Researchers and sponsors committed to the development of new and improved treatments for bipolar disorder have had to balance generalizability with the need to achieve assay sensitivity (separation from placebo) and, as a result, have made real but modest incremental progress. However, clinicians frequently note that the effectiveness of new medications appears to be substantially less than their documented efficacy in pivotal data sets submitted to the Food and Drug Administration for regulatory approval. Why is it that data from the most recently conducted randomized controlled trials look better than the drug’s performance in the real world?
Since data suggest that most acutely manic patients do not respond adequately to monotherapy, antimanic agents have more recently been studied in combination with other medications to improve generalizability. In this issue of the Journal, Vieta and colleagues (1) report on the adjunctive use of aripiprazole for the treatment of mania or mixed states in patients who are only partially responsive to lithium or valproate monotherapy. The findings from this trial, particularly effect size and tolerability, appear to be more credible and more consistent with what clinicians observe in the real world. The methodology employed is a welcome departure from previously conducted trials of mania, since high rates of attrition as a result of the lack of monotherapy efficacy were avoided through the use of an adjunctive design. In contrast to prior studies of combination therapy (2, 3), Vieta and colleagues prospectively assessed partial nonresponse to lithium or valproate over a 2-week period using well-defined criteria. This design feature confirmed that patients were partially nonresponsive to mood stabilizer therapy, rather than relying on unreliable patient self-report. After a minimum of 2 weeks of treatment with therapeutic blood levels of lithium or valproate, only 4.8% of patients responded.
Consistent with prior randomized controlled trials that combined an atypical antipsychotic with lithium or valproate, 6 weeks of adjunctive aripiprazole significantly reduced manic symptoms as assessed by the change in the Young Mania Rating Scale total score. However, reflective of the investigators’ diligent efforts to minimize missing data, completion rates were as high as 79% when aripiprazole was used adjunctively and 85% for the use of lithium or valproate alone. Still, the 2.6-point reduction over placebo on the Young Mania Rating Scale suggests that aripiprazole provided only modest efficacy. Although every patient received at least 2 weeks of therapeutic serum levels of lithium or valproate before entering the randomized phase, clinically useful historical information regarding the duration of lithium/valproate use was not reported. Additionally, the mean blood levels of lithium and valproate among patients who were deemed partial nonresponders were <0.8 mmol/l and <78 μg/ml, respectively. When faced with a partially responsive bipolar I manic patient with similar blood levels, we expect that many clinicians would increase the dose of lithium or valproate further, prior to adding a second medication. However, new perspectives on the dosing and tolerability of lithium have emerged, suggesting that previous trials of lithium used target doses and blood levels that were associated with substantial side effects and dropout rates (4). This perspective has more recently led to the study of lithium in lower doses that prioritize effectiveness, i.e., higher response rates in all patients who are assigned to receive the drug, over efficacy, and higher response rates in those patients who can fully comply with the drug regimen.
Regarding methodological innovations relevant to the reporting of clinical trials data, this article offers a series of post hoc analyses in the service of helping practicing clinicians better understand the clinical relevance of the published data by going beyond the conventional use of p values of less than 0.05, which are to a great extent driven by sample size. The article reports on the magnitude of aripiprazole’s adjunctive therapeutic effect size by utilizing Cohen’s d (improvement with drug over placebo divided by the pooled standard deviation). This construct is especially useful because it converts arbitrary rating scale numbers into a meaningful index of change. In the behavioral sciences, an effect size of around 0.2 is small, around 0.5 is moderate, and those greater than 0.8 are large (5). An effect size of 1 equals one standard deviation of improvement (assuming normality) and is essentially an effect that is twice as good as placebo. Number-needed-to-treat analysis was also conducted, which is an alternative estimate of effect size used for discrete data, such as response and remission rates, and is somewhat easier to understand from a clinician’s point of view. The reported Cohen’s d of 0.33 is consistent with a modest benefit, and the number needed to treat of seven indicates that seven patients would have to be treated with aripiprazole adjunctively in order to have one additional patient benefit from a marked response or remission (6).
Another post hoc analysis offered in this article that informs the real-world treatment of bipolar disorder is an item analysis from the Young Mania Rating Scale to ascertain if aripiprazole ameliorated the core symptoms of mania, which it did, including aggressive/disruptive behavior, irritability, euphoria, sexual interest, talkativeness, and insight.
Although conventional definitions were used to calculate remission rates, the authors offered another methodological innovation by acting on recent recommendations (7) to use more rigorous bimodal definitions of remission (Young Mania Rating Scale scores ≤12 and Montgomery-Åsberg Depression Rating Scale scores ≤8), since switching into depression from mania should not be allowed to count as evidence of efficacy. In fact, aripiprazole exhibited significantly lower rates of treatment-emergent depression, although the compound was unable to show superiority over placebo in two recent adequately powered acute bipolar depression trials (8).
Regarding tolerability, dropout rates as a result of intolerable side effects were relatively low but still significantly increased with aripiprazole (aripiprazole, 9% versus placebo, 5%). Confirming prior reports on aripiprazole, akathisia was the most frequently reported extrapyramidal symptom-related adverse event and was significantly increased (aripiprazole, 18.6% versus placebo, 5.4%), a finding that appears to be more commonly seen with less sedating atypical antipsychotics. Based on the presence of only small nonsignificant changes in body weight, aripiprazole’s metabolic profile appeared to be benign (1.2 lb of weight gain over 6 weeks of treatment). However, given the increased prevalence of metabolic syndrome in bipolar disorder (9) and the class labeling that atypical antipsychotics have received from regulatory agencies around the world in this area, we believe the burden of reporting is incumbent upon a study of this type to discuss all of the individual parameters of metabolic syndrome, including fasting blood insulin levels, so that surrogate measures of insulin resistance can be calculated. If these data are available, significance testing should be conducted and the findings should be made public.
Despite the unique contributions of this study, this editorial would be remiss to not highlight the exclusion of large numbers of patients with bipolar disorder commonly seen in clinical practice, including those with bipolar II disorder, rapid cycling, and a history of substance abuse/dependence, which might have diminished the real but already modest effect size.
Last, we should note that the past 20 years of industry-sponsored treatment trials of bipolar disorder have brought us a plethora of new treatments for mania but just a few new treatments for bipolar depression, where our patients live most of their symptomatic lives (10). There exists a tremendous need to develop new lifelong treatments for bipolar disorder that “stabilize mood from below baseline” (11).