Expressed Emotion and Behavior Therapy Outcome: A Prospective Study With Obsessive–Compulsive and Agoraphobic Outpatients

Expressed Emotion and Behavior Therapy Outcome: A Prospective Study With Obsessive–Compulsive and Agoraphobic Outpatients
[Regular Articles]
Chambless, Dianne L.1,3; Steketee, Gail2
1Department of Psychology, University of North Carolina at Chapel Hill
2School of Social Work, Boston University.
3Correspondence concerning this article should be addressed to Dianne L. Chambless, Department of Psychology, University of North Carolina, Chapel Hill, North Carolina 27599-3270. Electronic mail may be sent to chambles@email.unc.edu.
This work was supported by Grant R01-MH44190 from the National Institutes of Health.
We thank statistical consultant Leona Aiken; expressed emotion consultant Jill Hooley; therapists Leslie Shapiro, Elaine Williams, Ulrike Feske, and Judy Lam; Camberwell Family Interview (CFI) interviewers Marty Gillis, Cheryl Sheffler Rubenstein, Hope Warden, Sheila Woody, and Whitney Wycoff; CFI coders Henry Chung, Delphine Mattison, Scott Maynard, Brian McCorkle, and Maureen Patten; and data analysts Giao Tran, Kimberly Wilson, Tom Rodebaugh, and Carol Woods for their important contributions to this research.
Received Date: May 13, 1998; Accepted Date: February 8, 1999
Abstract
The relationship of expressed emotion (EE) to behavior therapy outcome for obsessive–compulsive disorder (n = 60) and panic disorder with agoraphobia (n = 41) was investigated. Relatives' emotional overinvolvement and hostility predicted higher rates of treatment dropout. Higher hostility, as assessed by the Camberwell Family Interview (CFI), was related to poorer outcome for target ratings and for the Social Adjustment Scale; higher perceived criticism was also predictive of worse response on target ratings. In contrast, nonhostile criticism on the CFI was associated with better outcome on the behavioral avoidance test. In general, the relationship of EE to outcome was not moderated by type of relative, diagnosis, amount of contact with the relative, or use of psychotropic medication.
________________________________________


Although the development of exposure-based treatment marked a breakthrough in treatment for people with agoraphobia or obsessive–compulsive disorder (OCD), the results of this treatment are not consistently satisfactory. The majority of patients completing treatment may improve (Jacobson, Wilson, & Tupper, 1988; Stanley & Turner, 1995), but only a minority recover normal functioning (e.g., Jacobson et al., 1988). Consideration of treatment dropout and refusal further lowers the apparent success rate (Jansson & Ost, 1982; Stanley & Turner, 1995). Thus, it is important to identify variables associated with poor treatment response that may point to additional interventions to improve outcome. In the present investigation, we examined the utility of expressed emotion (EE) as such a predictor.
EE refers to family members' feelings about an identified patient and includes emotional overinvolvement (EOI), criticism, and hostility. In a meta-analysis of EE research, Butzlaff and Hooley (1998) determined the average effect sizes for EE prediction of relapse for patients with affective disorder and with schizophrenia to be substantial and statistically significant (rs = .39 and .30, respectively). Similar findings were observed in the few studies on eating disorders (r = .51).
EE's relationship to treatment outcome has been infrequently examined for patients with anxiety disorders. In a sample of 31 patients with posttraumatic stress disorder (PTSD), Tarrier, Sommerfield, and Pilgrim (in press) reported that higher criticism and hostility (but not EOI) predicted less change at posttest after treatment with imaginal exposure. Most relatives were spouses. Assessing the role of EE in spouses of agoraphobic patients, Peter and Hand (1988) reported unusual findings: Higher criticism tended to predict better outcome 1 to 2 years after behavior therapy; posttest findings were not significant. The authors proposed that more critical spouses might tolerate less dependency in their agoraphobic partners and pressure them to improve more. Although these findings are intriguing, their interpretation is made difficult by the small sample (N = 20) and questionable data-analytic strategies.
Two prospective studies of predictive factors in OCD have included measures that directly or indirectly assessed EE-related constructs. Using unvalidated interview measures, Steketee (1993) examined patients' and relatives' perceptions of familial interactions in relation to outcome 9 months after behavior therapy. Poor family functioning assessed before treatment, as well as negative household interactions (especially anger and criticism), predicted fewer gains at follow-up. In a study of children with OCD treated with pharmacotherapy, Leonard et al. (1993) found that children living with high-EE parents had poorer global adjustment at 2- to 7-year follow-up compared with children in low-EE households, but EE failed to predict OCD symptoms per se. Because only one third of the children received exposure and response prevention, no clear conclusions can be drawn about family factors as predictors of outcome for behavior therapy from this research.
These few reports provide a mixed picture of the association of EE to outcome for anxiety disorders. Findings for agoraphobic patients do not accord with those for OCD and PTSD patients, but too little data from studies conducted with sound methodology and standard measures of EE are available to permit solid conclusions. To further this research, in the present study we examined the effects of EE on treatment outcome in a prospective study of response to outpatient behavior therapy for agoraphobia and for OCD. In addition, we examined the effects of two possible moderators of EE's relationship to outcome drawn from prior research on other patient populations: medication use and number of hours of contact with the relative weekly (see review by Hooley, 1985). Like Tarrier et al. (in press) and Peter and Hand (1988), we used the benchmark EE measure, the Camberwell Family Interview (CFI; Vaughn & Leff, 1976). This measure is derived from an interview with the relative in the patient's absence and coded by trained raters for relative's voice tone and comments. In addition, we included a measure of the patient's perception of the relative's criticism. Hooley and Teasdale (1989) have argued that an objective rater's report of relative's criticism may be less important than the patient's perception of that criticism because the latter is likely to be the proximal cause of any effects of EE on the patient. Although minimally correlated with the CFI measure of criticism, patients' ratings of perceived criticism have proved predictive of outcome in a number of investigations (Hooley & Teasdale, 1989; Lebell et al., 1993; Tompson et al., 1995; but see Okasha et al., 1994).
