Improving Outcomes for Patients With Medication-Resistant Anxiety: Effects of Collaborative Care With Cognitive Behavioral Therapy

Introduction

Pharmacological and psychological treatments can be highly effective for anxiety disorders (Arch & Craske, 2009; Hofmann et al., 2012; Ravindran & Stein, 2010). However, as in depression, response rates of just 50–60% and remission rates of only 25–35% are common in trials of first-line treatments (Bystritsky, 2006; Roy-Byrne, 2015). Relapse following remission of symptoms also may occur. Depending on the particular anxiety disorder, the rate of relapse has been estimated as 10–15% at 1 year, 25–45% at 5 years, and 40–60% at 12 years postrecovery (Bruce et al., 2005).

Limited empirical data are available to guide clinicians in determining appropriate intervention for patients with treatment-resistant anxiety (Roy-Byrne, 2015). Large-scale investigations—analogous to the STAR-D study of sequenced treatments for major depression (Rush et al., 2006; Thase et al., 2007)—have not been conducted for anxiety disorders. With a few notable exceptions (Foa et al., 2013; Pollack et al., 2014; Simpson et al., 2013), evidence supporting specific approaches to treatment resistance are limited to clinical case series (Otto, Pollack, Penava, & Zucker, 1999; Pollack et al., 1994) and small randomized clinical trials (RCTs; reviewed below). More empirical data are needed to inform next-step strategies for patients whose anxiety symptoms do not respond to standard treatments.

A judicious choice made by many clinicians faced with nonresponse is to add or switch to another treatment with documented efficacy for anxiety. Two RCTs support trying pharmacotherapy when panic disorder (PD) is resistant to cognitive behavioral therapy (CBT). One examined outcomes of 43 patients with CBT-resistant PD who received either continued CBT plus paroxetine or continued CBT plus placebo (Kampman, Keijsers, Hoogduin, & Hendriks, 2002). Patients who received paroxetine displayed significant decreases in anxiety and agoraphobia, whereas those who received placebo did not. Another study found that switching to a selective serotonin reuptake inhibitor (paroxetine or citalopram) produced better acute response than extending duration of CBT in 58 patients whose PD did not respond to a standard course of CBT (Payne et al., 2016). In contrast, a study of 23 post-traumatic stress disorder (PTSD) patients with incomplete response to prolonged exposure failed to demonstrate incremental benefit of adding paroxetine-CR over extending the duration of exposure therapy (Simon et al., 2008).

With respect to pharmacotherapy nonresponse, options include CBT or augmentation with another efficacious medication. An RCT of 181 patients with sertraline-resistant social anxiety disorder (SAD) suggested that addition of clonazepam produced greater clinical benefits than extending the duration of sertraline treatment or switching to venlafaxine (Pollack et al., 2014). Evidence supporting addition of CBT for medication-resistant anxiety is positive, but generally based on much smaller samples (Rodrigues et al., 2011). One trial found that CBT and clonazepam were similarly effective in reducing symptoms of SSRI-resistant PD, but the sample for this comparison was comprised of only 19 patients (Simon et al., 2009). Several studies tested culturally adapted CBT for medication-resistant PTSD (Otto et al., 2003; Hinton et al., 2004, 2005, 2011); the largest compared CBT to delayed treatment in a sample of 40 Cambodian refugees and found that 60% of the immediate treatment group no longer met criteria for PTSD at posttreatment (Hinton et al., 2005). One study also supports use of prolonged exposure to augment treatment for PTSD that is partially responsive to sertraline (Rothbaum et al., 2006). Finally, several RCTs have demonstrated superiority of exposure and response prevention over stress-management therapy (Foa et al., 2013) and risperidone (Foa et al., 2015; Simpson et al., 2013) for medication-resistant OCD.

