Augmentation Strategies for Treatment-Resistant Anxiety Disorders: A Systematic Review and Meta-Analysis

The search was done in duplicate (by Beth Patterson and Michael van Ameringen) and encompassed publications from January 1990 to January 2015. There were no restrictions on language. The following databases were searched: MEDLINE, EMBASE, PsycINFO, and the Cochrane Central Register of Controlled Trials. Search terms included combinations (and truncated terms) of the populations, interventions, and study design of the following: anxiety or anxious or panic or phobia or generalized anxiety disorder or GAD or panic disorder or social phobia or social anxiety or anxiety disorders and treatment or drugs or med or cognitive behave ther or CBT or therapy or resistance or treatment-resistant or drug-resistance or augmentation or refractory or treatment-refractory or antipsychotics or benzodiazepines or anticonvulsants or antidepressants or next-step and randomized controlled trial or controlled clinical trial or randomized controlled trials or double-blind or clinical trial or placebo or random or comparative study or evaluation studies or prospective studies or cross-over studies. To identify unpublished studies, ongoing and completed trials were located using the ClinicalTrials.gov (http://www.clinicaltrials.gov/). In addition, conference proceedings from the following organizations were hand-searched for abstracts presented at annual meetings over the past 5 years (Jan. 2010–Jan. 2015): Anxiety and Depression Association of America, American Psychiatric Association, European College of Neuropsychopharmacology, and the World Federation of Societies for Biological Psychiatry. The bibliographies/reference lists of all identified trials and relevant review papers were also reviewed.

Data extraction forms were completed in duplicate. The reviewers contacted all investigators by e-mail, if additional information was required to properly rate the article. Risk of bias of each study was evaluated using the Cochrane Collaboration risk of bias tool.^[11] Authors rated each study independently and the combined information was then entered into RevMan software for analysis.^[12] Dichotomous outcomes (response and dropouts due to adverse events) were analyzed by calculating risk ratios for each trial with the uncertainty in each result expressed using confidence intervals. For the analysis of change in symptom severity, where three different assessment tools were used, the standardized mean difference (SMD) was determined. This method of analysis standardizes the differences between the means of the treatment and control groups in terms of the variability observed in the trial.^[5] Mean differences (MD) were calculated for continuous summary data derived from the same scale (SDS). The outcomes were expressed in terms of an average effect size, as well as by 95% confidence intervals. Categorical and continuous treatment effects were determined using a random effects model (this includes both within-study sampling error and between-study variation in determining the precision of the confidence interval around the overall effect size).

All analyses of dichotomous data were intention-to-treat (ITT). When there were missing data, the first approach was to contact the authors for additional information. For studies missing standard deviations, we were able to calculate standard deviations using other data in the study results.

Heterogeneity of included studies was analyzed using RevMan software,^[12] which provided both a chi-squared test of heterogeneity (Cochrane’s Q statistic) and an I² statistic to estimate the degree that the result may be explained by chance alone and quantify the inconsistency of trial results within subgroup analyses.^[13] We anticipated a high percentage of heterogeneity as we pooled data of different disorders and different treatments. We hypothesized that much of the heterogeneity would be attributed to differences between specific anxiety disorders, and between specific augmentation strategies. To that end, we planned several a priori subgroup analyses to compare specific disorders and specific drug classes (considered a “class”) for each outcome measure. The current literature had lead us to hypothesize that GAD would have higher rates of response to augmentation strategies than SAD or panic disorder,^[5,14,15] and that subgroup analysis of drug class would reveal more robust effects for atypical antipsychotics and anticonvulsants than for other drug classes, including CBT.^[14]

The literature search identified 625 potentially relevant articles (Fig. 1), of which 610 were excluded (Fig. 1). The main reasons for exclusion were the following: population was not treatment resistant, the design was not RCT, review articles, and studies of OCD and PTSD. One study of topiramate augmentation in treatment-resistant SAD was ongoing.^[16] No studies of CBT augmentation of an SRI met inclusion criteria. The unweighted Kappa score for this stage of the review was 0.73; SE = 0.17, 95% CI = 0.39 to 1.0.

