The Role of the Therapeutic Alliance in Psychotherapy and Pharmacotherapy Outcome: Findings in the National Institute of Mental Health Treatment of Depression Collaborative Research Program

The research design and methods of the TDCRP have been described in detail elsewhere (Elkin, Parloff, Hadley, & Autry, 1985) and will only be summarized briefly. At each of three research sites, patients were randomly assigned to one of four treatment conditions: interpersonal psychotherapy (IPT), cognitive behavior therapy (CBT), imipramine plus clinical management (IMI-CM), or placebo plus clinical management (PLA-CM). The therapies were carried out at George Washington University (Washington, DC), the University of Pittsburgh, and the University of Oklahoma (Oklahoma City).

This study is the first empirical investigation to compare and contrast levels of therapeutic alliance and their relationship to outcome not only in distinctly different types of psychotherapy but also in pharmacotherapy. The results showed that clinical raters assessed comparable levels of alliance across all of the treatments under investigation. Also worth noting is the high level of mean alliance and small standard deviations in alliance for each therapeutic approach. These uniformly high alliance scores, with means ranging between average and good for all therapies, probably reflect the care with which therapists were selected for the TDCRP and the quality of the training the therapists received. Alliance levels in less carefully monitored studies or in community-based treatments might be lower on average or cover a wider range.

The results also showed a significant relationship between total therapeutic alliance ratings and treatment outcome across modalities, with more of the variance in outcome attributed to alliance than to treatment method. There were virtually no significant treatment group differences in the relationship between therapeutic alliance and outcome in interpersonal psychotherapy, cognitive behavior therapy, and active and placebo pharmacotherapy with clinical management. The one Early Alliance × Treatment Method interaction effect using the BDI suggested a weaker relationship between early total alliance and clinical outcome for CBT than for the other treatment methods. Because this result was not replicated with the HRSD nor the mean total alliance and outcome, however, it may represent a chance finding.

The association between alliance and outcome was significant whether ratings of alliance were based solely on a single early session or on a mean of alliance across early, middle, and late sessions. This association and the magnitude of outcome variance explained were considerably stronger, however, when mean rather than early alliance alone was used. The inclusion of later sessions, which might reflect clinical response, may partially account for the stronger relationship between mean alliance and outcome. However, Gaston, Marmar, Gallagher, and Thompson (1991) demonstrated that substantial amounts of outcome variance are uniquely accounted for by alliance scores, over and above initial symptomatology and symptomatic change in behavior, cognitive, and brief dynamic psychotherapy. It should be noted that this study, like the TDCRP on which it was based, largely defined outcome in terms of symptom reduction. In contrast, some recent literature has begun to focus on a broader set of outcome domains.

Our findings were consistent with those of Marmar et al. (1989) and Salvio et al. (1992), who reported comparable levels of therapeutic alliance in different types of psychotherapy, including a placebo or minimally supportive condition. These findings support Thurer and Hursch’s (1981) supposition that, even within different approaches, the therapeutic relationship may be established in similar ways. Our findings regarding alliance levels conflict with those of Raue et al. (1993), however, who found significantly different levels of alliance in cognitive behavior and psychodynamic interpersonal sessions; they studied nonstandard therapies in naturalistic settings, unlike our standardized treatments within the context of a research protocol.

Our findings with regard to the relationship of alliance to outcome within different treatment approaches differ not only with those of Marmar et al. (1989), who found a significant relationship only for cognitive therapy, but also with our own pilot study, in which an association was found only for interpersonal therapy. The most likely explanation lies in differences in methodologies. Marmar et al. used patient and therapist ratings to assess the strength of the alliance, whereas we relied on independent observer ratings; observer ratings and patient and therapist reports have shown different relationships of alliance to outcome across a range of studies (e.g., Marziali, 1984). The findings in this article need to be understood in the context of this variability. The TDCRP pilot study (Krupnick et al., 1994) used only a subset of extreme outcome cases, whereas this main study used the full range of participants and outcomes and a much larger number of participants (N = 225), producing a more representative sample; thus, it is a better test of the hypotheses.

An advantage of our methodology over that of earlier alliance studies is that ratings were based on observations of full-session videotapes rather than brief segments of sessions, which has been the typical method. Results are influenced by the data that are available, and full sessions represent a more accurate picture of the alliance than more limited, and hence more arbitrary, sampling. Another distinction of this study was the raters’ training to recognize the features of common relational behaviors across distinctive treatments, despite very different therapist interventions and styles. Because measures that have been developed in the context of dynamic therapy may assess constructs that reflect primarily the specific techniques of that approach rather than generic therapist-patient relational behavior, our revision of the VTAS and its rating manual may have also contributed to a clearer test of the role of the relationship variable in other approaches.

