Cognitive Behavior Therapy for Generalized Anxiety Disorder Among Older Adults in Primary Care: Title and subTitle BreakA Randomized Clinical Trial FREE

Generalized anxiety disorder (GAD) is common in late life, with prevalence up to 7.3%¹ in the community and 11.2% in primary care.² Late-life anxiety predicts increased physical disability,³ memory difficulties,⁴ decreased quality of life,⁵ increased service utilization,⁶ and mortality.⁷ Coexistent depressive disorders are common.⁸ Generalized anxiety disorder often precedes depression, suggesting it as a risk factor.⁹

Late-life anxiety is usually treated with medication. Benzodiazepines and antidepressants are effective,¹⁰ but associated risks (eg, falls, hip fractures, memory problems) and patient fears of adverse effects limit their usefulness.¹¹ Older patients also prefer psychosocial interventions.¹²

Initial outcome studies of cognitive behavior therapy (CBT) for late-life GAD focused on group-based interventions in academic settings.^{13 - 14} Effects were moderate relative to wait-list and minimal-contact controls (d = 0.71-0.75) and small relative to alternative psychotherapy (d = 0.20-0.29).¹⁴ Treatment response rates were low (28%-45%)¹⁴ but increased over follow-up. Individually administered CBT produced larger effect sizes (0.78-1.3) and improved response, though sample sizes in these trials were smaller.^{15 - 16}

Generalizability of these findings is limited because participants were mostly white, well educated, and healthy. Results also may not be relevant for patients in primary care, where older adults typically present for treatment. Cognitive behavior therapy within a collaborative-care framework in primary care has been effective for younger patients with panic disorder, GAD, or both¹⁷ as well as for older patients with depression,¹⁸ although effect sizes and treatment-response rates were lower than in academic clinical trials. The late-life anxiety literature lags behind these other areas, with only 2 pilot studies addressing treatment in primary care.^{19 - 20} These studies suggest benefits for CBT, but sample sizes were small (n = 12 and 31), and conclusions were limited.

We report results of the first randomized clinical trial of CBT for late-life GAD in primary care. We hypothesized that CBT would improve outcomes relative to enhanced usual care (EUC) at 3 months posttreatment and that gains would be maintained or enhanced over long-term follow-up (3-15 months).

The study was approved by the institutional review boards of The University of Texas Health Science Center at Houston and Baylor College of Medicine. Participants provided written informed consent for all procedures and received compensation of $20 for each completed assessment.

From March 2004 to August 2006, we recruited participants 60 years or older. Total recruitment duration, however, was only 26 months because of two 2-month interruptions. In August 2004, the primary academic affiliation of the project changed from The University of Texas Health Science Center at Houston to Baylor College of Medicine, and recruitment resumed in October 2004; recruitment was discontinued from September 2005 through October 2005 because of the significant impact of hurricanes Katrina and Rita in the Houston area.

Most recruitment occurred through the Kelsey-Seybold Clinic, a large, multispecialty medical organization in the greater Houston, Texas, area. For 6 months, participants were also recruited from the Baylor College of Medicine Family Medicine Clinic (Houston, Texas). Recruitment occurred through patient self-referral and physician referral. Educational brochures were placed in waiting and examination rooms, health information centers, and patient newsletters. Informational letters also were mailed to randomly selected patients in the appropriate age range. Physician referrals were encouraged by presentations at staff meetings, electronic communications with clinicians, and individual networking.

Referred patients were asked 2 anxiety screening questions from the Primary Care Evaluation of Mental Disorders.²¹ Individuals responding affirmatively were scheduled to review informed consent, collect demographic data, and receive the Mini-Mental State Examination²² and the Structured Diagnostic Interview for the Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition).²³ All interviewers administering the structured interview (psychology staff, graduate students, interns, and fellows) received extensive training. In addition, all interviews were audiotaped, and 20% were rated by a second clinician. Adequate diagnostic agreement was suggested for GAD (κ = 0.64), depression (major depression and dysthymia, κ = 0.75), social phobia (κ = 0.81), and specific phobia κ = 0.64). Patients with a principal or coprincipal diagnosis of GAD according to the Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition) were included. Patients with Mini-Mental State Examination scores less than 24 were excluded, as were patients with active substance abuse, psychosis, or bipolar disorder. Race and ethnicity were self-identified by the patient according to categories provided by research staff. These data were collected to facilitate conclusions about generalizability of the data.

