Journal of Affective Disorders 181 (2015) 33–40

Contents lists available at ScienceDirect

Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad

Research report

Predictors of short- and long-term avoidance in completers of inpatient group interventions for agoraphobia Asle Hoffart a,b,n, Tuva Øktedalen a,b, Karol Svanøe c, Liv M. Hedley d, Harold Sexton d a

Research Institute, Modum Bad, N-3370 Vikersund, Norway Department of Psychology, University of Oslo, PB 1094, Blindern, N-0317 Oslo, Norway c Buskerud and Vestfold University College, PB 2243, N-3103 Tønsberg, Norway d Research Institute, Modum Bad, N-3370 Vikersund, Norway b

art ic l e i nf o

a b s t r a c t

Article history: Received 19 December 2014 Received in revised form 14 March 2015 Accepted 6 April 2015 Available online 11 April 2015

Background: Little is currently known about predictors of follow-up outcome of psychological treatment of agoraphobia. In this study, we wished to examine predictors of short- and long-term avoidance after inpatient group interventions for agoraphobia. Methods: Ninety-six (68%) of 141 agoraphobic patients (74% women) who had completed treatment in two open and one randomized controlled trial (RCT) were followed up 13 to 21 years after start of treatment. Results: Major depression at pre-treatment predicted less short-term (up to one year after end of treatment) improvement in agoraphobic avoidance. Working and being married/cohabiting at pretreatment predicted greater long-term (across one-year, two-year, and 13–21 years follow-up) improvement. In contrast, the duration of agoraphobia, amount of Axis I and II co-morbidity, being diagnosed with avoidant, dependent, and obsessive-compulsive personality disorder, and the use of antidepressants and benzodiazepines the month before intake to treatment, were unrelated to short-term as well as long-term outcome. Limitations: As many as 31.9% of the included patients did not attend long-term follow-up and the power of the study was limited. The long time period between the two and 13–21 year follow-ups is a limitation, in which it is difficult to assess what actually happened. Although all the patients received some form of CBT, there was variability among the treatments. Conclusions: The only short-term predictor identified represented a clinical feature, whereas the longterm predictors represented features of the patients' life situation. The limited power of the study precludes the inference that non-significant predictors are unrelated to follow-up outcome. & 2015 Elsevier B.V. All rights reserved.

Keywords: Agoraphobia Psychological treatment Latent growth curve modeling Long term follow-up

1. Introduction Agoraphobia – with or without panic disorder – has a lifetime prevalence rate of 5–6% (Kessler et al., 2005). It is associated with an increased risk of suicide, severe social impairment, and significant comorbidity with other mental and medical disorders (Milrod and Busch, 1995). Recent meta-analyses suggest that both cognitive therapy and exposure therapy are efficacious (Norton and Price, 2007; Stewart and Chambless, 2009) appearing to be equally effective (Norton and Price, 2007). Some evidence also supports the efficacy of psychodynamic therapy for panic disorder (Milrod et al., 2007).

n Corresponding author at: Research Institute, Modum Bad, N-3370 Vikersund, Norway. Tel.: þ 47 32 74 98 61; fax: þ 47 32 74 98 68. E-mail address: [email protected] (A. Hoffart).

http://dx.doi.org/10.1016/j.jad.2015.04.015 0165-0327/& 2015 Elsevier B.V. All rights reserved.

However, it is important to know to what degree initial treatment gains are maintained over time. Furthermore, knowledge of pretreatment predictors of long-term outcome may inform theory and assist in treatment planning (Brown and Barlow, 1995). No predictor studies of psychological therapies for panic disorder/agoraphobia have extended beyond a two-year follow-up. Durham et al. (2005) examined a wider sample of anxiety disorder patients (including 189 panic disorder patients, 116 generalized anxiety disorder patients, and 31 post-traumatic stress disorder patients) at 2–14 years follow-up. These patients had participated in eight trials comparing cognitive-behavioral treatment (CBT) and non-CBT. Unemployment and severity of functional impairment predicted poorer outcome in terms of clinical significant change. Marriage/cohabiting correlated negatively with a dimensional long-term symptom factor. In short-term follow-up studies, no variables have consistently been found to predict follow-up outcome. However, at least some studies have provided support for the

34

A. Hoffart et al. / Journal of Affective Disorders 181 (2015) 33–40

following clinical and demographic predictors of poorer outcome in panic disorder/agoraphobia patients: duration (Faravelli et al., 1995), co-morbid major depression (Cowley et al., 1996), benzodiazepine use (Brown and Barlow, 1995), antidepressant use (Barlow et al., 2000), and the presence of DSM Cluster C personality disorders (avoidant, dependent, obsessive-compulsive; Hoffart and Hedley, 1997). Also treatment response variables such as degree of compliance with treatment, quality of therapeutic alliance, and immediate treatment response may influence follow-up outcome, but their ability to predict such outcome in agoraphobia has hardly been studied (for generalized anxiety disorder, however, see Durham et al., 2009). Thus, little is currently known about predictors of follow-up outcome of psychological treatment of agoraphobia. Studies of more short-term follow-up have yielded inconsistent results and no study has examined predictors of long-term outcome in a purely agoraphobic sample. In the present study, we wished to contribute to amending this lack of knowledge by examining pre-treatment clinical and demographic predictors of short- and long-term (13–21 years) avoidance in agoraphobic patients who had previously participated in one of three trials at our site. These were the randomized controlled trial (RCT) of 5weeks with cognitive therapy or guided mastery therapy for agoraphobia (Hoffart, 1998) and two open trials. One of the open trials utilized exposure in a first 5-week phase and psychodynamic therapy in a second 6-week phase (Hoffart et al., 1995) while the other utilized cognitive therapy in a first 5-week phase and schema therapy in a second 6-week phase (Hoffart et al., 2002). Psychodynamic and schema therapy addressed potential vulnerabilities for developing agoraphobia and co-morbid disorders and were intended to protect against relapse. A report on the long-term outcome in the RCT has been published elsewhere, showing that 56.5% of those attending follow-up no longer had a panic disorder/agoraphobia diagnosis (Hoffart et al., in press). In this study, our focus was the combined sample of agoraphobic patients from all the three trials. Common to these patients was that they had received standardized medication-free group-based and agoraphobiafocused CBT. We examined predictors that had received some support in previous studies. Our research questions included the following: will not working, not being married or cohabiting, a longer duration of the disorder, more Axis I and II co-morbidity, major depression, avoidant, dependent, obsessive-compulsive personality disorder, using benzodiazepines, and/or using antidepressants at pre-treatment predict poorer short-term (up to one year after end of treatment) and/or long-term (from one to 13–21 years after end of treatment) course of avoidance?

