Overview of studies
Psychodynamic therapy. As will be subsequently described, the 14 studies of psychodynamic therapy used different forms. Psychoanalysis was not applied. In most of the studies, psychodynamic therapy was time-limited. The effects of outpatient individual psychodynamic therapy were examined in eight studies. Woody et al. (
13) compared the effects of either 12 sessions of manualized supportive-expressive psychodynamic therapy (which used the Luborsky manual) or cognitive behavior therapy (per Beck), which were added to standard drug counseling. Stevenson and Meares (
15) examined the outcome of a 1-year psychoanalytically oriented (self-psychological) regimen of outpatient psychotherapy. Therapy was conducted by trainees who received weekly supervision. Hoglend (
16) studied the outcome of manualized psychodynamic focal therapy, which lasted an average of 27.5 sessions. Adherence to the manual was ensured. Winston et al. (
17) compared short-term anxiety-provoking therapy (after Davenloo) to both another form of psychodynamic therapy (brief adaptive psychotherapy) and to a waiting list control condition. Each form of psychotherapy was administered for 40 weekly sessions. Manuals were used, and adherence was tested and ensured. Monsen et al. (
19) studied a form of psychodynamic therapy that focused on object relations and self-psychology. Therapists were specially trained and supervised. Therapy lasted an average of 25 months. Munroe-Blum and Marziali (
20) compared a time-limited group treatment with individual psychodynamic therapy. Each form of therapy lasted an average of 17 sessions. Manuals were used and adherence was ensured. Since the authors did not find significant differences between individual and group therapy, Munroe-Blum and Marziali (
20) pooled the results of both conditions. However, the authors reported the results separately for the two conditions (personal communication, June 25, 2001). Since group therapy was based on an interpersonal rather than on a psychodynamic model, only the results reported for individual psychodynamic therapy were included in this review. Hardy et al. (
10) compared the effects of manualized psychodynamic/interpersonal therapy and cognitive behavior therapy in depressed patients with and without cluster C personality disorders. In both forms of therapy, eight versus 16 sessions were applied. Diguer et al. (
18) studied the effects of 16 sessions of supportive-expressive psychodynamic therapy (which followed Luborsky’s manual) in depressed outpatients with and without a personality disorder.
In three studies, psychodynamic treatment was applied within a partial hospitalization program for patients with personality disorders. Karterud et al. (
12) examined psychodynamically oriented day hospital community treatment, which lasted for an average of 6 months. Wilberg et al. (
21) reported the results of psychodynamic community day treatment with subsequent outpatient psychoanalytic group therapy for patients with borderline personality disorder. These results were compared with a treatment-as-usual condition, i.e., psychodynamic day hospital treatment without group therapy. Since the authors had found significant differences between the two kinds of psychodynamic treatments, we assessed the effect sizes of the two treatments separately. Bateman and Fonagy (
22) compared an analytically oriented partial hospitalization treatment program for borderline patients (maximum 18 months) with standard psychiatric care. Outcome data for an 18-month follow-up were recently reported (
37).
Three studies examined the outcome of an inpatient psychodynamic treatment. Liberman and Eckman (
9) compared insight-oriented therapy with behavioral therapy in the treatment of patients with repeated suicide attempts, each given for 32 sessions within a 10-day hospital treatment. Tucker et al. (
23) studied the results of inpatient psychotherapeutic treatment that focused on interpersonal relations and intrapsychic organization. Patients were treated for an average of 8.4 months. Antikainen et al. (
24) looked at the effects of inpatient psychotherapeutic community treatment for borderline patients that focused on the so-called split-type defense mechanisms. Treatment lasted for an average of 3 months.
