Randomized Clinical Trial of Matching Client Alcohol Use Disorder Severity and Level of Cognitive Functioning to Treatment Setting: A Partial Replication and Extension

Abstract

Objective

In a tightly-controlled, clinical research environment, Rychtarik et al. (2000) found that individuals with an alcohol use disorder (AUD) benefited more from inpatient (IP) than outpatient care, if they presented with high alcohol problem severity and/or low cognitive functioning. This study sought to: (a) validate and extend these findings within the uncontrolled environment of a community-based treatment center, and (b) test whether inpatients had fewer days of involuntary abstinence (e.g., incarcerations), controlling for differences in treatment expectancy across care settings.

Method

Clients (N = 176) with an AUD were deterministically assigned to inpatient-need group (Needs IP = high severity and/or low cognitive functioning; No need for IP = neither high severity nor low cognitive functioning). Within need group, participants were randomly assigned to 21 days of routine inpatient or outpatient care, plus six months of continuing care. Primary outcomes of percentage of days abstinent (PDA), monthly point prevalence abstinence (PPA), and drinks per drinking day (DDD), and the secondary outcome of involuntary abstinence were assessed over 18 months.

Results

Among high-severity participants, inpatients significantly reduced DDD; outpatients did not. Neither problem severity nor cognitive functioning moderated other setting effects. Treatment expectancy, assessed after being informed of scheduled setting, was higher among inpatients than outpatients. High-expectancy inpatients maintained the highest PDA throughout follow-up, had the highest PPA for most of follow-up, and delayed peak prevalence of involuntary abstinence by six months.

Conclusions

The Alcohol Problem Severity X Setting interaction on DDD appears robust. The role of expectancy warrants further study.

Only 10% of those in treatment for a substance use disorder (SUD) in the U.S. are now treated in non-hospital residential (9%) or hospital-based inpatient (1%) programs (Substance Abuse and Mental Health Services Administration, 2014). The limited availability of residential or inpatient care for SUD today is in marked contrast to practices thirty years ago, when treatment occurred almost exclusively in the inpatient setting. This near universal move to outpatient treatment has largely been driven by managed care and by research that found little or no difference in outcomes between the two settings. Though outpatient treatment is likely to remain dominant for the foreseeable future, recent reviews of this research area suggest that this trend may now have gone too far (Finney, Moos, & Wilbourne, 2014; Reif et al., 2014), with inpatient treatment becoming inaccessible to some who could potentially benefit most from it. In particular, these reviews suggest that individuals with higher alcohol/drug problem severity benefit more from inpatient than outpatient care. However, the size of these matching effects and the level of severity at which inpatient treatment should be considered remain unclear, leading Finney et al. (2014) to emphasize the need to validate specific placement criteria for allocating individuals with SUDs to different clinical care settings (inpatient vs. outpatient).

In a tightly-controlled study of matching clients to inpatient versus outpatient care, Rychtarik et al. (2000) found that individuals with more severe alcohol problems achieved a higher percentage of days abstinent (PDA) and drank fewer drinks per drinking day (DDD) with inpatient than outpatient care, an advantage that persisted throughout an 18-month observation period (i.e., during 6 months of continuing outpatient care for both conditions plus 12 months of follow-up). Independent of this finding, exploratory analyses of five other potential setting moderators (motivation, self-efficacy, social support for drinking, psychiatric severity, and cognitive functioning) found that only cognitive functioning moderated setting effects. Those low in cognitive functioning achieved higher PDA in inpatient than outpatient treatment throughout the 18 months. Alcohol problem severity and cognitive functioning cut points were then developed to better guide setting-placement decisions. Yet, Rychtarik et al. conducted their study in a clinical research center environment, standardized treatment content, controlled for clinical staff effects, and recruited by advertising. Therefore, the question remains whether the effects observed, their size, and the cut points developed are equally valid among clients presenting for treatment in the real-world environment of a community treatment program.

The primary aim of the current study, therefore, was to test whether the treatment-matching effects observed in Rychtarik et al. (2000) on the longitudinal trajectories of alcohol-related outcomes could be validated within the uncontrolled environment of a large, community-based, chemical-dependency-treatment program. Clients were prospectively matched or mismatched to inpatient or outpatient treatment based upon the respective upper and lower cut points established in the earlier study for problem severity and cognitive functioning. Rychtarik et al. found similar matching effects, regardless of outpatient intensity; outpatient intensity in the current study was therefore made comparable to less intense, standard care. Of note, cognitive functioning in the prior study was measured at the end of the primary treatment phase, whereas in the current study it was measured pretreatment, making it a potentially more useful measure for treatment planning. In addition, data analyses on PDA and DDD were updated to reflect more contemporary approaches to statistical modeling.

Rychtarik et al. also found fewer involuntary abstinent days among inpatients than outpatients across the entire 18-month observation period, and this effect was not moderated by alcohol involvement or cognitive functioning. An involuntary abstinent day is a day on which drinking was restricted due to hospitalization, residential treatment, or incarceration. Hence, involuntary abstinence serves as a measure of posttreatment severe negative consequences associated with healthcare and criminal justice utilization. A voluntary abstinent day, on the other hand, is defined as an abstinent day on which the client’s drinking was not restricted by hospitalization, residential treatment, or incarceration. Following the methods of the earlier study, the current study used voluntary abstinent days and voluntary abstinence as the primary abstinence-related measures. Secondary and tertiary aims, respectively, assessed the replicability of the involuntary abstinence setting effect and, in the absence of a moderation effect, examined any main effects of Setting and Setting X Time interactions.

Method

Participants

Participants were 176 clients presenting for treatment of alcohol use disorder (AUD) from July 2003 through September 2006 in the outpatient and inpatient medical detoxification programs of the Division of Chemical Dependency of a large, community-based healthcare network in the same geographical catchment area as the prior study. Eligible participants had to (a) be at least 18 years of age, (b) have drank alcoholic beverages within the last 90 days, (c) have a current address (not homeless), (d) be free of legal stipulations that would affect treatment decisions, (e) not have received treatment for substance misuse within the last 30 days, (f) have a past 3 months Alcohol Use Disorders Identification Test (AUDIT; Babor, de la Fuente, Saunders, & Grant, 1989) score ≥ 10, and (g) have a primary alcohol, or alcohol and other drug problem as assessed by clinical staff. See Table 1 for sample characteristics. The study protocol was reviewed and approved by the University at Buffalo Institutional Review Board prior to implementation.