2. Method 2.1. Participants Participants were selected among patients who had been treated in a specialized program for agoraphobia at an inpatient clinic. Due to scattered settlement in Norway, such residential adaptations of programs developed in an outpatient setting assist in providing specialized and intensive treatment opportunities for the entire Norwegian population. Thus, the patients were referred due to lack of specialized local treatment or because outpatient treatment attempts had failed. Patients from three different trials, two open and one RCT, were included in the present study of long-term outcome. In the first open trial, participants were allocated from January 1989 to November 1990 according to following inclusion criteria: (a) DSM-III-R criteria for panic disorder with agoraphobia or agoraphobia without history of panic disorder, (b) symptoms of agoraphobia were considered to be the main problem, and (c) age 65 years. These patients received exposure (E) in a first 5-week phase and psychodynamic therapy (PT) in a second 6-week phase. In the RCT (allocation from April 1992 to April 1993), the participant inclusion criteria were (a) DSM-III-R criteria for panic

disorder with agoraphobia, (b) degree of agoraphobia was moderate or severe, (c) the symptoms of agoraphobia were considered to be the main problem, and (d) age 20–65 years. These patients were randomly allocated to five weeks of either cognitive therapy (CT) or guided mastery therapy (GMT). Six groups were admitted sequentially, alternating between cognitive or guided mastery therapy. In the second open trial (allocation from January 1996 to February 1997), the participant inclusion criteria were (a) DSMIV criteria for panic disorder with or without agoraphobia or agoraphobia without history of panic disorder, (b) 22–65 years old, and (c) they presented at least one problem related to DSM-IV Cluster C personality disorder, i.e., they described chronic problems independent of their Axis I disorders and that expressed the core content of the Cluster C personality disorders. These patients received CT in a first 5-week phase and schema therapy (ST) in a second 6-week phase. As a majority of the participants had tried medication previously without satisfactory effect and the use of medication could potentially interfere with cognitive and behavioral learning, psychotropic medication was not utilized during the inpatient treatment period. Medications were tapered and discontinued prior to admission. Patients completing treatment were eligible for the present study. The chart of patient flow in Fig. 1 shows that the dropout rates from the treatments were low (13 of 154, on average 8.4%) and that a total of 141 patients from the three trials were included. After their two-year follow-up, these patients were contacted and asked to give informed consent to allow for subsequent follow-up. The patients were fully informed and the follow-up was conducted in compliance with the Regional Ethical Committee. Forty-five (31.9%) of the 141 included patients did not attend this follow-up. Reasons for non-attendance, listed in Fig. 1, indicate that 17 (12.1%) withdrew, while 28 (19.9%) did not attend for other reasons (death, sickness, unavailability). The 96 patients who attended the long-term follow-up were compared on pre-treatment characteristics to the 45 non-attending patients to determine the representativeness of the attenders in relation to the original sample (see Table 1). The two groups of patients were similar, with the exception that relatively more attenders were working (42.7% vs. 24.4%), X²(1) ¼4.58, p o.05; and initially had dysthymia (18.8% vs. 4.4%), X²(1) ¼5.15, p o.05. The attendance rates were similar across the subsamples, 38 of the 53 (71.7%) patients who received Eþ PT, 16 of the 23 (69.6%) patients who received CT, 15 of the 23 (65.2%) patients who received GMT, and 27 of the 42 (64.3%) patients who received CTþST; X²(3) ¼..61, ns. Thirteen (34.2%) of the 38 Eþ PT and 3 (11.1%) of the 27 CTþ ST patients had agoraphobia without a history of panic disorder at pre-treatment. The original inclusion criteria of the RCT excluded this diagnosis. The proportion married/cohabiting at pre-treatment, 68.8% in the total sample, was similar to the one of the Norwegian population at the time, 67% (Norwegian Social Science Data Services, 1995). However, the proportions having a lower occupational level, 72.9% of the attending and 86.7% of the non-attending patients, were higher than the 63% estimated in the population (Norwegian Social Science Data Services, 1995). This suggests that the sample had a bias toward social disadvantage. 2.2. Treatments The patients were admitted to closed treatment groups with 8 members in each. Except for the intake session, patients had no individual sessions during the treatments that focused on panic disorder and agoraphobia. The first week consisted of education concerning the treatment model with patients setting specific goals for their treatment. Also, within this first week, any

A. Hoffart et al. / Journal of Affective Disorders 181 (2015) 33–40

35

Fig. 1. Patient enrollment, randomization, and treatment.

2.4. Interview-based measure

remaining psychotropic medication was discontinued. In weeks 2 to 4 agoraphobia was focused upon according to the specific therapy model applied (E, CT or GMT). In the fifth week, patients returned home to test their newly acquired skills. Subsequently, the E þPT and CTþST patients attended 6-week of PT or ST, respectively, while the CT and GMT patients returned simply for 2 days, discussing experiences and setting personal goals for the next year. More information about the therapies, therapists, training, supervision, and independent ratings of the therapists' skill levels are available (Hoffart et al., 1995., 2002; Hoffart, 1998).