Cognitive behavior therapy. Three of the aforementioned studies also reported data for the outcome of behavioral or cognitive behavior therapy (
9,
10,
13). Eight other studies examined cognitive behavior therapy in the treatment of personality disorders. Linehan et al. (
25) studied the interpersonal effects of dialectical behavior therapy. The manualized treatment lasted for 1 year and was compared with treatment as usual. This was true of another study of dialectical behavior therapy, which compared the effects of dialectical behavior therapy to treatment as usual in patients with borderline personality disorder who had comorbid drug dependence (
26). Bohus et al. (
27) applied dialectical behavior therapy to the inpatient treatment of female parasuicidal borderline patients that lasted for 3 months. In a randomized controlled study, Springer et al. (
14) compared a modification of dialectical behavior therapy in an inpatient setting with a control condition. In this study, short-term group treatment was given to patients with personality disorders for 13 days (mean=6.3 sessions). Since it was not clear whether the applied form of therapy represented dialectical behavioral therapy, we prefer to regard it as a form of cognitive behavior therapy. Alden (
28) examined three specific forms of cognitive behavior therapy treatments that were compared to a waiting list control condition. The manualized treatment consisted of 10 weekly group sessions. Adherence to the manuals was demonstrated by ratings of therapy sessions. Fahy et al. (
29) studied cognitive behavior therapy in bulimia nervosa patients with and without personality disorders. The treatment lasted for 8 weeks. In a study by Stravynski et al. (
30), outpatients with diffuse social phobia and avoidant personality disorder received 12 sessions of social skills training, either alone or combined with cognitive modification. Manuals were used, and ratings of therapy sessions were applied to ensure adherence to the manuals. Outcome did not differ significantly between the two therapies. For this reason, we assessed a mean overall effect size for the treatments. Brown et al. (
31) studied the effects of cognitive behavior therapy in three groups of patients: patients with generalized social phobia (those with and those without avoidant personality disorder) and patients with nongeneralized social phobia. For this study, no data that allowed calculation of effect sizes in the form of Cohen’s d were provided; however, data referring to recovery were reported.
Psychodynamic and cognitive behavior therapy combinations. Johnson et al. (
35) studied a treatment that integrated cognitive behavior treatment of bulimic symptoms with psychodynamic therapy. They compared the outcome of borderline and nonborderline bulimic patients. As the pretreatment standard deviations were not published, we estimated the effect sizes conservatively by using the published t values (
35, pp. 620–622). Ryle and Golynkina (
36) examined the effects of time-limited cognitive-analytic therapy on a group of outpatients with borderline personality disorder. The outcome data of these two studies of combined therapies were treated separately in our meta-analysis.
Summary of study characteristics
Table 2 presents the length, number of sessions, followup duration, and other features of the psychodynamic and cognitive behavior studies. Treatment manuals were used in five studies of psychodynamic therapy (
10,
13,
16–
18) and in four studies that used cognitive behavior therapy only (
25,
26,
28,
30). Stevenson and Meares (
15) used weekly therapist supervision instead of a manual to ensure adherence to the applied form of therapy. This was true for Ryle and Golynkina (
36), who in addition examined a measure of therapist competence. With a few exceptions (
9), therapists had been specially trained in psychodynamic therapy or cognitive behavior therapy. In two studies of psychodynamic therapy, therapies were conducted mostly by trainees (
15,
21). Concurrent use of medication was reported in nine studies (
10,
12,
13,
22,
24–
27,
36).
Patient groups. Patients with borderline personality disorder were treated in seven of the psychodynamic therapy studies (
9,
15,
20–
24). Patients with borderline personality disorder, schizotypal personality disorder, and other personality disorders were studied by Karterud et al. (
12). Mixed types from all three DSM-III personality disorder clusters were treated in two studies (
17,
19). The patients in the study by Hoglend (
16) had cluster B or C personality disorders. Hardy et al. (
10) studied patients with major depression and a concomitant diagnosis of a cluster C personality disorder. Diguer et al. (
18) reported the results of patients with major depression and a concomitant diagnosis of a personality disorder without specifying the types of personality disorders. Woody et al. (
13) reported the results for opiate addicts with and without antisocial personality disorder. Thus, predominantly the severe forms of personality disorders (clusters A and B) were treated in most of the studies. Personality disorders predominantly of a neurotic level of dysfunction (cluster C) were treated in only two studies of psychodynamic therapy (
10,
17). The study of Hoglend (
16) included almost identical proportions of cluster C and cluster B patients (about 50% each).
