Relapse Research: Predicting Treatment Outcome

The concept of “relapse” was borrowed from medicine, and has long been in popular usage to describe any recurrence of substance use after abstinence.   Over the years I increasingly found this term disturbing, in part because of its judgmental overtones in English, but also because it connotes an inaccurate all-or-none view of substance use (Miller, 1996).  This has been ameliorated a bit by the use of an intermediate concept like “slip” to indicate something in between abstinence and a “relapse.”

            My discomfort with the term gelled when, in a multisite study, we had to come up with a research definition of relapse in order to predict its occurrence (Lowman, Allen & Miller, 1996).  The sheer arbitrariness of any binary definition became obvious.  How long did someone have to be abstinent in order to qualify to have a relapse?  Is a single use the same as a relapse?  If not, how much use or how many days of use are required in order to constitute a relapse?  We had already done substantial follow-up tracking in studies of treatment for alcohol problems in order to predict outcome (e.g., Miller & Joyce, 1979; Miller et al., 1992), and knew that post-treatment drinking is highly variable.  There are indeed abrupt recurrences of heavy drinking, but also many shades of grey.  It is not uncommon for people who have been abstaining to have a drink or two and then think the better of it and return to abstinence (Miller et al., 1992).

            What I realized is that the term “relapse” connotes, at least in American context, the very all-or-none disease model that limited thinking in this field for so long.  To educate and warn people about “relapse” has implied that any use equals loss of control.   It was a strategy, as pointed out by Albert Bandura, to increase abstinence self-efficacy by teaching zero self-efficacy for self-control after the first drink.  Either you abstain or you drink out of control.  That is not how drinking outcomes actually look in treatment research (Westerberg et al., 1998), but it can become a self-fulfilling prophecy.  Marlatt’s “abstinence violation effect” – that after first use one has lost control and, in effect, has nothing left to lose – is ironically embedded in the very term “relapse.”  That is why I now seek to avoid using the term in my writing, and instead just describe the behavior.

            Having entered into the multisite Relapse Replication and Extension Project (RREP, Lowman et al., 1996), we did nevertheless come up with an arbitrary definition of a “slip” and “relapse” and sought to discover what client variables might predict the occurrence of post-treatment drinking.  We were specifically funded to evaluate a popular developmental model of relapse first developed by Allan Marlatt.

Here is what we found (Miller, Westerberg, Harris & Tonigan, 1996).  The first component of Marlatt’s model involves exposure to high-risk situations.  Here we found no relationship to drinking recurrence.  In essence, everyone was so exposed to high-risk situations that degree of exposure did not discriminate.   The next step of the model proved fruitful, however: the person’s level of skills for coping with high-risk situations.  Those who relied more on avoidant coping were more likely to relapse, whereas those with more active coping skills had lower risk.  After coping skills were taken into account, neither negative affect nor self-efficacy added any predictive value, but there was one more surprise.  As a measure of the abstinence violation effect we included a measure of endorsement of the American all-or-none disease model of alcoholism.  People who more strongly endorsed this view were also more likely to relapse – the self-fulfilling prophecy referred to above.  Endorsement of the disease model added significant predictive value even after everything else in the model had already been taken into account.

Terence Gorski had described another prospective model that became popular in addiction treatment, hypothesizing a long chain of events leading up to relapse.  Because this model was in popular use but lacked empirical evaluation, we developed a simple questionnaire with items corresponding to the 37 hypothetical warning signs in Gorski’s chain, and administered it at each 2-month follow-up interview to see whether it would predict recurrence of drinking at subsequent points.  To our surprise, it did (Miller & Harris, 2000).  Two items of the scale loaded in the opposite direction as predicted by Gorski.  The item “I am convinced that I will stay sober” was meant to mark Gorski’s early warning sign of overconfidence.  Instead, the confidence reflected in this item was associated with decreased risk of relapse, consistent with self-efficacy research.   Second, saying “No” to “I have many problems in my life” was meant to tap Gorski’s early warning sign of denial.  Instead, it was admitting to rather than denying problems that was associated with higher relapse risk.  Otherwise, the scale did successfully predict subsequent relapse at follow-up points 2 months later, even after taking into account current drinking status.