The interview-based Phobic Avoidance Rating Scale – Separation subscale (PARS-Sep; Hoffart et al., 1989) measures avoidance of situations involving separation from safe persons (e.g., walk away from home alone, travel by bus or train alone) on a 0 to 4 scale. The two long-term follow-up interviewers (third and fourth author) independently rated tapes of 10 of each others' randomly selected PARS-Sep interviews. The intraclass correlation (ICC (3,1); Shrout and Fleiss, 1979) was .97.

2.3. Self-report measures

2.5. Diagnostics

The self-report Mobility Inventory for Agoraphobia (MI; Chambless et al., 1985) measures agoraphobic avoidance of a range of situations, both when the patients are alone (MI-AAL) and when they are accompanied (MI-ACC), on a 1–5 scale. The reliability and the concurrent and construct validity of the MI has been supported (Chambless et al., 1985), particularly for the MI-AAL subscale (Chambless et al., 2011) that was used as the primary outcome measure in the present study. In the present sample, Cronbach's alpha ranged from .95 to .97 across the five assessments. The self-report State-Trait Anxiety Inventory-Y2 (STAI-Y2; Spielberger, 1983) provides scores for trait anxiety. The self-report Beck Depression Inventory (BDI; Beck et al., 1988) measures degree of depressive symptoms. Alpha ranged from .91 to .96 on these two measures across assessments.

The Structural Clinical Interview for DSM-III-R (Spitzer et al., 1988) or DSM-IV (First et al., 1995, 1994) (SCID-I and II) was used for the assessment of Axis I and II disorders. The DSM-III-R interview was used in the Eþ PT, CT, and GMT samples and the DSM-IV interview was used in the CTþ ST sample. There was satisfactory inter-rater reliability for the baseline interviews (Hoffart et al., 1995, 2002; Hoffart, 1998). The diagnostic interviews at the long-term follow-up focused upon the period since the two-year follow-up and were conducted by the third and fourth author. Twenty videotaped interviews – 10 from each interviewer – were randomly drawn for a reliability check. The kappa values were .98 for panic disorder with agoraphobia, social phobia, and specific phobia, and .75 for agoraphobia without panic disorder. As very

36

A. Hoffart et al. / Journal of Affective Disorders 181 (2015) 33–40

Table 1 Participants characteristics at pre-treatment. Variable

Age at intake Duration of agoraphobia

Female Married/cohabiting Lower occupational levela Working last year ( 450%) Previous psychiatric treatment Use of benzodiazepines last month Use of antidepressants last month Axis I disorders Panic disorder with agoraphobia Agoraphobia without panic disorder Social phobia Generalized anxiety disorder Obsessive compulsive disorder Simple phobia Hypochondriasis Somatoform pain disorder Undifferentiated somatoform disorder Major depression, current Major depression, lifetime Dysthymia Alcohol dependence/abuse Non-alcohol dependence/abuse Axis II disorders Paranoid Schizoid Schizotypal Antisocial Borderline Histronic Narcissistic Avoidant Dependent Obsessive compulsive a

Attenders (n¼ 96)

Non-attenders (n¼ 45)

M

SD

M

SD

39.3 13.0

(9.0) (10.1)

40.3 13.1

(9.5) (11.1)

n

(%)

n

(%)

68 70 70 41 83 60 25

(70.8) (72.9) (72.9) (42.7) (86.5) (62.5) (26.0)

36 27 39 11 40 29 18

(80.0) (60.0) (86.7) (24.4) (88.9) (64.4) (40.0)

80 16 37 22 25 25 14 8 5 40 59 18 10 23

(83.3) (16.7) (38.5) (22.9) (26.0) (26.0) (14.6) (8.3) (5.2) (41.7) (61.5) (18.8) (10.4) (24.0)

40 5 18 8 11 14 5 6 0 23 30 2 8 10

(88.9) (11.1) (40.0) (17.8) (24.4) (31.1) (11.1) (13.3) (.0) (51.1) (66.7) (4.4) (17.8) (22.2)

4 0 0 1 3 4 1 40 12 11

(4.2) (.0) (.0) (1.0) (3.1) (4.2) (1.0) (42.7) (12.5) (11.5)

3 1 1 0 4 2 1 19 8 6

(6.7) (2.2) (2.2) (.0) (8.9) (4.4) (2.2) (42.2) (17.8) (13.0)

Unemployed, skilled/unskilled worker, uneducated functionaries.

few other diagnoses were assigned in the 20 interviews, kappa values were not calculated for these. A measure of co-morbidity was computed by summing the number of co-morbid Axis I and II diagnoses. The mean number of such diagnoses was 3.17 (SD¼2.12, range¼ 0–8) and this variable was fairly normally distributed. Level of co-morbidity did not differ among the four subsamples/treatment groups, F(3, 92)¼1.10, ns. 2.6. Procedure Patients had been assessed at evaluation, at admission to hospital/ treatment start (pre-treatment, on average two months after the evaluation), and immediately after the agoraphobia-focused treatment (5 weeks). The patients were invited as a group to short (2–4 days) follow-up stays at the hospital and assessed one and two years after completing treatment. One exception to this was that patients of the RCT received mailed questionnaires to complete at two-year follow-up. At long-term follow-up, patients were approached by mail, inviting them to contact the researcher for an appointment. If no response occurred within two weeks, the patients were contacted by phone. Participants could come to the clinic or meet at another place of their choice. Travel expenses and any loss of income incurred as a result of their participation were refunded. No incitements were given for participation. If it seemed apparent that a participant had a current need for treatment, the interviewer assisted with a referral.