In our approach to this research, we deviated from the traditions of the EE field in several ways. First, our focus was on treatment completion and initial outcome of outpatient treatment for anxiety disorders conducted while patients continued to live with their relatives. The more typical focus in research on depression and schizophrenia is on relapse of inpatients once they have been discharged to live with their families. Second, in accord with comments by a number of other authors (e.g., Gottschalk & Keatinge, 1993), we have treated EE variables as separate and continuous measures, rather than establishing a category of high EE in which EOI, critical comments, and hostility are combined. Given the dearth of previous research on anxiety disorders, it seemed unwise to assume that EOI, criticism, and hostility would all have the same relationship to outcome for this patient population. Indeed, Tarrier et al. (in press) found that critical comments and hostility predicted poor outcome for PTSD patients, whereas EOI did not. Retaining the predictor variables in their continuous form rather than dichotomizing provides greater statistical power (Cohen, 1983) and avoids the problem of relying on cut scores for high and low EE for which there is little empirical or theoretical foundation for anxiety disorders (cf. Gottschalk & Keatinge, 1993).
We predicted that higher perceived criticism, hostility, and EOI would be associated with poorer treatment outcome defined (a) as terminating treatment while still highly symptomatic or (b) as poor response to treatment for those who completed at least 10 sessions of behavior therapy. Given conflicting findings on the effects of CFI critical comments for anxiety disorders, no predictions were made for this variable.
Method
Participants
Participants were 101 outpatients with OCD (n = 60) or panic disorder with agoraphobia (PDA, n = 41) treated at American University in Washington, DC, or McLean Hospital in Belmont, Massachusetts, and all English-speaking adult relatives living in the household with the patients. Initially, 104 patients were accepted into the program, but 1 moved unexpectedly, 1 recovered before treatment began, and a 3rd decided she needed treatment more urgently for her marital problems than her OCD. Fifty-three other applicants were interviewed and appeared to meet criteria for the project but declined to participate. In 11 cases, refusal was directly pertinent to the purposes of this investigation: Relatives were unwilling (n = 7) or too elderly and frail (n = 1) to participate, or patients did not wish to involve their family members (n = 3). Others were too fearful of entering behavior therapy (n = 7), unwilling to restrict other treatment (n = 16), or did not return for further assessment for unknown reasons (n = 19).
All of the patients met criteria for a primary diagnosis of OCD or PDA. Additional criteria for inclusion were at least moderate avoidance for PDA patients and at least 1 hr of rituals per day for OCD participants. Patients were excluded if they had a current diagnosis or history of psychosis, reported evidence of organic etiology of the disorder, or met criteria for alcohol or substance dependence. Additional criteria for inclusion were being 18 to 65 years of age, having symptoms for at least 1 year, having lived with the relative for at least 3 months prior to beginning the study (the time frame for relatives' responses to the CFI), and intending to continue to do so at least until the 6-month follow-up. Further, patients had to agree to suspend any current psychotherapy, to keep any psychotropic medication constant, and to refrain from other psychotherapy or change in medication until the 6-month follow-up.
Seventy percent of patients were married or cohabiting, whereas the remainder were living with parents or, more rarely, other family members. Patients averaged 35 years in age (range = 18–62) and reported a mean symptom duration of 15 years for OCD (range = 1–37) and 10.4 years for PDA (range = 1–41). The ethnic/racial distribution was 87% White, 10% African American, and 3% “other”; 72% were female. A total of 143 relatives participated. The households spanned the full range of Hollingshead's Four-Factor Index of Social Position (Hollingshead, 1975); median socioeconomic status (SES) was 48 (e.g., minor professional, owner of a medium-sized business). Fifty-two percent, equally distributed across diagnoses, entered the project taking psychotropic medication; of these, 29% were taking anxiolytics, 40% antidepressants, and 31% both.
Measures
Symptom and Diagnostic Measures
Structured Clinical Interview for DSM–III–R—Patient Version (SCID–P;Spitzer, Williams, Gibbon, & First, 1989).
The SCID–P was used to determine participants' diagnoses. All of the SCIDs were audiotaped, and approximately 25% of these tapes (n = 33) were randomly selected, mixed with tapes from other diagnostic groups, and independently coded by a rater at the other treatment site. Kappas for interrater reliability were 1.0 and .94 for diagnoses of OCD and of panic disorder, respectively. Because of empty cells, we could not compute kappas for presence or absence of agoraphobia (rated only once the diagnosis of panic disorder had been assigned). In all 19 cases, both raters identified the patients as having panic disorder, but in 2 of these cases, the interviewer and the site principal investigator agreed on the presence of agoraphobia, whereas the rater at the other site assigned a label of limited avoidance. SCID interviewers also rated clients according to the Global Assessment of Functioning Scale (GAF), a 0–90 rating of general functioning. Cross-site interrater reliability for the GAF was adequate (r = .76).
Target symptom ratings.