Table 1.

Participant pretreatment characteristics by treatment setting

Characteristic	Inpatient	Outpatient	Test statistic	p
n	84	92
Age, M years (SD)	40.89 (10.36)	40.40 (9.35)	t(174) = .33	.74
Gender, n (% Female)	22 (26)	25 (27)	χ2 = .02	1.00
Race/ethnicity, n (%)			χ2 = .48	.78
White	55 (66)	56 (61)
Black	26 (31)	33 (36)
Other	3 (4)	3 (3)
Education, M years (SD)	12.06 (1.75)	12.24 (2.15)	t(174) = −.57	.57
Employment status, n (%)			χ2 = .28	.87
Full/part-time	22 (26)	24 (26)
Unemployed	54 (64)	57 (62)
Not in work force	8 (10)	11 (12)
Marital status n (%)			χ2 = .36	.83
Single/widowed	39 (46)	39 (42)
Separated/divorced	20 (24)	25 (27)
Married/cohabitating	25 (30)	28 (30)
Occupational statusa M (SD)	3.45 (1.75)	3.58 (1.86)	t(174) = −.47	.63
Ever received inpatient AD treatment, n (%)	44 (52)	52 (57)	χ2 = .30	.58
Ever received outpatient AD treatment, n (%)	53 (63)	67 (73)	χ2 = 1.92	.17
Clinic Source n (% from detoxification unit)b	39 (46)	49 (53)	χ2 = .82	.37
Alcohol Use Inventory-AIS, M (SD)	33.11 (13.34)	33.11 (12.94)	t(174) = −.001	1.00
Symbol Digit Modalities Test, M (SD)	43.60 (10.69)	42.52 (10.26)	t(174) = .68	.50
URICA readiness to change, M (SD)	10.58 (1.61)	10.70 (1.83)	t(174) = −.50	.62
Treatment expectancyc, M (SD)	8.29 (1.46)	7.72 (2.05)	t(174) = 2.15	.03
Voluntary alcohol abstinent days/mo.d (%)	31.55	27.49	z = −.90	.37
Voluntary alcohol/drug abstinent days/mo.d (%)	22.35	22.02	z = .08	.94
Totally abstinent from alcohol/mo.e (%)	8.01	4.27	z = −1.62	.11
Drinks per drinking dayf	13.39	14.36	z = .70	.41
Involuntary abstinent days/mog (%)	4.46	2.38	z = −2.02	.04
Involuntary abstinence at least 1 day/moh (%)	13.98	11.50	z = 1.14	.25

Note: AD = Alcohol/Drug; AUI-AIS = Alcohol Use Inventory, Alcohol Involvement Scale; URICA = University of Rhode Island Change Assessment scale. Estimates of drinking, involuntary abstinence, drug use, self-help attendance, and outpatient treatment variables derived from the 12-month Timeline Follow-back pretreatment interview were obtained using Generalized Estimating Equations (GEE).

^aHollingshead and Redlich (1958), coded 1 (low status) to 9 (high status);

^bStudy referrals from the detoxification unit;

^cMean item score on the Feelings About Your Scheduled Treatment scale;

^dMonthly percentage of days across the 12 months pretreatment on which the participant was abstinent and substance use was not restricted due to hospitalization, residential treatment, or incarceration;

^ePercentage of participants totally abstinent in any one month across the 12 months pretreatment (months in which the participant was restricted from drinking on all days were excluded from these analyses);

^fNumber of standard drinks consumed on days the participant was drinking in the pretreatment months;

^gPercentage of days in any pretreatment month in which the participant’s drinking was restricted due to hospitalization, residential treatment, or incarceration;

^hPercentage of participants with at least 1 restricted-drinking day in a pretreatment month.

Procedure

Outpatient treatment staff performed a brief, cursory prescreening for study eligibility and interest as part of the agency’s own intake process. Those preliminarily eligible, interested, and willing to consider randomization to treatment setting then completed a full screening interview by on-site research staff; detoxification clients completed the full screening close to discharge. Eligible and consenting individuals subsequently participated in a pretreatment assessment that included measures of motivation, pretreatment drinking, involuntary abstinence, problem severity, and cognitive functioning. To aid the randomization process, results were used to create a presumed inpatient-need category assignment (Needs Inpatient = high severity and/or low cognitive function; No need for Inpatient = neither high severity nor low cognitive functioning) using the Rychtarik et al. (2000) cut points. An urn randomization algorithm then randomly assigned participants within need category to their primary treatment, consisting of 21 days of either inpatient or outpatient care; the algorithm balanced settings within need category on race (white; nonwhite), gender (male; female), marital status (married: Yes; No); prior SUD treatment (Yes; No); and incarceration history (Yes; No). On completion of the 21-day treatment period, all participants received an additional 6 months of outpatient continuing care.

Participants completed an in-person assessment at the end of the 21-day primary treatment period, follow-up phone assessments at 3, 9, and 15 months post primary treatment, and in-person assessments at 6, 12, and 18 months. All study follow-ups were conducted in the period from November 2003 through April 2008. Except for pretreatment assessment of participants in the detoxification unit just prior to discharge, all assessments occurred at a research location separate from the treatment sites. Research staff blind to participants’ setting assignment conducted all assessments. The Consort diagram is presented in Figure 1.

Figure 1.

Consort diagram. Over a 39-month recruitment period, approximately 8,604 cursory pre-screenings were conducted in the outpatient and detoxification programs, combined; 3,842 (45%) of these preliminarily met basic eligibility requirements; 533 (14%) of the latter agreed to a full screening interview; 491 (92%) completed it.