To expand the scope of investigation of CBT for medication-resistant anxiety, we analyzed data from a large RCT of the Coordinated Anxiety Learning and Management (CALM) intervention. The CALM study differs from efficacy trials reviewed above in that its effectiveness design more closely approximated delivery of treatments for anxiety in clinical practice settings (Sullivan et al., 2007). The CALM intervention featured individual CBT that could be adapted to treat any of the four most common anxiety disorders in primary care: generalized anxiety disorder (GAD), PD, PTSD, and SAD (Kroenke, Spitzer, Williams, Monahan, & Lowe, 2007). Another innovation of the CBT program was its computer-assisted format, which enabled it to be implemented by trained but novice therapists situated at patients' primary care clinics. The CALM intervention proved superior to usual care (UC) for reducing overall anxiety and disability (Roy-Byrne et al., 2010) and many disorder-specific symptoms (Craske et al., 2011) in patients with PD, SAD, GAD, or PTSD. In this investigation, we tested whether CALM's CBT program (CALM-CBT) benefited patients whose anxiety disorders were resistant to pharmacotherapy. Medication-resistance was defined retrospectively as meeting study entry criteria (i.e., having PD, SAD, GAD, or PTSD and at least moderate anxiety severity) despite ongoing pharmacotherapy of appropriate type, dose, and duration. We predicted CALM-CBT would be more effective than UC for patients with medication-resistant anxiety. Within-group analyses further explored impacts of treatment choice (CBT vs. CBT plus medication management) and CBT dose on outcomes; as well as maintenance, loss, and delayed acquisition of treatment response among CALM-CBT patients with medication-resistant anxiety.

Materials And Methods

Participants

The CALM study recruited 1,004 patients from geographically diverse primary care clinics during June 2006 through April 2008 (Roy-Byrne et al., 2010; Sullivan et al., 2007). Referred patients were eligible if they were aged 18–75; spoke English or Spanish; met DSM-IV criteria for GAD, PD (with or without agoraphobia), PTSD, or SAD on the Mini International Neuropsychiatric Interview (Sheehan et al., 1998); and scored ≥8 on the Overall Anxiety Severity and Impairment Scale (OASIS (Campbell-Sills et al., 2009)). Patients were excluded if they were actively suicidal; had psychotic, Bipolar I, or substance use disorders (except alcohol and marijuana abuse); or were receiving CBT or psychiatric treatment. All participants provided written informed consent and protocols were approved by Internal Review Boards of all participating institutions.

During the baseline interview, participants provided names and dosages of each medication they took in the prior 6 months for anxiety, depression, or sleep (Stein et al., 2011). They indicated whether each medication was current, how long they had taken it for, and how they took it (regularly as prescribed, regularly with occasional missed doses, or irregularly). Medications that were not current, taken irregularly, taken for <2 months, or taken at sub-therapeutic dose were excluded from further consideration, as we could not confidently infer that anxiety symptoms were medication resistant under those circumstances. Remaining medications were examined in relation to baseline diagnoses and the following were retained as appropriate: any antidepressant (except bupropion) for GAD, PD, PTSD, or SAD; any benzodiazepine for GAD, PD, or SAD; buspirone for GAD; gabapentin for GAD, PD, or SAD; and pregabalin for GAD, PD, or SAD. A list of qualifying medications and corresponding minimum and median doses is provided in Supporting Information Table S1.

Patients who were eligible for the CALM study and whose pharmacotherapy at the time of entry met all of the above criteria were classified as “medication-resistant” (n = 227). Of these patients, 117 were randomized to the CALM intervention and 110 to UC. For outcome analyses, the intervention group was restricted to patients who chose CBT either with or without medication management (CALM-CBT; n = 104). This resulted in exclusion of 10 patients who chose medication management only and three who had no postrandomization contact with the study therapist (i.e., treatment choice was never established).

Measures

Treatment choice and CBT dose.

Information about which treatment(s) were chosen (CBT, medication management, or both) and number of CBT visits was extracted from the study's web-based monitoring system. A prior investigation of all CALM patients who selected CBT demonstrated that a dose of 6 or more CALM-CBT sessions was associated with improved treatment response across a variety of outcome measures (Glenn et al., 2013). Thus, for the current study of response among medication-resistant patients, high CBT dose was defined as 6 or more CBT visits.