To assess the specific symptoms targeted in treatment, the therapist and patient jointly selected three main situations that would provoke anxiety and avoidance and, for OCD patients, rituals. Patients rated these situations on 0–8 scales, where 8 indicated highest severity; the three scores were averaged to yield the overall score for each patient. These scales have satisfactory interrater reliability and convergent validity with standardized measures of phobia and OCD and have proved highly sensitive to treatment effects (see reviews by Feske & Chambless, in press, and Lelliott, 1988). Retest reliability across the first 2 weeks of information gathering (before introduction of exposure treatment) for 29 patients in the present study was .79.
Behavioral Avoidance Test (BAT).
A BAT was administered by project therapists or research assistants before exposure treatment began and by research assistants at posttest. Three behaviors that were difficult or impossible for patients to do without significant anxiety (or rituals in the case of OCD) were identified and attempted. The test administrator recorded degree of avoidance or ritualizing (0 = no avoidance/rituals, 1 = partial avoidance/rituals, 2 = unable to do task). Scores were averaged across the three tasks. Because BATs have been infrequently used for OCD, the convergent and discriminant validities of this measure were demonstrated for OCD patients included in the present sample (Steketee, Chambless, Tran, Worden, & Gillis, 1996).
Social Adjustment Scale—Self-Report (SAS; Weissman & Bothwell, 1976).
The SAS assesses social adjustment in multiple areas of functioning. For this investigation, the Marital and Family Functioning subscales were excluded because these were confounded with some EE measures. The work, economic, and social and leisure scores were averaged to yield the SAS—Non-Family (SAS–NF) score.
Measures of EE
CFI (Vaughn & Leff, 1976).
The CFI is a 1- to 2-hr semistructured interview of a psychiatric patient's relative, which is audiotaped and later scored to yield indexes of the relative's attitudes toward the patient. The measure of critical comments is scored as a simple frequency count. In the present sample, critical comments ranged from 0 to 38, with a median of 5. EOI and hostility are each rated on a single global scale. Hostility ranges from 0 (no hostility) to 3 (generalization of criticism and rejection), whereas the EOI rating ranges from 0 (none) to 5 (marked). These variables each had a median of 0 in the present sample. The entire range of hostility was represented, whereas EOI ranged from 0 to 4. Because the hostility variable was extremely positively skewed in this sample, for the purposes of multiple regression analyses, hostility was recoded as 0 (no hostility) versus 1 (any hostility). The CFI was administered by trained interviewers and coded by raters trained to reliability (defined as minimum interrater reliability of .80 on each measure) with their instructor before coding for the present study. During the study, a second rater independently coded 12% of CFI tapes. Interrater reliability was excellent (intraclass correlation coefficient [1,1] = .74–.91; see Shrout & Fleiss, 1979).
Perceived Criticism Measure (PCM; Hooley & Teasdale, 1989).
Patients rated each adult relative with whom they lived on a scale ranging from 1 (not at all critical) to 10 (very critical). PCM–Criticism has shown good discriminant (Hooley & Teasdale, 1989; Riso, Klein, Anderson, Ouimette, & Lizardi, 1996) and predictive validity (Hooley & Teasdale, 1989). However, convergent validity evidence is limited and conflicting (Chambless, Bryan, Aiken, Steketee, & Hooley, 1999; Chambless et al., 1997; Hooley & Teasdale, 1989). Patients in the present investigation completed the PCM on three occasions: twice before starting behavioral treatment for their disorder (at pretest assessment and before exposure commenced about 2 weeks later) and at the end of treatment (approximately 4 months later). Test–retest reliability was acceptable, whether calculated before exposure (r = .75, n = 80) or from pre- to posttest (r = .66, n = 66). PCM scores in this sample ranged from 1 to 10, with a mean of 5.21 (SD = 2.67).
Relative's Reactions Questionnaire (RRQ; Steketee, 1987).
The RRQ includes items tapping the degree to which relatives feel responsible for the patients' anxiety problem; feel guilty about their role in the patients' problems; and believe that, when they criticize the patients, they are trying to help them. Each item is rated on a scale ranging from 0 (not at all) to 6 (completely, or very often). The items proved adequately reliable for one-item measures (rs = .50–.69, n = 82) to be used in a composite measure (see below) for research purposes.
Composite measure of EOI.
As is typical for samples in which relatives are predominantly spouses, few relatives (12%) were rated as emotionally overinvolved on the CFI, resulting in a restricted range and very positively skewed distribution. Because this variable was not suitable for the planned parametric data analyses, a broader EOI measure was constructed on the basis of a confirmatory factor analysis of EOI indexes conducted on data from the present sample (Chambless et al., 1999). This included the CFI rating and the three RRQ items previously described, all of which loaded significantly (p < .05) on an EOI factor. These were transformed into T scores and summed to yield the EOI factor score used in the present study.
Other Measures
Biographical data (e.g., ethnicity, education, occupation, and duration of the disorder) were collected before treatment began. To monitor medication use during treatment, we had patients record the type and dose of medication in a weekly diary. Hours the patient and relative spent together per week were extracted from the CFI interview with the relative.
Procedure
Assessment.
Subsequent to telephone screening, patients were interviewed by either Dianne L. Chambless or Gail Steketee, who made an initial determination of eligibility. Those who passed this further screen and agreed to participate provided written informed consent and attended a second, independent diagnostic interview (the SCID–P) before a final decision concerning admission to the study was made. Subsequently, patients completed questionnaires, except for the target ratings. The patient and therapist together determined the content of the target ratings and of the BAT after two information-gathering sessions prior to exposure.