The long-term follow-up interviews occurred on average 17.55 (SD¼2.73, range 12.8–21.2) years post-treatment, were conducted individually and videotaped. A comprehensive interview was conducted in addition to the administration of the MI and other selfreport measures. It included the SCID-I and –II (DSM-III-R or DSM-IV, depending on which DSM version was used at inclusion), targeting the period since the previous SCID interview (one-/two-year followup), the PARS-Sep (administered only at long-term follow-up), and assessment of the frequency of panic attacks for the two last weeks and for the last month. As for the one- and two-year follow-up, psychosocial treatments in the previous year and use of medication the previous month were tracked. The one interviewer (the fourth author) is a Ph.D. psychologist experienced in performing diagnostic interviews and follow-investigations. She was blind to the results of diagnostic assessments at baseline, but not always blind to treatment conditions of the patients, as she had participated in the treatment and supervision related to some patients. The other interviewer (the third author) is a Clinical Social Worker (MSW), who had been trained in conducting DSM diagnostic interviews in previous studies at another site. She was blind to all previous results of this study and had not been involved in the treatments. 2.7. Statistical analysis Latent growth curve modeling with piecewise functions for repeated measurement data was used to investigate the trends for both short-term and long-term outcomes and predictors of these trends. Parameter estimates were derived through maximum likelihood using m-plus version 7.3. Full information maximum likelihood (FIML) was used to handle missing data. The Little's test of missing completely at random (MCAR) (Little, 1988) indicated that data were missing at random (Little’s MCAR test: χ2(4) ¼ 14.161, p ¼.224), and this was therefore assumed. Individuals were allowed to have different intercepts (initial status) and linear slopes (growth) in the analysis, known as the unconditional model if predictors are not included in the model. If the unconditional models fit the data well and the variances of the initial status and linear slopes were statistically significant, we included covariates to predict the two linear slopes in the piecewise model, known as the conditional model. To assess the appropriateness of the growth models, we used the comparative fit index (CFI), Tucker-Lewis index (TLI), standardized root-mean-square residual (SRMR), and the root-meansquare error of approximation (RMSEA). A model is judged as reasonably fitting the data when CFI and TLI is larger than .95 and RMSEA is .06 or smaller (Hu and Bentler, 1999). Different degrees of change were expected for short-term and long-term follow-up and the time intervals between assessments varied greatly. To find the model that best captured the non-linear rate of change in avoidance, we tested three models for the primary outcome measure, the MI-AAL. In the first model, time was represented with a piecewise latent curve model consisting of two latent growth curves with a knot point at one-year follow-up. In this piecewise model, the first latent growth curve represents pre-treatment, post-treatment and 1 year follow-up MI-AAL scores (loadings set to 1, 2 and 3 on this first latent time term), while the second latent growth curve represents the follow-ups (1 year, 2 years, 13–21 years; loadings set to 1, 2, 3 on this second latent time term). In the second model, a quadratic term (time2) was added to a linear elapsed time term. In the third model, the initial loading on the time factor was set to 0 and the final loading to 1, whereas the intermediate loadings were freely estimated. The second model with the quadratic effects of time showed poor fit to the data (χ 2 6 (N ¼ 140) ¼19.078, po .001 with CFI¼.96, TLI¼.96, SRMR¼.052 and RMSEA ¼.125 with 90% confidence interval from.064 to.190). The third model with freely estimated loadings

A. Hoffart et al. / Journal of Affective Disorders 181 (2015) 33–40

37

Table 2 The means (SDs)a and effect sizesb on symptom measures across assessments and treatment conditions. Measure

Pre-treatment

5 weeks

1 year

2 years

13–21 years

Mean

(SD)

n

Mean

(SD)

n

d

Mean

(SD)

n

d

Mean

(SD)

n

d

Mean

(SD)

n

d

MI-AALg E þ PTc CTd GMTe CT þSTf Total

3.73 3.59 3.62 3.34 3.57

(.85) (.64) (.88) (1.06) (.90)

53 23 23 42 141

2.70 2.76 2.90 3.15 2.88

(.86) (.73) (.87) (1.14) (.95)

53 23 23 42 141

 1.19  1.15  .79  .17  .74

2.20 2.28 2.40 2.60 2.36

(1.04) (.82) (.87) (1.16) (1.02)

53 23 23 42 141

 1.58  1.70  1.35  .65  1.25

2.21 2.38 2.42 2.44 2.34

(1.03) (.82) (.84) (1.18) (1.02)

49 20 19 42 130

 1.59  1.77  1.28  .78  1.28

1.86 2.05 2.04 2.41 2.07

(.90) (.99) (1.02) (1.07) (1.00)

38 16 15 27 96

 2.15  1.60  1.53  .81  1.55

STAI-Y2g E þ PTc CTd GMTe CT þSTf Total

57.8 60.0 65.1 62.9 60.9

(10.5) (13.5) (9.5) (12.0) (12.0)

53 23 23 42 141

52.8 51.7 53.0 62.0 55.4

(12.2) (14.1) (15.0) (12.9) (13.8)

53 23 23 42 141

 .43  .58  .89  .07  .42

45.0 47.7 51.5 51.1 48.3

(12.8) (17.1) (16.7) (15.9) (15.3)

53 23 23 42 141

 1.07  .76  .90  .82  .90

41.4 45.7 47.3 46.9 44.8

(12.5) (14.6) (16.3) (15.5) (15.0)

49 20 19 42 130

 1.39  .98  1.24  1.13  1.17

38.0 44.6 46.9 46.4 42.8

(12.1) (16.6) (16.1) (13.4) (14.3)

38 16 15 27 96

 1.78  .92  1.45  1.29  1.40

BDIg E þ PTc CTd GMTe CT þSTf Total

17.1 19.5 24.5 26.0 21.4

(9.4) (10.3) (10.7) (11.3) (11.0)