Primary Outcomes. Primary outcomes assessed worry and GAD severity. Worry severity was measured with the Penn State Worry Questionnaire (PSWQ),²⁴ a 16-item, self-report scale. GAD severity was evaluated with the Generalized Anxiety Disorder Severity Scale (GADSS),²⁵ a 6-item, clinician-rated scale. The PSWQ has good psychometrics among older adults²⁶ and has been a primary outcome measure in psychosocial trials of late-life GAD.^{27 - 28} No minimal clinically significant difference has been established for the PSWQ, but prior positive clinical trials have demonstrated a change of 8.5 points on the measure following CBT.^{28 - 29} This value was used here to indicate meaningful change. The GADSS also has adequate psychometric support for older adults,²⁹ although no data from this measure have been reported from clinical trials of GAD. Here, a change of 2.0 points on the GADSS was considered meaningful, given that a difference of 2 points significantly differentiated patients referred to a clinical trial who did and did not meet criteria for GAD.²⁹

Secondary Outcomes. Secondary outcomes assessed coexistent anxiety and depressive symptoms as well as physical/mental health status. Anxiety severity was measured with the Structured Interview Guide for the Hamilton Anxiety Scale (SIGH-A).³⁰ The SIGH-A has adequate psychometric support among older adults.³¹ Severity of depressive symptoms was measured with the Beck Depression Inventory II (BDI-II).³² The BDI-II has been used with community samples of older adults³³ and older adults with anxiety.²⁰ Health quality of life was measured with the Medical Outcomes Study 12-item Short Form Health Survey, which yields 2 summary scores developed from the original measure,³⁴ the mental component score (MCS) and the physical component score.³⁵ Reliability and validity of the short form survey have been established in older adults.³⁶

Other Treatments. Medication use was assessed with patient self-report questions about the type and frequency of medications used over the prior 3 months.^{37 - 38} Medications classified as antianxiety (benzodiazepine, buspirone, or other) or antidepressant (selective serotonin reuptake inhibitor, nonselective serotonin reuptake inhibitor, tricyclic antidepressant, or other) were considered. Patients also were asked to report the number of outpatient medical and mental health visits and to indicate whether they had spoken with their primary care physician (PCP) about emotional issues or received a mental health referral over the past 3 months.

Data Collection. All outcome measures were administered via telephone by a master’s- or postdoctoral-level independent evaluator unaware of treatment assignment. (Given the nature of the intervention and control, patients and therapists were not blinded to group assignment.) Independent evaluators readministered outcome evaluations at 3 months posttreatment and at 3-month intervals over a 12-month follow-up period. The third author (D.M.N.) held regular calibration meetings with independent evaluators. All assessment sessions were audiotaped, and a random 10% were reviewed by a different independent rater. Interrater agreement on the GADSS²⁹ and the SIGH-A³¹ was excellent (intraclass correlation coefficient = 0.99 for GADSS; intraclass correlation coefficient = 0.95 for SIGH-A). Psychometric properties for telephone-based assessment instruments were comparable to in-person versions.³⁹

Five experienced therapists provided CBT in up to 10 individual sessions over 12 weeks. Therapists were 3 master’s-level therapists with at least 2 years of CBT experience, 1 predoctoral intern with more than 3 years' experience in CBT for anxiety, and 1 post−bachelor’s-level therapist (P.D.W.) with 5 years' experience in CBT for late-life anxiety.

Therapist training involved treatment of 2 nonstudy patients, with competence and adherence ratings of at least 6 (on a 0-8 scale). Cognitive behavior therapy included education and awareness, motivational interviewing, relaxation training, cognitive therapy, exposure, problem-solving skills training, and behavioral sleep management.⁴⁰ Brief telephone booster sessions were offered at 4, 7, 10, and 13 months. Sessions were audiotaped, and 20% were rated by 2 independent treatment-integrity experts, Hopko and Diefenbach, who coauthored the treatment manual⁴⁰ on which the CBT used in this study was based. Neither participated in the study in other ways. Ratings suggested excellent adherence (7.7 [SD, 0.55]; range, 0-8) and competence (7.3 [SD, 0.67]; range, 0-8).

Patients randomized to receive EUC were telephoned biweekly during the first 3 months of the study by the same therapists to provide support and ensure patient safety. Calls lasted approximately 15 minutes. Therapists reminded patients to call project staff if symptoms worsened and suggested contacting their PCPs for medical problems. A supervisor (M.A.S., N.L.W., M.E.K.) and the patient's PCP were notified of patients needing immediate psychiatric care. All calls were audiotaped, and 20% (n = 93) were reviewed by the first author (M.A.S.). Only 3 protocol deviations were identified.

