Introduction
Selective serotonin reuptake inhibitors (SSRIs) and serotonin norepinephrine reuptake inhibitors (SNRIs) are first-line pharmacologic treatments for anxiety disorders (eg, generalized anxiety disorder (GAD), social anxiety disorder, separation anxiety disorder, panic disorder).Reference Stein and Sareen1, Reference Strawn, Geracioti and Rajdev2 By contrast, benzodiazepines, once the first-line pharmacotherapy for anxiety disorders, have been supplanted by SSRIs and SNRIs over the last several decades but are still commonly prescribedReference Agarwal and Landon3 and are now used for acute episodic anxiety or in conjunction with SSRIs/SNRIs.Reference Pollack, Van Ameringen and Simon4, Reference Seedat and Stein5 Despite the substantial empirical support for and availability of these medications, many patients do not receive adequate treatment.Reference Alonso, Liu and Evans-Lacko6 Consequently, anxiety disorders remain a source of morbidity and continue—internationally—to represent a leading cause of disability-adjusted years lost.Reference James, Abate and Abate7 As such, understanding how the response varies among these pharmacologically distinct classes of medications could enable clinicians to tailor treatments for specific types of anxiety disorders and patient populations as well as to provide guidance to patients in terms of when improvement may emerge. Understanding these differences in the trajectory of improvement across pharmacologic treatments in anxiety disorders could indirectly inform clinicians as to the adequate duration of a trial based on when most improvement is likely to occur.
Meta-analyses of SSRIs, SNRIs, and benzodiazepines have largely focused on endpoint responsesReference Hsu, Tseng and Wang8-Reference Gosmann, De Abreu Costa and De Barros Jaeger10 rather than the course of response with few exceptions.Reference Jakubovski, Johnson and Nasir11, Reference Stimpfl, Mills and Strawn12 In a meta-analysis of treatment response trajectory in SSRI and SNRI-treated adults with anxiety disorders (
$ K $
= 57, N = 16056), SSRIs produced improvement beginning at week 4, compared to placebo, while SNRIs produced improvement beginning at week 2. Additionally, treatment response was linear with SSRIs and logarithmic with SNRIs.Reference Jakubovski, Johnson and Nasir11 Further, in this meta-analysis, higher SSRI doses produced greater improvement, whereas dose–response relationships were not observed for SNRIs. In contrast to SSRI and SNRI treatment response trajectories, a meta-analysis of 65 studies involving 7110 patients found that benzodiazepine-related improvement emerged early—within the first week of treatment. However, benzodiazepine response plateaued by week 4.Reference Stimpfl, Mills and Strawn12
This broad focus of prior meta-analyses on the magnitude of improvement at endpointReference Hsu, Tseng and Wang8-Reference Gosmann, De Abreu Costa and De Barros Jaeger10 limits our understanding of the heterogeneity of treatment response—which patients get better fastest, which medication classes (or disorders) are associated with early improvement, and which treatments require longer to gain traction. Further, our current understanding of how the class of medication influences the trajectory of response has primarily been evaluated in meta-analyses of benzodiazepinesReference Stimpfl, Mills and Strawn12 and SSRIs/SNRIsReference Jakubovski, Johnson and Nasir11 separately, although these analyses suggest that benzodiazepines work more quickly than do SSRIs or SNRIs. Beyond this, understanding differences in the trajectory of improvement across pharmacologic treatments in anxiety disorders could indirectly inform decisions related to medication switching. For example, knowing when the likelihood of continued improvement dissipates could allow clinicians to abandon a trial of a medication that is likely to be ineffective earlier and, in doing so, decrease morbidity related to a patient being treated with a medication that is unlikely to produce relief.
