Is there a sleeper effect of exposure-based vs. cognitive-only intervention for anxiety disorders? A longitudinal multilevel meta-analysis

Highlights

• No evidence to support a sleeper effect in psychotherapy for anxiety disorders

• No evidence to support that CT's effect on anxiety was more enduring than ET's or vice versa.

• Results favor a uniform efficacy hypothesis regarding ET and CT's enduring efficacy in time.

Abstract

There is a longstanding debate in the cognitive behavioral literature whether exposure-based methods produce more sustainable outcomes relative to cognitive methods or vice versa. This debate concerns particularly the time after treatment termination (at follow-up assessments), also referred to as the sleeper effect. Therefore, the aim of the current meta-analysis was to examine the enduring efficacy of Exposure Therapy (ET) in comparison to Cognitive Therapy (CT) from treatment termination to follow-up in anxiety disorders. Available literature also allowed for the assessment of their long-term additive benefits relative to ET only. Traditional random effects analyses with restricted maximum likelihood estimators and multilevel longitudinal analyses were conducted on 39 randomized controlled trials (N = 1878). Traditional analyses revealed no differential efficacy at post-treatment or follow-up. Similarly, the multilevel longitudinal analyses identified no differential growth in efficacy from treatment termination to follow-up. The majority of the variables investigated did not moderate the results. However, there was evidence suggesting that CT was superior to ET when treatment was delivered individually, while ET was superior to CT when delivered as group therapy. Overall, the findings did not validate a number of assumptions, such as the existence of a sleeper effect. Several strengths and limitations are further discussed in the paper.

Introduction

Anxiety disorders are among the most common mental health problems. The current global prevalence of anxiety disorders ranges from 5.3% in African cultures to 10.4% in Anglo-European cultures (Baxter, Scott, Vos, & Whiteford, 2013). In other words, 1 in 14 people worldwide meet the diagnostic criteria for an anxiety disorder at one point in life (Baxter et al., 2013). Moreover, anxiety disorders are characterized by a considerable rate of relapse, as defined by a return of anxiety symptoms after treatment termination, with estimates of relapse ranging from 19% to 62% (Craske & Mystkowski, 2006; Fava et al., 2001; Newman, Llera, Erickson, Przeworski, & Castonguay, 2013).

It has been argued that considering the relapse rate of anxiety disorders proposed treatments should ideally have lasting effects not limited to post-treatment measurements (Gibby, Casline, & Ginsburg, 2017). There is an ongoing debate about whether particular psychotherapy interventions produce more sustainable outcomes in the months or years following therapy as opposed to other psychotherapies (e.g., Flückiger, Del Re, Munder, Heer, & Wampold, 2014; Flückiger, Del Re, & Wampold, 2015; Kivlighan III et al., 2015). This growth in effect from treatment termination to follow-up is referred to as the sleeper effect (Bell, Marcus, & Goodlad, 2013; Flückiger et al., 2015); and it remains relatively understudied in the field of psychotherapy research (Flückiger & Del Re, 2017; Wampold & Imel, 2015). The sleeper effect posits that symptom alleviation as a result of intervention may require additional time to materialize.

The sleeper effect focus may have in part contributed to the split within the psychotherapy community where some treatment techniques are believed to ensure more benefits in time relative to others (e.g., Shedler, 2010; Stopa & Clark, 1993; Tolin, 2010); such is the case of Cognitive-Behavioral Therapy (CBT). In CBT, the standalone treatment success of Exposure therapy (ET) and Cognitive therapy (CT) - two of the most widely used complementary CBT methods - raises the issue of whether (a) challenging maladaptive thoughts rather than (b) using standalone behavioral interventions ensures a more durable relief from anxiety symptoms (Hofmann, 2008; Kazantzis et al., 2018; Longmore & Worrell, 2007). This issue is all the more important as it is central to any further inquiry into specific mediators of CBT effects.

Currently, there are only two other notable traditional meta-analyses that investigated the relative efficacy of ET against CT in the treatment of anxiety disorders (Norton & Price, 2007; Ougrin, 2011). Both meta-analyses generally concluded that there appears to be no evidence of differential efficacy between CT and ET with regard to the treatment of anxiety disorders with the exception of social phobia where patients benefited more from CT. Aside from reaching common conclusions, these meta-analyses also shared common limitations that are enlisted below.