Patients randomized to receive CBT completed a mean of 7.4 (SD, 1.91) sessions in the primary care clinic. Patients randomized to receive EUC received a mean of 4.3 (SD, 1.26) telephone check-ins. After the first session in each condition, patient expectancy was assessed with a single item asking patients to rate the amount of improvement they expected (0% = none, 100% = complete improvement).⁴¹ Communication with the PCP occurred via notes filed in the research section of the written medical record. Initial notes indicated diagnoses assigned and inclusion/exclusion status. For excluded patients, potential referrals were provided to the PCP and the patient. For included patients, notes encouraged care as usual and indicated whether patients had received a CBT session or an EUC contact.

Patients were initially randomized according to a 1:1 ratio within blocks of 10 to receive CBT or EUC. More patients were randomized to CBT (n = 70) than to EUC (n = 64). Inspection of data at study midpoint (50% recruitment completed) revealed a disproportionate random assignment of Hispanic patients to receive EUC. A stratified randomization schedule was then instituted for Hispanic patients, with 80% assignment to CBT, to ensure equivalence across groups. Non-Hispanic patients were maintained on the original 1:1 randomization schedule. The randomization scheme was generated by the study statistician (H.M.R.) using a random-number generator, and assignments were placed in numbered, sealed envelopes. At completion of baseline assessment, a study research assistant opened the next envelope in sequence and assigned the participant to a treatment condition.

Before conducting outcome analyses, we compared patients in the CBT and EUC subgroups on pretreatment demographic variables, clinical characteristics, and medication use, using χ² analyses and t tests (Table 1). Primary analyses then examined posttreatment outcomes by comparing group differences at 3 months, using a between-groups analysis of covariance, with pretreatment assessment as the covariate. Initial analyses were by intention-to-treat (ITT), using the Proc MI and MIANALYZE multiple imputation procedures in SAS version 9.2 (SAS Institute Inc, Cary, North Carolina) to address missing data. Another set of analyses repeated these comparisons only with observed data, using random regression methods (ie, completer analysis using the SAS Proc Mixed procedure). Secondary analyses examined long-term outcomes with a repeated-measures analysis of covariance procedure (SAS Proc Mixed), again with the pretreatment assessment as the covariate. To control for multiple comparisons within clusters of outcomes (primary outcomes, coexistent anxiety-depression, health quality of life), each of which included 2 variables, critical α was set at P<.025. All significance testing was 2-sided.

Because 3 couples participated in the study (with both partners randomized to receive the same intervention; 2 couples to receive CBT, 1 couple to receive EUC), all analyses were rerun with one partner from each couple randomly removed to assess the possible influence of correlated data. These analyses did not result in different findings and therefore are omitted. Changes in the use of antidepressant and antianxiety medications during the trial were examined with χ² analyses, and exploratory analyses examined the role of expectancies in predicting outcome (0% = none, 100% = complete improvement).

Treatment response was defined by meaningful change in scores on the PSWQ and GADSS (8.5 for the PSWQ, 2.0 for the GADSS; see “Primary Outcomes”). The proportions of patients classified as treatment responders according to these cutoffs were determined at 3 and 15 months. Group differences in the ITT and completer samples were tested using χ² analyses. For ITT analyses, patients with missing data were classified as nonresponders.

Power calculations were based on the PSWQ, which is a primary outcome in CBT trials of late-life GAD. The median SD across these trials is 10.1. This value was considered in combination with an expected moderate effect size (d = 0.60; minimal detectable difference, 6.2), desired 80% power, and α = .025. Given these parameters and a potential 30% attrition, our goal was to include 150 participants so that 53 per group would be available for analysis. The included sample (n = 148) was 99% of this target. The randomized sample (n = 70 CBT, n = 64 EUC) and the completer sample (n = 115) both exceeded the required 53 per group.

A total of 968 potential participants were referred (75% self-referred), and 381 provided consent (Figure). Of these, 68 (18%) dropped out or were excluded before diagnostic evaluation because of negative responses to screening questions on the Primary Care Evaluation of Mental Disorders (n = 14), lack of interest (n = 35), or logistic problems (n = 19). Of the remaining 313 patients, 154 were excluded and 11 were included as nonstudy clinical training cases but excluded from analyses. A total of 148 patients met inclusion criteria (95% self-referred); of these, 14 (9%) dropped out before randomization, leaving 134 patients with principal (n = 86) or coprincipal (n = 48) GAD for randomization.