Beyond medication response, placebo response has received considerably less attention in anxiety disorders compared to depressive disorders.Reference Stein, Baldwin and Dolberg13, Reference Rutherford, Mori and Sneed14 Placebo-related improvement is generally thought to be less in anxiety disorders compared to mood disorders,Reference Benedetti, Carlino and Pollo15, Reference Dunlop, Thase and Wun16 especially in specific populations (ie, youth).Reference Dobson and Strawn17, Reference Mossman, Mills and Walkup18 To our knowledge, there have been no studies examining how placebo response differs across specific anxiety disorders in adults, although it appears to differ across disorders.Reference Khan, Kolts and Rapaport19 Yet this information could inform how clinicians harness placebo response as a component of active treatment and how clinical trials are designed. In fact, understanding and addressing disorder-specific placebo response could enhance our ability to detect medication-placebo differences—the goal of clinical trials. This is particularly important in that placebo response—rather than active medication response—has been shown to undermine our ability to detect efficacy signals for new medications, potentially increasing the likelihood that effective treatments are not identified and not brought to our clinics.
Recent computational advances, including Bayesian hierarchical models (BHMs), have made it possible to evaluate the improvement trajectory and dose–response relationships.Reference Stimpfl, Mills and Strawn12 As a result of the variability that exists on different levels when comparing studies to evaluate treatment effects, BHMs can integrate multi-level information to estimate the change in symptoms or probability of improvement.Reference McGlothlin and Viele20 BHMs make it possible to isolate observed variability, which better differentiates between random and true differences in treatment outcomes. Applying BHMs to anxiety disorders treated with SSRIs, for example, allows us to control for factors such as patient-to-patient variability or medication-specific factors that may distort the true treatment effect. Ultimately, BHMs allow us to ascertain treatment effects while considering the information available in aggregate for all studies.
With these considerations in mind, this study aimed to examine the magnitude and trajectory of response to SSRI, SNRI, and benzodiazepines using BHMs in adults with anxiety disorders and to characterize placebo response to determine when placebo response first emerged and how study-specific factors or disorder type influenced placebo response. We hypothesized that (a) younger patients would have less response across medication classes and (b) SSRIs and SNRIs would have greater treatment response after 8 weeks than benzodiazepines and that benzodiazepines would have earlier improvement compared to SSRIs and SNRIs. We also hypothesized that placebo response would be low in patients with social anxiety disorder and panic disorder compared to GAD.
Methods
Search strategy
Studies were obtained through an electronic search of English language articles in PubMed (1966 through May 2023) and this search was repeated on May 17, 2023 (Figure 1). In addition to the Cochrane Database as well as the government clinical trials registry, www.clinicaltrials.gov using the search strategy: (SSRI* OR SNRI* OR benzodiazepine* OR selective serotonin reuptake inhibitor OR serotonin and norepinephrine reuptake inhibitor OR serotonin uptake inhibitor OR escitalopram OR citalopram OR fluoxetine OR fluvoxamine OR paroxetine OR sertraline OR vilazodone OR venlafaxine OR desvenlafaxine OR duloxetine OR alprazolam OR chlordiazepoxide OR clonazepam OR diazepam OR lorazepam OR midazolam OR oxazepam OR temazepam OR triazolam OR clorazepate OR bromazepam OR nimetazepam OR clobazam OR loprazolam OR nitrazepam OR etizolam) AND (anxiety OR anxious OR anxiolytic OR panic* OR panic disorder OR generalized anxiety OR generalized anxiety disorder OR GAD OR social anxiety disorder OR social phobia OR SoP) AND (placebo) NOT (“systemic review” OR “systematic review” OR review OR meta-analysis OR meta analysis OR editorial OR case report OR letter OR children OR adolescent OR pediatric OR mice OR zebrafish OR rat OR dog OR insomnia OR sleep OR dental OR delirium OR memory OR seizure OR withdrawal).
PRISMA flow diagram for trials of selective serotonin reuptake inhibitors (SSRIs) and serotonin norepinephrine reuptake inhibitors (SNRIs) (top) and benzodiazepines (bottom). This search was repeated on May 17, 2023, and no additional articles were identified.
The results of the search were then manually limited to randomized, placebo-controlled trials. The references of all eligible trials were searched for additional clinical trials. All meta-analytic methods and sensitivity analyses were specified before conducting the meta-analysis proper.