First, prior works were traditional meta-analyses pooling effect sizes separately for post-treatment and follow-up. However, change effects in time, such as the sleeper effect, can only be tested through a longitudinal meta-analysis that can effectively investigate treatment by time interactions as argued by (Flückiger & Del Re, 2017; Maas, Hox, & Lensvelt-Mulders, 2004). In this respect, there are only a handful of longitudinal meta-analyses of this kind in psychotherapy research (Flückiger et al., 2014, Flückiger et al., 2015; Flückiger, Del Re, Wampold, Symonds, & Horvath, 2012; Kivlighan III et al., 2015). In a nutshell, the meta-analysis by Kivlighan III et al. (2015) compared psychodynamic treatment with a broad range of alternative treatments in regard to anxiety and depression; while Flückiger et al. (2014) used longitudinal analyses to investigate the broad spectrum of evidence-based psychotherapies vs. treatment as usual in regard to anxiety and depression. The other two longitudinal meta-analyses by Flückiger et al., 2012, Flückiger et al., 2015 focused on the relationship between alliance and outcome efficacy across a wide spectrum of treatments, respectively, compared psychotherapies with and without additional components in terms of long-term efficacy. In contrast to prior work that focused on the generalizability of results across various psychotherapy trials and frameworks, the current meta-analysis is specifically aimed at investigating the enduring efficacy of CT and ET for anxiety disorders. This issue should be addressed given the still standing controversies within the CBT framework (e.g., Hofmann, 2008; Kazantzis et al., 2018; Longmore & Worrell, 2007).

Second, although some of the studies investigated combined treatments (i.e., CT plus ET) relative to exposure only, none of the previous meta-analyses explored these potential additive effects at treatment termination or in time. A combined treatment approach would be all the more important as in the CBT literature it is assumed that both behavioral and cognitive components are necessary for an enduring treatment effect and that using both methods would provide better symptom relieve relative to a standalone cognitive or behavioral treatment. However, this idea requires additional support as there are currently no meta-analyses investigating the additive advantage of a combined treatment relative to standalone cognitive or behavioral therapy (e.g., Norton & Price, 2007; Ougrin, 2011).

Third, heterogeneity was assessed only in the Ougrin (2011) meta-analysis and, although moderate to high heterogeneity was found in some cases, the sources of heterogeneity were not explored by subsequent subgroup analyses. A thorough assessment of heterogeneity is required to pinpoint potential moderator variables. Moreover, both meta-analyses require updating since they were published more than six years ago.

Based on the shortage of studies in the literature, we argue that longitudinal meta-analyses that focus on growth of effect sizes from post-treatment to follow-up are long due as many competing approaches claim maximized effects over time with limited evidence supporting these claims (e.g. Flückiger et al., 2014; Flückiger & Del Re, 2017; Kivlighan III et al., 2015).

In order to address the limitations of the previous literature, we aim to conduct several traditional and multilevel longitudinal meta-analyses: (a) A traditional random effects meta-analysis was employed in order to update prior work in light of more recent publications with regard to the relative efficacy of ET relative to CT at separate points in time (i.e., post-treatment and follow-up). (b) Due to the limited research properly examining enduring treatment efficacy, a multilevel longitudinal meta-analysis examined the sustainability of the effect of CT relative to ET from treatment termination to follow-up. (c) In light of the available literature, we were also able to investigate whether a combined treatment format (ET + CT) provided an additive advantage in time relative to standalone ET following treatment termination. For all these purposes, we considered anxiety-specific measures as outcomes (Wampold et al., 2011).

The study also examined a number of 7 variables that we expected to have a moderating effect on the results, as listed below. Their selection was guided by both theoretical considerations and previous research.