Relatives also provided written informed consent and, like patients, completed a packet of questionnaires. Patients and relatives were asked to complete their forms separately and privately and to return these to the research assistant at their next office visit. Depending on scheduling requirements and the number of relatives living in the home, families attended one or more pretreatment assessment sessions to complete the CFI.
Treatment.
All of the treatment was conducted according to detailed manuals for OCD and PDA treatment.1 Therapists were master's- or doctoral-level clinicians, trained and supervised weekly by the local site investigators, with occasional consultation with the investigator at the other site in complicated cases. Treatment began with two 90-min sessions for treatment planning. Relatives were invited to the latter part of the second of these, during which therapists explained the treatment procedures and invited questions and comments. Relatives did not participate further in treatment sessions. Active treatment for the patient included 12 twice-weekly 90-min therapist-assisted exposure sessions (plus response prevention for OCD participants), followed by 4 weekly 90-min exposure sessions. These sessions took place in relevant exposure locations. Treatment ended with 4 weekly 60-min in-office sessions during which the focus switched to encouraging patients to devise their own plans for continued self-directed exposure and to developing plans for maintenance and relapse prevention. These sessions included no therapist-assisted exposure. In all, patients received a total of 22 sessions over 16 weeks. Throughout treatment, therapists were enjoined from discussing family problems.
Treatment integrity checks.
To verify the adequacy of the treatment delivered, we required therapists to complete routine forms indicating the content of each session and to record patients' discomfort every 10 min throughout the session. Supervisors reviewed these forms in weekly sessions. Adherence to the treatment protocol was otherwise examined only for the final four in-office sessions of treatment because audiotaping of exposure sessions conducted outside of the office was technically difficult. The therapist at the other site listened to randomly selected audiotapes of 25% of treatment sessions to check for violations of the protocol. Violations were examined using a Treatment Integrity Checklist, which inquired about use of prescribed interventions (e.g., review of weekly activities involving contact with feared situations) and proscribed ones (e.g., continued in vivo exposure and suggestions for resolution of family problems). Protocol adherence in all of the sessions was within treatment guidelines.
Posttreatment assessment.
Dependent measures were collected again at the end of treatment and at follow-up. (Follow-up procedures and results will be reported elsewhere.) Once they had already received treatment (the major inducement for undergoing assessment procedures), not all of the patients were willing to provide complete data. Rather than risk total loss of data, if the patient was unwilling to complete the BAT or SAS–NF, we focused on obtaining target ratings, which required minimal effort on the patient's part. These were collected from all of the treatment completers (n = 83), whereas posttest BAT data were obtained from 68 completers and SAS–NF data from 72 completers.
Selection of the primary relative for prediction analyses.
To avoid introducing dependence into the data, when patients lived with more than one relative (n = 26), we selected one relative per household for inclusion in prediction analyses. In choosing this relative, we relied on our a priori definition of the importance of the relationship: (a) Spouses/domestic partners (hereinafter spouses) were selected over other relatives; (b) a parent was selected over other nonspousal relatives; (c) when two parents were available (n = 11), the more critical parent according to the CFI was selected; and (d) in the one case in which the relative was neither a parent nor spouse, we selected the more critical relative. In all but two cases, use of these criteria also resulted in the selection of the most critical relative according to the CFI, making the results of our analyses highly comparable with results that would have been obtained had we selected the most critical relative, as is typical in EE research. Among these primary relatives, 73% were spouses, 19% were parents, and 8% were other adult relatives (e.g., adult child or sibling). Diagnostic groups did not differ on primary relative type, [chi]2(2, N = 101) = 0.88, p > .75.
On the basis of traditional EE cutoff scores, 55% of primary relatives met criteria for a designation of high EE. According to individual CFI variables, 40% were highly critical (6 or more critical comments), 33% were hostile (rating of 1 or more), and 12% were emotionally overinvolved (rating of 3 or more). These findings are quite similar to the frequencies reported by Tarrier et al. (in press) for relatives of PTSD patients (of whom a comparable percentage were spouses), with the exception of a lower rate of EOI in the present study (12% vs. 26%).
Results
Preliminary Analyses
Before analyses of prediction of outcome, we conducted checks for site effects and diagnosis effects on treatment outcome. We conducted 2 (diagnosis [OCD vs. PDA]) × 2 (site [Washington, DC vs. Boston]) × 2 (occasions [pretest vs. posttest]) mixed model analyses of variance (ANOVAs) with target ratings, BAT avoidance, and SAS–NF as dependent variables. All of the pertinent effects (Site × Diagnosis interaction, Diagnosis × Time interaction, and Site × Diagnosis × Time interaction) were not significant (all ps > .11). Because one therapist treated the great majority of patients at each site, the absence of site effects is tantamount to a demonstration of lack of therapist effects as well. Given that these findings indicated that pooling of data across sites was justified, we conducted analyses for hypothesis testing on the combined Boston and Washington, DC, data sets. As expected, the ANOVA main effects for occasion, testing change with treatment on dependent variables, were all significant (all ps < .02). Effect sizes for treatment outcome, calculated with Cohen's d' for paired observations, were large for target ratings and the BAT. The effect for SAS–NF was small for PDA and medium for OCD (see Table 1).