Relative to patients who provided consent but were not randomized (n = 247), randomized patients (n = 134) were younger (mean age, 66.9 years [95% confidence interval {CI}, 65.9-67.9] vs 69.0 years [95% CI, 68.2-69.9]; P = .002), more educated (mean years of education, 15.9 [95% CI, 15.4-16.4] vs 15.1 [95% CI, 14.7-15.5]; P = .02), and more likely to be women (75% [105/134] vs 68% [168/246], P = .04). Randomized patients also had higher baseline PSWQ scores than other patients for whom baseline severity data were available (n = 167) (57.2 [95% CI, 55.4-59.0] vs 46.5 [95% CI, 44.6-48.3], P < .001).

Dropout during active treatment (0-3 months) was significantly lower for patients receiving CBT than for those receiving EUC (5.9% [4/70] vs 21.9% [14/64], P = .006), primarily because of self-reported dissatisfaction with random assignment (CBT, n = 0; EUC, n = 9). Attrition over long-term follow-up (3-15 months) was comparable for CBT and EUC (CBT, 12.9% [9/70]; EUC, 9.4% [6/64]; P = .52). Total attrition over 15 months was 24.6% (n = 33) and did not differ significantly between groups (CBT, 18.6% [13/70]; EUC, 31.3% [20/64]; P = .09). Reasons for dropout are described in the Figure. Analyses of variance revealed no group × drop status differences on any baseline measure.

Sample characteristics are included in Table 1. Patients in the CBT and EUC groups differed with regard to baseline PSWQ scores. Analyses of primary outcomes (assessed using the PSWQ and GADSS) included baseline PSWQ as a covariate.

Mean observed scores at 0 and 3 months are presented in Table 2. Imputed means and CIs were not statistically or substantially different from observed values.

Primary Outcomes. Intention-to-treat analyses indicated significantly greater improvement on the PSWQ in patients completing CBT than in those completing EUC (45.6 [95% CI, 43.4-47.8] vs 54.4 [95% CI, 51.4-57.3], P < .001) (Table 2). Mean change was 7.7 points in the CBT group and 3.2 points in the EUC group. Group differences on the GADSS were not significant, with a mean change of 2.8 points in the CBT group and 1.4 points in the EUC group. Completer analyses resulted in the same pattern of statistical results.

Secondary Outcomes. Intention-to-treat analyses suggested significantly greater improvement on the BDI-II in patients completing CBT than in those completing EUC (10.2 [95% CI, 8.5-11.9] vs 12.8 [95% CI, 10.5-15.1], P = .02); this was true for the MCS as well (49.6 [95% CI, 47.4-51.8] vs 45.3 [95% CI, 42.6-47.9], P = .008) (Table 2). Changes on the SIGH-A and PCS were not significantly different between groups. Completer analyses suggested similar patterns of results.

Other Treatments. Patients who completed CBT and EUC did not differentially add or increase dosages of antianxiety medication during active treatment (4.6% [3/65] vs 2.0% [1/50], P = .45). Rates of discontinuation or reduced dosages of these medications also were equivalent across groups (7.7% [5/65] vs 4.0% [2/50], P = .41). Similarly, rates of increased doses or new antidepressant medication (12.3% [8/65] vs 6.0% [3/50], P = .25) and decreased doses or medication discontinuation (7.7% [5/65] vs 2.0% [1/50], P = .17) were similar in patients completing CBT and EUC.

Patients in the CBT and EUC groups also had equivalent numbers of outpatient medical visits (2.7 [95% CI, 2.1-3.2] vs 2.3 [95% CI, 1.6-3.0], P = .85) and similarly infrequent mental health visits (0.5 [95% CI, 0.0-1.0] vs 0.7 [95% CI, 0.2-1.3], P = .82). Very few patients in either group received a mental health referral (7.7% [5/65] vs 8.0% [4/50], P = .95) or spoke with their PCP about emotional problems (20.0% [13/65] vs 18.0% [9/50], P = .82).

Primary Outcomes. Analyses indicated continued improvement in PSWQ scores over long-term follow-up for patients in both groups (time effect [Table 2]). Covariate-adjusted posttreatment group differences also continued over the long term for the PSWQ (group effect [Table 2]). These differences suggest that no additional differential group response was present and that the initial treatment effectiveness at the end of active treatment was maintained throughout the 12-month follow-up period.

