In a recent paper, Dr Taylor raises an important issue: how long does it take for people with bipolar disorder to respond to lithium treatment?^{Reference Taylor1} In meta-analysis of data from three clinical trials, he found that patients randomised to lithium had significantly lower relapse rates than those receiving placebo, even in the first 2 weeks of treatment. This conclusion, however, does not answer the more relevant question as to how long a treatment trial should last before it can be established whether it is effective. In other words, is it worth waiting for let us say a year before switching to another option?

Clinical experience would suggest that there is a great range of time to response, which may relate to diagnostic and genetic heterogeneity.^{Reference Manchia, Cullis, Turecki, Rouleau, Uher and Alda2} Some patients respond within a few weeks whereas others may continue having major mood symptoms during the first year of treatment. Patients in the latter group will be inevitably categorised as ‘non-responders’ if even a single relapse is the criterion of treatment failure.

In Dr Taylor's study all three trials were based on discontinuation designs and were enriched for acute response to quetiapine or lamotrigine. However, enriched discontinuation designs with time to relapse as the outcome variable are less than ideal for evaluation of treatments of an illness that runs a lifelong course that is often highly unpredictable. Furthermore, most recent studies of long-term treatment of bipolar disorder (including the three trials discussed here) evaluate continuation treatment rather than recurrence prevention.

With respect to the minimal necessary length of treatment trial, there is practically no systematic data and the existing bipolar treatment guidelines stay away from the subject as well. In an earlier study, Ahrens et al attempted to estimate the time needed for patients to benefit from the suicide-reducing effect of lithium; they concluded that a treatment period of at least 2 years was necessary to return suicide risk to population baseline.^{Reference Ahrens, Muller-Oerlinghausen and Grof3} Given this, a more realistic design of maintenance studies might consider different outcome criteria such as affective morbidity assessed periodically over a sufficiently long observation period. As for a practical decision as to how long a treatment trial needs to last, it may become easier with advances in personalised treatment and discoveries about predictors of treatment response. Then it should be possible to individualise the length of a treatment trial – longer in those people expected to benefit from a specific treatment and abandon unsuccessful treatment earlier in those where the likelihood of response is equivocal.