In their review of drug treatments for borderline personality disorder, Lieb et al, ^{Reference Lieb, Völlm, Rücker, Timmer and Stoffers1} despite considering similar evidence, draw largely different conclusions from those we drew when developing the National Institute for Health and Clinical Excellence (NICE) guideline. ² Lieb et al recommend a range of drugs. These include anticonvulsants for affective dysregulation symptoms (topiramate, valproate semisodium and lamotrigine) and impulsive–behavioural dyscontrol symptoms (lamotrigine and topiramate); and anti-psychotics (aripiprazole and olanzapine) for cognitive–perceptual symptoms. In contrast, we do not recommend drug treatment other than for the treatment of comorbid disorders.

There are a number of reasons for the disparity. First, we did not consider the evidence from some studies to be usable. ^{Reference Nickel, Nickel, Mitterlehner, Tritt, Lahmann and Lieberich3–Reference Nickel, Muehlbacher, Nickel, Kettler, Pedrosa and Bachler7} These trials tended to find large effect sizes favouring treatment compared with effect sizes from other trials. Following further investigation, we considered this evidence for topiramate, lamotrigine or aripiprazole to be unreliable and excluded the trials from our analysis (see p. 218 of the full guideline ² ).

Second, most other recommendations made by Lieb et al are based on weak and/or low-quality evidence. We do not agree with the interpretation of the evidence for valproate, which Lieb et al claim shows a reduction in interpersonal problems and depression. The apparent effect on interpersonal problems is derived from a trial of 30 participants with more than 60% drop out. The effect on depression, which we noted as not statistically significant (s.m.d. = −0.61 (95% CI −1.29 to 0.07)), is derived from a larger trial with skewed data, in which over 60% of participants were not diagnosed as having borderline personality disorder. We therefore graded this evidence ‘low quality’.

The authors also claim ‘favourable results’ for haloperidol and the other antipsychotics on symptoms of affective dysregulation, and for omega-3 fatty acid supplementation and flupentixol decanoate. It is unclear for which ‘symptom constellation’ these latter drugs are recommended. We calculated similar effect sizes, but tended to grade the quality of evidence ‘low’ because of single studies, skewed data and wide confidence intervals. We excluded the trial of flupentixol ^{Reference Montgomery, Roy and Montgomery8} because its inclusion criterion was not specifically a diagnosis of borderline personality disorder.

Third, NICE guidelines are developed as a practical synthesis of clinical recommendations based on a pragmatic analysis of the evidence for the clinical effectiveness and cost-effectiveness, including evidence of harm, of particular treatments and approaches to a problem. As far as possible we do not rest NICE guideline development on speculative theory. The American Psychiatric Association ⁹ based their recommendations about selective serotonin reuptake inhibitors and low-dose antipsychotics on a speculative theoretical model which has never been tested in hypothesis-driven studies. Treatment recommendations thus derived are based on post hoc reconstructions rather than primary evidence. Lieb et al implicitly use this model to understand the evidence and to develop recommendations.

Fourth, Lieb et al made recommendations regarding a number of drugs on the basis of single trials in which positive findings are restricted to one or two symptoms. They place greater emphasis on simple statistical significance without sufficient consideration of clinical significance, whether the outcome measures used were appropriate – in many cases they are not – or indeed the potential for harm. For example, valproate semisodium is an especially dangerous drug for women of child-bearing years who may unexpectedly become pregnant; and antipsychotics have a wide range of neurological side-effects, some of which can be permanent, as well as metabolic effects leading to weight gain and an increased risk of diabetes.

Finally, the NICE guideline considered evidence for non-drug treatments, for example psychological therapies, and looked at the care pathway within the National Health Service (NHS) in England and Wales. Recommendations relating to drug treatment were therefore developed in the context of evidence for the whole range of treatments for, as well as the clinical management of, borderline personality disorder. Consensus-based recommendations for the management of crises and sleep problems, experiences which in the NHS commonly lead to excessive reliance on various pharmacological solutions, were also included.

No drug has been licensed in the UK for borderline personality disorder. It is important that drugs that are used commonly within the NHS are subject to post-licensing surveillance by the Medicines and Healthcare products Regulatory Agency. It is therefore unusual for a NICE guideline to recommend the use of any unlicensed drug. There are exceptions to this, for example where there are no other treatments or other treatments are associated with significant harm. These remain exceptions, nevertheless. We hope that readers can see that, with these considerations in mind, the guideline group was correct in deciding not to recommend drug treatments for either the core symptoms of borderline personality disorder or indeed for any symptom clusters. More good-quality evidence is required.