Treatment settings

Inpatient treatment

Participants assigned to inpatient treatment were scheduled for 21 days of care in the agency’s 20-bed inpatient treatment program located in the healthcare network’s large urban medical center. Following admission, clients met individually with their counselors on the day of admission, and approximately twice a week thereafter. They also attended various program-based education and specialized therapy groups, Alcoholics Anonymous/Narcotics Anonymous (AA/NA) meetings, and other services provided by the medical center, as appropriate. On completion of inpatient care, they were allowed 24 sessions of continuing outpatient care over the subsequent 6 months.

Outpatient treatment

Participants assigned to outpatient treatment could attend up to 30 sessions over approximately 7 months (21 days of primary treatment; 6 months of continuing care). Treatment occurred in either the outpatient clinic where originally screened or, if screened in the detoxification unit, the clinic most convenient for the participant. The primary counselor determined the frequency and mix of the sessions (i.e., individual, family, specialty group, etc.), based on the client’s need, although the recommended schedule was 6 sessions over the course of the 21-day primary treatment period, and 24 sessions during continuing care. Weekday, evening, or weekend AA/NA meetings also were available, but did not count toward the session total.

Treatment orientation

The program coordinator of each setting was administered the Drug and Alcohol Treatment Program Inventory (DATPI; Swindle, Peterson, Paradise, & Moos, 1995) prior to study initiation. The DATPI measures the degree to which goals and activities related to eight specific treatment orientations are similar to those used in the respondent’s program. The inpatient (IP) and outpatient (OP) coordinators’ DATPI subscale scores (possible score range 0–24; higher scores represent higher use of the approach) were as follows: AA/12 step (IP =14; OP = 14), cognitive-behavioral (IP = 20; OP = 21), insight/psychodynamic (IP = 18; OP = 17), marital/family (IP = 17; OP = 13), therapeutic community (IP = 14; OP = 10), rehabilitation/vocational (IP = 3; OP = 18), dual diagnosis (IP = 15; OP = 15), and medical (IP = 19; OP = 19). A cognitive-behavioral approach appeared slightly favored in both settings, but an overall eclectic mix was present in each. Vocational rehabilitation services, however, were markedly higher in outpatient relative to inpatient care.

Staffing and clinical management

Agency clinical staff at the respective settings provided all care; the primary clinician assigned to provide and coordinate participant care was determined by the respective program’s existing staff assignment procedure. A quasi-managed care model then insured that study time and session limits were maintained. Specifically, the primary clinician received the participant’s planned completion date, and the number of days and sessions allowed. Research staff then carefully monitored treatment to insure compliance, and provided clinical staff with notices when a client was reaching study-defined limits. The primary clinician could request additional days/sessions under predefined criteria (e.g., acute suicidal ideation, deterioration). Such requests were reviewed by the study’s Clinical Care Committee, which could approve a specific amount of additional treatment, or recommend an alternative approach for dealing with the participant’s issue. The study paid for all study care not covered by third-party payers.

Measures

Alcohol Involvement and cut-point

Raw scores ≥ 30 on the Alcohol Involvement Scale (AIS; range = 0–68) of the Alcohol Use Inventory (AUI; Horn, Wanberg, & Foster, 1990) defined a need for inpatient treatment; lower scores defined no need for inpatient treatment. The AIS has good internal consistency (.93) and construct and criterion validity (Horn et al., 1990; Skinner & Allen, 1983). Higher scores represent greater obsession with drinking, sustained drinking, perceptual withdrawal, somatic withdrawal, social role maladaptation, and loss of control of behavior when drinking. Derived by Rychtarik et al. (2000), the cut-point is at the 35^th percentile of AIS norms for AUD populations, and corresponds to the 44^th percentile of the current sample.

Cognitive Functioning and cut-point

A correct response score < 42 on the Symbol Digit Modalities Test (SDMT; Smith, 1982; range = 0–90) defined low functioning and need for inpatient treatment; scores ≥ 42 were classified as high functioning. Derived by Rycharik et al. (2000), the cut-point is at the 45^th percentile of this sample; the 19^th and 36^th percentiles of Project MATCH outpatient and aftercare clients, respectively (Bates, Pawlak, Tonigan, & Buckman, 2006), and the 3^rd percentile of a community sample of males in the participants’ age range (Jorm, Anstey, Christensen, & Rodgers, 2004).

Treatment expectancies

A 6-item, Feelings About Your Scheduled Treatment scale was administered immediately after participants were informed of their treatment setting assignment, but before treatment initiation. On this measure, developed in-house by Rychtarik et al. (2000), participants privately rated their scheduled treatment on a 10-point scale ranging from 1 (not at all) to 10 (extremely) with respect to its reasonableness, their confidence that it will be helpful, their confidence in recommending it to a friend, its similarity to that expected, the expected ease of participating, and their overall satisfaction with the treatment as scheduled. Internal consistency was .87. All staff was blind to expectancy level.

Motivational Readiness to Change

The University of Rhode Island Change Assessment Scale (URICA; McConnaughy, Prochaska, & Velicer, 1983) was used to derive a composite readiness score (Project MATCH Research Group, 1997; range = −2 to 14).

Setting preference

A single-item, administered at the post primary treatment point, assessed the participant’s preference for the alternate setting from that assigned on a 6-point scale (1 = There is no way I would have participated in [the alternate setting] even if scheduled; 6 = I very much wanted to receive the [the alternate setting], and I definitely would have gone).