In total, six RCTs were included in the review (see Table 1) yielding a total of 557 randomized participants: three studies of GAD^[17–19] (n = 390), one study of SAD^[20] (n = 122), and two studies of panic disorder^[21,22] (n = 45). All studies were parallel design. All studies examined pharmacological augmentation; three studies used antipsychotics,^[17,18,22] two studies used benzodiazepines,^[20,21] and one study used the novel anticonvulsant pregabalin.^[19] One of the studies was unpublished: a research poster presented at the Anxiety and Depression Association of America’s 2012 annual conference.^[22] This study was included in this review and meta-analysis with written permission from the corresponding author.

Three trials began with an open-label phase where participants were treated with an antidepressant, with nonresponders randomized to augmentation treatment or placebo^[17,20,21]; the other three trials used a predefined definition to assess historical treatment resistance prior to randomization. The duration of RCT treatment was either 6,^[17] 8,^[18,19,22] or 12 weeks.^[20,21]

One study randomized SAD nonresponders to sertraline to three phases: augmentation with placebo, augmentation with clonazepam, and switch to the SNRI venlafaxine.^[20] Only the clonazepam and placebo arms of this study were included in the pooled analysis. Two studies of treatment-resistant SAD were initially included but had to be excluded from analysis due to insufficient information for outcome comparisons.^[23,24]

Seven studies were excluded from the review (Table 1): four studies of treatment-resistant GAD,^[25–28] one study of treatment-resistant SAD,^[29] and two studies of treatment-resistant panic disorder.^[30,31] The most common reason for exclusion was allowing participants to take different baseline drug classes (i.e., antidepressants, anticonvulsants, benzodiazepines, antipsychotics, etc.) as well as allowing multiple concomitant medications as long as the dose was stable.^[25,26,28] For example, Brawman-Mintzer et al.^[25] reported on a study using risperidone augmentation versus placebo in GAD (N = 40, positive study). Participants in this study were taking a wide range of base agents, many of which were at doses considered to be subtherapeutic and below typical starting doses, according to most treatment guidelines. These included alprazolam 0.25 or 0.5 mg/day, diazepam 2.5 mg/day, buspirone 5–20 mg/day, imipramine 75 mg/day, and citalopram 5 mg/day. Many participants had been at these doses for many years (some close to 10 years). Agents not considered treatments for anxiety such as bupropion were also base agents.^[25] Similarly, Pandina et al.^[26] reported on another risperidone augmentation RCT in GAD (N = 417—negative study). In this study, 45% of baseline agents were benzodiazepines. Agents not indicated for the treatment of GAD were also included as base agents such as bupropion and trazadone. Treatment resistance was defined; however, the duration of treatment with the baseline agent or information whether participants had reached a maximally tolerated dose was not explained. Also in the study, concomitant treatment with the cholinergic agent benztropine was allowed, as were hypnotics used as needed for insomnia. These agents could be considered confounders as they may have mitigated adverse events from the combination treatment and, in the case of hypnotics may have treated a common symptom of GAD. The conclusions of both studies noted that the heterogeneity of the baseline agents and concomitant agents limited the generalizability of their findings and were a significant limiting factor. A study by Lohoff et al.^[28] (N = 62, negative study) was excluded for similar reasons as participants were allowed to be on an SSRI, SNRI, benzodiazepine, or a combination of these agents. In addition, only a portion of the sample was augmented.^[28] In an RCT of quetiapine augmentation in GAD (N = 22, negative study), nonremitters to 10 weeks of open-label paroxetine CR were randomized into the augmentation phase of the study.^[27] Since this sample included individuals who may have responded to paroxetine, but did not achieve remission status, we decided to exclude this study. In the anxiety disorders literature, response is typically defined as a percentage change in baseline symptoms’ severity scale score, whereas remission is usually defined by a cut-score. Remitters should be virtually free of symptoms and represent a different population from those who have responded and not remitted. A study of open-label paroxetine augmentation in SAD (N = 21, negative study) was excluded as the paroxetine was added to all nonresponding participants of a double-blind trial of mirtazepine versus placebo.^[29] An RCT of pindolol augmentation (N = 25, positive study) was excluded as no primary outcome measure nor definition of response was specified.^[30] In addition, the treatment groups were not specified in the description of the study dropouts.^[30] Finally, an RCT of paroxetine augmentation in panic disorder (N = 161, positive study) was excluded because CBT was the base agent.^[31] We had specified in the inclusion criteria that all baseline agents had to be an SRI, the type of treatment most commonly used in the community. This was done in order to limit the heterogeneity of the pooled data.