________________________________________

[Help with image viewing]
[Email Jumpstart To Image]
Table 1 Pre- and Posttest Means and Standard Deviations and Paired Sample Effect Sizes for Obsessive–Compulsive and Agoraphobic Outpatients Treated With Behavior Therapy
________________________________________
Intercorrelations among EE variables revealed that perceived criticism and the EOI factor were each relatively independent of the other variables (rs = -.07 to .19, ps >= .06). As would be expected, CFI critical comments and hostility were substantially and positively correlated (r = .46, p < .001). For dependent variables, intercorrelations among the pretest scores (rs = -.03 to .36) and pretest–posttest residual change scores (rs = .20–.36) indicated that they were also substantially independent. In light of this degree of independence, and because complete data were only available for target ratings, we conducted separate analyses on the three dependent variables. Similarly, rather than combining EE variables in one summary high–low EE category, we treated them as individual predictors.
Major Analyses
Premature Termination
Eighteen patients dropped out of treatment, which was defined as leaving treatment before receiving at least 10 sessions of exposure. These were equally likely to have OCD or PDA (18% of each group) and in all cases withdrew from treatment without significant improvement.
We conducted preliminary analyses to identify potentially confounding variables (gender, SES, GAF, initial severity, and duration of disorder), which were defined as variables related both to one or more EE variables and to dropout. Only GAF was identified as a confounding variable and statistically controlled in subsequent analyses.
Data-analytic strategy.
We conducted logistic regression equations with dropout/completer status as the criterion and the EE variables of interest as predictors, with GAF included to control for its effects. First-order effects for relative type (spouse vs. other) and for diagnosis (OCD vs. PDA) and their interactions with EE were included in preliminary tests; OCD and PDA were dummy coded 0 and 1, respectively, as were nonspouses and spouses. When these effects were clearly not significant (defined as p >= .20), we eliminated them to avoid multicollinearity and unnecessary diminution of power and then calculated a reduced model that included only the variables of interest and any control variables tending to significance.
In light of the restricted sample size and the large number of predictors and potential interactions, we first conducted separate analyses of EE variables rather than all of the predictors at once. The exceptions were CFI critical comments and hostility. Because of their moderate intercorrelation, these were entered in the same equation to permit examination of the independent contribution of each. Once these simpler equations had been tested, we calculated a summary equation that included all of the EE predictors and any necessary control variables detected in the preliminary equations. Findings from preliminary analyses are only reported when these differ from the summary equations.
Findings.
In the summary analysis (n = 95), higher EOI factor scores predicted higher rates of dropout (B = 0.16, SE = 0.06, odds ratio [OR] = 1.17, p = .006), as did higher hostility (B = 1.78, SE = 0.88, OR = 5.94, p < .05). In addition, there was a significant interaction of perceived criticism with relative type (p < .02). Probes of this interaction indicated that the direction of the relationship of perceived criticism differed for patients living with spouses (B = 0.15, SE = 0.12, OR = 1.16, p > .24) versus nonspousal relatives (B = -0.37, SE = 0.25, OR = 0.69, p > .13), but the findings were not statistically significant for either group.
EE and Prediction of Response to Treatment for the Completer Sample
Potentially confounding variables were defined as those correlated with both (a) one or more EE variables and (b) residual gain scores on one or more dependent variables. SES was identified as a confounding variable and therefore included in all of the initial tests of the relationship of EE to treatment outcome.
We performed prediction analyses with multiple regression equations according to the same strategy used for prediction of dropout. The criterion was the posttest score on target ratings (or the SAS–NF or the BAT), with the pretest score on that variable forced in as the first predictor to control for initial severity. Predictors were centered to reduce collinearity with the interaction terms (see Aiken & West, 1991). Examination of condition indexes for all of the equations verified that this strategy was successful. We also examined plots of residuals to ascertain that the assumptions of multiple regression were met and conducted tests of influence to determine whether aberrant data points might unduly affect the results. These diagnostic tests yielded no evidence of such problems with the data, with one exception: One outlier was detected and removed from the analysis of moderation of the effects of the EOI factor by contact time with the relative for one dependent variable (the BAT).
The semipartial correlations of the predictors with the three criteria in the summary equations are given in Table 2.2 These depict the relationship of the predictor to outcome, controlling for the effects of all of the other variables in the equation, with the square of the semipartial correlation equaling the variance in outcome accounted for by the residualized predictor. Estimations of power vary because of differences in sample size and the predictors included in the final equations. For the summary equations, the smallest semipartial correlation for which power was estimated to be adequate (.80) ranged from .22 to .28.

________________________________________

[Help with image viewing]
[Email Jumpstart To Image]
Table 2 Semipartial Correlations of Measures of Expressed Emotion and Prediction of Change From Pretest to Posttest
________________________________________
Target ratings.
The summary regression including all of the EE variables yielded one significant predictor: Higher perceived criticism predicted poorer outcome with a moderate effect size. Findings of the preliminary equations differed in only one regard: Without perceived criticism in the equation (but controlling for CFI criticism), hostility predicted significantly less change with treatment (sr = .22, p < .05).
BAT avoidance.
Only CFI criticism emerged as a significant predictor of change on the BAT. Higher criticism, controlling for hostility and other EE variables, was moderately associated with better outcome.
SAS–NF.
A significant first-order effect for CFI hostility emerged, modified by a Hostility × Relative Type interaction. Probes of this interaction, conducted according to procedures detailed by Aiken and West (1991), indicated that when the relative was a spouse, hostility was not related to functioning (sr = -.01). However, when relatives were not spouses (most often parents), higher hostility was associated with poorer outcome (sr = .22, p < .05).