Primary drinking outcomes

The primary outcomes were: (a) percentage of voluntary abstinent days (PDA) per month, that is excluding days in which a client was involuntarily detained; (b) monthly point prevalence abstinence (PPA), representing complete abstinence (yes/no) on all unrestricted days of a month; and (c) standard drinks consumed on a drinking day (DDD). PDA and DDD, representing both frequency and intensity drinking measures, respectively, have commonly been used as AUD primary treatment outcomes (e.g., Project MATCH Research Group, 1997), and were the primary, predefined outcomes of Rychtarik et al. (2000). Though PPA typically has not been reported, and was not analyzed by Rychtarik et al. (2000), we included it among our primary outcomes, as it is the first step in, and integrally related to, the more contemporary two-part DDD analysis model used in the current study (see Data Analyses). As clients had limited to no access to drinking during days spent hospitalized, in residential treatment, or being incarcerated, complete voluntary abstinence in any 30-day observation period (PDA=100) was equivalent to monthly abstinence in the dataset (PPA=1). All of these measures were derived from the Timeline Follow-back Alcohol and Substance Use interview (Sobell & Sobell, 1992), administered monthly for all 12 months pretreatment, once at the end of the 21-day primary treatment period, and with quarterly frequency across 18 months of follow-up. Alternate PDA and PPA measures were also derived for alcohol and other drug use combined. Study findings for these two variables closely mirrored those for alcohol alone and, therefore, are not reported separately here.

Secondary involuntary abstinence outcome

The monthly point prevalence of involuntary abstinence (i.e., of experiencing at least one day in any 30-day period in which the opportunity to drink was restricted due to hospital admission, inpatient or residential treatment, or incarceration) also was derived from the Timeline Follow-back interview.

Blood biochemistry

Elevations in the serum transferin protein, carbohydrate deficient transferrin (%CDT; cut score set at 2.6% for both genders), and the liver enzyme, γ-glutamyl transferase (GGT; elevation level determined using laboratory-specific, gender-based cut-scores) were measured at in-person assessments to check on the degree of accuracy of self-reported heavy drinking (Anton, Lieber, and Tabakoff, 2002). A heavy-drinking blood marker index was coded as negative (no elevated test) or positive (elevated CDT and/or GGT).

Statistical power

The current study was powered to detect Setting X Alcohol Involvement and Setting X Cognitive Functioning interaction effects on percentage of voluntary abstinence days averaged across the first 12 months of follow-up. Power calculations used residual covariance matrices from arcsine-square-root transformed PDA data from Rychtarik et al. (2000), analyzed using random-intercept normal linear models. Regression models included the following variables: baseline Voluntary Abstinent Days, Time, Setting (inpatient vs. outpatient), Moderator (i.e., alcohol involvement or cognitive functioning), Setting X Time and Setting X Moderator interaction. Non-differential attrition by setting was set at 3% per quarter of follow-up. Based on Hedeker, Gibbons, and Waternaux (1999), and using Cohen’s (1988) effect-size nomenclature, a sample size of N=166 (83 per setting) at end of 12-month follow-up was needed to achieve at least 80% power for detecting the small-to-moderate Setting X Alcohol Involvement (d = 0.36) and moderate Setting X Cognitive Functioning interactions (d = 0.47) observed by Rychtarik et al. (2000).

Data analyses

Generalized linear regression models with canonical link functions were used to analyze the monthly longitudinal trajectories of each outcome of interest across the 18 months. Generalized Estimating Equations (GEE) with participant ID as the cluster identifier and a working independence correlation matrix were used to account for within-subject correlations across time points. All GEE computations were implemented using the correlatedData library of Splus 8.2 (TIBCO Spotfire Inc., 2010). Preliminary analyses indicated that assigned treatment outcome expectancies differed significantly between settings, with inpatients expressing higher confidence in their scheduled treatment than outpatients (see Table 1). To preclude the possibility that expectancy effects could confound setting effects on drinking outcomes, expectancy terms were added to all models, including main effects and 2-way interactions with alcohol involvement, cognitive functioning, and time. Expectancy X Setting interactions were also added to all models, so as to explore whether participants with high outcome expectancy for their assigned treatment setting performed better in inpatient relative to outpatient care.

The initial model specification explored treatment setting effects on the longitudinal trajectories by randomized (i.e., alcohol involvement and cognitive functioning) and non-randomized (i.e., treatment expectancy) moderators, and included recruitment setting (i.e., outpatient clinic vs. detoxification unit). This model was simplified using backwards elimination, while ensuring that no lower-order terms were deleted, if higher-order terms were statistically significant at the alpha = .05 level. Model findings were adjusted for 12-month pretreatment values of the outcome via regression modeling, and for differences in exposure through the use of an offset. Time was coded as 0–17 months since the first follow-up; cognitive functioning was coded as 0 = Low (SDMT < 42) versus 1 = High (SDMT ≥ 42); alcohol involvement was coded as 0 = Low (AUI < 30) versus 1 = High (AUI ≥ 30); baseline treatment expectancy was centered at the mean, and scaled by the distance between the mean (M=8) and the maximum expectancy (Max=10); baseline PDA was transformed to the logit scale, centered by the median, and scaled by the difference between the median (logit 0.21) and the 3^rd quartile (logit 0.56); baseline DDD were log-transformed, centered by the median, and scaled by the difference between the median (ln 12) and the 3^rd quartile (ln 16). Outcome-specific information is given below.

Percentage of voluntary abstinence days per month (PDA)

Logistic regression was used to estimate the proportion of voluntary abstinent days per 30-day observation period during follow-up, controlling for PDA differences at baseline. Days with drinking access were adjusted on a monthly basis for the number of days participants spent in restricted status, the modeled response being (number of voluntary days abstinent)/(30-involuntary abstinent days). An overdispersed binomial model was used to account for within-subject correlations in daily abstinence outcomes measured during the same month.

Monthly Point Prevalence Abstinence (PPA) and Drinks per Drinking Day (DDD)

In this two-part hurdle model (Olsen & Schafer, 2001; Atkins, Baldwin, Zheng, Gallop, & Neighbors, 2013), we first estimated the probability of complete abstinence during each month of follow-up using logistic regression for Bernoulli outcomes. We then estimated DDD on a monthly basis for participants who failed to be completely abstinent during the month (PPA = 0), controlling for differences in baseline DDD. The total amount drank per month was used as the outcome, while the number of actual drinking days per month was used as an offset. An overdispersed Poisson model was used to account for within-subject correlations in daily drinking outcomes measured during the same month. Although a zero-truncated Poisson distribution would have been more appropriate for our strictly positive DDD outcome, observed daily drinking rates were high enough to make such a correction unnecessary.

