Mediational Analyses of the Effects of Social Behaviour and Network Therapy on Alcohol Use Open Access

Alcohol treatment trials are primarily designed to address whether the evaluated interventions work or not, and less attention is given to how they may work and under what circumstances [1]. The study of mediators and moderators has long been recognized as an important means of advancing understanding of how behaviour change works, and this can in turn assist in developing more effective interventions [2‒4].

The United Kingdom Alcohol Treatment Trial (UKATT) study tested the null hypothesis that there were no differences in outcomes between a less intensive, motivationally based treatment (motivational enhancement therapy [MET]) and a more intensive socially based treatment (social behaviour and network therapy [SBNT]). The study was undertaken in a sample of 742 clients in treatment for alcohol problems [5, 6]. No group differences in alcohol outcomes were found between treatments after 3 and 12 months from study entry [6].

MET is a globally recognized treatment and has been broadly evaluated worldwide [7, 8]. MET uses a combination of motivational interviewing (MI) with personal feedback of assessment results [9]. Process studies have investigated mediators in MET and MI [10‒17], making this a relatively well-understood treatment. SBNT’s possible mediators have not been similarly studied.

SBNT is based on an integration of effective strategies found in other network and behavioural treatments. The essence of SBNT is the active involvement of the practitioner in recruiting a support network for change, focussed on reducing or stopping drinking and giving attention to life contexts and relationships, with the aim of replacing social support for drinking with social support for change [18]. Evidence shows that network support treatments can effectively change drinker’s social network and improve long-term drinking outcomes [19, 20].

Our previous analyses of UKATT treatment processes found that the quality of MET skills significantly predicted 12-month alcohol outcomes across both MET and SBNT [10]. In addition, our recent study found that a change in social support for drinking by the end of the treatment did not operate as a mediator of SBNT skills on alcohol outcomes [21]. These findings underline the need for further study of how SBNT exerted its effects in UKATT.

A large literature attests to the importance of therapist skilfulness for positive treatment outcomes [22], including prior studies we have conducted on UKATT data (e.g., Gaume et al. [10]). Qualitative data from UKATT showed that three-quarters of the SBNT therapists and nearly half of the clients referred to the involvement of others, a specific therapist skill, as the most useful aspect of sessions [23]. This was also the factor to which clients most frequently attributed change in alcohol use [24]. Therefore, if not through altered social network, there is a need to understand how SBNT works as it secured treatment outcomes not inferior to MET in UKATT Research Team [6]. Therapist skill in delivering this treatment appears promising to investigate further.

Considering the main effects of the UKATT and that studies supporting the effects of treatment processes on 12-month alcohol outcomes, the aim of the present study was to examine whether SBNT effects on drinks per drinking day (DDD) and percentage of days abstinent (PDA) 12 months after treatment entry, in comparison with MET, were mediated by SBNT-specific therapist skills. Considering there are few studies on this specific topic and limitations of the available evidence on similar treatments, the present study examines the full range of quantity, quality, and combined quantity-quality measures in an exploratory fashion.

This study is a secondary analysis of data from UKATT, a multicentre, pragmatic, randomized controlled trial, and the largest trial of treatment for alcohol problems conducted in the UK. A detailed description of the original trial study is reported elsewhere [5, 6, 25].

Participants in the UKATT study were randomized to MET or SBNT. SBNT is a social network-based treatment comprising multiple components drawn from well-established treatment principles. It employs a range of cognitive and behavioural strategies through a collaborative therapeutic relationship aimed to support clients to mobilize and develop positive social networks that are supportive of a change in drinking [5, 18]. In UKATT, SBNT was carried out in up to eight 50-min sessions over a maximum of 12 weeks [5]. In contrast to SBNT, MET is a shorter, less intensive, and less costly intervention, based on the principles of the MI [26] combined with feedback [9]. Clients randomized to MET received up to three sessions, the first two a week apart and the third approximately 6 weeks later, again within a maximum of 12 weeks [5].

UKATT primary outcomes were the number of DDD and PDA, both recorded at the final study outcome assessment 12 months after baseline [6]. Abstinent clients (i.e., having PDA = 100) were allocated a DDD of zero.

Therapist skills were measured using the UKATT Process Rating Scale, a validated instrument specifically designed to monitor treatment fidelity and assess treatment processes in UKATT [27]. All UKATT therapy sessions were video recorded with the client’s permission and one video per participant was randomly sampled, stratified by treatment condition, treatment centre, and session number (1–3 for MET; 1–8 to SBNT). The number of coded sessions by intervention group is presented as online supplementary Table 3 (for all online suppl. material, see https://doi.org/10.1159/000535200). Replacement of videos was used to maintain balance between the criteria above.