Potential Moderators
Logistic and multiple regression equations were used to test the interaction of EE variables with amount of contact with the relative per week and with use of medication (see Footnote 2). There were no trends for dropout to be moderated by either variable. Medication use did not interact with EE in predicting posttreatment outcome on any dependent variable (all ps > .20). Comparable results were obtained for contact time (ps > .13), with the exception of a significant interaction with perceived criticism (p < .01). Counterintuitively, perceived criticism was associated with negative outcome on target ratings for those patients with relatively low contact with their relatives (sr = .36, p = .001) but not for those with relatively high contact (sr = .03). For the BAT, there was a trend (p < .06) for the same interaction.
Power estimates for examination of moderator effects vary because of missing data and differences in the predictor variables included in equations. The smallest interaction semipartial correlations for which power was estimated to be adequate ranged from .23 to .30.3
Discussion
In conjunction with findings reported by earlier investigators, the results of this investigation demonstrate EE's utility as a predictor of negative treatment outcomes for anxiety disorders. In addition to examining the relationship of EE to treatment outcome, we examined two variables occasionally found to be moderators of this relationship: medication use and amount of contact with the relative. Further, given the presence of subsamples among our patient group, we examined whether diagnosis or type of relative with whom the patient resided would act as moderators. On the whole, these variables had no significant effect. In total, there were 64 tests of interactions conducted across the four measures of outcome, of which only 3 were statistically significant, exactly the number to be expected by chance. Thus, in this study, it seems reasonable to interpret the EE–treatment outcome relationship without qualifying reference to these moderators.
The most consistent predictor of negative treatment outcome was hostility as measured by the CFI. When relatives were hostile, the odds were approximately six times greater that patients would drop out than when relatives were not hostile (if all of the other EE variables were controlled). When patients completed treatment while living with a hostile (as opposed to a nonhostile) relative they changed less on target ratings of fear and avoidance and on self-reported functioning in work, school, economic, and social and leisure spheres. Hostility was also associated with poorer outcome on the behavioral avoidance test, but this effect was small and not significant. Using procedures described by Rosenthal (1983) to recast the effect size for prediction of target ratings and the SAS (srs >= .20) as a binomial effect size display, we found that this effect size was equivalent to a drop in treatment success rate from 60% to 40%.
In the present study, contrary to the typical finding in EE research, higher rates of nonhostile critical comments on the CFI were predictive of significantly better treatment outcome on the BAT. These findings are not unique for anxiety disorders (cf. Peter & Hand, 1988). To understand these data, it is important to consider the definitions of criticism and hostility in the CFI scoring system. Criticism is noted whenever relatives indicate, through voice tone or a clear statement, that they dislike something about the patient. Such criticism is not necessarily hostile. For example, one relative stated, “If she's stuck on a particular issue that I deem as OCD-related, yeah, I nag. It bugs the hell out of me.” Hostility is scored when the relative's criticism becomes global, or when the relative indicates dislike of the patient as a person rather than disapproval of some aspect of the patient's behavior. For example, one wife described telling her husband, “If that's the way you're going to be, that's it. We're through. I can't stand it.”
Given that, in our data, CFI criticism had a positive effect when hostility was controlled, and given that Peter and Hand's (1988) sample did not include hostile relatives, these findings might be interpreted to mean the following: When relatives express dissatisfaction with specific aspects of patients' behavior (e.g., symptoms) but do not reject the patients themselves (defined as hostility on the CFI), critical comments may have motivational properties (cf. Peter & Hand). Although relatives in our sample were critical of a wide variety of patients' behaviors, the most common focus of their criticism was the patients' anxiety disorder symptoms (Rodebaugh, Chambless, Renshaw, & Steketee, 1998). In exposure-based therapies, patients must confront anxiety-provoking stimuli, although their natural inclination is to avoid these. Relatives' expressed dissatisfaction with the patients' disability and symptoms may help to keep patients engaged in this difficult process. However, given that we observed significant positive effects of nonhostile criticism only for one dependent variable, and given that Tarrier et al. (in press) found CFI critical comments to be related to negative treatment outcome for PTSD patients, these findings must be considered quite tentative.4
The findings of this investigation add to the small body of research demonstrating the utility of patients' perceptions of relatives' criticism as a predictor of poor treatment outcome. Like Hooley and Teasdale (1989), we found perceived criticism to be a significant predictor of poor outcome, even after controlling for the contributions of criticism and hostility coded from the CFI. However, this finding was observed for only one of our three outcome variables (target ratings) and thus must be considered tentative. Quite puzzling is our finding that perceived criticism and CFI critical comments both predicted outcome, but in opposite directions. Confirming Hooley and Teasdale's report, we observed little relationship between objective ratings of relative's criticism of the patient in the patient's absence (the CFI) and the patient's perception of the relative's criticism in general (r = .15). These data drive home the importance of learning more about patients' perceptions of (and response to) criticism.
In the present study, as is typical of samples in which most relatives are spouses, EOI as determined by the CFI was infrequent. Accordingly, EOI was represented here in a unique fashion by an empirically derived four-item factor score comprising relatives' self-ratings as well as the CFI rating. Although our findings must be considered with some caution because of this atypical measure, they are roughly consistent with the only other study we have seen of EOI as a predictor of treatment dropout (Szmukler, Eisler, Russell, & Dare's, 1985, study of eating disorder) and congruent with logical predictions from the construct of EOI in families of anxious patients: When relatives are emotionally overinvolved and experience guilt and self-blame, they may have difficulty realizing or acting on the knowledge that it is in the patients' best interest to endure the short-term distress of exposure treatment so as to improve in the long run. Indeed, Hansen, Hoogduin, Schaap, and de Haan (1992) provided retrospectively collected evidence that pressure from family members was significantly related to OCD patients' treatment completion. In the present study, for every increase of 1 SD in the EOI factor score, the odds of patients' dropping out of treatment increased threefold.