Bias of RCTs was assessed in duplicate using the Cochrane Collaboration risk of bias tool.^[11] Any disagreements between raters (Beth Patterson and Michael van Ameringen) were resolved by discussion (Fig. 2). The weighted kappa was 0.86; SE = 0.0914, 95% CI = 0.69 to 1.0.

Only three studies included a measure of disability or functional impairment^[19,20,22] and all three used the SDS.^[10] Overall, a nonsignificant effect was determined. The authors queried whether this result was driven by the study by Simon et al.,^[21] where augmentation with CBT resulted in greater improvement in SDS score than did augmentation with clonazepam. However, a post hoc sensitivity analysis where the Simon et al.^[21] study was eliminated from the analysis did not support this hypothesis (SMD = –1.62, 95% CI = –3.86—0.62, I² = 76%). A priori subgroup analyses did not reveal any significant subgroup effects for this outcome.

All studies reported the number of dropouts due to adverse events; two studies had no dropouts.^[18,21] The pooled analysis of the four remaining studies revealed a nearly significant risk in favor of controls (P = 0.08). The RR for dropout due to adverse events with pharmacological augmentation as compared with controls was 2.02, 95% CI 0.91 to 4.46. No significant heterogeneity was found between studies (I² = 0).

No significant subgroup effects were found in any of the a priori subgroup analyses. In addition, no significant within group effects were found when this outcome was analyzed by drug class or disorder. The study duration subanalysis revealed a small significant risk in favor of controls for studies with a 6- to 8-week duration (RR = 2.64, 95% CI = 1.05–6.61).

These results should be interpreted cautiously as the sample was comprised of only one large study and several small studies, decreasing confidence in the estimated effects. However, all post hoc sensitivity analyses of response revealed a similar null difference in effect, which may be an indicator of the strength of the findings. In addition, despite combining different disorders and different augmentation agents, heterogeneity in the pooled sample was quite low, and no significant subgroup effects were determined. The findings are also potentially limited by the exclusion of two studies, which met entry criteria but lacked sufficient outcome data for inclusion. Seven studies were excluded from the analysis due to the rigorous inclusion criteria (Table 1), which would have added 748 participants to the meta-analysis. Interestingly, four of the seven studies were negative (n = 522); therefore, it is possible that even if entry criteria had been less stringent, the results would have been more strongly negative.

The overall quality of the evidence examining augmentation of a first-line pharmacological agent in treatment-resistant anxiety disorders would be considered modest according to quality assessment using GRADE (Grading of Recommendations Assessment, Development and Evaluation) criteria.^[33] This is mainly due to the risk of bias associated with methodological limitations in the largest study^[19]; namely, the early termination for benefit. This type of limitation may contribute to an overestimation of effect, particularly in a meta-analysis where the study carries substantial weight.^[32] However, a post hoc sensitivity analysis did not reveal an appreciable difference in effect. The authors rated each outcome of the meta-analysis based on risk of bias, inconsistency, indirectness, imprecision (random error), and other considerations such as publication bias. Given that only three to six studies were included in each outcome, publication bias was difficult to accurately assess by funnel plot. However, the sample included two unpublished studies, one of which was negative, which limited the suspicion of publication bias.

The main finding in this analysis (of limited benefit for augmentation) may be explained by factors related to study patient populations such as comorbidity, duration of illness, number of previous treatment trials, or severity of the anxiety disorder. Further subgroup analyses in a larger pooled estimate would be necessary to address these issues. Nevertheless, our findings differed from those reported in a previous meta-analysis of augmentation in treatment-resistant anxiety disorders,^[5] where treatment with any augmentation agent was associated with a significant risk of CGI-I response compared to placebo. However, considering that all nine studies included in the primary outcome analysis of that study were in OCD, and findings were similar to those reported in a recent meta-analysis of OCD,^[34] the difference in results between our study and the paper by Ipser et al.^[5] is not surprising. These differences in meta-analytic findings lend support to the organization of the DSM-5, where OCD was separated from the anxiety disorders into its own chapter, as the neurobiological underpinnings of anxiety disorders are likely quite different.