In the seven studies using cognitive behavior therapy only, patients with borderline personality disorder (
25–
27), bulimia nervosa with and without personality disorder (
29), and patients with avoidant personality disorder (
28,
30,
31) were studied. Linehan et al. studied parasuicidal women with borderline personality disorder (
25) and women with borderline personality disorder and drug dependence (
26). Bohus et al. (
27) reported the outcome for borderline inpatients. Springer et al. (
14) examined a sample of inpatients with personality disorder, of which 50% were diagnosed as having borderline personality disorder.
Two studies required an axis I diagnosis of major depression as an inclusion criterion (
10,
18), and for two studies opiate addiction or substance use disorder were required (
13,
26). In one of the studies that used cognitive behavior therapy only, a diagnosis of bulimia nervosa was required (
29); in two other studies, a diagnosis of social phobia was required (
30,
31). In seven studies, information was given about the prevalence of comorbid axis I diagnoses (
12,
13,
16,
17,
22,
26,
36). The most prevalent comorbid axis I diagnoses, in descending order, were depression (major depression or dysthymia), adjustment and anxiety disorders, substance abuse, somatoform disorder, and eating disorders.
Design and outcome measures. Three studies of psychodynamic therapy used randomized controlled designs with a waiting list or a nonspecific treatment condition (
13,
17,
22). Four studies included randomized comparisons of two active treatments (
9,
10,
13,
20). Of the studies using cognitive behavior therapy only, five studies applied randomized controlled designs (
14,
25,
26,
28,
30). The other studies were naturalistic observations of therapy groups.
Three studies used only self-report outcome measures (
9,
14,
17), two studies used only observer-rated measures (
16,
24), and the other studies used both. The most frequently used self-report instruments were the Beck Depression Inventory (
9,
10,
18,
20,
22,
24,
27) and the SCL-90-R (
10,
12,
17,
19,
22,
27). The most frequently used observer-rated measures were the Health-Sickness Rating Scale and the Global Adjustment Scale (
12,
16,
18,
23,
25). In all studies, observer-rated measures were assessed by using structured interview methods that were applied by independent assessors.
Effect size analyses
Psychodynamic therapy studies. As seen in Table 3, psychodynamic therapy yielded an unweighted mean overall effect size of 1.46 (SD=0.73), which significantly differed from zero (t=7.73, df=14, p=0.0001). For self-report measures, the unweighted mean effect size was 1.08 (SD=0.36); for observer-rated measures, it was 1.79 (SD=1.07). These mean effect sizes both differed significantly from zero (self-report: t=10.51, df=11, p=0.0001; observer-rated: t=5.78, df=11, p=0.0001). Adjusted for sample size (
34), the corresponding effect sizes were 1.40 (t=7.85, df=14, p<0.001), 1.03 (t=10.95, df=11, p=0.0001), and 1.71 (t=5.91, df=11, p=0.0001). In the two randomized controlled studies that reported data for the control condition (
17,
22), active psychodynamic therapy was significantly more effective than the control conditions. For these two studies, the differences in within-condition effect sizes between psychodynamic therapy and the control condition yielded an unweighted mean difference of 1.32 for self-report measures. For observer-rated measures, only one study provided data that allowed comparison of active psychodynamic therapy with a control condition (
22). We used the t and chi-square statistics reported by Bateman and Fonagy (
22) to calculate effect sizes (
32 [pp. 67, 223, 225]). Psychodynamic therapy yielded medium to large between-condition effect sizes for mean duration of inpatient episodes (d=4.29), number of individuals no longer self-mutilating (w=0.45), and number of individuals no longer parasuicidal (w=0.34).
