In their study of treatment for borderline personality disorder, Doering et al state that their results ‘demonstrate the significant superiority of transference-focused psychotherapy with regard to the primary outcome criteria of drop-out rate and suicide attempts during the treatment year’ when compared with treatment by experienced community psychotherapists. ^{Reference Doering, Hörz, Rentrop, Fischer-Kern, Schuster and Benecke1} They report that ‘significantly fewer participants dropped out of the transference-focused psychotherapy group (38.5% v. 67.3%) and also significantly fewer attempted suicide (d = 0.8, P = 0.009)’.

In our view, this interpretation of primary outcome criteria might lead to misunderstandings. As regards suicidality, the authors suggest that the P-value of 0.009 would relate to absolute numbers of attempted suicides during the treatment year. However, the actual difference in suicide attempts during the treatment year (7/52 v. 11/52) is not significant (P = 0.44, continuity-corrected χ²-test, LOCF analysis). The significant P-value reported by the authors seems to correspond to change scores (defined as 1/0/–1 by the authors), not to suicide attempts during the treatment year, which seems to be the outcome as defined in the study protocol (trial NCT00714311). The authors further suggest that the effect size of 0.8 would refer to the between-group comparison. However, the reported effect size seems to correspond to the within-group comparisons reported in Table DS2. The between-group effect size for suicide attempts during the treatment year would be small.

The other primary outcome to assess efficacy was defined as the rate of participants not receiving the allocated treatment plus the rate of participants who discontinued the allocated treatment. Even when accepting that patients not even starting treatment were included in a measure of treatment efficacy, it seems problematic to ascribe differences in this criterion to the efficacy of transference-focused psychotherapy without excluding accumulative effects of alternative explanations. The higher rate of non-starters among patients randomised to community therapists (the control condition) compared with those randomised to transference-focused psychotherapy (25% v. 13%) and the substantially higher rate of patients stopping treatment in the control group within the first month (Fig. 2 of the paper) might reflect a general preference of participants for transference-focused psychotherapy rather than its superior efficacy. Furthermore, the authors have not mentioned that this criterion combining non-starters and ‘drop-outs’ as primary outcome of efficacy was introduced post hoc (for post-hoc changes in the definition of primary outcome criteria see http://clinicaltrials.gov/archive/NCT00714311). Without addressing this issue, the statistical implications of this proceeding are difficult to evaluate.

As for the secondary outcome measures, the picture seems to be mixed. Some of the LOCF analyses indicated lower scores after transference-focused psychotherapy (e.g. number of borderline criteria, level of personality organisation). Other scores (e.g. general psychopathology, depression) were numerically higher after transference-focused psychotherapy and did not improve significantly more under it (P = 0.92 and P = 0.85 for general psychopathology and depression respectively).

Recapitulating, it seems that the claimed efficacy of transference-focused psychotherapy does not follow from the primary outcome criteria. Accordingly, further research seems necessary to establish the efficacy of this therapy in the treatment of borderline personality disorder.