when comparing CBI alone versus CBI + placebo + MM. However, this cannot account for the superior abstinence outcomes for placebo + MM when compared with CBI alone because the CBI-alone group averaged nearly three times more treatment hours than placebo + MM participants. Thus, increased attention does not appear to account for the placebo response in this study.

Some explanations for a positive placebo effect are better characterized as a response to trial participation itself. For example, some participants are said to respond to placebo in part because of the naturally fluctuating course of certain disorders that have periods of relapse and remission (Walsh et al., 2002). The increased level of awareness about one's disorder as a result of study participation is also sometimes cited as a cause of placebo response (Ernst and Resch, 1995). However, the differences among the three groups discussed in this report cannot be explained by these phenomena because the groups were equivalent with regard to these factors.

What, then, can account for the better outcomes among the placebo groups in this study? The group receiving CBI alone was distinguished from the other two groups by the fact that all the other participants took pills (albeit placebos) and saw a medical professional regularly. It is thus possible that the act of taking pills three times daily was an affirmation of participants' desire to change their drinking behavior. Although the need to take acamprosate three times a day can be a disadvantage because more frequent dosing may be correlated with lower adherence (Claxton et al., 2001), the frequency of use may actually be helpful, independent of the pharmacological effects of acamprosate, because it serves as a more frequent reminder of the reason for the pill taking.

Placebo was not delivered in a vacuum but was given in the context of MM visits. Indeed, placebo medication, like its active counterpart, is never delivered alone because participants also receive clinician contact, medication instruction, side effect assessment, and disease education. Even brief advice and attention from a health care professional regarding drinking may be therapeutic, as evidenced by the effectiveness of brief interventions in primary care settings (Bertholet et al., 2005). Thus, the content of MM itself may have contributed to our findings.

The difference in AA attendance among participants receiving placebo suggests that repeated encouragement of AA by a health care professional can increase the likelihood of meeting attendance and, in turn, lead to better drinking outcomes. These data suggest that this simple and straightforward intervention can have a beneficial impact. However, increased AA attendance among participants receiving placebo (and thus MM), although beneficial, does not account for the entire difference between the participants receiving CBI alone and those receiving placebo.

Because the COMBINE Study was designed so that eight of nine participants received pills (and two thirds received active medication), it is reasonable to posit that most study participants were favorably disposed toward taking medication. It is thus plausible that participants randomized to receive CBI alone may have been disappointed with their assignment and may have experienced a “negative placebo effect”; they might have fared worse than they otherwise would have, either in a pure behavioral therapy trial, in which they would not have expected medication, or in a standard medication trial, in which they would have believed that they might be receiving active medication. The increased rate of treatment withdrawal among participants receiving CBI alone may reflect their unhappiness with their treatment assignment.

In conclusion, the placebo effect in this trial likely consisted of a combination of factors: optimism about the potential benefits of the medication, the act of pill taking itself, the effect of receiving advice from a health care professional, and the impact of regular encouragement to attend AA meetings. Thus, a potentially important finding from this study is that treatment of alcohol dependence in a general medical or psychiatric setting may increase its effectiveness beyond the efficacy of the specific medication being administered. One must be cautious about generalizing too much from these results because the study attracted participants who were likely to be favorably disposed toward medication use, and the length of the MM visits exceeded that of most visits to primary care physicians (Blumenthal et al., 1999). However, this study does suggest that there is a rather robust placebo effect when using medications for alcohol dependence.