Bayesian approach to bipolar guidelines. Academic Article uri icon


  • same medication in the past. Given the danger of non-response to a new treatment we suggest that it is better to rely on the individual patient’s history when this is available than to rely only on a concept of quality of evidence in the literature. Clinicians have long believed that longitudinal data is as important for specific patients as knowledge of the latest randomized trial with a large number of unselected patients. If a patient with current bipolar depression responded in six previous depressions over 25 years to imipramine with full remission and no manic relapse, it would be unwise to give him treatment based on an algorithm of hierarchically organized current controlled trials. A common case is that of a patient who received lithium and haloperidol during each of two previous manic attacks, one 8 years ago and one 4 years ago, with no worrisome side effects. Would it really be good advice to use the algorithm presented in Figure 1 of WFSBP 2009 BP Guidelines suggesting the equivalence of the various level 1 monotherapies? Given the gradually increasing standards of psychopharmacology research, newer compounds are likely to have stronger quality of evidence even though they may have in truth no larger an effect size. Adequate attention to individual past response could remedy this distortion. Past history and controlled data could be integrated in a Bayesian approach (Goodman 1999).

publication date

  • January 1, 2010