Pharmacotherapy at follow-up.

During follow-up interviews at 6, 12, and 18 months, patients provided information about pharmacotherapy received during the prior 6 months. The following dichotomous variables were used to examine whether pharmacotherapy differed substantially in CALM-CBT vs. UC patients: any antidepressant taken at therapeutic dose for 2 months or more; any benzodiazepine taken at therapeutic dose for 2 months or more; and other antianxiety medication (buspirone, gabapentin, or pregabalin) taken at therapeutic dose for 2 months or more.

Response and remission.

Following the main CALM outcome study (Roy-Byrne et al., 2010), response and remission were measured by summing the 12 anxiety and somatic symptom items of the Brief Symptom Inventory (Derogatis, 2001). Response was defined as ≥50% reduction in BSI-12 or meeting the remission criterion. Remission was defined as BSI-12 <6, or average per-item score <0.5 (between “none” and “mild”).

Procedures

Assessment.

Eligible patients completed a baseline telephone interview conducted by RAND Survey Research Group and were subsequently randomized to CALM or UC (stratified by clinic and MDD). Outcome assessments were conducted at 6,12, and 18 months by RAND interviewers blinded to treatment assignment.

CALM.

Patients randomized to CALM could choose CBT, medication management, or both. More treatment with the same or alternative modality was allowed if response to initial treatment was unsatisfactory. When response was achieved [OASIS <5 or patient declined further treatment], patients were transitioned to a continued care phase during which their Anxiety Clinical Specialist (ACS (Rose et al., 2011)) phoned monthly to assess symptoms and reinforce CBT skills and/or medication adherence. Continued care was withdrawn at 12 months.

CALM-CBT (Craske et al., 2009) was delivered by the ACS (typically over 6–8 weekly sessions) and followed a computerized program covering psychoeducation, self-monitoring, breathing retraining, cognitive restructuring, exposure, and relapse prevention. Homework practice was strongly encouraged. For medication management, ACSs promoted adherence and relayed recommendations from study psychiatrists to primary care providers. Supervising psychiatrists did not directly adjust the regimen but encouraged use of an algorithm specifying first-line use of SSRIs or SNRIs and second/third step combinations of two antidepressants or an antidepressant and benzodiazepine (Roy-Byrne et al., 2009).

Usual Care.

UC consisted of unrestricted care by primary care providers, which could include pharmacotherapy, referral for counseling (7 of 17 clinics had in-house mental health resources), or referral for specialty care (including CBT). Providers at participating clinics received training in the pharmacotherapy algorithm recommended by study psychiatrists overseeing patients assigned to CALM (Roy-Byrne et al., 2009). Following randomization, UC patients had no contact with study personnel except for the telephone assessments conducted at 6,12, and 18 months by the RAND Survey Research Group.

Data analysis

Pearson's X² (for categorical variables) and t-tests (for continuous variables) were conducted to test for baseline differences between medication-resistant and other CALM participants. Pearson's X² also was used to evaluate differences in pharmacotherapy reported by CALM-CBT versus UC patients at 6, 12, and 18 months. Proportions of CALM-CBT and UC patients who responded and remitted were calculated based on samples with valid BSI-12 scores at 6, 12, and 18 months. Logistic regression was used to estimate effects of CALM-CBT (vs. UC) on response and remission at 6, 12, and 18 months. Because preliminary analyses showed site effects for several outcomes, study site was adjusted for in all regression models. Within the CALM-CBT group, a two-sample test of proportions was used to compare proportions responded/remitted by treatment participation (CBT only vs. CBT plus medication management). A similar test was performed to determine if high-CBT dose (≥6 visits) produced greater response/remission than low CBT dose. One-tailed P < .05 was considered significant for high versus low dose comparisons. Two-tailed P < .05 was considered significant for all other tests.