A high rate of placebo response was found in our study (64% for augmentation vs. 59% for placebo), and although this was mainly driven by the study by Rickels et al.^[19] (63% placebo response rate), the placebo response rates reported in the other studies included in the analysis were also relatively high: 52.5% (Pollack et al.)^[20] and 46% (Goddard et al.)^[22]. Looking at the broader treatment-resistant literature and excluding studies in this pooled analysis, placebo response rates using the CGI-I ≤2 range from 35% in SAD^[24] to 47% in panic disorder.^[31] High rates of placebo response have been long been a problem among the anxiety disorders, and appear to be higher in SAD, GAD, and panic disorder than they do in disorders such as OCD. Methodological enhancements such as the use of blind raters may lower the rates of placebo response currently seen in the anxiety disorders treatment-resistant literature. However, it is also possible that the use of the CGI-I scale erroneously inflated placebo response. Since it is a global measure, any improvement in patients’ comorbid conditions would also be reflected in the rating and may explain the magnitude of response in both treatment and placebo conditions. Perhaps an alternative measure of response, such as a cut score on a symptom severity scale as measure of response may have yielded different results. Alternatively, the high placebo response rates demonstrated in augmentation trials in this analysis may be attributed factors related to the natural course of pharmacotherapy. For example, it is possible that the response demonstrated in placebo groups was due to a longer duration of baseline antidepressant treatment and not due to the placebo augmenter.

Pharmacological augmentation did not demonstrate significant improvement in functional impairment in this analysis. This may be explained by a lag between symptom improvement, and improvements in functioning, as the duration of the included studies may have been too short to demonstrate improvements on the SDS. However, the negative findings in terms of response and functional impairment may also be interpreted as an issue of “refractoriness.” Perhaps patients who are refractory to first-line anxiety disorder treatments will also be refractory to subsequent treatment trials. In the prospective longitudinal STAR*D study, depressed patients who did not achieve initial response to citalopram demonstrated decreased rates of response with each successive treatment. In addition, there was little difference found between various next-step augmentation agents in the STAR*D study.^[35] Despite our predictions of a larger effect for atypical antipsychotics, particularly in GAD, we did not find any difference between agents. Atypical antipsychotics have demonstrated superior efficacy as augmenters in treatment-resistant GAD, OCD, and depression in previous meta-analyses.^[5,14] The studies using these agents in the present analysis were very small, however; and this likely contributed to the lack of findings for both the subanalysis by drug class and the subanalysis by disorder.

Where then do these results leave the field in terms of what to do with patients who do not achieve response to initial SRI treatment for anxiety disorders? At this stage, there appear to be more questions rather than answers. In contrast to the OCD and depression literature, the state of the treatment-resistant anxiety disorder literature is less evolved; therefore, the conclusions drawn from this analysis should be tentative. However, although the results of this meta-analysis were suggestive of a modest reduction in symptom severity, they do not support pharmacological augmentation as a next-step treatment in refractory anxiety patients in the key realms of treatment response and functional improvement. Despite this finding, without the support of adequate scientific evidence, at this stage, clinicians have little choice but to use second- or third-line treatment strategies when first-line treatments fail. Such strategies would accordingly include, but not be limited to augmentation with various agents and treatment modalities. Although no or partial response to first-line SRI treatments is common, there are many disparate treatments that are considered first-line monotherapeutic agents in the anxiety disorders, leaving clinicians with a number of options before augmentation need be considered. There is strong evidence to support monotherapy with pharmacological treatments such as serotonin noradrenalin reuptake inhibitors (SAD, GAD, panic disorder), pregabalin (SAD and GAD), benzodiazepines (panic disorder, GAD, SAD), or the atypical antipsychotic quetiapine (GAD) as well as CBT (GAD, SAD, panic disorder) in nonrefractory populations.^[36] The literature would greatly benefit from more algorithm-based longitudinal studies such as STAR*D, which would help clinicians understand how to sequence first-line treatments prior to moving onto second- or third-line strategies. The findings of this study highlight the great need for the field to pay more attention to refractory populations, which have been virtually ignored. This may be accomplished in part through larger randomized controlled trials in treatment-resistant anxiety disorders, using a variety of augmentation agents, to adequately address the efficacy of augmentation strategies in this population.