Meta-analytic approach
Studies were included if they were prospective, randomized, parallel-group, placebo-controlled trials that evaluated the efficacy of an SSRI, SNRI, or benzodiazepine in the treatment of social, generalized, and/or separation anxiety disorder or panic disorder or a DSM-IV or DSM-III-R forerunner of these disorders in adults, and used the Hamilton Anxiety Rating Scale (HAM-A). We selected the HAM-A, a validated rating scale to measure the severity of the anxiety symptoms, for several reasons. First, the HAM-A has been collected across most clinical trials of anxiety disorders—including social anxiety disorder and panic disorder—which increases the data available for the meta-analysis. Second, the HAM-A has been sensitive to treatment-related change across anxiety disorders and treatment-related improvement on HAM-A and other anxiety measures such as the Leibowitz Social Anxiety Scale (LSAS) is strongly correlated.Reference Kummer, Cardoso and Teixeira21 Third, despite the common use of the HAM-A in generalized anxiety studies, overlapping symptoms (both cognitive and physical) are present in other anxiety disorders. Clinical trials were excluded if they met any of the following criteria: included children and adolescents (age < 18 years); utilized a cross-over design; did not study an SSRI, SNRI, or benzodiazepine; were not randomized; were not placebo-controlled; provided adjunctive psychotherapy or adjunctive SSRI, SNRI, or benzodiazepine to active or control group; or included <10 patients per treatment group.
Data were extracted into an Excel (Microsoft, Redmond, WA) spreadsheet by 3 authors (E.M.M. and J.R.S. and J.N.S. for benzodiazepine studies). Additional data related to the methods, demographics, dosing, duration of the trial, and other relevant aspects and results of the studies were collected. Consistent with a prior meta-analysis of the efficacy and tolerability of SSRIs and SNRIs, the outcome measurements selected from each included clinical trial was the mean improvement in the treatment and placebo groups. This mean improvement was determined for the clinical rating scale measuring anxiety symptom severity at each reported time point for the continuous measure of anxiety symptoms.
Statistical methods
For each treatment type (benzodiazepine, SSRI, SNRI, placebo) and disorder (GAD, social anxiety disorder, panic disorder), posterior distributions for average treatment response at each point in time (weeks 1, 2, 4, 6, and 8) were obtained from Hamiltonian Monte Carlo (HMC) posterior simulation (using the Turing.jl package in the Julia programming language),Reference Ge, Xu and Ghahramani22 allowing computation of the posterior distribution of the difference in average treatment response,
$ \overline{\delta} $
, across categories at each point in time. The functional form within each category was,
$$ {y}_{it}={\sum}_{w\in W}{\delta}_w{D}_{wit}+{X}_i\beta +{u}_{it}, $$
where
$ {y}_{it}= $
treatment response in study
$ i $
at time point
$ t $
,
$ W=\left\{0,1,2,4,6,8\right\} $
,
$ {D}_{wit}=1\; if\;t=w,0\hskip0.45em \mathrm{otherwise}, $
$ {X}_i= $
a vector of covariates (average age, percent female), and
$ {u}_{it}\sim N\left(0,{\sigma}^2/{N}_{it}\right) $
,
$ {N}_{it}= $
number of patients in study
$ i $
at time period
$ t $
.
As previously described, posterior distributions for the average trajectory of response for each treatment type and disorder were also obtained from HMC posterior simulation of a linear-logarithmic time trend functional form,Reference Stimpfl, Mills and Strawn12, Reference Suresh, Mills and Croarkin23 where for study
$ i $
at time point
$ t $
,
$ \log {T}_{it}=\log (t),\mathrm{when}\;t\ne 0,\mathrm{and}\;\log {T}_{it}=0,\mathrm{when}\;t=0 $
,
$ {X}_i= $
any covariates included (eg, age, sex),
$$ {y}_{it}=\delta \log {T}_{it}+\log {T}_{it}\times {X}_i\beta +{u}_{it},{u}_{it}\sim N\left(0,{\sigma}^2\right). $$
Results
Descriptive statistics
The meta-analysis involved 15 760 patients (7662 treated with benzodiazepines, 5191 treated with SSRIs, and 2934 treated with SNRIs). Characteristics of the studies are shown in Table 1.