Our findings were consistent with earlier reports (Frieswyk et al., 1986; Horowitz, Marmar, Weiss, DeWitt, & Rosenbaum, 1984; Marmar et al., 1989; Marziali et al., 1981) that found patient contribution to the therapeutic alliance to be a significant predictor of outcome. It is plausible that therapist contribution is also an important factor but the low reliability of the Therapist factor limited our ability to detect a potential association. An examination of variance components (Fleiss, 1986) helps explain the low interrater reliability for the Therapist factor. Similar measurement error rates were found for the Therapist and Patient factors, but there was a more constricted range of alliance ratings for the Therapist factor (true score variance was .09 for the Therapist factor and .26 for the Patient factor). The truncated range of scores, with very few below-average ratings of therapist contribution to the alliance, would make it statistically difficult to detect an effect even if one were present. This was probably because we used only experienced therapists who had extensive training and who met competency criteria and because we monitored their conduct of treatment throughout the study. It is possible that if there had been more variability among the therapists (e.g., inclusion of less skilled or less experienced therapists), an effect on outcome of therapist contribution to the alliance might have been detected. Another possibility is that the therapist contribution to the alliance may be salient only with a particular subset of patients (e.g., those who have the least capacity to form a therapeutic relationship) and that the effect would be diluted by pooling ratings with those of other cases for whom the therapist influence on outcome might be less important.

Perhaps of most interest was the finding of a strong association between alliance and outcome in pharmacotherapy, in both the imipramine and placebo conditions. What this finding suggests is that the therapeutic alliance may strongly influence the placebo response embedded in pharmacotherapy as a nonspecific factor above and beyond the specific pharmacologic action of the drug. The therapeutic alliance may help to create a “holding” environment in which the acceptance of taking a drug may be enhanced and permit concerns to be addressed and worked through within the context of a supportive and collaborative relationship (such as fears of dependence on medication, resistance, and demoralization regarding the delayed or variable effects of medication or placebo and the difficulty tolerating the discomforts of drug side effects). The variability of the doctor-patient relationship may account for the considerable variability in placebo and active drug response rates across studies of tricyclic antidepressants (Morris & Beck, 1974; Sotsky, Barnard, & Docherty, 1983). Thus, the role that the therapeutic alliance plays in affecting outcome extends not only beyond psychodynamic psychotherapy to cognitive behavior therapy, but also beyond psychotherapy itself, with implications for the way in which pharmacotherapy is conceptualized and practiced.

Our findings regarding the relationship between therapeutic alliance and outcome in pharmacotherapy appear particularly salient in light of the recent controversy over the proper settings and providers for the treatment of depression. (Munoz, Hollon, McGrath, Rehm, & VandenBos, 1994). According to the recently published Agency for Health Care Policy and Research (AHCPR) guidelines focused on depression in primary care settings, primary care physicians are encouraged to initiate treatment for their patients with depression in lieu of making a referral to a mental health specialist. As Munoz and his colleagues (1994) pointed out, treatment for depression by primary care physicians is likely to be pharmacological in nature (whether or not that is an appropriate treatment decision). Furthermore, the guidelines suggest that referral to a mental health specialist need not occur until after two unsuccessful antidepressant trials have been carried out by the primary care doctor. This recommendation may be questionable given the results of the present investigation, which provides empirical support for an idea initially asserted by Rickels (1968), that there is more to pharmacotherapy than simply choosing an appropriate medication and dosage level. Munoz et al. (1994) raised doubts as to whether primary care physicians have the requisite training and experience to adequately and effectively work through medication issues with depressed patients. They also note that, given changes that are occurring in the delivery of primary care services (e.g., increased use of large group practices and health maintenance organizations [HMO’s] in which patient information may be provided to an array of personnel), patients and primary care doctors are now less likely to have a close personal relationship than may have been the case in the past. At the very least, it seems that if primary care physicians will indeed be providing a good deal of pharmacological treatment for depression, research into the quality and effect of the therapeutic alliance between patients and physicians in primary care settings is warranted.

In summary, our findings showed that the level, pattern, and effect on clinical outcome of depression of the therapeutic alliance were comparable not only across theoretically and technically different psychotherapies, but also across active and placebo pharmacotherapy conditions. These results are most consistent with the view that the therapeutic alliance is a common relational factor across modalities of treatment for depression that is distinguishable from the specific technical or pharmacological factors within the treatments.

(Reprinted with permission from the Journal of Consulting and Clinical Psychology 1996; 64:532–539; © American Psychological Association; all rights reserved)