Results

Comparison of CALM study participants with and without medication-resistant anxiety

Table 1 presents demographic, clinical, and treatment characteristics of medication-resistant versus other CALM study participants. Medication-resistant patients differed in racial/ethnic composition [X² (3, 1004) = 17.91, P < .001], in that a smaller proportion identified as Hispanic and a larger proportion identified as White. Patients with medication-resistant anxiety also had higher baseline BSI-12 scores [t (1002) = 2.88, p = .004] and a greater proportion had MDD at baseline [X² (1, 1004) = 7.61, P = .006]. Medication-resistant patients did not differ from other CALM study participants on any other variable under consideration.

Effects of CALM-CBT on response and remission of patients with medication-resistant anxiety

Table 2 presents logistic regression results and proportions of medication-resistant CALM-CBT and UC patients who responded and remitted at 6, 12, and 18 months. Relative to UC, CALM-CBT was associated with greater response at 6 months (AOR = 3.78, 95% CI 2.02–7.07) and 12 months (AOR = 2.49, 95% CI 1.36–4.58) but not at 18 months (AOR = 1.55, 95% CI 0.85–2.85). CALM-CBT was associated with increased odds of remission at 6, 12, and 18 months (AORs = 2.44–3.18).

Outcomes in Table 2 diverged from those of the main CALM outcome study (Roy-Byrne et al., 2010) in the lack of intervention effect on 18-month response and absence of growth of response/remission rates from 6 to 12 months. These discrepancies raised the possibility that medication resistance moderated effects of CALM-CBT on response and remission. To evaluate this, we expanded the analysis sample to include nonresistant CALM-CBT and UC patients and conducted supplementary regression analyses that modeled Treatment (CALM-CBT vs. UC) x Medication-Resistance (resistant vs. non-resistant) interaction effects on 6-, 12-, and 18-month response (Supporting Information Table S2) and remission (Supporting Information Table S3). Medication resistance did not moderate effects of CALM-CBT on any outcome (AORs = 0.66–1.41; Ps > .25).

Many patients in CALM-CBT also received some medication management, raising the possibility that pharmacotherapy differences could explain the advantages of CALM-CBT over UC. To explore this, we compared the proportions of CALM-CBT and UC participants who reported minimum 2-month treatment with appropriate doses of (1) antidepressants; (2) benzodiazepines; and (3) other antianxiety medications at the 6-, 12-, and 18-month assessments. CALM-CBT and UC patients did not differ on any of these indices of pharmacotherapy at 6, 12, or 18 months [X² (1) = 0.00 to 1.80, Ps > .18].

Impact of treatment choice and CBT dose within the CBT-CALM group

Table 3 displays proportions of medication-resistant CALM-CBT patients who responded and remitted by treatment participation and CBT dose. Proportions for CBT versus CBT plus medication management did not differ [zs = 0.20-1.54; Ps > .12]. The lack of added benefit of medication management was not discernably related to greater illness severity of patients who chose both treatments, as their baseline BSI-12 scores [t(98) = 0.59, P = .55] and rates of MDD [X² (1, 100) = 0.50, P = .48] were similar to those of patients who chose CBT only.

Median number of CBT visits within the CALM-CBT group was 7 (IQR = 6–9), with more than three-quarters (77.88%) of medication-resistant patients receiving a high CBT dose. Patients with a high CBT dose displayed greater response at 6 months (z = 2.06, P = .020) and remission at 6 and 18 months (z = 1.76, p = .039 for 6 months; z = 1.87, p = .031 for 18 months). Significant advantage of high CBT dose was not observed for the other outcomes (zs = 0.10 to 1.13; Ps > .12).

Maintenance, loss, and delayed acquisition of response and remission within the CALM-CBT group

Response.

Of the 56 responders to CALM-CBT at 6 months, 38 (67.86%) maintained response, 15 (26.79%) lost response, and 3 (5.36%) were missing outcome data at 12 months. At 18 months, 37 (66.07%) maintained response, 16 (28.57%) lost response, and 3 (5.36%) were missing outcome data. Of the 41 nonresponders to CALM-CBT at 6 months, 10 (24.39%) acquired response, 25 (60.98%) remained nonresponders, and 6 (14.63%) were missing data at 12 months. By 18 months, 14(34.15%) had acquired response, 21 (51.22%) remained non-responders, and 6 (14.63%) were missing outcome data.