Characteristics of Included Studies
Abbreviations: SD, standard deviation; SNRI, serotonin norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor.
SSRI, SNRI, and benzodiazepine response trajectory
The models for treatment response—regardless of medication type (eg, benzodiazepine, SSRI, or SNRI) were Bayesian time indicator “fixed effects” models and linear-logarithmic models for the trajectory of improvement. Benzodiazepine-related improvement in anxiety was greatest initially and statistically superior to both SSRIs and SNRIs until after the fourth week of treatment (Table 2); this rate of improvement decreased over successive weeks. By the eighth week of treatment, response converged with no significant differences among benzodiazepines, SSRIs, and SNRIs (SSRIs vs. benzodiazepines, mean difference 6.63, p = 0.103; SSRIs vs. SNRIs, mean difference 7.13, p = 0.057; SNRIs vs. benzodiazepines, mean difference −0.50, p = 0.911) (Table 2 and Figure 2C).
Differences in Medication Response in Adults with Anxiety Disorders
Positive
$ \overline{\delta} $
values indicate that the second medication class comparison is superior (eg,
$ \overline{\delta} $
of 5.40 in the “SSRI vs. SNRI” column indicates that the SNRI produces a greater mean improvement of 5.4% in SNRI-treated compared to SSRI-treated patients).
Abbreviations: CrI, credible interval;
$ \overline{\delta} $
, mean difference; SNRI, serotonin-norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor.
Treatment response in adults with anxiety disorders. The change in anxiety symptom severity for patients receiving placebo is shown in (A). Placebo-related change in anxiety symptoms was lower in patients with social anxiety disorder compared to those with generalized anxiety disorder and panic disorder (B). Change in anxiety symptoms for benzodiazepines, SSRIs, and SNRIs was logarithmic in the best-fitting model and was more rapid initially in patients receiving benzodiazepines (C). Across treatments, change in anxiety symptoms in patients with generalized anxiety disorder (D), social anxiety disorder (E), and panic disorder (F) exhibited similar trajectories. In all panels, the response is shown by solid lines, and dotted lines reflect the 95% credible interval.
Impact of age and sex on treatment response across medication classes
Across disorders, SSRIs and SNRIs produced a greater and faster response in studies with more female participants (SSRI: β = −0.282 + 0.067, CrI: −0.413 to −0.152, p < 0.001; SNRI: β = −0.445 + 0.082, CrI: −0.605 to −0.289, p < 0.001) and studies with older patients had decreased response (SSRI: β = −0.628 + 0.012, CrI: 0.389 to 0.862, p < 0.001; SNRI: β = 0.790 + 0.014, CrI: 0.521 to 1.062, p < 0.001). By contrast, for benzodiazepines, studies with more older participants had greater and faster improvement (β = −0.391 + 0.014, CrI: −0.656 to −0.113, p = 0.003) while the percentage of women participants did not affect response magnitude/trajectory (p = 0.148).
The trajectory of placebo response
Placebo response—like response to active treatment—emerges within the first week (SMD: −0.142, 95% CrI −0.135 to −0.150) and plateaus by the fourth week of treatment (SMD: −0.244, 95% CrI: −0.232 to −0.256, Figure 2A). The trajectory and magnitude of placebo response differes significantly across disorders. Patients with social anxiety disorder have slower and less improvement compared to patients with panic disorder or GAD (Figure 2B and Table 3). At week 8, trials of GAD have a larger placebo response compared to trials focused on social anxiety disorder (p < 0.001), and trials with panic disorder have a greater placebo response compared to trials focused on social anxiety disorder (p = 0.025). Finally, at week 8, placebo response is greater in GAD trials compared to panic disorder (p = 0.002).