The above analyses offer advances over the linear mixed effects models employed by Rychtarik et al. (2000). In that study, DDD was treated as a continuous measure, with abstinent days coded 0, and then arcsine square root transformed. The inclusion of 0 DDD, however, underestimates actual DDD, and results in a positively skewed distribution with a mass at the origin that violates distributional assumptions of the normal linear mixed models, even after a normalizing transformation. In addition, this approach conflates two separate distributions represented in monthly DDD data: (a) the zero-nonzero distribution representing abstinence versus drinking in a given month—PPA in the current study, and (b) the amount of alcohol consumed when there is drinking (DDD), each of which could be differentially influenced by the moderator variables under study. The 2-part hurdle model employed in this study allows for separately analyzing these two distributions, and, in conjunction with PDA analyses, provides a more complete picture of participants’ abstinence and drinking intensity on drinking days.

Supplemental analyses

The 2-part model has been recently criticized by Liu et al. (2016) for ignoring a possible dependence of DDD on PDA, as subjects more likely to drink on a given day (i.e., those with lower PDA), are also more likely to consume larger amounts of alcohol (i.e., to have higher DDD). For this reason, setting effects on DDD may partly reflect setting effects on PDA. To ensure that our DDD analyses capture direct setting effects not mediated via effects on PDA, we also ran supplementary analyses with logit PDA as an additional model covariate. Also, when Alcohol Involvement X Setting or Cognitive Functioning X Setting interactions were not supported based upon dichotomization of the AIS and SDMT scales at cutoffs previously established in a clinical research setting by Rychtarik et al. (2000), continuous AIS and SDMT scores were substituted instead, and the resulting interactions assessed in terms of magnitude and significance. Similarly, when an a priori interaction effect was supported for the dichotomous AIS and SDMT scales, sensitivity analyses were conducted to determine whether alternate cutoffs might have been more appropriate for this sample.

Separate, longitudinal logistic regression models estimated using GEE methods examined the relationship between the blood index and each of PDA, PPA, and DDD in the last 30 days of the 6, 12 and 18-month follow-up periods. The interval in days between a period’s end and blood testing was dichotomized at the median of 18 days and entered in the regression model as a moderator of the association between objective and self-reported drinking measures, because the blood tests more accurately reflect recent drinking. Only for PDA did testing interval moderate the relationship (shorter interval = greater association). Hence, the PDA-blood index relationship reported from the model is for assessments completed within 18 days. The correspondence between participant and collateral reports on drinking measures was not evaluated, due to a large amount of collateral report uncertainty. The percentage agreement of participant-collateral reports on the point prevalence of involuntary abstinence was calculated at the pair level across all 18 follow-up months, and summarized via sample tertiles (median and inter-quartile range).

Pearson correlation assessed the relationship between expectancy and motivation; Spearman correlations examined expectancy associations with preference and number of continuing care sessions attended.

Results

Figure 1 provides participant flow, treatment participation, and research follow-up rates. Table 1 provides baseline characteristics by setting condition. Stratified randomization was successful in achieving balance between the settings (see Table 1). Overall, 133 were found to be in need of inpatient treatment; 43 as having no need for it, an approximate 3:1 ratio that did not differ significantly between conditions, and was comparable to the approximate 3.7:1 ratio in Rychtarik et al. (2000). Following randomization, inpatients reported significantly higher expectancies, M = 8.29, SD = 1.46, than outpatients, M = 7.72, SD = 2.05 (see Table 1). Still, expectancies showed substantial overlap across assigned treatment setting, with 98% of inpatients and 90% of outpatients having scores in the 5–10 range.

The significance of Setting X Moderator interaction terms is reported in the text. To facilitate interpretation of interactions found significant, Table 2 presents our final models for both primary and secondary outcomes, as applied separately to inpatient and outpatient settings. In this way, the table allows one to examine (a) differences in the level of influence of a parameter (e.g., alcohol involvement) depending on the setting, and (b) setting differences at different parameter levels. For example, a replication of the Rychtarik et al. (2000) Setting X Alcohol Involvement interaction on PDA would be exemplified in Table 2 by (a) a non-significant effect of alcohol involvement in the inpatient setting, but a significant effect in the outpatient setting, and (b) significant setting effects in both inpatient and outpatient columns at high, but not at low, alcohol involvement levels. If the Setting X Alcohol Involvement interaction was not significant, and a main effect of alcohol involvement also did not contribute significantly to the model, alcohol involvement would not be retained in the model, and would not appear in the Tabled PDA results.

Table 2.

Final analysis models by setting for primary and secondary drinking outcomes.