53 23 23 42 141

9.3 11.4 14.7 15.6 12.4

(9.0) (10.4) (11.7) (13.3) (11.4)

49 20 19 42 130

 .83  .75  .83  .82  .80

7.1 12.6 10.7 11.2 9.8

(6.5) (11.2) (12.5) (9.3) (9.4)

38 16 15 27 130

 1.24  .52  1.21  1.32  1.12

a

Computed for those attending each assessment. Effect sizes computed as Hedges g using pre-treatment as baseline. E þPT ¼ Exposure and Psychodynamic Therapy. d CT¼Cognitive Therapy. e GMT ¼ Guided Mastery Therapy. f ST¼ Schema Therapy. g MI-AAL¼ Mobility Inventory-Avoidance Alone; STAI-Y2 ¼ State-Trait Anxiety Inventory-Y2 (trait version); BDI¼ Beck Depression Inventory. b c

showed somewhat better fit (χ 2 7 (N ¼140) ¼14.890, po .001 with CFI¼.98, TLI¼.97, SRMR ¼.129 and RMSEA ¼.090 with 90% confidence interval from .021 to .153). The piecewise model with two latent growth curves fitted best (χ 2 9 (N ¼ 140) ¼9.208, p o.162; CFI¼.99, TLI ¼.99, SRMR ¼.050, RMSEA ¼.062 with 90% confidence interval from .000 to .136) and was utilized in the further analyses. Thus, this study used a piecewise function for describing segmented change of a latent construct over time. We then examined the pre-treatment predictors in the piecewise model with two latent growth curves representing short-term and long-term change, respectively. The predictors were included in separate models as time-invariant covariates for both latent growth curves and the latent intercept. As this is a first exploratory study of predictors of long-term outcome in agoraphobia after CBT treatment, a significance level of .05, two-tailed, was selected and no adjustment for family-wise error rate was made. Groups were compared at a single assessment point using a one-way ANOVA or independent samples t-test for continuous data and chi-square (Fisher's Exact Test when an expected cell number was below 5) for categorical data. McNemar test was used for dependent (longitudinal) categorical data. Effect sizes (ESs) were computed as Hedges' g for dependent samples (Borenstein et al., 2009).

3. Results 3.1. Improvements and end states The means (SDs) and the effect sizes on the MI-AAL, the STAIY2, and the BDI across assessments for each treatment condition and the overall sample are provided in Table 2. The 96 attending patients had improved from 3.56 (SD ¼.90) at pre-treatment to 2.07 (SD ¼1.00) at long term follow-up, an ES of 1.55, 95% confidence interval [  1.92,  1.18]. The number of patients free of panic attacks the last two weeks before long-term follow-up examination was 68 (70.8%) and the number of patients not having

a panic disorder and/or agoraphobia diagnosis was 60 (62.5%). The rates of patients no longer fulfilling a panic disorder/agoraphobia diagnosis were 27 (71.1%) of the 38 EþPT patients, 9 (56.3%) of the 16 CT patients, 10 (66.7%) of the 15 GMT patients and 16 (59.3%) of the 27 CT þST patients, X² (3)¼1.54, ns. The frequency of comorbid diagnoses exceeding a number of 1 was 12 (12.5%) for social phobia, 4 (4.2%) for generalized anxiety disorder, 4 (4.2%) for obsessive compulsive disorder, 15 (15.6%) for simple phobia, 2 (2.1%) for hypochondriasis, 5 (5.2%) for somatoform pain disorder, 14 (14.6%) for major depression, 4 (4.2%) for non-alcohol dependence/abuse, 12 (12.5%) for avoidant and 2 (2.1%) for obsessive compulsive personality disorder. Thus, most of these diagnoses were reduced by more than two third compared to pre-treatment (see Table 1). The mean level on the PARS was 1.02 (SD ¼1.13). There was less use of psychosocial treatments at the follow-ups than before intake, and only 10 patients (10.4%) had received psychosocial treatment during the year prior to follow-up. Also the use of benzodiazepines (BZs) was reduced from before treatment to all the follow-ups, whereas the use of antidepressants was unchanged. There was a rather prevalent use of BZs (n ¼38, 39.6%) and antidepressants (n ¼35, 36.5%) at long term follow-up. Of those 38 using BZs, 14 were on stable doses, whereas the other 24 used them on an intermittent basis, 8 of these more seldom than monthly. There were no between-condition differences on these variables. At long-term follow-up, those who used psychosocial treatment and those who used BZs had increased MI-AAL scores, t(95) ¼3.62 and t(95) ¼ 3.40, respectively, whereas those who used antidepressants did not differ from those who did not, t(95) ¼ .89. 3.2. The unconditional latent growth curve model The unconditional piecewise model with two latent growth curves showed reasonable fit to the data (χ 2 6 (N¼140)¼9.208, po.162 with CFI¼ .99, TLI¼.98, SRMR¼.050 and RMSEA¼.062 with 90% confidence interval from .000 to .136). The MI-AAL scores changed differently in the two latent growth curves representing

38

A. Hoffart et al. / Journal of Affective Disorders 181 (2015) 33–40

Table 3 Model fit indices for the conditional piecewise linear growth curve models. Model

Standardized values

Major depression Working Married/ cohabiting

pχ2 Value

df RMSEAa SRMRb CFIc TLId

.277

.002

9.398 8

.035

.043

.996 .993

 .358  .376

.026 .023

13.246 8 12.609 8

.068 .064

.043 .049

.986 .974 .988 .977

a

RMSEA ¼ root mean squared error of approximation. SRMR¼ standardized root mean residuals. c CFI¼ comparative fit index. d TLI¼ Tucker Lewis index. b