Change in Symptom Severity Placebo Response in Adults with Anxiety Disorders
Abbreviations: CrI, credible interval;
$ \overline{\delta} $
, mean difference in symptom severity from baseline.
Placebo response was greater in studies with more women participants (β = −1.591 + 0.179, CrI: −1.941 to −1.238, p < 0.001), but was not associated with the mean age of study participants (p = 0.127) and was not associated with publication year (p = 0.636, Table 3).
Heterogeneity
The Q statistic provides a measure of the degree of heterogeneity in the observed effect across studies. For context, if all included trials found a similar effect, Q is low. In contrast, if there are large differences in effect size across the trials, Q is high. If Q is statistically significant, this indicates statistical evidence of substantive differences in the effect size across the included trials. The BHM used herein explicitly models and estimates heterogeneity across trials and applies a shrinkage factor to the individual study estimates, reducing the impact of heterogeneity on the overall estimate, so while the less heterogeneity, the more reliable the results, the results obtained using a BHM are still valid regardless of Q (and hence the degree of heterogeneity). Nonetheless, as a standard measure of heterogeneity, we have reported Q and I 2.
Heterogeneity measures for baseline symptom severity in benzodiazepine studies were Q = 66.72 (p < 0.001), I 2 = 50.54. For SSRI studies, baseline symptom severity heterogeneity was Q = 2851 (p < 0.001), I 2 = 98.84, and for SNRI studies was Q = 48.20 (p < 0.001), I 2 = 60.58. For patients randomized to placebo, in panic disorder studies, Q = 54.77 (p < 0.001), I 2 = 50.83, while for GAD studies, Q = 62.08 (p = 0.046), I 2 = 27.51, and for studies of social anxiety disorder was Q = 135.76 (p < 0.001), I 2 = 87.48. Thus, the significant heterogeneity supports the use of the BHM.
Discussion
This meta-analysis of randomized, placebo-controlled trials of SSRIs, SNRIs, and benzodiazepines in adults with anxiety disorders revealed: (a) each medication class’s superiority to placebo; (b) no difference in SSRI or SNRI efficacy in adults with anxiety disorders; (c) maximal benzodiazepine effect emerges more quickly than SSRI or SNRI improvement and that (d) medication response is similar across disorders (except for benzodiazepines which produced greater improvement over the first 4 weeks of treatment compared to SSRIs and SNRIs in panic disorder). We also observed that placebo response was lower in patients with social anxiety disorders compared to trials involving patients with GAD and panic disorder. These findings and their clinical implications warrant additional discussion.
Our analysis replicates the findings of previous studies regarding the efficacy of SSRIs, SNRIs, and benzodiazepines for the treatment of anxiety disorders.Reference Gomez, Barthel and Hofmann9 That is to say that all 3 medication classes are superior to placebo in reducing anxiety in GAD, social anxiety disorder, and panic disorder.Reference Andrisano, Chiesa and Serretti24, Reference Hedges, Brown and Shwalb25 However, our findings introduce nuance with regard to when these improvements occur. The response to benzodiazepines was seen early in treatment compared to SSRIs and SNRIs. While benzodiazepines are associated with class-specific limitations in clinical practice, these data raise the possibility that they may have a distinct role during early treatment, perhaps while awaiting tardive response seen with SSRIs and SNRIs. Consistent with this possibility, a recent study of Italian adults with panic disorder found that the duration of untreated illness was significantly shorter for patients beginning a benzodiazepine compared to those receiving an antidepressant (64 vs. 35 months, p < 0.0001).Reference Piccoli, Bergamaschini and Molteni26 The timing with which these treatments are introduced is also important with regard to longer-term outcomes, given that adults with a shorter period between disorder onset and beginning treatment with an SSRI predict a better response to pharmacologic treatment in patients with GAD.Reference Altamura, Dell’Osso and D’Urso27
Most current treatment guidelines and recommendations focus on overall improvementReference Katzman, Bleau and Blier28-Reference Bandelow, Werner and Kopp31 rather than on the trajectory of improvement. Given that endpoint responses of SSRIs, SNRIs, and benzodiazepines are similar, the earlier improvement seen with benzodiazepines is lost in these guidelines. Yet, knowing how quickly a patient feels better is critically important. Finally, regarding this nuance, our analysis reveals that SSRI and SNRI-related improvement emerges later and that by week 8, there are no statistically significant differences between the 3 treatments. This was similarly observed in a 2019 Cochrane review of patients with depression in which a combined benzodiazepine and an antidepressant produced a greater response compared to an antidepressant alone in the first 4 weeks of treatment, but this benefit was no longer apparent after the fifth week of treatment.Reference Ogawa, Takeshima and Hayasaka32 As such, while there is rapid response associated with benzodiazepines, the continued use of them past week 8, even when used adjunctively, does not confer an advantage over SNRIs or SSRIs alone.