Cognitive behavior therapy studies. For the 11 studies that used cognitive behavior therapy, there was a mean unweighted mean overall effect size of 1.00 (SD=0.48), which significantly differed from zero (t=6.56, df=9, p=0.0001). For self-report measures, the unweighted mean effect size was 1.20 (SD=0.38); for observer-rated measures, it was 0.87 (SD=0.71). These mean effect sizes both differed significantly from zero (self-report: t=8.95, df=7, p=0.0001; observer-rated: t=3.43, df=7, p=0.01). Adjusted for sample size (
34), the corresponding effect sizes were 0.95 (t=6.67, df=9, p=0.0001), 1.14 (t=9.11, df=7, p=0.0001), and 0.82 (t=3.47, df=7, p=0.01). In three of the five randomized controlled studies (
25,
26,
28), active cognitive behavior therapy was significantly more effective than the control conditions. In one study, no significant difference was found between cognitive behavior therapy and a discussion control group (
14). For the three studies (
25,
26,
28) that reported the relevant data, the differences in within-condition effect size between cognitive behavior therapy and the control condition yielded an unweighted mean difference of 0.81 for self-report measures and 0.50 for observer-rated measures. While these values correspond to medium to large effect sizes (
32), the mean differences in effect sizes did not differ significantly from zero (t=4.76, df=1, p=0.13 and t=4.00, df=2, p=0.06, respectively) on account of the small number of studies. In the study of Linehan et al. (
38), dialectical behavior therapy yielded a significantly greater reduction in parasuicidal acts than the control condition. We used the means and standard deviations reported by Linehan et al. to calculate effect sizes. For the reduction in parasuicidal acts, dialectical behavioral therapy yielded a medium between-condition effect size of 0.53.
In Table 4, the mean unweighted effect sizes for the most frequently used measures are presented. Effect sizes for both patients with and without personality disorders were assessed. For psychodynamic therapy, the largest effect sizes in personality disorders were found with the Health-Sickness Rating Scale or Global Adjustment Scale. For cognitive behavior therapy, the largest effect sizes were found with the Beck Depression Inventory.
Psychodynamic and cognitive behavior therapy combination studies. For the two studies using combinations of psychodynamic therapy and cognitive behavior therapy (
35,
36), we found a mean unweighted effect size for self-report measures of 0.79. Observer-rated measures data that allowed calculation of mean effect sizes in the form of Cohen’s d were not available.
Correlations of measures and recovery rate
Self-rated and observer-rated measures showed a positive but nonsignificant correlation in studies of both psychodynamic therapy (rs=0.32, N=9, p=0.41) and cognitive behavior therapy (rs=0.26, N=6, p=0.62). However, for cognitive behavior therapy, only six studies provided both self-rated and observer-rated measures.
Three studies reported data referring to recovery from personality disorder after psychodynamic therapy, defined as no longer fulfilling the full criteria for personality disorder (
15,
16,
19). Using these data, we calculated a mean recovery rate from personality disorders of 59% after a mean of 15 months of treatment. Concerning cognitive behavior therapy, only Brown et al. (
31) reported data referring to recovery rates. After treatment, 47% of the patients were no longer diagnosed with avoidant personality disorder. These results and the definition of recovery, which is debatable, will be subsequently discussed.
Factors influencing outcome and effect size analyses
Diagnosis. In eight studies, the effects of psychodynamic therapy in patients with borderline personality disorder were reported (
9,
12,
15,
20–
24). The mean unweighted overall effect size was 1.31 (SD=0.71). The mean unweighted effect sizes for self-rated and observer-rated measures were 1.00 (SD=0.25) and 1.45 (SD=1.09), respectively. These effect sizes were significantly different from zero (t=5.25, df=7, p=0.001; t=10.47, df=6, p=0.0001; t=3.28, df=5, p=0.02). The mean treatment duration was 33 weeks (SD=26.85). Adjusted for sample size, the corresponding effect sizes were 1.27 (t=5.18, df=7, p=0.001), 0.96 (t=10.27, df=6, p=0.0001), and 1.35 (t=3.47, df=5, p=0.02).