Intercept

Major depression

Married/coh abiting

Working

Short-term latent slope

-.36 (.161)

Long-term latent slope

Fig. 2. Significant pre-treatment predictors of the initial level and the rate of change in agoraphobic avoidance. The short-term latent slope represents the pretreatment, post-treatment, and 1 year follow-up Mobility Inventory-Avoidance Alone scores whereas the long-term latent slope represents the 1 year, 2 years and 13–21 years follow-up scores.

short term improvement and improvement in the long-term followup period (from one-year to 13–21 years after end of treatment). The factor means for the first latent growth curve was  .599 (SE¼.07, po.001), and  .055 (SE¼ .04, p¼.182) for the second latent growth curve. The variances of the initial statuses and first slope were significant at α ¼.01, indicating individual variation in initial statuses and trajectory of change, while the second slope displayed a trend towards individual variation (po.10). No significant relationship was found between the initial level and the rate of change in agoraphobic avoidance. That is, the initial level of agoraphobic avoidance was unrelated to the rate of change. The R square statistics indicate the proportion of variance in the observed measures that is explained by the latent growth curve factors. Small R-square indicates that most of the observed change is not related to change over time. In the unconditional model, all obtained R-squares are large, meaning that the growth curve factors explain 86%, 79%, 94%, 76%, and 48% of the observed variance in agoraphobic avoidance at pre-treatment, posttreatment, 1-year, 2-year, and 13 to 21 year follow-up, respectively. 3.3. Predictive analyses We included the covariates for both short-term and long-term growth curves in separate models and compared the improvement of fit relative to the unconditional model. The time-invariant pretreatment covariates were working, being married/cohabiting, duration of agoraphobia, Axis I and II co-morbidity, major depression, and avoidant, dependent, and obsessive-compulsive personality disorder.

In addition, use of benzodiazepines and of antidepressants during the month before start of treatment and treatment condition were examined as covariates. Table 3 shows the fit indices and the standardized parameters for the conditional piecewise growth models. Only statistical significant effects on the MI-AAL in models with satisfactory fit are provided. Also Fig. 2 depicts the significant relationships. Major depression at pre-treatment was positively related to the short-term rate of change, but unrelated to long-term improvement. Also duration of agoraphobia was negatively related to short-term but not to longterm rate of change. However, the model including this covariate showed poor fit to the data (χ 2 8 (N¼140)¼19.558, po.012 with CFI¼.97, TLI¼.94, SRMR¼ .055 and RMSEA¼.101 with 90% confidence interval from .044 to .159). Likewise, there was a treatment condition effect on the short-term growth curve, but not on the long-term growth curve. However, also the model including treatment condition as a covariate showed poor fit to the data (χ 2 8 (N¼140)¼32.489, po.001 with CFI¼.94, TLI¼.88, SRMR¼.061 and RMSEA¼ .148 with 90% confidence interval from .097 to .203). The inclusion of treatment condition resulted in significantly worse model fit relative to the unconditional model (χ 2 Δ (2)¼23.3, po.001). Accordingly, the models that included duration of agoraphobia and treatment condition were evaluated to not fit the data. The only significant effects for the long-term follow-up were working and being married/cohabiting. Being married/cohabiting was related to the latent intercept (start level) of MI-AAL scores. Both working and being married/cohabiting were negatively related to the slope of MI-AAL scores across the follow-ups (see Fig. 2), indicating greater reduction of MI-AAL scores over time among those who were working or living with a partner. Co-morbidity, being diagnosed with avoidant, dependent, and obsessive-compulsive personality disorder, use of antidepressants and/or benzodiazepines were unrelated both to short-term and long-term improvement.

4. Discussion This is the first study of predictors of a long-term (13–21 years) outcome study after standardized medication-free psychological treatment for agoraphobia of which we are aware. We assessed and predicted short- and long-term agoraphobic avoidance by using latent growth curve modeling. Participants in two open trials and one RCT were combined and studied. For those who attended long-term follow-up, there was a large effect size of  1.55 on the primary self-report avoidance measure in the total sample. Large effect sizes were observed also on measures of trait anxiety ( 1.40) and depression ( 1.12). The primary interview-based outcome measure (mean 1.02 at long term follow-up) indicates that on the whole patients were not avoiding separation situations – those that are most feared by agoraphobic patients (e.g., walk away from home alone, travel by bus or train alone). However, they still had avoidant tendencies (e.g. hurrying through them) (Hoffart et al., 1989). Nearly two-third (62.5%) lost their panic disorder/agoraphobia diagnosis. The prevalence of most co-morbid diagnoses was reduced by even more than two-third. The interviewers were stuck by how many of those retaining a diagnosis of panic disorder/agoraphobia spontaneously reported that they lived better with the disorder. One may speculate that the extra reduction of co-morbid diagnoses reflects this process. In any event, these figures are likely to represent an overestimation of improvement as non-attenders at follow-up could be expected to improve less. Moreover, more of the non-attenders were not working at pre-treatment, which turned out to be a predictor of poorer long term course. Variables previously found to predict response to psychological treatment among agoraphobic patients in at least some studies were