Our findings also shed light on placebo response in adults with anxiety disorders—a phenomenon that has received limited attention.Reference Benedetti, Carlino and Pollo15, Reference Dunlop, Thase and Wun16 To our knowledge, this is the first study to examine how placebo response differs across specific anxiety disorders in adults. We observed that trials focusing on GAD had a larger placebo response compared to trials focused on social anxiety disorder and panic disorder, and trials with panic disorder had a greater placebo response compared to trials focused on social anxiety disorder. These findings may inform clinical trial design and assist the clinician in understanding placebo response as a component of active treatment, particularly knowing that—in general—it may be greater in patients with GAD compared to other anxiety disorders. This process may relate to neurobiological differences in placebo response across anxiety disordersReference Benedetti, Carlino and Pollo33, Reference Lu, Li and Mills34 or to psychological aspects of the disorder that respond to placebo. In this regard, GAD may be associated with less avoidance compared to panic disorder and social anxiety disorder. Its ruminative nature may be more responsive to the non-specific aspects of clinical care during a clinical trial that produces improvement (ie, discussion of symptoms, therapeutic hope, and collaboration).Reference Rutherford and Roose35 As compared to GAD, panic disorder and social anxiety disorder may also be associated with a different experience of physical symptoms that could impact the placebo response. Interestingly, Lu et al.Reference Lu, Li and Mills34 proposed that placebo responders may have increased sensitivity to moment-to-moment shifts between internal and external stimuli, owing to higher neurofunctional flexibility—which may differ among anxiety disorders.Reference Lu, Li and Mills34
The differences in placebo response observed herein replicate and extend the results of studies of placebo response in anxiety disorders compared to other psychiatric disordersReference Khan, Kolts and Rapaport19, Reference Rutherford, Pott and Tandler36 and have important implications for researchers and clinicians. Understanding and addressing disorder-specific placebo responses could enhance our ability to detect medication-placebo differences in trials.Reference Dobson and Strawn17, Reference Rutherford and Roose35, Reference Sneed, Rutherford and Rindskopf37 This is particularly important because placebo response undermines our ability to detect efficacy signals for new medications, potentially increasing the likelihood that effective treatments are not identified nor made available to our patients.