Remission.

Of the 43 CALM-CBT patients who had remitted at 6 months, 30 (69.77%) maintained remission, 10 (23.26%) lost remission, and 3 (6.98%) were missing data at 12 months. At 18 months, 27 (62.79%) maintained remission, 14 (32.56%) lost remission, and 2 (4.65%) were missing outcome data. Of the 54 CALM-CBT patients who had not remitted at 6 months, 11 (20.37%) achieved remission, 37 (68.52%) remained unremitted, and 6 (11.11%) were missing data at 12 months. By 18 months, 13 (24.07%) had achieved remission, 34 (62.96%) remained unremitted, and 7 (12.96%) were missing data.

Discussion

Collaborative care with CBT was effective for primary care patients with medication-resistant anxiety. Compared to UC, CALM-CBT was associated with more than tripled odds of response and remission at 6 months and doubled odds of response at 12 months and remission at 12 and 18 months. Some 58% of CALM-CBT patients with medication-resistant anxiety achieved satisfactory response and 46% remitted by the final assessment.

Patients classified as medication-resistant were taking appropriate medications (at sufficient dosage and for 2 months or more) at the time of their baseline assessment for the CALM study. They were demographically similar to other CALM study participants except that a smaller proportion identified as Hispanic. This finding might reflect ethnic differences in attitudes toward psychotropic medication use (Cooper et al., 2003; Green-Hennessy & Hennessy, 1999; Hunt et al., 2013) or variations in clinical presentation and provider recognition of anxiety disorders that relate to ethnicity (Brenes et al., 2008). Despite strong evidence of adequate pharmacotherapy (mostly with antide-pressants), medication-resistant patients had higher baseline anxiety and rates of co-occurring MDD than other CALM study participants.

Consistent with rates of treatment selection in the full CALM study sample (Roy-Byrne et al., 2010), medication-resistant patients randomized to CALM demonstrated a strong preference for CBT either as their sole treatment within the study or in combination with medication management. The largest advantage of CALM-CBT over UC was observed at the 6-month assessment, which was in closest temporal proximity to active treatment for most CALM-CBT patients. CALM-CBT's effect on treatment response diminished but remained significant at 12 months, and was nonsignificant by 18 months. We found no evidence that effects of CALM-CBT were at any point moderated (e.g., weakened) by medication resistance. The most plausible explanation for the lack of effect on 18-month response is the marked improvement in the UC group at the 18-month assessment. That increase suggests that some medication-resistant UC patients eventually responded to their primary care provider's treatment or found other helpful resources. However, even with this improvement in response, only about one-quarter of UC patients had remitted by 18 months. CALM-CBT's advantage with respect to remission was robust and significant at all assessments.

Proportions of medication-resistant CALM-CBT patients who responded (58%) and remitted (44%) by 6 months were basically equivalent to the corresponding proportions reported for all patients randomized to CALM (Roy-Byrne et al., 2010). This level of treatment response is encouraging in light of the more severe anxiety, greater depressive comorbidity, and medication resistance of this patient subgroup. Response and remission rates also compare favorably to average rates for efficacy trials of many first-line treatments (Roy-Byrne, 2015), despite the more real-world context allowed by the effectiveness study design (e.g., inclusive patient sample; treatment delivered flexibly within the primary care setting by therapists of varied clinical background).

Although medication resistance did not appear to moderate the effects of CALM-CBT, the longer-term response of the medication-resistant subgroup was somewhat less impressive than that observed in the main CALM outcome study. In the full sample, response and remission rates of the intervention group climbed to 64% and 51% at 12 months (and remained at that level at 18 months) (Roy-Byrne et al., 2010). Among medication-resistant CALM-CBT patients, response and remission rates remained stable across the three assessment waves. Although additional treatment steps were permitted up through 12 months, only about one-quarter of 6-month nonresponders to CALM-CBT were classified as responders at 12 months. Those cases were offset by a comparable proportion of 6-month responders to CALM-CBT who lost response at 12 months. Overall, however, maintenance of treatment gains was satisfactory with about two-thirds of 6-month responders and remitters having BSI-12 scores at 12 and 18 months that indicated maintenance of those levels of treatment response.