Findings from the present investigation suggest that EE plays a role in response to treatment for anxiety disorders, as is the case for other psychiatric disorders that have been studied. However, these correlational data cannot be taken as proof of a causal relationship between EE and outcome. Despite the longitudinal design of this research, it is possible that some unknown characteristic of the patient might lead to both poor outcome and EE. Nonetheless, we have ruled out or statistically controlled for the effects of a wide variety of potential confounds, such as amount and type of treatment, initial symptom severity and global functioning, unstable doses of medication, chronicity, SES, and type of relative with whom the patient lived. Thus, our findings, coupled with those of prior researchers, suggest that behavior therapists should devise family interventions to mitigate the effects of EE on treatment completion and response. On the whole, the inclusion of spouses in behavior therapy for PDA and OCD has not led to improved treatment outcome (see review by Emmelkamp & Gerlsma, 1994). However, unlike family interventions developed for schizophrenia and bipolar disorder (e.g., Barrowclough & Tarrier, 1992; Miklowitz & Goldstein, 1997), these interventions were not designed with an explicit focus on EE. In the two investigations including interventions that would be expected to reduce EE (e.g., problem solving and communications training), the combination of couples or family interventions along with anxiety-focused behavior therapy led to better outcome than anxiety-focused treatment alone for PDA patients (Arnow, Taylor, Agras, & Telch, 1985) and for anxious children (Barrett, Dadds, & Rapee, 1996). Hence, research directed at changing EE or the patient's response to EE has promise for improving the treatment outcome for severe anxiety disorders by reducing dropout and increasing improvement with behavior therapy.
References
Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage. [Context Link]
Arnow, B. A., Taylor, C. B., Agras, W. S., & Telch, M. J. (1985). Enhancing agoraphobia treatment outcome by changing couple communication patterns. Behavior Therapy, 16, 452–467. [Context Link]
Barrett, P. M., Dadds, M. R., & Rapee, R. M. (1996). Family treatment of childhood anxiety: A controlled trial. Journal of Consulting and Clinical Psychology, 64, 333–342. [Context Link]
Barrowclough, C., & Tarrier, N. (1992). Families of schizophrenic patients: Cognitive behavioural intervention. London: Chapman & Hall. [Context Link]
Butzlaff, R. L., & Hooley, J. M. (1998). Expressed emotion and psychiatric relapse: A meta-analysis. Archives of General Psychiatry, 55, 547–552. Bibliographic Links [Context Link]
Chambless, D. L., Bryan, A. D., Aiken, L. S., Steketee, G., & Hooley, J. M. (1999). The structure of expressed emotion: A three-construct representation. Psychological Assessment, 11, 67–76. Ovid Full Text [Context Link]
Chambless, D. L., Fauerbach, J., Floyd, F. J., Wilson, K. A., Remen, A., & Renneberg, D. (1997, November). Behavioral observation of the marital interaction of agoraphobic women: A comparison with community couples. Paper presented at the meeting of the Association for Advancement of Behavior Therapy, Miami Beach, FL. [Context Link]
Cohen, J. (1983). The cost of dichotomization. Applied Psychological Measurement, 7, 249–253. [Context Link]
Emmelkamp, P. M. G., & Gerlsma, C. (1994). Marital functioning and the anxiety disorders. Behavior Therapy, 25, 407–429. Bibliographic Links [Context Link]
Feske, U., & Chambless, D. L. (in press). A review of assessment measures for obsessive–compulsive disorder. In W. K. Goodman, M. Rudorfer, & J. Maser (Eds.), Obsessive compulsive disorder: Contemporary issues in treatment. Hillsdale, NJ: Erlbaum. [Context Link]
Gottschalk, L. A., & Keatinge, C. (1993). Influence of patient caregivers on course of patient illness: “Expressed emotion” and alternative measures. Journal of Clinical Psychology, 49, 898–912. [Context Link]
Hansen, A. M. D., Hoogduin, C. A. L., Schaap, C., & de Haan, E. (1992). Do drop-outs differ from successfully treated obsessive–compulsives? Behaviour Research and Therapy, 30, 547–550. [Context Link]
Hollingshead, A. B. (1975). Four-Factor Index of Social Status. (Available from A. B. Hollingshead, Department of Sociology, Yale University, P.O. Box 1965, New Haven, CT 06520) [Context Link]
Hooley, J. M. (1985). Expressed emotion: A review of the critical literature. Clinical Psychology Review, 5, 119–139. [Context Link]
Hooley, J. M., & Teasdale, J. D. (1989). Predictors of relapse in unipolar depressives: Expressed emotion, marital distress, and perceived criticism. Journal of Abnormal Psychology, 98, 229–235. Ovid Full Text [Context Link]
Jacobson, N. S., Wilson, L., & Tupper, C. (1988). The clinical significance of treatment gains resulting from exposure-based interventions for agoraphobia: A reanalysis of outcome data. Behavior Therapy, 19, 539–554. Bibliographic Links [Context Link]
Jansson, L., & Ost, L. G. (1982). Behavioral treatments for agoraphobia: An evaluative review. Clinical Psychology Review, 2, 311–336. [Context Link]