While our findings have implications for the vexing consequences of placebo response, strategies to address placebo response rarely focus on the disorder being treated and instead, focus on broader strategies such as independent raters and placebo-lead-ins.Reference Rutherford and Roose35, Reference Sneed, Rutherford and Rindskopf37 These approaches produce variable success and increase the cost of clinical trials. Our data suggest that placebo response is associated with the type of disorder and patient characteristics (eg, sex), which raises the possibility that clinical trialists might use targeted interventions to address placebo response. One of the main drivers of a larger placebo response is the baseline severity of illness, which may result from patients with contextual anxiety—that resolves over time—remitting regardless of treatment. Further, factors such as trial size, whether or not the trial was federally funded, and the number of study sites correlate with the size of the placebo effect and so can act as a proxy for implementation effects.Reference Benedetti, Carlino and Pollo15–Reference Dobson and Strawn17, Reference Strawn, Dobson and Mills38 The extant literature suggests the importance of limiting the number of sites, including patients with more severe illness, and allowing for greater comorbidity, which is seen in patients with greater severity of illness. Beyond this, many trials have focused on retention rates, and it is noteworthy that in many of the earlier studies included in this meta-analysis, wherein placebo effects were lower, retention rates were low by today’s standards. This may reflect the greater severity of illness and functional impairment in these trials as patients and their treating clinicians were hesitant to continue an ineffective treatment—whether medication or, more likely, placebo—in such an ill population. Further, we argue that this may not necessarily reflect a problem, particularly given shifts to imputation strategies/last observation carried forward approaches that have become standard in clinical trials. Finally, using a Bayesian hierarchical modeling approach, it is possible to weight trials according to these observable proxies, which could potentially improve inference from a meta-analysis, and such an analysis could be a complementary addition to the results currently provided by meta-analyses. Without this admittedly somewhat subjective additional analysis, the size of the placebo effect across trials can be overstated and attenuate the observed medication-placebo difference in response, particularly when examining conditions such as anxiety.
Whether to prescribe an SSRI or an SNRI for a patient with an anxiety disorder often comes down to side effect profiles since the comparative efficacy of SSRIs and SNRIs is similar for anxiety disorders in adults.Reference Jakubovski, Johnson and Nasir11 We did not observe statistically significant differences in efficacy between SSRIs and SNRIs. However, this finding should be interpreted with caution. The SSRIs in these trials vary considerably in their metabolism. For sertraline,Reference Huddart, Hicks and Ramsey39, Reference Rudberg, Hermann and Refsum40 escitalopram,Reference Strawn, Poweleit and Ramsey41–Reference Jukić, Haslemo and Molden43 citalopram,Reference Chang, Tybring and Dahl44 and, to a lesser extent, fluoxetineReference Altamura, Moro and Percudani45, Reference LLerena, Dorado and Berecz46 exposure (medication concentrations over time) varies within populations based on variation in pharmacokinetic genes (eg, CYP2C19, CYP2D6, CYP2B6). Thus, dosing for some SSRIs in these trials may not reflect equivalent exposure across patients, which contributes to greater heterogeneity of response among these SSRI trials. Additionally, in some anxiety disorders (eg, panic disorder), patients may have exquisite sensitivity to and activating effects to some antidepressants—even at low doses—early in treatment.
While this study is the first to examine how the trajectory and magnitude of response differ among SSRIs, SNRIs, and benzodiazepines in anxiety disorders, there are several important limitations. In addition to the differences in metabolism discussed above, other factors may influence trajectory, including additional pharmacologic, clinical, and study characteristics, which are difficult to account for in meta-analyses. For example, given the shorter onset of action expected with benzodiazepines as compared to SSRIs and SNRIs, benzodiazepine trials were observed to have more early assessment points than SSRI and SNRI trials. It is possible that trial participants experienced additional psychotherapeutic benefits of more early assessment points, which could have contributed to the early improvement observed in this group. Second, the risk of bias may vary among studies and may influence findings and is especially pertinent to our findings pertaining to differential placebo response among anxiety disorders, as variation in blinding procedures can have a marked influence on placebo response. Also, there may be an additional risk of bias in benzodiazepine trials, given that benzodiazepines work quickly and