Outcome Variable	Retained Component	Inpatienta			Outpatientb
(Covariate Reference Value)
PDA		OR	95% CI	p	OR	95% CI	p
Median baseline PDA Mean expectancy	Intercept	12.17	[6.92, 21.39]	<.001	3.58	[2.43, 5.27]	<.001
	Baseline PDA	1.23	[1.10, 1.37]	<.001	1.23	[1.10, 1.37]	<.001
	Expectancy	2.37	[1.54, 3.67]	<.001	0.96	[0.70, 1.31]	.789
	Setting	0.29	[0.15, 0.58]	<.001	3.40	[1.72, 6.72]	<.001
	Linear time	0.83	[0.73, 0.94]	.005	1.01	[0.93, 1.09]	.870
	Quadratic time	1.01	[1.00, 1.02]	.013	1.00	[1.00, 1.00]	.881
Median baseline PDA High expectancy	Intercept	28.88	[13.32, 62.63]	<.001	3.43	[1.95, 6.03]	<.001
	Baseline PDA	1.23	[1.10, 1.37]	<.001	1.23	[1.10, 1.37]	<.001
	Expectancy	2.37	[1.54, 3.67]	<.001	0.96	[0.70, 1.31]	.789
	Setting	0.12	[0.05, 0.31]	<.001	8.42	[3.24, 21.87]	<.001
	Linear time	0.83	[0.73, 0.94]	.005	1.01	[0.93, 1.09]	.870
	Quadratic time	1.01	[1.00, 1.02]	.013	1.00	[1.00, 1.00]	.881
VOLUNTARY PPA		OR	95% CI	p	OR	95% CI	p
Median baseline PDA Mean expectancy	Intercept	1.62	[1.08, 2.43]	.020	0.86	[0.57, 1.29]	.460
	Baseline PDA	1.15	[1.02, 1.30]	.019	1.15	[1.02, 1.30]	.019
	Expectancy	2.00	[1.21, 3.28]	.007	0.99	[0.71, 1.38]	.953
	Setting	0.53	[0.30, 0.94]	.029	1.89	[1.07, 3.35]	.029
	Linear Time	0.97	[0.93, 1.00]	.042	1.02	[0.98,1.05]	.318
Median baseline PDA High expectancy	Intercept	3.23	[1.85, 5.66]	<.001	0.85	[0.49, 1.48]	.561
	Baseline PDA	1.15	[1.02, 1.30]	.019	1.15	[1.02, 1.30]	.019
	Expectancy	2.00	[1.21, 3.28]	.007	0.99	[0.71, 1.38]	.953
	Setting	0.26	[0.12, 0.57]	<.001	3.81	[1.75, 8.31]	<.001
	Linear Time	0.97	[0.93, 1.00]	.042	1.02	[0.98,1.05]	.318
DDD		RR	95% CI	p	RR	95% CI	p
Median DDD Low involvement	Intercept	7.92	[5.71, 10.98]	<.001	6.76	[5.29, 8.65]	<.001
	Baseline DDD	1.08	[1.03, 1.13]	.002	1.08	[1.03, 1.13]	.002
	Alcohol Involvement	0.93	[0.64, 1.37]	.722	1.76	[1.31, 2.38]	<.001
	Setting	0.85	[0.58, 1.26]	.430	1.17	[0.79, 1.73]	.430
Median DDD High involvement	Intercept	7.39	[6.08, 8.98]	<.001	11.92	[10.20, 13.94]	<.001
	Baseline DDD	1.08	[1.03, 1.13]	.002	1.08	[1.03, 1.13]	.002
	Alcohol Involvement	1.07	[0.73, 1.57]	.722	0.57	[0.42, 0.77]	<.001
	Setting	1.61	[1.25, 2.07]	<.001	0.62	[0.48, 0.80]	<.001
INVOLUNTARY PPA		OR	95% CI	p	OR	95% CI	p
Mean expectancy	Intercept	10.30	[6.00, 17.68]	<.001	6.34	[3.83, 10.49]	<.001
	Expectancy	2.95	[1.30, 6.71]	.010	1.22	[0.81, 1.83]	.335
	Setting	0.62	[0.31, 1.21]	.162	1.63	[0.82, 3.21]	.162
	Linear Time	0.88	[0.80, 0.97]	.009	0.93	[0.84, 1.02]	.136
	Quadratic Time	1.01	[1.00, 1.01]	.008	1.01	[1.00, 1.01]	.008
High expectancy	Intercept	30.36	[10.85, 84.96]	<.001	7.73	[3.79, 15.75]	<.001
	Expectancy	2.95	[1.30, 6.71]	.010	1.22	[0.81, 1.83]	.335
	Setting	0.25	[0.08, 0.84]	.025	3.93	[1.19. 12.98]	.025
	Linear Time	0.81	[0.71, 0.92]	.001	0.93	[0.83, 1.06]	.285
	Quadratic Time	1.01	[1.00, 1.01]	.008	1.01	[1.00, 1.01]	.008

Note: OR = Odds Ratio; RR = Rate Ratio; PDA = percentage of abstinent days per month; Expectancy = mean item score of the Feelings About Your Scheduled Treatment questionnaire; Voluntary PPA = monthly point-prevalence of complete voluntary abstinence; DDD = monthly standard drinks consumed on a drinking day (log-transformed); Involuntary PPA = monthly point-prevalence of no involuntary abstinence (hospitalization, inpatient treatment, incarcerations; Any=0, None=1); Alcohol involvement measured via Alcohol Involvement Scale (low: < 30; high: ≥ 30).

^aSetting coded: 0 = inpatient, 1 = outpatient;

^bSetting coded: 0 = outpatient, 1 = inpatient. Time is coded 0–17 months since the first follow-up month.

Did alcohol involvement and cognitive functioning moderate setting effects?

PDA

Neither alcohol involvement nor cognitive functioning significantly moderated setting effects on PDA trajectories. Supplemental analyses also failed to find significant moderation effects using continuous measures of the moderators.

PPA and DDD

Cognitive functioning and alcohol involvement also failed to moderate monthly PPA rates. Supplemental analyses based on continuous measures did not change this finding. However, consistent with Rychtarik et al. (2000), a significant Setting X Alcohol Involvement interaction (p = .007) did emerge for DDD. Expressed as a drinking rate ratio (RR), among those with higher alcohol involvement, the rate of drinking on a drinking day was 1.61, 95% CI [1.25, 2.07] times higher among outpatients than inpatients; whereas among lower involvement participants the drinking rate of outpatients was not significantly different from inpatients (RR = .85, 95% CI [.58, 1.26]. To better understand the magnitude of these differences, we note that model intercepts in Table 2 show DDD at follow-up for typical study participants consuming approximately 12 DDD at baseline. Large drinking reductions were observed among low-involvement participants, whether as an inpatient, M = 7.92, 95% CI [5.71, 10.98], or an outpatient, M = 6.76, 95% CI [5.29, 8.65]. In contrast, high-involvement participants experienced drinking reductions only in the inpatient setting, M = 7.39, 95% CI [6.08, 8.98]; outpatients showed no change from baseline, M = 11.92, 95% CI [10.20, 13.94]. These protective effects of inpatient care persisted across the follow-up period (see Figure 2A).

Figure 2.

Primary follow-up drinking outcome results depicting (A) the Setting X Alcohol Involvement interaction on monthly drinks per drinking day, (B) Setting X Time², and Setting X Expectancy interactions on monthly percentage of days voluntarily abstinent, (C) the Setting X Time, and Setting X Expectancy interactions on the point prevalence of monthly abstinence. IP = inpatient; OP = outpatient; Exp = expectancy.

