This study sought to find mediating mechanisms of a previously demonstrated positive treatment effect for a telephone-based continuing care intervention (McKay et al., 2004
). Evidence of longitudinal mediation was found with three variables: self-help behaviors during the last month of treatment (i.e., assessed at 3-months), and self-efficacy and commitment to abstinence assessed 3 months later (i.e., 6-months). The telephone condition sustained higher levels of self-help involvement during continuing care than standard group counseling, and a longitudinal model indicated that this self-help involvement predicted better abstinence outcomes across the 4 to 24-month follow-up. A significant mediation effect was supported by the test of joint significance. Only marginal support however, was shown for 3-month self-help behaviors mediating treatment outcomes for 4 to 24 months using the asymmetrical confidence limit test and normal theory method.
Similarly, the telephone condition led to higher self-efficacy scores and higher rates of commitment to abstinence at the 6-month follow-up, compared to standard group counseling, and scores on these two measures predicted abstinence outcomes over months 7 to 24. Moreover, self-efficacy and commitment to abstinence at 6 months both passed the test of joint significance, the normal theory test, and the asymmetrical confidence interval test, indicating that they mediated the treatment effect on abstinence outcomes during months 7 to 24.
Since significant treatment differences between TEL and STND care were evident as early as the 6-month time-line-follow-back assessment (i.e., 4 to 6 months) we suspected that the mediating effects of self-efficacy and commitment to abstinence were part of a complex sequence of events and processes. We tested this post hoc explanation in supplementary analyses and found support for one of the two mechanisms examined. Self-help behaviors assessed at 3-months were associated with 6-month self-efficacy scores, but not 6-month commitment to abstinence scores. Perhaps this was due to the method of measuring commitment to abstinence (i.e., a dichotomous measure as opposed to continuous). Nevertheless, the mediation effect of commitment to abstinence should be interpreted with caution. Although there is presumably a chain of events occurring, it is difficult to determine the nature of the causal effects. It is possible that early outcomes were driving commitment to abstinence as opposed to the other way around. However, it should be noted that all the mediation models controlled for abstinence either during continuing care (i.e., for the 3-month mediator models) or during continuing care and the subsequent 3 months (i.e., for the 6-month mediator models). This strengthens the likelihood that mediation effects are in fact present.
Overall, these results suggest that initially, the greater therapeutic effect of the telephone condition compared to the standard group counseling approach is partially accounted for by a differential change in behavior. Namely, telephone participants indicated more involvement in self-help meetings and related activities during the period of the intervention than their standard care counterparts. After the continuing care intervention ended, a difference also emerged on self-efficacy, once again favoring the telephone condition over standard care. Therefore, the treatment first yielded differences in behavior, which were then followed by differences in efficacy beliefs. These results are generally in agreement with models in the addictions (Marlattt & Gordon, 1985
) and other areas (Bandura, 1997
), in which successful coping (i.e., behavior change) is thought to produce increases in self-efficacy, which in turn influences subsequent outcomes. Therefore, our data imply the presence of sequential effects with self-help and self-efficacy. Determining if sequential effects are also present with commitment to abstinence will require further study.
Perhaps the most notable finding of the study is that the telephone-based continuing care, which provided approximately half the minutes of therapeutic contact as the other two conditions (McKay et al., 2005
) and considerably less face-to-face contact (average of 5 versus 14 sessions) produced higher scores on three mediators (e.g., self-help behaviors, self-efficacy, and commitment to abstinence) than the comparison condition. This could be in part due to the content of the TEL intervention, which placed greater emphasis on the need for patients to be the active change agents in their recoveries. By comparison, patients in the STND condition may have relied more on the therapist and other group members for making progress in recovery. Because the patient who received telephone continuing care may have more strongly attributed the changes in behavior to his or her own efforts, the patient's self-efficacy and commitment to abstinence were maintained, and in fact improved somewhat. It should be noted that this effect may not have been related to the content of the telephone intervention, per se, but rather to the fact that this more minimal intervention might have prompted recipients to do more to support their recoveries, including participating more actively in self-help programs.
The present findings were equivocal with respect to the effects of the process measures (i.e., the putative mediators) on substance use outcomes. At the 3-month assessment, only commitment to abstinence and self-help behaviors predicted subsequent abstinence outcomes, whereas at 6 months, all putative mediators except self-help beliefs predicted outcome. These findings are at least in part contradictory to previous research where post-treatment measures of self-help beliefs, self-efficacy, and social support predicted treatment outcomes (Connors et al., 1996b
; Hall et al., 1991
; Longabaugh et al., 1998
; McKay et al., 2001
; Miller et al., 1996
; Morgenstern et al., 1997
). However, the fact that our models included current substance use at the time the mediator was assessed likely reduced the predictive power of these process measures.