A. Hoffart et al. / Journal of Affective Disorders 181 (2015) 33–40

examined. Consistent with Cowley et al. (1996), pre-treatment major depression was related to reduced short-term improvement. The studied inpatient group treatment may be less suited for patients who suffer from the combination of agoraphobia and major depression. In particular, the presence of major depression may inhibit engagement in treatment through the symptoms of loss of interest, lack of energy and/or pessimistic thinking. Pre-treatment working predicted greater long-term improvement, that is, across the follow-ups. This is consistent with the findings of Durham et al. (2005) among anxiety patients. Employment may indicate a greater capacity to function and/or it may help patients overcome agoraphobia by providing regular exposure to feared situations. Also consistent with Durham et al. (2005), being married/cohabiting predicted long-term improvement. In addition, being married/ cohabiting was related to a lower initial level of avoidance scores. This is consistent with all the literature on the importance of stable attachment relationships for mental health (Schore, 2001). Being in a close relationship to another person may also be beneficial for reducing symptoms of agoraphobia in several ways: by providing social support, reducing the risk for isolation, and encouraging the agoraphobic person not to avoid feared stimuli. Use of antidepressants and benzodiazepines the month before intake to treatment was not related to levels of or improvements in agoraphobic avoidance. As the treatments were medication-free, we could not study the influence of concurrent use of medication on longterm outcome. Although the frequency of co-morbid DSM-III-R and DSM-IV Axis I and II disorders was high on average (mean43) and varied among the patients, co-morbidity predicted neither short-term nor long-term avoidance. Overall, it is noteworthy that a person-related clinical feature (major depression) was predictive of short-term change, whereas external situation variables were related to long-term change. Thus, it appears that environmental variables are increasingly influential on agoraphobic avoidance over time. Several limitations of the present study are worth noticing. Although these long term results suggest that the treatments were effective in the long term, we do not know what the patients' course would have been without treatment. However, the use of a no-treatment control group in such a very long term follow-up study would be both unethical and unfeasible. We must instead rely on longitudinal community studies, which suggest that agoraphobia is a chronic disorder and few attain natural remission (Wittchen et al., 2008). Nevertheless, the findings that long-term predictors were limited to environmental variables and that there was no treatment condition effect on long-term avoidance are consistent with the proposition that long-term avoidance is unrelated to the original treatments. It is also difficult to evaluate the influence of treatment received in the follow-period. The utilization of psychosocial treatments was reduced from 52.1% during the last year at one-year follow-up to only 10.4% at long term follow-up. The use of BZs was reduced from before treatment to all the follow-ups, whereas the use of antidepressants remained stable. There was a rather prevalent use of BZs (39.6%) and antidepressants (36.5%) at long term follow-up, although the BZs use was mostly on an intermittent per-as-needed basis. The use of BZs but not of antidepressants was associated with more avoidance at longterm follow-up. Another limitation of this study is that 31.9% of the included patients did not complete long-term follow-up, leading to reduced power and threat to the internal representativity of the sample of respondents. Only 12.1% actually withdrew from the study while 19.9% did not complete because they were deceased, physically ill, or not possible to contact. However, there were few differences in baseline characteristics between those we could and those we could not follow up long term. Moreover, in the predictive analyses, missing

39

data were handled by the FIML procedure, which is supposed to lead to unbiased estimates of the parameters. The long time gap between the two and 13–21 years follow-ups is a limitation, in which it is difficult to assess what actually happened during this period. Another shortcoming is that the length of the follow-up period varied between the patient cohorts. There were significant individual variations in initial statuses and trajectory of change for short-term avoidance, while the trajectories of change in long-term avoidance displayed only a trend towards individual variation. However, despite less individual variation for long-term outcome of treatment, the variables employment and living with a partner/spouse still predicted longterm effects of treatment. Although all the patients received some form of CBT, there was variability among these. It appears that this variability did not influence avoidance as the model including treatment condition led to a poorer fit. Furthermore, some of the treatments had additional vulnerability-focused parts. It remains uncertain whether these parts influenced short-term and/or long-term avoidance. With regard to external validity, these patients had a relatively long duration of their disorder, a high frequency of co-morbid disorders, and most of them had previously not responded to treatment. Therefore, they were probably more severely affected and more treatment resistant than the average patient seeking anxiety treatment. Moreover, the inpatient context of the treatments is unusual for anxiety disorder patients. Compared to the Norwegian population, their demographics suggested a tendency toward social disadvantage.

5. Conclusions Nearly two-third of those attending follow-up had lost their panic disorder/agoraphobia diagnosis, and the prevalence of most co-morbid diagnoses was reduced by even more than two third. The only short-term predictor of avoidance identified represented a clinical feature (i.e. major depression), whereas the long-term predictors represented features of the patients’ life situation, that is, working and being married/cohabiting.

Role of funding source This research was funded by the South-Eastern Norway Regional Health Authority (Project no. 2009045), who had no role in conducting the research or in writing and submitting the report.

Conflict of interest All the authors declare that they have no conflicts of interest.

References Barlow, D.H., Gorman, J.M., Shear, M.K., Woods, S.W., 2000. Cognitive-behavioral therapy, imipramine, or their combination for panic disorder. J. Am. Med. Assoc. 283, 2529–2536. http://dx.doi.org/10.1001/jama.283.19.2529. Beck, A.T., Steer, R.A., Garbin, M.G., 1988. Psychometric properties of the Beck depression inventory: twenty-five years of evaluation. Clin. Psychol. Rev. 8, 77–100. Borenstein, M., Hedges, L.V., Higgins, J.P.T., Rothstein, H., 2009. Introduction to Meta-analysis. Wiley, Cornwall. Brown, T.A., Barlow, D.H., 1995. Long-term outcome in cognitive-behavioral treatment for panic disorder: clinical predictors and alternative strategies for assessment. J. Consult. Clin. Psychol. 63, 754–765. http://dx.doi.org/10.1037/ 0022-006X.63.5.754. Chambless, D.L., Caputo, G.C., Jasin, S.E., Gracely, E.J., Williams, C., 1985. The mobility inventory for agoraphobia. Behav. Res. Therapy 23, 35–44. Chambless, D.L., Sharpless, B.A., Rodriguez, D., McCarthy, K.S., Milrod, B.L., Khalsa, S.R., Barber, J.P., 2011. Psychometric properties of the mobility inventory for agoraphobia: convergent, discriminant, and criterion-related validity. Behav. Therapy 42, 689–699. http://dx.doi.org/10.1016/j.beth.2011.03.001. Cowley, D.S., Flick, S.N., Roy-Byrne, P.P., 1996. Long-term course and outcome in panic disorder: a naturalistic follow-up study. Anxiety 2, 13–21.