Supplemental DDD Analyses

Adding logit PDA as an additional covariate in the DDD model revealed a negative, but not statistically significant, association (p = .45) that failed to deflate the Setting X Alcohol Involvement interaction. Further, detailed interaction probing revealed that the largest interaction effect did indeed occur at the AIS cut point of 30, as derived in the previous study.

Did outcome expectancies contribute to the analysis models?

PDA

A significant Setting X Expectancy interaction (p < .001) emerged, such that the benefit of inpatient over outpatient was nearly two and a half times stronger among high vs. mean expectancy participants throughout follow-up, OR = 2.48, 95% CI [1.42, 4.22]. Interaction probing over the full range of expectancy scores revealed that inpatients had higher PDA than outpatients at the first month of follow-up for scores above 6.2, representing 85% of the sample. The inpatient benefit attained statistical significance for scores above 7.5, representing 69% of the sample. Setting also interacted with Time, such that inpatient PDA rates decreased during early follow-up and then rebounded, while outpatient rates rose steadily, but at a non-significant pace. Figure 2B shows the combined effects of the above Setting X Expectancy and Setting X Time relationships for participants with median baseline PDA of 21% at mean (M = 8) and high (Max = 10) expectancy levels. As shown, large differences favoring mean-expectancy inpatients at Month 1, OR = 3.40, 95% CI [1.72, 6.72], became deflated by Month 6, OR = 1.58, 95% CI [0.97, 2.59]. Differences in PDA rates continued to diminish until Month 12, OR = 1.10, 95% CI [0.66, 1.85]. Although they rebounded slightly in magnitude at Month 18, OR = 1.41, 95% CI [0.73, 2.72], they remained non-significant. In contrast, high-expectancy inpatients derived large and statistically significant benefits throughout follow-up: Month 1, OR = 8.42, 95% CI [3.24, 21.87]; Month 6, OR = 3.91, 95% CI [1.82, 8.41]; Month 12, OR = 2.72, 95% CI [1.31, 5.68]; and Month 18, OR = 3.49, 95% CI [1.52, 7.99].

PPA and DDD

A significant Setting X Expectancy (p = .021) interaction also emerged on monthly PPA, with inpatient setting effects twice as strong across time among high vs. mean expectancy participants, OR = 2.02, 95% CI [1.11, 3.66]. Paralleling our PDA findings, Figure 2C shows that inpatients with high treatment expectancy were more likely to exhibit periods of sustained abstinence throughout follow-up than outpatients. For mean-expectancy participants, the odds of abstinence converged quite rapidly over time across treatment settings, with inpatients’ odds of abstinence almost twice that for outpatients at Month 1, OR = 1.89, 95% CI [1.07, 3.35], one and a half times as large at Month 6, OR = 1.47, 95% CI [0.92, 2.34], and a fifth lower at Month 18, OR = 0.80, 95% CI [0.42, 1.50]. In contrast, convergence was slower for high-expectancy participants, who experienced strong benefits from inpatient treatment at both Month 1, OR = 3.81, 95% CI [1.75, 8.31], and Month 6, OR = 2.96, 95% CI [1.48, 5.91], though these differences were statistically deflated by Month 18 as well, OR = 1.61, 95% CI [0.73, 3.51]. Expectancy was not associated with DDD.

Motivation, preference, and continuing care sessions

Weak, but significant, correlations emerged between expectancy and motivational readiness, r = 0.17, 95% CI = [0.02, 0.31]; post-treatment preference for an alternate clinic setting, r_s = −.21, 95% CI [−0.34 −0.06]; and continuing care sessions attended, r_s = 0.22, 95% CI [0.07, 0.35]. These relationships did not vary by treatment setting. Hence, treatment expectancy did not appear to be simply a proxy for either motivation or preference, and differential aftercare attendance could not easily account for the interactions of expectancy and setting observed on primary and secondary outcomes.

Did involuntary abstinent days (i.e., incarcerations, hospitalizations, residential treatment) differ between settings?

Table 2 models the probability of experiencing no involuntary abstinent days during any 30-day period during follow-up. Overlaying a quadratic time effect (p < .008), a significant Setting X Expectancy X Linear Time interaction (p = .027) was detected, such that expectancy moderated the linear time slope among inpatients (p = .014), but not outpatients (p = .653). Figure 3 plots this effect on the prevalence any of involuntary abstinence. Trajectories among high and low expectancy outpatients are similar; but, among inpatients, high versus mean expectancy lowered the prevalence initially and delayed the peak prevalence by six months (i.e., from Month 10 to Month 16).

Figure 3.

Monthly point prevalence of any involuntary abstinence (i.e., incarceration, hospitalization, residential treatment) across follow-up by setting and expectancy level. IP = inpatient; OP = outpatient; Exp = expectancy

Did blood chemistry and collateral reports agree with self-reported alcohol measures?

Considerable agreement was observed between 30-day self-reported outcomes and blood chemistry assessments. Among 144 participants with such assessments, the odds of a negative blood index doubled with each 10% increase in PDA, OR = 2.00, 95% CI [1.48, 2.70]; quadrupled with 30-day PPA, OR = 4.09, 95% CI [2.43, 6.91]; and decreased by 6% per DDD, OR = .94, 95% CI [.92, .97]. Median 30-day participant-collateral agreement on point prevalence of involuntary abstinence (N = 160) was 94% (Q1 = 83%; Q3 = 100%).

Discussion

Two key questions arise when considering the relative benefits of inpatient versus outpatient care. Below we discuss these questions in the context of the current study, and the extent to which they are informed by the study results.

Does inpatient treatment produce better alcohol treatment outcomes than outpatient treatment for all comers, and, if so, how big is the advantage?

