The Evidence-Based Medicine framework [28
] also provides ways of quantifying the strength of association between antidepressants and mania. These include the relative risk (RR; the ratio of the rates of mania in the antidepressant exposed group, divided by the rate in the nonantidepressant comparison group), the odds ratio (odds of mania in the antidepressant group divided by odds of mania in the comparison group) and the number needed to harm (NNH). The NNH is the reciprocal of the difference in the rates of mania in the antidepressant exposed versus comparison group. Conceptually, the NNH is the number of patients that would need to be exposed to antidepressants before one more on average would develop mania. The NNH has a corresponding measure of the magnitude of beneficial treatment effects, the number needed to treat (NNT) in order for one more patient to have a good outcome. The NNT and NNH can be combined to create a likelihood of help versus harm (LHH; the reciprocals of the NNT divided by the reciprocal of the NNH, so that larger numbers indicate greater probability of benefit).
What would these metrics suggest about the relationship between antidepressants and mania? We focus on fluoxetine as an example, because: first, fluoxetine has the best evidence of efficacy for pediatric depression [6
]; second, fluoxetine appears to be associated with somewhat higher rates of mania than other SSRIs [90
]; and third, fluoxetine has multiple published RCTs where the rates of mania can be compared with rates in a randomly assigned placebo arm, providing some of the best evidence available on the issue [3
]. Pooling the results of the three RCTs, the rate of mania on fluoxetine was 2.8%, versus a rate of 1.0% on placebo. The RR of mania is 2.8, approaching the range where we should be concerned (RR < 4 is ambiguous if the designs are weak; but for an RCT RR of > 3 would be ‘convincing’) [28
]. The NNH is 56, meaning that for every 56 youths exposed to fluoxetine, on average one more would develop mania. The NNH estimated separately for each RCT ranged from 24 to 145. On the other hand, fluoxetine appeared efficacious across all three studies, with NNT estimates ranging from 3.8 to 6.5, and a pooled estimate of 4.7. The LHH is 11.8 when all three studies are combined, and ranged from 6 to 22 for the studies separately. Thus, when focusing on the compound with the most evidence for efficacy and also the greatest concern about mania as an adverse event, patients appear roughly 12-times more likely to benefit than to experience mania; and even the worst case scenario is that patients would still be six-times more likely to benefit. Estimating the LHH separately for several studies is one way of conducting a ‘sensitivity analysis’, or examining the extent to which the LHH estimate changes depending on the starting values. Sensitivity analyses can also include ‘what if’ scenarios, where the LHH is recalculated using more liberal or conservative estimates of risk, so that the decision-makers can understand the effects of changing assumptions on the final estimates [28
It is possible to further customize the LHH, adding additional information about individual factors affecting the probability of risk or benefit, as well as incorporating patient preferences. Straus et al.
provide detailed examples of adding these factors into the equation (see page 139–143 [28
]). On average, including patient preferences will typically shift the balance even further in favor of benefit instead of harm. Two major reasons for this are the evidence that depression creates more burden than mania, and has a worse effect on quality of life [91
]; and also because depression is the more lethal phase of the illness (including being a major component of mixed states) [44
]. It also may be possible to shift the balance further by providing psychoeducation to the family, using careful monitoring to detect mania, and establishing a plan for managing manic symptoms should they emerge. By increasing the knowledge and the external supports available, the risks associated with mania may be more contained. With close monitoring, it is more likely that treatment could be changed during hypomania rather than waiting until full blown mania manifests. At the same time, the irritable mood and aggression frequently associated with pediatric mania are often a chief concern of families and a major treatment target [92
], so interventions that increase the risk of disinhibited, aggressive behavior may require more caution.
There are several important limitations to keep in mind, both about the state of the literature, and also about this particular review. Decisive studies about the relationship between antidepressants and mania have not been conducted, so it is impossible for a review to draw decisive conclusions. Some limitations of the literature include:
- Most published articles do not satisfy most of the criteria for providing a valid source of information about risk of harm . Case reports are not adequate, unless perhaps they used rechallenge, A–B–A–B designs, but these are clinically not conducted. See Box 1 for a list of the guidelines and other comments on the state of the literature;
- There are large differences in the rates of mania identified across studies and inconsistencies in the size of the findings, making it plausible that methodological issues (such as the Vigilance or False Alarm hypotheses) or the natural course of bipolar disorder (consistent with the Medication as Irrelevant or the Bipolar Depression Hypotheses) explain the findings, instead of one of the proposed iatrogenic mechanisms;
- There are major differences in the operational definition of mania across studies, and equally large differences in how mania is assessed; These make it much more difficult to interpret findings;
- There may be local and historical changes in patterns of diagnosis. The steep increases in the rate of bipolar diagnoses shows that there have been dramatic changes in practice that are independent of changes in actual rate of incidence. Clinicians may become hyper-vigilant to the possibility of mania. Cognitive heuristics and publication bias both will lead to over emphasizing the degree of risk;
- Many chart reviews and clinical trials excluded bipolar youth. We do not know rates of switching for these youth (versus youth with anxiety or MDD that has not yet shown itself to be bipolar) [cf. 76], although the adult literature suggests it could be higher than what we are seeing here. Conversely, studies with youths exposed to antidepressants who did not have a history of depression have consistently found low rates of mania [93–95]. The two competing explanations for this pattern of findings are that assessment of mania in these studies is not sufficiently systematic to be sensitive to instances that actually are occurring (a variant of the Vigilance hypothesis), or that it is previously undiagnosed bipolar disorder that was being treated for depression, and then subsequently manifested its associated manic symptoms (the Bipolar depression hypothesis). The insufficient sensitivity argument is less of a concern given that one of the studies finding no increase in risk was a secondary analysis of longitudinal data from one of the leading investigations of pediatric bipolar disorder .
An additional limitation of this review is that it does not attempt to meta-analyze the existing studies, relying instead on qualitative ways of appraising and describing the findings. The literature does not yet seem large enough to support a meaningful meta-analysis, particularly with so few studies reporting adequate information to estimate relative risks.