The Process Rating Scale measured both the frequency (to what extent the therapist carried out each specific item, rated from 0 = not at all to 4 = extensively) and quality (rating of the therapist’s behaviour, ranging from 0 = not at all well to 4 = very well) of treatment-specific task completion. Quality items could not be assessed when the frequency measure was 0 (i.e., the specific item was not observed at all in the given session). Of 20 Process Rating Scale items, eleven were related to MET (not used in the present analysis) and nine were related to SBNT (homework, alternative activities to drinking, social support for change, involvement of others in behaviour change, identification of sources of support for change, therapist as task oriented, therapist as an active agent for change, collaboration, and interpersonal focus). A description of each SBNT process rating item is provided in online supplementary Table 1.

Videos were scored by two or three independent raters and the primary rater was blind to the therapy type. The rater was trained by the UKATT team and supervised using independently rated tapes to enhance rating manual adherence and score consistency over time. Inter-rater reliability was measured using intra-class correlations and overall agreement for individual items is considered high overall; additional details on the Process Rating Scale validity and reliability are available in Tober et al. [27].

Since quality items had large amounts of missing data (from 15% to 90%, particularly for MET sessions – see online suppl. Table 2), we created a new score combining frequency and quality. Quality scores were recoded to be negative for low quality (0 = −3, 1 = −2), neutral for medium quality (2 = 1), and positive for high quality (3 = 2, 4 = 3). These scores were then multiplied by the unchanged frequency score. The skills scores on this measure thus range from −12 (e.g., high frequency [4] of low quality skills [−3]) to +12 (e.g., high frequency [4] of high quality skills [3]). When the skill was not present (frequency = 0), a score of 0 was attributed. Online supplementary Table 2 presents the frequency, quality, and skills scores for each SBNT item by treatment groups.

The UKATT study sought to include clients who would normally receive an offer of treatment for alcohol problems from British treatment facilities. The sample was recruited from seven treatment sites and included clients aged 16 and over [6]. Of the 742 participants in the UKATT, 290 had no video available for Process Rating Scale coding or were unrateable. Among the 452 participants with available videos, we only considered data from participants with 12-month follow-up assessment, resulting in a sample of 376 (N = 159 for SBNT and N = 217 for MET).

We used a simple mediation approach to identify potential mediators of SBNT effects on DDD and PDA compared to MET. Mediation models provide information on direct, indirect, and total effects. The direct effect (known as the c’ path) refers to the effect of X on Y independently of all mediators (i.e., controlling for all mediators) [28]. The indirect effect represents how Y is influenced by X through a causal dependence where X influences the mediator M (known as the a path) and M influences Y (known as the b path) [28]. Said another way, the indirect effect represents the influence of X on Y passing through M. The total effect is the combined effect of the direct and indirect effects and corresponds to the main effect of X on Y (i.e., without controlling for M).

First, we modelled Process Rating Scale frequency items as simple mediators. Second, we repeated these models using our new skills scores (combined frequency and quality). We used separate models for each outcome (PDA and DDD). All mediation models considered the randomization group (0 = MET; 1 = SBNT) as the X variable and all models were controlled for age, sex, the baseline measure of the outcome, and the number of days between baseline and the coded session (to control for time in treatment). We used the robust ML estimator and confidence intervals were calculated using bootstrap sampling with 5,000 draws. Analyses were performed using R v4.1.1 [29] and RStudio [30] via package lavaan [31] and with the minimum significance level of 5%.

We used a Bayesian approach estimator for sensitivity analysis, which combines prior distributions to form posterior distributions for the parameter estimates [32]. This estimator uses Markov chain Monte Carlo algorithms to create approximations to the posterior distributions and accounts for the skewness of the outcomes. Analyses used non-informative priors and 95% credible interval, which is the default option in Mplus [32]. Model fit considered the posterior predictive p value [33], and values close to 0.5 suggest that the model is true or close to true [34]. We set the number of iterations to 30,000 and model convergence was checked using the potential scale reduction values after increasing the number of fixed iterations to 60,000. Sensitivity analyses were performed using Mplus v7 [35] with 5% significance level.

Table 1 presents sample characteristics and outcome data used in the present analyses. There were no significant differences between treatment groups on sample characteristics and outcomes. Descriptive data for each mediator by intervention group is presented in online supplementary Table 2. SBNT group had significantly higher scores for all frequency items, and for all combined skills scores, except for collaboration which showed a non-significantly different combined skills score across groups (p = 0.096).

The mediation models using Process Rating Scale frequency items are shown in Table 2. The SBNT group showed significantly higher frequency of all mediators compared to MET (a paths), providing further evidence of the distinct nature of the delivery of the treatments. Only two mediators, however, were significantly related to the outcome (b paths). First, therapist as active agent was associated with higher PDA. Second, social support for change was unexpectedly associated with higher and not lower DDD at 12 months.

