Prediction of medium-term outcome in Cluster B personality disorder following residential and outpatient psychosocial treatment

Chiesa, M., Cirasola, A., & Fionagy, P. (2017). Four years comparative follow‐up evaluation of communitybased, step‐down, and residential specialist psychodynamic programmes for personality disorders. Clinical Psychology and Psychotherapy, 24, :1331–1342.


There is a paucity of research concerning the identification of individual characteristics predictive of outcome in the treatment of personality disorders as there are only a handful of reports that have attempted to identify patients’ clinical characteristics predictive of treatment outcome. 

In this study, we carried out a predictor analysis of a relative homogeneous group of hospitalized patients (n=73) with a standardized diagnosis of cluster b personality disorder (borderline, histrionic and narcissistic), and we attempted to locate the presence of significant predictive factors that influenced positive and negative medium-term outcome. These patients were admitted to two different psychoanalytically-oriented psychosocial programs for personality disorder: (a) long-term inpatient treatment, and (b) a step-down program. Because of the high co-morbidity of diagnosis in the sample, we also set out to evaluate whether specific combinations of diagnostic categories within Axis-I and Axis-II were significantly associated with outcome 24-month after intake in the dimensions of severity of symptoms presentation (SCL-90-R-GSI), social adjustment (SAS) and global assessment of functioning (GAS). In addition, we explored predictors of outcome specific to each treatment in order to refine clinical recommendations for selection for specific programs.

The stepwise logistic regression analysis with improvement status as the dependent variable revealed that the model including self-mutilation the year prior to intake, avoidant PD, intake GAS scores, age at intake and length of treatment was predictive of improvement at 24 months. Cluster B patients with no previous self-mutilation, who did not have a co-morbid avoidant PD, with higher GAS intake scores, longer treatment exposure and younger age were more likely to improve.  Absence of self-mutilation and co-morbid avoidant PD improved 6 and 4 folds the chances to achieve positive outcome, respectively. Six years below the mean age of 30 years, 31 weeks more treatment from the mean of 53 weeks and 6.5 points above the GAS mean score of 46.5 double the chances of improvement two years after treatment intake. Although deliberate self-injury was found to be a negative predictor, improvement rates in self-mutilating patients were significantly different in the two different treatment programs (60% in the step-down program versus 24% in the long-term residential program). A cluster analysis on Axis-I diagnoses identified a larger group whose primary Axis-I diagnosis was major depression and a smaller more heterogeneous group with anxiety or substance misuse diagnoses. No association with improved status at 24 months was found. 

Of the two homogeneous clusters of co-morbid personality disorder diagnoses the large borderline and self-defeating group had achieved significantly greater level of improvement compared to the smaller borderline, avoidant, paranoid, dependent cluster. The difference in outcome between the two PD diagnostic clusters appears to be accounted for by a differential treatment response in program allocation: whereas almost 90% of B-SF patients allocated to the step-down model improved, only 52% of those allocated to the long-term residential program did so.  There was no similar difference between the improvement rates in the two treatment arms for the B-P-A-D cluster. Thus, it seems that therapeutic advantage came especially from the step-down treatment of the self-defeating borderline group of patients.


The findings may carry potential clinical implications concerning patient selection and treatment delivery for inpatient and outpatient psychosocial programs for Cluster B personality disorder.  Limitations include a relatively low sample size for a regression analysis, and a larger sample of Cluster B patients may be needed to ensure greater reliability of results.


Marco Chiesa MD, FRCPsych,

The Cassel Hospital & University College London, Richmond, TW10 7JF, UK

Tel + 44 (0) 20 8237 2902


Peter Fonagy PhD, FBA

University College London & The Anna Freud Centre