Therapist and researcher allegiance refers to the therapist's and, respectively, the researcher's belief that one treatment is preferable over another (e.g., Gaffan, Tsaousis, & Kemp-Wheeler, 1995; Luborsky et al., 1999; Munder, Brütsch, Leonhart, Gerger, & Barth, 2013; Wampold & Imel, 2015). Allegiance might be assessed through the therapist's and/or researcher's advocacy of the preferred therapy at the expense of others or even through the therapist's familiarity with a certain type of therapy. Additionally, direct contribution to the development of a certain form of therapy or to its empirical testing can increase the likelihood of allegiance bias (Falkenström Allegiance Control for Therapists measure; FACT; Falkenström, Markowitz, Jonker, Philips, & Holmqvist, 2013). The two concepts are different, albeit interrelated: while therapist allegiance might manifest itself through treatment delivery and implementation, researcher allegiance may go as far as to affect study design, preferential reporting, and the interpretation of the results. Both therapist and researcher allegiance have been shown to affect the outcomes of clinical trials in the sense that the lack of proper control for allegiance may result in artificially increased effect sizes (Falkenström et al., 2013; Luborsky et al., 1999; Munder, Flückiger, Gerger, Wampold, & Barth, 2012; Munder et al., 2012). Not accounting for allegiance effects might compromise the validity of the conclusions drawn in comparative psychotherapy studies (Munder et al., 2013; Spielmans & Flückiger, 2018; Wampold & Imel, 2015).

The treatment delivery format, namely individual or group sessions, might be a potential moderator based on previous literature. For example, in the case of generalized anxiety disorder or obsessive-compulsive disorder, meta-analytic findings suggest that individual therapies outperform group therapies (Covin, Ouimet, Seeds, & Dozois, 2008; Eddy, Dutra, Bradley, & Westen, 2004), while in the case of social anxiety no such moderation effect was found (Acarturk, Cuijpers, Van Straten, & De Graaf, 2009). Hence, given that there is no consensus over the optimal treatment delivery format in the treatment of anxiety, we decided to include this categorical variable in the current meta-analysis.

The type of anxiety disorder is another relevant variable that may impact the results of clinical trials and meta-analyses. From a theoretical standpoint, each disorder presents different symptomatology and, thus, may be differentially impacted by therapy. This theoretical perspective also becomes apparent in the empirical literature. For example, certain trials hinted at the superiority of CT over ET and/or other behavioral techniques in the treatment of social anxiety (Clark et al., 2006; Hofmann, 2004), while for other anxiety disorders (e.g. specific phobia), meta-analyses as well as clinical guidelines highlight exposure as the treatment of choice (Moriana, Gálvez-Lara, & Corpas, 2017; Wolitzky-Taylor, Horowitz, Powers, & Telch, 2008). Hence, the type of anxiety disorder might be a relevant moderator for the purposes of the current meta-analysis.

Anxiety disorders record a high and well known comorbidity rate both with depression and other anxiety disorders. Indeed, a report from the Netherlands Study of Depression and Anxiety (Klein Hofmeijer-Sevink et al., 2012). The presence of comorbid disorders has been known to affect outcomes of clinical trials for specific anxiety disorders (e.g., Olatunji, Cisler, & Tolin, 2010), which is why most trials tend to control for the presence of comorbidities, either by means of exclusion criteria or otherwise. For the scope of this meta-analysis, the presence of comorbid conditions alongside the primary anxiety diagnostic will be taken into consideration as a dichotomous moderator (i.e., present or absent).

Treatment dose mainly refers to whether the number of sessions was matched or not across treatment conditions. The required dose for efficient psychotherapy interventions is subject to ample debates (e.g. Baldwin, Berkeljon, Atkins, Olsen, & Nielsen, 2009). Some meta-analyses identified a positive dose-response relationship, while others found no such connection regarding the treatment of anxiety (e.g., Norton & Price, 2007). In light of these inconclusive results, most comparative efficacy meta-analyses account for this potential moderator in their analyses. As per recommendations by Spielmans and Flückiger (2018), the number of psychotherapy sessions will be treated in the current research as a continuous moderator.

The year of publication might also be a relevant moderator for the current meta-analysis in light of the changing reporting standards of individual trials potentially impacting the efficacy of remission rates (Johnsen & Friborg, 2015) or longitudinal efficacy more generally (Flückiger et al., 2014). With the addition of novel reporting standards, such as intent-to-treat samples, more recent trials might have higher quality reporting standards which may differentially impact in comparison with older trials the end results of the meta-analysis. As the studies included span over a larger period of time (1988–2015), the year of publication is a variable that needs to be considered in the current meta-analysis.

Based on the above-mentioned literature, we expect that (a) the traditional analyses will render no differential results in efficacy between ET and CT at post-assessment and follow-up similar to prior meta-analytic evidence and that (b) multilevel longitudinal analyses will further extend these results in the sense that there will be no differences regarding the enduring effect in time and no evidence favoring a sleeper effect.