Though only a tertiary focus of the current study, the significant Setting X Linear Time and Setting X Quadratic Time interactions for PPA and PDA, respectively, are consistent with findings of a meta-analysis suggesting an initial, but decreasing benefit of inpatient over outpatient care in the early months of follow-up (Finney & Moos, 1996). In this study, however, the effect appeared stronger than that previously reported. At mean expectancy levels, the advantage of inpatient treatment for PDA was quite large (OR = 3.40) in the first month posttreatment, but was halved by Month 6 (OR = 1.58). For PPA, the advantage for inpatients was not as large (OR = 1.89) at Month 1, and was attenuated further by Month 6 (OR=1.47). These results highlight the importance of frequent measurement for assessing time trends in outcome research. This study did not evaluate potential mediators of these effects. Specifically, the mechanism through which individuals treated in the inpatient setting achieve better abstinence outcomes, at least initially, is unclear, as are the reasons that this relative benefit is not sustained across time. Finney, Hahn, and Moos (1996) have hypothesized potential mechanisms of the inpatient effect (e.g., the inpatient respite from drinking and stressors), but these were not evaluated in this randomized trial. Identifying mediators of these effects may lead to improvements in both inpatient and outpatient care.

Does inpatient treatment produce better alcohol treatment outcomes than outpatient treatment among identifiable subgroups of clients and, if so, how big is the advantage?

This study partially validated, in an uncontrolled community treatment environment, the Rychtarik et al. (2000) finding that individuals with higher levels of alcohol problem severity benefit more from inpatient than outpatient care. However, the replicated finding was limited to a Setting X Alcohol Involvement interaction on DDD. Namely, inpatient treatment significantly reduced DDD among participants at or above the severity cut point, while outpatient treatment did not. This specific finding appears to be a robust interaction effect that generalizes across both research and clinical environments. However, contrary to Rychtarik et al., neither problem severity nor cognitive functioning moderated setting effects on abstinence.

Noteworthy, though serendipitous, findings indicated that treatment expectancies were higher among inpatients than outpatients, and that these expectancies moderated setting effects, being positively prognostic of PDA and PPA among inpatients, but not outpatients. The result was markedly higher and sustained PDA and PPA among high expectancy inpatients, but not outpatients. Expectancies also moderated setting effects on the incidence of hospitalization, incarceration, or residential treatment, with high expectancy appearing to delay the onset of such involuntary abstinence among inpatients but not outpatients. Expectancies have long been known to influence outcomes in psychological treatments, and such effects can be quite large and robust (e.g., Greenberg, Constantino, & Bruce, 2006). But, expectancy effects typically have not been assessed in AUD outcome research (Raylu & Kaur, 2012). The current results suggest that, at least under certain as yet unspecified conditions, expectancy may significantly influence AUD outcomes, as well. However, whether outcome expectancies could be useful in clinical, level-of-care decision-making is not known. The expectancy assessment in this study occurred only after the individual knew of his or her setting assignment, not before, as would be required for making placement decisions. As in most expectancy research, the assessment relied on a not-widely studied expectancy measure (Younger, Gandhi Hubbard, & Macky, 2012). Using this same measure, Rychtarik et al. (2000) found no main expectancy effect, and did not examine expectancy moderation effects. Future research is needed that implements a more clinically-relevant expectancy assessment, before recommendations can be made as to expectancy’s clinical value in level of care decision making.

This report does not address a third key question in this area of research. That is, whether the initial, though decreasing benefits from inpatient treatment and the reduction in the rate of drinking among high involvement inpatients versus outpatients outweigh the increased costs of inpatient care. This report did not examine the relative cost-effectiveness of the two treatment settings, nor did it examine any cost savings of placing individuals with higher severity AUDs in inpatient versus outpatient care. Importantly, inpatient treatment in this study was provided in a medical facility, where treatment costs would be expected to be higher in the first place. Further research needs to assess whether the relative benefit of inpatient care early on, and for those with higher severity is generalizable to lower-cost, non-medical residential treatment facilities.

A number of additional qualifications and limitations should be noted with regard to the current findings. First, this study compared an initial period of inpatient treatment followed by continuing outpatient care only, as did Rychtarik et al. (2000). So, it was not a “pure” inpatient versus outpatient comparison. Yet, the design reflects current treatment standards in which inpatient treatment is rarely curtailed abruptly, without referral to continuing outpatient care. Second, the AIS measures a broad severity dimension that includes both negative consequences and physical dependency symptoms. It remains unclear whether similar findings would be found using alternate severity measures. Third, the sample represented a relatively small portion of clients presenting for care, and as with other research on treatment settings was likely biased toward greater severity (Rychtarik, McGillicuddy, Connors, & Whitney, 1998). Hence, study findings may not be generalizable to all clients presenting for AUD treatment. Fourth, only one of six a priori moderation tests (2 moderators X 3 primary outcomes) yielded a significant finding. That the one significant moderation effect replicated a prior finding increases the confidence in its veracity, but it is still possible it resulted from chance. Fifth, due to the use of naturalistic settings in the study, we cannot rule out that between-setting differences in the services offered, or the clustering of participants within the treatment staff and settings accounted for the results obtained. Detailed records of individual services obtained were not recorded. However, both programs had the same treatment philosophy, much of treatment was group-based, and staff from both programs received joint training experiences. Sixth, the results also are tempered by the fact that the Treatment X Alcohol Involvement Severity interaction was observed only on DDD, and not PDA, as found in Rychtarik et al. (2000), and also not on the newer PPA variable used in the 2-part model. Further, while the number of DDD was nearly cut in half from the pretreatment level among high-involved inpatients relative to outpatients, the mean level of the former was still above safe consumption guidelines. Yet, this reduction in daily consumption might still have had a positive health impact.

To summarize, this study found support for (a) the initial but decreasing benefit of inpatient over outpatient care across time, and (b) the validity of alcohol involvement as a client placement criterion for determinations of level of care decisions, at least with respect to alcohol consumption rates. Assigned treatment expectancy also moderated setting effects, but additional research is needed before judgments can be made as to its clinical utility. Also, future research is needed that studies the mediators of the expectancy and other setting effects observed, and the effects in lower-cost residential care environments.