With regard to social support, we had hypothesized that the relative lack of direct social support provided in the TEL intervention, compared to that provided through twice-weekly group counseling, would lead patients who received TEL to more actively seek out other sources of social support. Although this occurred to some extent during the treatment intervention, the difference in scores was not statistically significant from those in the group counseling condition. The failure to generate support for this hypothesis could indicate a failure of action theory with respect to the effect of the telephone intervention on this construct, or limitations in the measure of social support. The Procidano and Heller scale (1983)
measures perceived general support, but it does not provide a more objective measure of actual social support or an indication of the amount of abstinence-oriented social support provided by peers. There is some evidence that measures which assess social support specifically for abstinence, such as Longabaugh, Wirtz, Beattie, Noel, & Stout's Important Peoples and Activities Inventory (1995)
are better predictors of outcome than more general social support measures (Beattie & Longabaugh, 1999
). Further work with better measures is needed to more fully understand the possible impact of the TEL condition on social support.
Limitations and Concluding Remarks
The mediation methods applied in this study have been tested most thoroughly with continuous measures. We used a binary outcome, and one of our mediators (commitment to abstinence) was binary as well. Further simulation studies are needed to determine the range of validity of the methods when variables with distributions from other exponential families are included as mediators or outcomes, particularly in the longitudinal mediation setting. This paper provides an initial step in the process of extending these more advanced models by using three methods to test the mediation models. Notably, while demonstrating rigor on the one hand, it also introduced complexity and underscored the differential levels of power to detect significant effects among the various methods. Hence, a significant mediation effect for 3-month self-help behaviors was obtained with the test of joint significance, while more marginal results were found for the mediation effect of 3-month self-help behaviors using the asymmetrical confidence limit test and the normal theory approach.
In a recent publication, Kazdin and Nock (2003)
recommended assessing values on putative mediators early and often in treatment. This approach is particularly applicable to studies of treatment for disorders such as depression, anxiety, or conduct problems, in which patients begin treatment with high levels of problem severity, and improve over the course of treatment. In the addictions however, many patients stop using prior to entering treatment, or in the case of our study, during the phase of treatment that precedes continuing care. Therefore, it is not generally possible to show that improvements in mediators precede improvements in symptoms. Nevertheless, our design would have been stronger if we had also assessed scores on the mediators at repeated points during the first six months of the follow-up (e.g., monthly or perhaps even weekly), in order to demonstrate more conclusively that declining scores on the mediators preceded any return to substance use during continuing care.
Another limitation involves the generalizibility of the results to a more heterogeneous population. Our participants were mostly middle-aged, unmarried, African American men with long histories of substance abuse. Future studies should examine the TEL intervention with a different population in order to see if the therapeutic processes operate in a similar fashion.
Missing data on the mediator variables and covariates also produced sample sizes that were slightly smaller than for the general outcomes results. Follow-up rates on these variables were not as strong due to the priority given to gathering the substance use data in interviews where phone contacts, which were typically more limited in time, were utilized for data collection. Nevertheless 74-82% follow-up rates (the rates our final longitudinal models yielded once all covariates were included) over a period of two years is standard in community-based substance use research. As with any study of this kind, loss to follow-up is a notable limitation. It is possible we would have seen different results without the missing data.
Finally, it is difficult to judge the clinical significance of the mediation effects. However, as shown in the raw data presented in , the fact that the telephone condition produced increases in self-efficacy (during the intervention) and commitment to abstinence (immediately following the intervention), whereas there were decreases on these measures in standard care, may be salient to patients and therefore important clinically.
In conclusion, our results suggest that the greater therapeutic effects of telephone-based continuing care are partially accounted for by participation in self-help meetings and related activities during the continuing care phase of treatment, and by subsequent increases in commitment to abstinence and the maintenance of self-efficacy. Furthermore, we were able to establish a chain of events to help elucidate the processes leading to the 6-month mediation models. Our results revealed that increases in self-help behaviors are associated with increases in self-efficacy, which accounted for the treatment differences from 7-24 months.
With more sophisticated analytic methods we are now able to fully utilize longitudinal datasets in order to answer questions about mechanisms of change over time. Since the 1986 publication of the widely adopted Baron and Kenny mediation model, understanding causal mechanisms became of growing interest and treatment processes a common theme in psychological research. Using several tests of mediation based on MacKinnon's and colleagues work (2002b
, the present paper was able to successfully demonstrate several longitudinal mediation models within the context of a logistic modeling framework. Further work is needed on study designs in addictions treatment research to facilitate more specific guidelines and criteria for mediation with complex analyses using binary outcomes as recommended by Kazdin and Nock (2003)