In a review of the relationship between childhood abuse history and substance use/problems (Simpson & Miller, 2002), studies indicated that it is not history of abuse per se so much as current PTSD that is associated with increased risk of substance use disorders.   Unfortunately, addiction treatment professionals are rarely trained in the exposure-based therapies that are most likely to help PTSD sufferers, and the usual forms of talk therapy or group counseling are unlikely to help or may exacerbate the problem.  The good news is that addiction treatment outcomes do not appear to be compromised by PTSD; that is, addiction treatment can succeed without having to first address PTSD.  Addiction counselors are thus best advised to leave PTSD treatment to those who are trained in exposure-based therapies, and focus on helping people get freed from substance use disorders, which can compromise the treatment of PTSD.

Finally, we developed a methodology to help with prediction of DWI recidivism (C’deBaca, Miller & Lapham, 2001).  Multiple and logistic regression formulas often do predict re-offending, but are of very limited usefulness to clinicians.  Examining each risk factor that was related to actual re-offense in a large sample, we established a cut score above which the risk of recidivism exceeded baseline.  We then used the number of risk factors over cut score (0-5) to predict recidivism, and found that this approach did as well in a hold-out sample as the logistic regression formula itself.  This is a practical method that is more likely to be useful to practitioners.

Publications on Relapse and Predicting Treatment Outcome (in chronological order)

Miller, W. R., & Joyce, M. A. (1979).  Prediction of abstinence, controlled drinking, and heavy drinking outcomes following behavioral self-control training.  Journal ofConsulting and Clinical Psychology, 47, 773-775.

Miller, W. R. (1980).  Maintenance of therapeutic change: A usable evaluation design.  Professional Psychology, 11, 660-663.

Ogborne, A. C., Annis, H. M., & Miller, W. R. (1982).  Discriminant analysis and the selection of patients for controlled drinking programs: A methodological note.  Journal of Clinical Psychology, 38, 213-216.

Miller, W. R., Leckman, A. L., Delaney, H. D., & Tinkcom, M. (1992).  Long-term follow-up of behavioral self-control training.  Journal of Studies on Alcohol, 53, 249-261.

Lowman, C., Allen, J., & Miller, W. R. (Eds.) (1996).  Perspectives on precipitants of relapse.  Monograph supplement to Addiction, Volume 91.

Connors, G. J., Longabaugh, R., & Miller, W. R. (1996).  Looking forward and back to relapse: Implications for research and practice.  Addiction, 91 (Supplement), S191-S196..

Miller, W. R. (1996).  What is a relapse?  Fifty ways to leave the wagon.  Addiction, 91 (Supplement), S15-S27.

Miller, W. R., & Marlatt, G. A. (1996).  Relapse Interview.  Addiction, 91 (Supplement), S231-S240.

Miller, W. R., Westerberg, V. S., Harris, R. J., & Tonigan, J. S. (1996).  What predicts relapse?  Prospective testing of antecedent models.  Addiction, 91 (Supplement), S155-S171.

Westerberg, V. S., Miller, W. R., Harris, R. J., & Tonigan, J. S.  (1998).  The topography of relapse in clinical samples.  Addictive Behaviors, 23, 325-337.

Miller, W. R., & Harris, R. J. (2000).  A simple scale of Gorski=s warning signs for relapse.  Journal of Studies on Alcohol, 61, 759-765.

C=de Baca, J., Miller, W. R., & Lapham, S. (2001).  A multiple risk factor approach to predicting DWI recidivism. Journal of Substance Abuse Treatment, 21, 207-215.

Simpson, T. L., & Miller, W. R. (2002).  Concomitance between childhood sexual and physical abuse and substance abuse: A review.  Clinical Psychology Review, 22, 27-77.