Secondary Outcomes. Posttreatment effects on the BDI-II and the MCS were maintained over the subsequent 12 months (group effects [Table 2]). Patients who completed CBT and EUC did not increase doses or add antianxiety medication at different rates over 3 to 15 months (12.5% [8/64] vs 16.7% [8/48], P = .53). Rates of decreasing doses or discontinuation of these medications also were equivalent (14.1% [9/64] vs 16.7% [8/48], P = .70). Antidepressant medications also were added or doses increased (18.8% [12/64] vs 14.6% [7/48], P = .56) and discontinued or doses reduced (18.8% [12/64] vs 12.5% [6/48], P = .37) at similar rates over long-term follow-up.

Mean percentages of treatment responders according to ITT analyses of meaningful change scores (8.5 for the PSWQ, 2.0 for the GADSS) are reported in Table 3. Treatment response rates according to the PSWQ were higher at 3 months in patients completing CBT than in those completing EUC. Completer analyses revealed no group differences at 3 or 15 months.

Average treatment expectancies were significantly higher for patients who completed CBT than for those who completed EUC (78% [95% CI, 74%-83%] expected improvement vs 66% [95% CI, 58%-74%], P = .007). When expectancies were added to the previously tested models as covariates, effects for treatment remained significant for the PSWQ (P < .001), but group differences on the BDI-II (P = .05) and the MCS (P = .04) only approached significance.

This study is the first to suggest that CBT can be useful for managing worry and associated symptoms among older patients in primary care. Patients receiving CBT had greater improvements in worry severity, depressive symptoms, and general mental health than those receiving EUC. Mean change in worry severity over time was slightly lower than in other clinical trials of CBT for late-life anxiety.^{28 - 29} Mean change in GAD severity following CBT was meaningful but not significantly different than that following EUC. Effect sizes for symptom improvement were comparable to or greater than those in primary care studies of younger adults with GAD or panic disorder^{17 ,42} and older adults with depression.¹⁸

Positive findings in this trial are particularly notable, given that other primary care studies involve full collaborative-care models incorporating medication as well as cognitive behavioral treatments. In this study, comparable effects were found with CBT alone, as in a subset of data from the Improving Mood Promoting Access to Collaborative Care Treatment (IMPACT) trial.⁴³ The findings of the current study also are the first to demonstrate positive effects of treatment for GAD in primary care. A subset of data from a study by Rollman et al¹⁷ showed no significant benefits of collaborative care for young patients with GAD only (without coexistent panic disorder), relative to usual care. In the current trial, posttreatment symptom improvements were maintained or improved over 1 year of follow-up.

Treatment response rates based on worry severity as measured using the PSWQ were higher at 3 months in patients receiving CBT than in those receiving EUC, but treatment-group differences were not maintained at 15 months. Patterns of response suggest maintenance of gains with CBT but a slight increase in treatment response with EUC. Increased booster sessions over the long term might be beneficial for continued improvement following CBT.

The GADSS failed to demonstrate treatment effects, suggesting a potential limitation of this scale as a measure for late-life GAD. Exploratory analysis also suggested a meaningful role of expectancies on secondary outcomes. Future work is needed to more fully examine possible moderator and mediator variables in response to CBT.

Significant strengths of this study include careful selection and diagnosis of patients, breadth of outcome assessment, excellent treatment integrity, low attrition (6%) in the CBT group, and significant improvement in primary as well as secondary outcomes. Certain design features, however, limit translational value and potential sustainability for primary care settings. First, the sample was not representative of older patients in primary care with regard to age, sex, and education. Moreover, randomized patients were younger, more highly educated, and more likely to be women than nonrandomized patients. These characteristics may reflect both the recruitment setting (insurance-based care) and the self-referred nature of most participants. Primary care patients with anxiety who are recruited through physician referral using prompts from electronic medical records are more diverse demographically and have higher levels of anxiety.⁴⁴

Second, clinicians providing CBT and EUC had significant expertise in late-life anxiety and CBT. Treatment with CBT was delivered during weekly in-person sessions of approximately 1 hour. This type of service is rarely available in primary care. Also, the telephone contacts in the EUC condition did not completely control for the increased attention for the patients receiving CBT. Communication with PCPs was limited to written notes in a research section of the medical record. No electronic medical record was available to facilitate integration of treatment with ongoing care.

Cognitive behavior therapy is useful for older adults with GAD in primary care. This study paves the way for future research to test sustainable models of care in more demographically heterogeneous groups. In future studies, it will be important to examine the impact of treatment delivered by clinicians without specialized CBT expertise. Improved integration with ongoing care would be facilitated through use of an electronic medical record to identify patients and communicate with clinicians,³⁹ and collaborative models of care that incorporate both CBT and medication need to be tested.