40

A. Hoffart et al. / Journal of Affective Disorders 181 (2015) 33–40

Durham, R.C., Chambers, J.A., Macdonald, R.R., Fisher, P.L., 2009. Predictive validity of two prognostic indices for generalized anxiety disorder. Int. J. Cogn. Therapy 2, 383–399. Durham, R.C., Chambers, J.A., Power, K.G., Sharp, D.M., Macdonald, R.R., Major, K.A., Dow, M.G.T., Gumley, A.I., 2005. Long-term outcome of cognitive behaviour therapy clinical trials in central Scotland. Health Technol. Assess. 9, 42. Faravelli, C., Paterniti, S., Scarpato, A., 1995. 5-year prospective, naturalistic followup study of panic disorder. Compr. Psychiatry 36, 271–277. First, M.B., Spitzer, R.L., Gibbon, M., Williams, J.B.W., 1995. Structured Clinical Interview for DSM-IV Axis I Disorders – Patient edition (SCID-I/P, Version 2.0). New York State Psychiatric Institute, New York. First, M.B., Spitzer, R.L., Gibbon, M., Williams, J.B.W., Benjamin, L., 1994. Structured Clinical Interview for DSM-IV Axis II Personality Disorders – Patient (SCID-II), (Version 2.0). New York State Psychiatric Institute, New York. Hoffart, A., 1998. Cognitive and guided mastery therapy of agoraphobia: long-term outcome and mechanisms of change. Cogn. Therapy Res. 22, 195–207. Hoffart, A., Friis, S., Martinsen, E.W., 1989. The phobic avoidance rating scale: a psychometric evaluation of an interview-based scale. Psychiatr. Dev. 7, 71–81. Hoffart, A., Hedley, L.M., 1997. Personality traits among panic disorder with agoraphobia patients before and after symptom-focused treatment. J. Anxiety Disord 11, 77–97. Hoffart, A., Hedley, L.M., Svanøe, K., Sexton, H. Cognitive and guided mastery therapies for panic disorder with agoraphobia: 18-year long-term outcome and predictors of long-term change. Clin. Psychol. Psychother., http://dx.doi.org/10. 1002/cpp.1934, in press. Hoffart, A., Thornes, K., Hedley, L.M., 1995. DSM-III-R Axis I and II disorders in agoraphobic inpatients with and without panic disorder before and after psychosocial treatment. Psychiatry Res. 56, 1–9. Hoffart, A., Versland, S., Sexton, H., 2002. Self-understanding, empathy, guided discovery and schema belief in schema-focused cognitive therapy of personality problems: a process-outcome study. Cogn. Therapy Res. 26, 199–219. http: //dx.doi.org/10.1023/A:1014521819858. Hu, L.T., Bentler, P.M., 1999. Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct. Equ. Model: Multidiscip. J. 6, 1–55. http://dx.doi.org/10.1080/10705519909540118.

Kessler, R.C., Chiu, W.T., Demler, O., Merikangas, K.R., Walters, E.E., 2005. Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the National Comorbidity Survey Replication. Arch. Gen. Psychiatry 62, 617–627. http://dx.doi.org/10.1001/archpsyc.62.6.617. Little, R.J.A., 1988. A test of missing completely at random for multivariate data with missing values. J. Am. Stat. Assoc. 83, 1198–1202. Milrod, B., Leon, A.C., Busch, F., Rudden, M., Schwalberg, M., Clarkin, J., Shear, M.K., 2007. A randomized controlled clinical trial of psychoanalytic therapy for panic disorder. Am. J. Psychiatry 164, 265–272. Milrod, B., Busch, F., 1995. Long-term outcome of panic disorder treatment: a review of the literature. J. Nerv. Mental Dis. 184, 723–730. Norton, P.J., Price, E.C., 2007. A meta-analytic review of adult cognitive-behavioral treatment outcome across the anxiety disorders. J. Nerv. Mental Dis. 195, 521–531. http://dx.doi.org/10.1097/01.nmd.0000253843.70149.9a. Norwegian Social Science Data Services, 1995. Labour Force Survey. University of Bergen, Bergen. Schore, A.N., 2001. The effects of early relational trauma on right brain development, affect regulation, and infant mental health. Infant Mental Health J. 22, 201–260. http://dx.doi.org/10.1002/1097-0355. Shrout, P.E., Fleiss, J.L., 1979. Intraclass correlations: uses in assessing rater reliability. Psychol. Bull. 86, 420–428. Spielberger, C.D., 1983. Manual for the State-Trait Anxiety Inventory (Form Y). Consulting Psychologists Press, Palo Alto. Spitzer, R.L., Williams, J.B.W., Gibbon, M., First, M.B., 1988. Structural Clinical Interview for the DSM-III-R – Patient version (SCID-P, 6/1/88). New York State Psychiatric Institute, New York. Stewart, R.E., Chambless, D.L., 2009. Cognitive-behavioral therapy for adult anxiety disorders in clinical practice: a meta-analysis of effectiveness studies. J. Consult. Clin. Psychol. 77, 595–606. http://dx.doi.org/10.1037/a0016032. Wittchen, H.-U., Nocon, A., Beesdo, K., Pine, D.S., Höfler, M., Lieb, R., Gloster, A.T., 2008. Agoraphobia and panic: prospective-longitudinal relations suggest a rethinking of diagnostic concepts. Psychotherapy Psychosom. 77, 147–157. http://dx.doi.org/10.1159/000116608.