Four studies reported the effects of cognitive behavior therapy in patients with borderline personality disorder (
14,
25–
27). The mean unweighted overall effect size was 0.95 (SD=0.31). The mean unweighted effect sizes for self-rated and observer-rated measures were 0.97 (SD=0.24) and 0.81 (SD=0.54), respectively. The mean treatment duration was 22 weeks (SD=26.84). Adjusted for sample size, the corresponding effect sizes were 0.89, 0.92, and 0.76.
Core pathology of personality disorders. Some of the studies included measures that were more specific to personality disorders. For psychodynamic therapy, two studies reported results for the Inventory of Interpersonal Problems (
10,
22). Interpersonal problems are of some importance, since they are regarded as one of the core problems in personality disorders (
39, also DSM-IV). For another measure of social functioning, the Social Adjustment Scale, outcome was reported by two other studies (
17,
20). Monsen et al. (
19) applied a measure of affect consciousness. They also reported data for the MMPI and for Morey’s Personality Disorder Scales, but these data were not included in our meta-analysis because effect sizes were not reported for all scales. Stevenson and Meares (
15) reported outcome data referring to a measure of DSM-III criteria. Hardy et al. (
10) applied a measure of self-esteem. Psychodynamic therapy yielded an unweighted mean effect size of 1.56 (SD=0.76) for these more specific measures of personality disorder pathology (Table 3). This mean effect size differs significantly from zero (t=5.02, df=5, p=0.004). Adjusted for sample size, the corresponding effect size is 1.50 (t=5.02, df=5, p=0.004).
For cognitive behavior therapy, two studies (
10,
25) reported large effect sizes in more specific measures of personality disorder pathology (Inventory of Interpersonal Problems, Social Adjustment Scale, self-esteem, and the anger trait subscale).
Treatment duration. For psychodynamic therapy, we found a positive correlation between the overall effect size and length of treatment, although it did not yield statistical significance because of the small number of studies (rs=0.41, N=13, p=0.16). For cognitive behavior therapy, no correlation was assessed because the number of studies providing data was too small (N=8). For the same reason, no correlations with the number of sessions were assessed.
Dropouts. Studies differed with regard to the kind of dropout data that were reported: subjects dropping out during the initial assessment phase (before therapy), during therapy, after therapy, or during the follow-up period. This is one reason why dropout rates varied considerably between studies. For psychodynamic therapy, the mean dropout rate was 15.36%; for cognitive behavior therapy, it was 16.89 (Table 3). For psychodynamic therapy, the dropout rate correlated significantly and positively with the presence of a cluster A or cluster B personality disorder (r
s=0.72, N=9, p=0.03); for cognitive behavior therapy, the correlation was insignificant (r
s=0.33, N=8, p=0.43). Contrary to Perry et al. (
8), we did not find a significant correlation between length of therapy and dropout rate for either psychodynamic therapy (r
s=−0.31, N=9, p=0.41) or cognitive behavior therapy (r
s=0.31, N=6, p=0.54). Since the number of studies is very small, these results can only be preliminary.
Other factors. We tested for patient gender, inpatient versus outpatient status, use of therapy manuals, clinical experience of therapists, and study design. For psychodynamic therapy, the use of a therapy manual correlated significantly and positively with effect sizes for self-rated measures (rs=0.64, N=10, p=0.05). For observer-rated measures, no such correlation was found (rs=−0.17, N=11, p=0.61). None of the other variables correlated significantly with the outcome for either psychodynamic therapy or cognitive behavior therapy.