Consistently, examination of indirect effects showed that the frequency of therapist as active agent mediated the effects of SBNT by increasing PDA after 12 months (indirect effect = 0.03, 95% CI: 0.003; 0.05, p = 0.03). Similarly, but in the opposite direction, the frequency of therapist advocacy of social support for change mediated the effects of SBNT by increasing DDD after 12 months (indirect effect = 0.10, 95% CI: 0.01; 0.18, p = 0.02). The findings were confirmed in the sensitivity analysis (online suppl. Table 4).

The models using the combined skills scores (combining frequency and quality of SBNT items) as mediators are presented in Table 3. Compared to MET, the SBNT group showed significantly higher scores on all tested mediators (a paths), except for therapist as active agent and collaboration, which were non-significant. Higher skills scores on involvement of others in behaviour change and therapist as active agent were related to increased PDA at 12 months (b paths). Consistently, there was only one significant indirect effect for involving others in behaviour change, which mediated SBNT effects on PDA (indirect effect = 0.06, 95% CI: 0.02; 0.10, p = 0.009). The findings were confirmed in the sensitivity analysis (see online suppl. Table 5).

The main findings of this study are that as follows: (1) Compared to MET, SBNT effects on alcohol outcomes were not mediated by treatment-specific therapist skills as measured by the Process Rating Scale; (2) therapists acting more frequently as an active agent supporting behaviour change and showing higher skills scores (combined frequency and quality) in planning the involvement of others in working towards behaviour change mediated SBNT effects on increasing percentage of days abstinent; and (3) in contrast, the frequency with which therapists stressed the importance of social support in achieving change unexpectedly mediated effects of SBNT in increasing drinking (i.e., increased DDD).

This study is the first investigation dedicated to understanding the role of treatment-specific therapist skills in how SBNT works. The results are derived from a large sample in a pragmatic trial, which enhances data generalizability to populations that typically receive treatment for alcohol problems [5, 6]. The sensitivity analysis using a Bayesian approach helps to attenuate type 1 error as it does not rely on large sample theory and data normality [32]. Results should however be interpreted in the light of study limitations. We tested only those SBNT-specific therapist skills that were expected to be not or minimally present in MET sessions in line with the UKATT design. Process Rating Scale measures were rated within only one treatment session over the 8 possible sessions in SBNT and 3 in MET and thus capture only a portion of the full treatment. Treatments were delivered over multiple sessions and we were not able to capture possible effects of treatment attrition, and relatedly, we may have missed therapeutic processes that further developed towards the end of the full treatment. It is important to note, however, that despite being a longer intervention, SBNT did not differ in effectiveness from MET, with both therapy groups showing reductions in drinking over 12 months [6]. Despite the longitudinal design supporting the temporality of the process measures, the results presented cannot be assumed as causal when one bears in mind these caveats.

The present study identified three therapist SBNT skills that significantly mediated the effect of SBNT on alcohol outcome, as compared to MET. When using skills scores combining frequency and quality, therapists’ skills in planning the involvement of others in behaviour change showed a significant indirect effect on percentage of days abstinent. Developing and consolidating a social network for change by involving the client’s social network is a basic principle of SBNT [18], and our findings support this basic concept and other evidence underscoring the importance of client’s network in achieving beneficial outcomes [36‒38]. A previous qualitative study in UKATT showed that the involvement of others was considered the most useful aspect of sessions for the majority of SBNT therapists and nearly half of the clients; clients also frequently attributed to this factor their change in alcohol use [23, 24]. There is clearly a need to develop a clearer understanding of how, and how well, the involvement of others is accomplished by the therapist within the treatment process.

When examining the findings on the frequency items, a clear picture emerges of a rather complex set of phenomena. The extent to which therapists actually carried out tasks on behalf of the client (therapist as active agent, e.g., telephoning or writing to the friends or family of the client about their drinking and treatment, with the consent of the client) showed a significant indirect effect on percentage of days abstinent. It should be noted in passing that the skills score combining frequency and quality for this item was related to increased percentage of days abstinent (b path), but treatment did not predict the mediator (a path, p = 0.06). Therapist as active agent for change is the Process Rating Scale item with lowest inter-rater reliability, low frequency ratings, and low loadings [27]. Many of the tasks related to this skill largely occur, by definition, between sessions (e.g., telephoning, visiting, or writing to network members) and the quality of performance, beyond the fact that they have been undertaken, may not be evident during the recorded session. For these reasons, this skill was only coded as being present in approximately 10% of the sessions.