have a distinct tolerability profile from SSRIs and SNRIs. Third, baseline anxiety ratings differed among studies, which raises the possibility of a floor effect in studies of patients with milder symptoms.Reference de Vries, Roest and Burgerhof47 Fourth, assessing the effect of covariates is challenging in meta-analyses when the average covariate value does not differ greatly across studies (average age of patients, percentage of female patients). Fifth, the subsets of data reduce the effective sample size; the number of patients represented at any one-time point is reduced when examining a specific treatment for a particular disorder, so the precision of the estimate is reduced. Sixth, the assessment of tolerability and dropout was limited, given the heterogeneity of adverse event reporting across studies and the lack of consistency and change over time in the terms by which adverse events were classified. Seventh, we focused on DSM-5 anxiety disorders, whereas earlier versions of the DSM included obsessive-compulsive disorder (OCD) and posttraumatic stress disorder (PTSD). The decision to focus on “fear-based” anxiety disorders relates to the fact that these disorders are often studied en bloc, have similar developmental epidemiology and neurobiology, and share risk factors. Further, this meta-analysis focused on “antidepressants” and benzodiazepines, which are the most common pharmacologic interventions for these disorders. In addressing the issue of heterogeneity and streamlining comparisons, we opted to exclude older medications like tricyclic antidepressants, monoamine oxidase inhibitors (MAOIs), and additional compounds such as gabapentin and pregabalin. It is crucial to emphasize, though, that several medications within these classes, particularly MAOIs and phenelzine in particular, exhibit substantial efficacy in treating persistent PD and social anxiety disorder.Reference Strawn, Geracioti and Rajdev2, Reference Versiani, Nardi and Mundim48, Reference Quitkin, Mcgrath and Stewart49 Furthermore, there has been a resurgence of interest in MAOIs in recent years, underscoring their potential role in the contemporary therapeutic landscape.Reference Van Den Eynde, Abdelmoemin and Abraham50, Reference Bodkin and Dunlop51 Finally, while we restricted our analyses to studies using the HAM-A to increase the reliability of cross-study comparisons, the HAM-A may differentially capture disorder-specific impairment and improvement across anxiety disorders.
This study is the first to characterize response trajectories for SSRIs, SNRIs, and benzodiazepines in anxiety disorders and, in doing so, has a variety of clinical implications. From these data, benzodiazepines, SSRIs, and SNRIs are each effective in treating anxiety disorders, with no significant difference between SSRIs and SNRIs. To augment early treatment, clinicians may consider adjunctive benzodiazepines initially along with an SSRI or SNRI, especially for patients with symptoms necessitating more immediate relief, as well as those suffering from panic disorder. However, after 8 weeks, there is unlikely to be additional benefit in continuing benzodiazepines concurrently with an SSRI or SNRI, so tapering should be considered at this point. Additional consideration may be given in the risk versus benefit discussion for the use of benzodiazepines in older patients, as they tend to show greater and faster improvement as compared to younger patients. Further, clinicians may both utilize and be wary of the observed placebo response which plateaus around week 4, especially when evaluating the efficacy of a medication and may expect less of a placebo effect in patients with social anxiety disorder. Taken together, these findings support the need for and importance of individualized approaches in treating patients with anxiety disorders.
Author contribution
Conceptualization: J.A.M., E.M., J.R.S.; Formal analysis: J.A.M., V.S., J.R.S.; Investigation: J.A.M., V.S., E.M., J.S., J.R.S.; Methodology: J.A.M., V.S., J.R.S.; Visualization: J.A.M., V.S., J.R.S.; Writing – original draft: J.A.M., V.S., E.M., J.N.S., J.R.S.; Writing – review & editing: J.A.M., V.S., E.M., J.N.S., J.R.S.; Data curation: V.S., E.M., J.N.S., J.R.S.; Validation: V.S.; Funding acquisition: J.R.S.; Project administration: J.R.S.; Supervision: J.R.S.
Financial support
This work was supported by the Yung Family Foundation (J.R.S., J.A.M.) and the National Institute of Child Health & Development R01HD098757 (J.R.S.).
Disclosures
J.R.S. has received research support from AbbVie, PCORI, and the National Institutes of Health. He has provided consultation to Cerevel, Intracellular Therapeutics, Otsuka, and will serve on a scientific advisory board for Genomind. He receives royalties from Springer Publishing, Cambridge University Press, and UpToDate and received material support from Myriad. He has also received honoraria from Medscape Live, Neuroscience Education Institute, the American Academy of Pediatrics, and the American Academy of Child and Adolescent Psychiatry. J.A.M. and J.R.S. receive support from the Yung Family Foundation. The other authors declare no potential conflicts of interest.
