Dropout from individual psychotherapy for major depression: A meta-analysis of randomized clinical trials

Highlights

• Meta-analysis on dropout from RCTs on individual outpatient therapy for depression.

• We used random-effects meta-analysis for treatment- and study-level dropout rates.

• Average dropout rate was 17.5% for psychotherapy conditions, 19.9% overall.

• Higher dropout for longer treatments, more diverse or personality disordered samples.

• Study-level dropout was higher if study included inactive control condition.

Abstract

Dropout from mental health treatment poses a substantial problem, but rates vary substantially across studies and diagnoses. Focused reviews are needed to provide more detailed estimates for specific areas of research. Randomized clinical trials involving individual psychotherapy for unipolar depression are ubiquitous and important, but empirical data on average dropout rates from these studies is lacking. We conducted a random-effects meta-analysis of 54 such studies (N = 5852) including 80 psychotherapy conditions, and evaluated a number of predictors of treatment- and study-level dropout rates. Our overall weighted dropout estimates were 19.9% at the study level, and 17.5% for psychotherapy conditions specifically. Therapy orientation did not significantly account for variance in dropout estimates, but estimates were significantly higher in psychotherapy conditions with more patients of minority racial status or with comorbid personality disorders. Treatment duration was also positively associated with dropout rates at trend level. Studies with an inactive control comparison had higher dropout rates than those without such a condition. Limitations include the inability to test certain potential predictors (e.g., socioeconomic status) due to infrequent reporting. Overall, our findings suggest the need to consider how specific patient and study characteristics may influence dropout rates in clinical research on individual therapy for depression.

Introduction

Premature discontinuation of psychotherapeutic treatment is a commonly occurring phenomenon that has a host of negative effects at all levels of patient care (Reis & Brown, 1999). Patients who drop out generally experience worse clinical outcomes (Klein, Stone, Hicks, & Pritchard, 2003), and their service providers may lose revenue, feel demoralized, or think their efforts have been wasted (Barrett et al., 2008, Sledge et al., 1990). At the societal or systems level, the problem of dropout affects the allocation of clinical resources and contributes to the extended burden of untreated mental illness (Barrett et al., 2008, Reis and Brown, 1999). Efforts to investigate and potentially ameliorate dropout have varied in their scope and scale, ranging from qualitative interviews with dropouts (e.g., Wells, Palinkas, Qiu, & Ell, 2011) to epidemiological studies of treatment utilization patterns (Olfson et al., 2009, Wang, 2007). Estimates of dropout vary substantially across populations, disorders, interventions, and treatment contexts. As such, comprehensive reviews (e.g., Barrett et al., 2008) and meta-analyses of this literature have provided valuable insights into average dropout rates, as well as various study characteristics that moderate these estimates.

In an early meta-analysis of psychotherapy dropout, Wierzbicki and Pekarik (1993) reviewed 125 studies published between 1974 and 1990 that represented a wide variety of treatments, diagnoses, and study designs. This early meta-analysis reported an unweighted mean estimate of dropout of 46.9% (SD = 22.3). The authors tested study, therapist, demographic, and psychological variables as predictors of dropout, and they reported that dropouts were more likely to be of minority ethnicity, have lower socioeconomic status (SES), and were less likely to be educated. The method of defining dropout was also associated with its occurrence, with lower rates observed in studies that used failure to attend a session as their criterion. Unfortunately, meta-analyses were not widely used or well-developed when this study was published, leading to substantial statistical and methodological problems. Swift and Greenberg (2012) highlighted these issues in the introduction to their updated meta-analytic review of adult psychotherapy dropout, noting how paradigm changes in the provision of care, standards of reporting in clinical research, and meta-analytic research techniques in the past three decades might substantially influence estimates of dropout. Their updated review included 669 studies and more than eighty thousand patients. Dropout was estimated at 19.7% (95% CI [18.7%, 20.7%]) across a wide range of diagnoses, interventions, and treatments. There was substantial variability in dropout rates across studies (range: 0% to 74%) and several client, provider, and study variables predicted variability in these rates. Higher rates of dropout were predicted by treatment setting (higher in university-based clinics), whether the treatment was time-limited (higher in unlimited duration), the focal diagnosis of the client (higher for personality and eating disorders), experience level of the provider (higher for trainees), and the method of defining dropout (higher for therapist judgment). In separate meta-regression analyses from a subset of studies reporting characteristics by dropout status, younger age and less education were also found to be associated with dropout.