Lebell, M. B., Marder, S. R., Mintz, J., Mintz, L. I., Tompson, M., Wirshing, W., Johnston-Cronk, K., & McKenzie, J. (1993). Patients' perceptions of family emotional climate and outcome in schizophrenia. British Journal of Psychiatry, 162, 751–754. Bibliographic Links [Context Link]
Lelliott, P. (1988). Target problems. In M. Hersen & A. S. Bellack (Eds.), Dictionary of behavioral assessment techniques (pp. 468–469). New York: Pergamon. [Context Link]
Leonard, H. L., Swedo, S. E., Lenane, M. C., Rettew, D. C., Hamburger, S. D., Bartko, J. J., & Rapoport, J. L. (1993). A two to seven year follow-up study of 54 obsessive compulsive children and adolescents. Archives of General Psychiatry, 50, 429–439. Bibliographic Links [Context Link]
Miklowitz, D. J., & Goldstein, M. J. (1997). Bipolar disorder: A family-focused approach. New York: Guilford Press. [Context Link]
Okasha, A., El Akabawi, A. S., Snyder, K. S., Wilson, A. K., Youssef, I., & El Dawla, A. S. (1994). Expressed emotion, perceived criticism, and relapse in depression: A replication in an Egyptian community. American Journal of Psychiatry, 151, 1001–1005. [Context Link]
Peter, H., & Hand, I. (1988). Patterns of patient–spouse interaction in agoraphobics: Assessment by Camberwell Family Interview (CFI) and impact on outcome of self-exposure treatment. In I. Hand & H. -U. Wittchen (Eds.), Panic and phobias: 2. Treatments and variables affecting course and outcome (pp. 240–251). Berlin, Germany: Springer-Verlag. [Context Link]
Riso, L. P., Klein, D. N., Anderson, R. L., Ouimette, P. C., & Lizardi, H. (1996). Convergent and discriminant validity of perceived criticism from spouses and family members. Behavior Therapy, 27, 129–139. Bibliographic Links [Context Link]
Rodebaugh, T. L., Chambless, D. L., Renshaw, K. D., & Steketee, G. (1998, November). The content of relatives' criticisms of anxious patients on the Camberwell Family Interview. Paper presented at the meeting of the Association for Advancement of Behavior Therapy, Washington, DC. [Context Link]
Rosenthal, R. (1983). Assessing the statistical and social importance of the effects of psychotherapy. Journal of Consulting and Clinical Psychology, 51, 4–13. [Context Link]
Shrout, P. E., & Fleiss, J. L. (1979). Intraclass correlations: Uses in assessing rater reliability. Psychological Bulletin, 86, 420–428. Ovid Full Text Bibliographic Links [Context Link]
Spitzer, R. L., Williams, J. B. W., Gibbon, M., & First, M. B. (1989). Structured Clinical Interview for DSM–III–R—Patient Version. New York: New York State Psychiatric Institute, Biometrics Research Department. [Context Link]
Stanley, M. A., & Turner, S. M. (1995). Current status of pharmacological and behavioral treatment of obsessive–compulsive disorder. Behavior Therapy, 26, 163–186. Bibliographic Links [Context Link]
Steketee, G. (1987). Predicting relapse following behavioral treatment for OCD: The impact of social support. Unpublished doctoral dissertation, Bryn Mawr College Graduate School, Bryn Mawr, PA. [Context Link]
Steketee, G. (1993). Social support and treatment outcome of obsessive compulsive disorder at 9-month follow-up. Behavioural Psychotherapy, 21, 81–95. Bibliographic Links [Context Link]
Steketee, G., Chambless, D. L., Tran, G. Q., Worden, H., & Gillis, M. M. (1996). Behavioral avoidance test for obsessive compulsive disorder. Behaviour Research and Therapy, 34, 73–83. [Context Link]
Szmukler, G. I., Eisler, I., Russell, G. F. M., & Dare, C. (1985). Anorexia nervosa, parental “expressed emotion” and dropping out of treatment. British Journal of Psychiatry, 147, 265–271. Bibliographic Links [Context Link]
Tarrier, N., Sommerfield, C., & Pilgrim, H. (in press). Relatives' expressed emotion (EE) and PTSD treatment outcome. Psychological Medicine. [Context Link]
Tompson, M. C., Goldstein, M. J., Lebell, M. B., Mintz, L. I., Marder, S. R., & Mintz, J. (1995). Schizophrenic patients' perceptions of their relatives' attitudes. Psychiatry Research, 57, 155–167. Bibliographic Links [Context Link]
Vaughn, C., & Leff, J. (1976). The measurement of expressed emotion in the families of psychiatric patients. British Journal of Social and Clinical Psychology, 15, 157–165. [Context Link]
Weissman, M. M., & Bothwell, S. (1976). Assessment of social adjustment by patient self-report. Archives of General Psychiatry, 33, 1111–1115. [Context Link]
1Treatment manuals are available from Dianne L. Chambless or Gail Steketee. [Context Link]
2Further details of the statistical analyses are available from Dianne L. Chambless. [Context Link]
3Effective power is probably less, in that interactions are more susceptible to loss of power from measurement unreliability than are first-order effects (Aiken & West, 1991). [Context Link]
4One notable difference between the present study and that of Tarrier et al. (in press) is that, in the latter case, hostility and critical comments on the CFI were highly correlated (r = .77 vs. r = .46 in the present sample). Hence, in comparison with our sample and that of Peter and Hand (1988), Tarrier et al.'s sample was much less likely to contain relatives who were critical but not hostile. [Context Link]
________________________________________

Accession Number: 00004730-199910000-00004
Copyright (c) 2000-2007 Ovid Technologies, Inc.
Version: rel10.5.8, SourceID 1.13281.2.32.1.0.2.197.1.4.1.5