That is only part of the picture, however, as the frequency with which therapists stressed the importance of social support in achieving change showed a significant indirect effect, yielding unwanted treatment outcomes. This adverse effect disappeared, however, in the analysis of the skill score, indicating that if this core treatment task is performed well, then there is no adverse effect. This finding, therefore, shows that there are sessions in which frequent emphasis on the importance of social support is not performed well, and this can detract from positive treatment outcomes. Other evidence suggests that social network intervention effects are mediated by changes in the proportion of non-drinkers in a client’s network, rather than how supportive of drinking is the network [20].

This particular finding should also be interpreted in the context of the other mediational findings. Taken together, therapists concentrating more on being active agents of change, by doing things that help clients with their consent, rather than being mere advocates for social support of change, leads to better treatment outcomes. These findings are also congruent with the other demonstrated mediator incorporating skill in involving others in change, suggesting that how well the closely related tasks are accomplished has important implications for treatment outcome. This is also in line with our previous study finding in which quality, but not frequency, of MET skills is on the causal pathway to observed outcomes, including for SBNT [10].

It is well established that social networks play an integral role in treating drinking problems and that influences may be positive, negative, or mixed [39]. The level of emotional support provided by network members and the quality of their support are related to abstinence and relapse rates [40]. Therefore, it has been suggested that if the client has little investment in his or her social network or the network has a strongly negative influence the therapist should attend less to involving social network members [39]. Thus, it could also be the case that the unexpected negative result may have occurred due to therapists stressing the importance of social support among clients without a clearly supportive network (i.e., in the absence of change in network composition). Although therapists can provide interventions that initiate behavioural changes, the client’s social network, if not reconfigured by them, could prevail over the longer term [41]. Therefore, future studies should explore whether the supported mediators are related to changes in social network configuration.

Overall, the results of this study should be interpreted in light of all other non-significant indirect effects, indicating that most hypothesized SBNT therapist skills did not mediate SBNT effects on alcohol outcomes. The frequency of all items and most skills scores (combining frequency and quality) of items theorized as SBNT skills were actually higher in SBNT (i.e., significant a paths). This suggests that even though they are not found only in SBNT, they occurred, as expected, with higher frequency and more skilfully within its sessions. The effects on alcohol outcomes after 12 months were, however, largely not related to these processes (i.e., non-significant b paths). Together with our recent study showing that changes in a client’s social support for drinking after treatment did not mediate the effects of SBNT skills on alcohol outcomes [21], the present findings should emphasize that processes of change relating to SBNT remain largely unknown.

Study findings are, nonetheless, interesting in various ways. They are treatment specific, belonging to the distinctive SBNT content, and at the same time likely to be one component of a repertoire of advanced skills possessed by well-trained therapists. Teasing apart the common factors at play and their interplay with treatment-specific elements is one of the major challenges that has faced the alcohol treatment field for some time, and this remains an ongoing challenge [42]. Therefore, as the analyses were exploratory in nature, we suggest that the present findings should be regarded as hypothesis generating and due for further confirmation in studies of the active ingredients in alcohol treatment.

In conclusion, in UKATT we found little support for the hypothesis that SBNT exerted effects on alcohol outcomes in ways that were clearly distinct from MET, despite the contrasting nature of the two treatments as they were delivered. Better outcomes were obtained through the application of more advanced therapist skills in planning the involvement of others in behaviour change and through more frequent activity on the part of the therapist on behalf of the client, including actions outside treatment sessions. Unexpected findings suggest stressing the importance of social support needs to be handled carefully. The main contribution of the present study is to identify specific targets for further investigation in alcohol treatment process studies.

Ethical approval was obtained through the local NHS Ethics Committee for each of the seven treatment centres and in accordance with national guidelines. Written informed consent was obtained from each participant prior to participation, and the authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

The authors have no conflicts of interest to declare.

A.B. received grants from São Paulo Research Foundation (FAPESP): grants #: 2017/13831-9, 2018/12729-9 and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq): grant # 427671/2018-3. M.L.O.S.F. received grant from CNPq: #315769/2018-1, Associação Fundo de Incentivo à Pesquisa (AFIP), and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES Print). UKATT (the United Kingdom Alcohol Treatment Trial) funding was funded by the Medical Research Council (Project Grant G9700729). The funding agencies did not play any role in study design, collection, analysis, and interpretation of data.

A.B., J.G., and J.M. outlined the present study. N.H., G.T., and D.R. were part of the research team who designed, got funding, and conducted the parent trial (UKATT Research Team). A.B. performed the analyses and wrote the first draft of the manuscript. J.M., J.G. and J.S. reviewed the analyses. M.L.O.S.F. contributed to data interpretation. All authors contributed to manuscript writing, editing, or data interpretation. All authors revised and approved the final manuscript.

Research data are not publicly available on legal or ethical grounds. Further enquiries can be directed to the corresponding author.