In summarizing their findings, Swift and Greenberg (2012) advocated for focused reviews on specific subject areas (e.g., by patient diagnosis or study type) to help provide greater clarity to researchers interested in these domains. In line with this recommendation, they then conducted a follow-up, targeted meta-analysis of disorder-by-treatment comparisons in dropout rates (Swift & Greenberg, 2014), finding limited evidence of differences across modalities except for eating disorders, depression and post-traumatic stress disorder (PTSD). Contemporary meta-analyses of dropout have emphasized specific diagnoses (e.g., PTSD; Imel, Laska, Jakupcak, & Simpson, 2013) or treatments (e.g., dialectical behavior therapy; Kliem, Kroger, & Kosfelder, 2010). Meta-analyses can also be targeted at transdiagnostic psychotherapy processes (e.g., therapeutic alliance; Sharf, Primavera, & Diener, 2011) or particular study design characteristics (e.g., non-randomized effectiveness trials; Hans & Hiller, 2013). Indeed, such targeted reviews may provide critical information on specific subjects that might be obscured in a more comprehensive review. For instance, a recent review of computer-based treatments for depression reported differences in dropout of 30 to 40% on the basis of the kind of support provided to participants, a variable likely to have been overlooked in a meta-analysis of broader scale (Richards & Richardson, 2012).

The past 20 years of research in clinical psychology has seen a tremendous increase in the use of randomized clinical trials (RCTs) and their role in meta-analytic reviews of treatment efficacy. RCTs of adult unipolar major depression are among the most ubiquitous, with hundreds of such trials published since the mid-20th century, reflecting a wide variety of interventions and comparisons. The treatment of depression remains an ongoing important area of clinical treatment research. A primary reason for this is that despite empirical evidence of moderate efficacy of pharmaceutical and psychological treatments, many patients fail to respond to treatment or experience a recurrence of symptoms (Cox et al., 2014). As in general clinical practice, dropout represents a major barrier to the achievement of successful treatment outcomes in RCTs. However, it also poses unique challenges in this research context, as patients who fail to complete study protocols can affect statistical analyses, outcomes, and interpretation of results (Coffman et al., 2007, Lane, 2008, Yang and Maxwell, 2013). Some features common to RCTs, such as use of manualized interventions, fixed duration, and specific eligibility criteria, have the potential to affect dropout rates versus typical clinical care. Furthermore, the structured nature of RCTs often helps provide more clear and objective guidelines for defining dropout, which has been shown to influence estimated rates (e.g., Swift and Greenberg, 2012, Wierzbicki and Pekarik, 1993). As some authors have noted (Hans and Hiller, 2013, Schottenbauer et al., 2008), these factors suggest the need to consider dropout from RCTs separately from clients in community treatment or even effectiveness studies.

Surprisingly, despite the ubiquity of RCTs for adult unipolar depression, there is no meta-analytic review of dropout from psychotherapy treatments in this context specifically. An empirically derived estimate of typical frequency of dropout would be useful to investigators conducting or developing RCTs in this area of psychotherapy research. Information about study or treatment characteristics that predict dropout may help inform decisions related to study design and provide context to observed dropout rates. Accordingly, we conducted a targeted meta-analytic review of dropout in these studies, including an investigation of potential predictors of dropout. In recognition of the potential value of adopting a more focused approach, we restricted our analyses to individual, outpatient treatments for major depressive disorder. As RCTs vary with respect to the conditions being tested (e.g., psychotherapy versus medication, psychotherapy versus psychotherapy), evaluating predictors of dropout at the study level may prove imprecise, as dropout rates may differ between psychotherapy and non-psychotherapy conditions (for example, see Dimidjian et al., 2006). We therefore coded dropout rates and potential predictors at both the study level (using overall dropout rate) and treatment level (using dropout associated with a particular psychotherapy arm of an RCT), with this latter group restricted to active psychotherapy interventions. We selected predictor variables that were used in prior reviews or highlighted as being potentially important predictors of dropout (e.g., Swift & Greenberg, 2012), including characteristics of the patient, provider, treatment, and study.