We examined effects of CBT in patients with medication-resistant anxiety because this analysis corresponds to the common clinical decision to try another modality when symptoms are resistant to pharmacotherapy. However, we acknowledge that the CALM study design does not permit isolation of effects of CBT. Patients randomized to the intervention could choose CBT, medication management, or both. While we excluded a small number of medication-resistant patients who participated in medication management only, many patients included in our outcome analysis participated in both CBT and medication management. For patients in medication management, ACSs encouraged adherence to the regimen and relayed recommendations from supervising psychiatrists to the primary care provider. These interventions may have partially or completely determined clinical improvement in some patients. However, in the aggregate, patients who participated in CBT plus medication management did not fare better than those who received CBT as their sole treatment within the study. CALM-CBT patients also did not differ from those who received UC at any assessment point on indices of pharmacotherapy with antidepressants, benzodiazepines, or other anti-anxiety medications.

Results of these supplemental analyses do not rule out the possibility that pharmacotherapy differences contributed to the better outcomes of CALM-CBT patients. However, available data do not support that interpretation, and the most obvious treatment event for these patients was the introduction of CBT. Most of the medication-resistant CALM-CBT patients completed enough sessions to meet our definition of high CBT dose. Though sample sizes for comparisons were modest, significant advantages of high CBT dose were observed for 6-month response and 6- and 18-month remission.

Usual limitations of secondary analysis of RCT data apply to this investigation. The CALM study was not designed to compare treatments for medication-resistant anxiety; therefore, conclusions are more tentative than for a design involving randomization of patients to “next-step” interventions. More studies of that type are needed. It is possible that medication-resistant patients who benefited from CBT provided directly by trained ACSs also would have improved with pharmacotherapy optimization provided directly by specialists with allegiance to that modality. Results achieved with computer-assisted CBT delivery (which is yet to be widely adopted in practice settings) may not generalize to more traditional CBT formats. Additionally, although within-group analyses showed some advantages of higher CBT dose— and failed to provide evidence that medication management improved outcomes—conclusions are limited by low power and lack of random assignment to those conditions.

For this report, we defined “medication-resistant anxiety” as qualifying for the CALM study (i.e., having PD, GAD, SAD, or PTSD and at least moderate anxiety severity) despite appropriate pharmacotherapy of minimum 2-month duration. However, research definitions have varied in terms of the number (one vs. two) and length of trials (generally 6–12 weeks) considered sufficient for labeling anxiety symptoms “medication-resistant” (Freire, Zugliani, Garcia, & Nardi, 2016; Rodrigues et al., 2011). We did not aim to evaluate effectiveness for severely resistant (i.e., treatment refractory) disorders, which is another critically important topic with its own literature. Our definition of adequate pharmacotherapy also could be debated. Prior reports from the CALM study used a similar definition (Roy-Byrne et al., 2010; Stein et al., 2011); however, we were more restrictive here in that we excluded certain medications from efficacious classes (e.g., bupropion) and imposed diagnosis-based rules. Overall, our definition has good concordance with the pharmacotherapy literature and published guidelines for treatment of anxiety (American Psychiatric Association Work Group on Panic Disorder 2010; Katzman et al., 2014).

Conclusion

Collaborative care with CBT was effective for patients with medication-resistant anxiety. Six-month response and remission rates were similar to those observed in the overall CALM sample, and compared favorably to average response rates of first-line treatments for anxiety. Treatment gains were adequately maintained. Medication-resistant patients who chose CBT plus medication management did not fare better than those who chose CBT only. Some evidence suggested that higher CBT dose (6 or more sessions) produced better response. Considered alongside the strong preference for CBT shown by CALM patients with medication-resistant anxiety, these findings support CBT as a promising avenue for improving outcomes of patients whose anxiety symptoms are resistant to standard pharmacotherapy.