When Hunter-Reel and colleagues  contacted us initially with a draft version of this manuscript we were thrilled with their proposed extension of the dynamic model of relapse  and their thoughtful comments on potential ways of testing interpersonal factors within a dynamic modeling framework. The original relapse prevention model  incorporated interpersonal factors, such as interpersonal conflict and social pressure to drink, as high-risk situations for relapse, but did not provide a discussion of interpersonal relationships and social support as a unique part of the model. Our dynamic version of the relapse model [2,4] provided a cursory overview of the potential interpersonal antecedents and consequences of relapse, and we were hopeful that other researchers with greater expertise in this research area would elaborate on the role of interpersonal factors in a relapsing system.
Indeed, there is much that could be said and that remains to be studied about the role of outside relationships influencing the internal experiences of an individual who is attempting to change their behavior. As highlighted by the authors, there are a number of potential points of intersection whereby an external interpersonal experience could provide the ‘tipping point’ that either pushes the person over the edge to relapse or turns the individual away from returning to an unhealthy behavior. Miller and colleagues  have talked about quantum change as a means for this latter type of positive behavior change, and based on the review provided by Hunter-Reel and colleagues , we speculate that encouraging positive interpersonal interactions could be a critical and necessary aspect of enduring health behavior change. A meta-analysis of relapse prevention  found that relapse prevention delivered in a group format produced larger effect sizes than the effect sizes for individual treatment, thus it could be that development of interpersonal relationships between group members results in better outcomes. Indeed, a recent review of the research on Alcoholics Anonymous (AA) concluded that social support was a consistently strong mechanism in the effectiveness of AA in improving abstinence rates . This review of 24 research studies and three studies that analyzed social support specifically as a mediator of AA outcomes [8–10] indicated clearly that friends' and AA members' support for abstinence, as well as a lack of networks supportive of drinking, mediated significantly the relation between AA involvement and drinking outcomes over time. Similarly, using data from Project MATCH, Longabaugh and colleagues  and Wu & Witkiewitz  found that individuals with social networks that were supportive of drinking prior to treatment had better outcomes if they had been assigned randomly to 12-Step facilitation treatment that encouraged AA participation, compared to those who were assigned to either cognitive–behavioral or motivation enhancement treatments. Other mutual help groups, such as Smart Recovery, are likely to be comparable to AA in their ability to foster healthy peer network support for abstinence.
The review by Hunter-Reel and colleagues  provided many examples of how researchers might address some of the knowledge gaps in the research on interpersonal influences of relapse. We would like to comment on one research technique, Markov modeling, which was mentioned by Hunter-Reel and colleagues , as well as one potential research design that was not addressed. A type of Markov model, associative latent transition analysis, was found recently to be very useful in studying the dynamic association between negative affect and alcohol lapses following treatment . Results indicated that changes in negative affect predicted changes in drinking, but changes in drinking influenced subsequent changes reciprocally in negative affect. Thus, individuals who experienced increases in negative affect were more likely to lapse (or transition from moderate to heavy drinking) and individuals who lapsed (or transitioned from moderate to heavy drinking) were also more likely to transition from lower to higher levels of negative effect. This modeling strategy could be particularly useful in studying whether changes in social networks (either in size, form or function) influence proximally subsequent behavior change and whether behavior change subsequently influences a change in social networks. Importantly, this research question requires multiple measures of social networks that are measured across multiple time-points. This type of analysis could provide very useful information about associated changes between social support and relapse; however, to truly test a cause and effect relation one must conduct an experiment in which social support is manipulated by the experimenter. On the surface this research question may seem untenable, given a variety of ethical reasons; however, an experimental study using an animal reinstatement model could provide much-needed insight. For example, an experiment that manipulated both social interaction and potential for drug relapse in animals that display behavioral and/or physiological reactions to social stressors (e.g. prairie voles, see ) might help to elucidate further the mechanisms of the association between interpersonal relationships and relapse.