Suicide-related outcomes in veterans with major depression or bipolar disorder
JAMA Psychiatry, Katz et al, 2021
This recent study reported that lithium did not differ from placebo in prevention of suicide-related outcomes over one year in 519 patients. The study was stopped early due to this lack of difference in interim analysis. The authors are to be commended for making the effort to conduct such an important and difficult study. Their extensive efforts in data collection deserve the highest quality of data analysis and scientific interpretation.
If the study had positive results, it would have been surprising, but definitive. Lithium would have been proven effective in reducing all kinds of suicidality – fatal, non-fatal, serious, mild, attempts, ideas – in all kinds of patients – bipolar, unipolar, substance abuse, PTSD. The study did not have that result. But a negative result does not mean the opposite, i.e, that lithium has no anti-suicidal effect at all, not only because of a larger scientific literature proving otherwise, but also because of some of this study’s own results.
To explain:
In contrast to the blanket negative interpretation of the study, the analysis of outcomes indicated half as many total suicidal events in bipolar subjects (15% of the sample) treated with lithium versus placebo. There were ten primary outcomes in the bipolar group treated with lithium (n=37) versus 20 in the non-bipolar group on placebo (n=43). The authors did not analyze this outcome, but this article does. This bipolar subsample was small, 80 subjects, but a doubling of an effect is large enough to warrant some data analysis.
The bipolar subgroup consisted of 10/37 lithium patients versus 20/43 placebo patients who had the primary outcomes in first and subsequent events, 27% for lithium vs 46.5% for placebo. Standard statistical software produces a RR of 1.44, with 95% confidence intervals of 0.98 to 2.15, which produces statistically meaningful results of a 44% increased risk without lithium, that barely include the null value.
It is important to note that this analysis includes ALL suicidal events, which in table 2 is listed as “first and subsequent events.” The primary outcome in the paper was number of first events, ignoring subsequent ones. On that outcome, the two groups were similar. Bu there is a clear difference when all suicidal events are included, and it is unclear why they should not be included. At least the paper should have discussed the issue and noted that there was benefit with lithium when all suicidal events were included. But it did not do so.
One might claim that this subgroup analysis has a high false positive possibility, since it was not a primary outcome and it is one of many possible comparisons. However, this group is exactly the group where lithium’s benefits have been shown most clearly for half a century. Hence this is not a random subgroup analysis, like an astrological sign, but one based on a huge scientific evidence based that is consistent with the observed result.
Thus, the most reasonable interpretation of the study is that lithium likely does reduce suicidal activity in bipolar illness, but perhaps not in non-bipolar illness.
Even that conclusion might be too strong, though, because it is important to distinguish suicide attempts from parasuicidal behavior. Suicide attempts are acts with intent to die, such as hanging or overdose. Parasuicidal behavior reflects self-harm without intent to die and with non-fatal means, such as self-cutting or cigarette burning on the skin. It would have been helpful if the study differentiated suicide attempts from parasuicidal behavior, but the published paper does not do so. The main meta-analysis upon which the causal relationship between lithium use and suicide prevention is based made this distinction: lithium was effective in prevention of completed suicide; it was not effective in reduction of parasuicidal behavior. Since parasuicidal behavior reflects at least part of the outcomes in this study, the result is consistent with the main prior meta-analysis that showed lithium prevents completed suicide but does not benefit parasuicidal behavior. The present study does not address, much less refute, the prior meta-analysis of randomized studies that lithium prevents completed suicide.
The key issue of completed suicide could not be addressed definitively by this study because, thankfully, such outcomes are rare. This is why randomized trials of suicidality really aren’t studies of suicide, but studies of suicide attempts or ideation, which is not the same thing. Nonetheless, it is important to note that here also, in completed suicide, this study showed a benefit with lithium.
There were four completed suicides in or around the study. One with lithium and three with placebo. This result is a three fold elevated risk with placebo over lithium. The study authors stated that the numbers were too few to calculate a difference. That’s not the case. Differences always can be calculated; whether they are meaningful or not is another question. Of course with these small numbers, 95% statistical significance is not achievable but that does not mean the results are meaningless. A 2 x 2 table analysis provides a relative risk of 2.92, with 95% lower and upper confidence intervals of 0.30 and 27.8. This means that the confidence with which one would see this result ranges from a 70% possibility of increased risk with lithium and an equally likely possibility of a 27-fold (or 2700%) probability of decreased risk with lithium. Since the null value is one, these confidence intervals are highly skewed in the direction of more harm with placebo, and protection with lithium. Of course, one could argue that the results are so few that one more case of completed suicide in the lithium arm would make the two groups equal. That is the case, but the possibility that this result is real should be considered given that four prior randomized trials found benefit with lithium in prevention of completed suicide, as meta-analyzed previously.
Scientific research does not happen in a vacuum. No single study is definitive. As taught in the Bayesian approach, one should interpret a new study in the context of prior knowledge, so as to revise one’s perspective based on that knowledge. Taking an all-or-nothing (“it works or it doesn’t work”) approach is not productive since it ignores the total literature.
It should be noted that lithium has been shown to prevent completed suicide in randomized clinical trials. This study is consistent with those results, not contradictory. Also, there is a large geological literature which finds much higher suicide rates in populations where lithium levels in the ground are near-absent. These data imply that very low lithium doses, equivalent to about 25 mg/d of lithium carbonate, could prevent suicide. Again this study, which had subtherapeutic levels for mood illness but high levels compared to the geological studies, does not refute that literature because it too found lower completed suicide rates with lithium than with placebo.
In sum:
Lithium has been shown to prevent suicide attempts in bipolar illness; this study supports the same finding, not a contradictory result.
The non-bipolar population in this study is heterogenous, not simply “unipolar” depressive, including a high substance abuse rate, so the lack of benefit with lithium in that group is difficult to interpret. However, lithium has been found not to be effective in reducing parasuicidal behavior, which tends to occur in populations without severe mood illness. This study is consistent with that literature, not contradictory to it.
This study has been misreported as somehow contradicting the prior lithium literature. Instead, it confirms it: There is lack of benefit for parasuicidal behavior in general, but there is likely benefit for suicidal events in bipolar illness, and there is numerical benefit for completed suicide. Unfortunately, the study authors did not analyze differences when seen, such as nearly two-fold benefit for prevention of suicidal events in the bipolar illness subgroup. And the study did not attempt to understand the relevance of completed suicide, as opposed to other suicidal behavior, where, with the few completed suicides that occurred, the results again did not contradict the prior literature showing lithium benefit.
Methodological conclusions:
This study should be used to learn important lessons to further clarify the evidence base regarding lithium and suicide. Randomized studies in suicidality are extremely difficult to do. Completed suicide is infrequent, and thus it is impossible to do a large enough study for a long enough time to adequately power a primary outcome of completed suicide. Such a study likely would be unethical in any case. So suicidality studies tend to use composite outcomes: suicide attempts with fatal intent, suicide attempts without fatal intent, suicide attempts with potentially fatal methods, suicide attempts without potentially fatal methods, parasuicidal self-harm, increase in suicidal ideation, new suicidal ideation – all these aspects of suicidality are combined in one composite outcome, along with the rare outcome of completed suicide. This mixing or suicidal ideas and behaviors in one outcome is understandable statistically; a higher frequency of an outcome is obtained, thereby reducing the sample size needed for statistical power. However, it comes at a cost: the outcome is noisy, and one outcome could go in one direction while another goes in the opposite direction, producing an average with no change. This possibility is the case especially with the general difference between suicide attempts or intent with potentially fatal methods, and self-harm of a non-fatal nature. This study confirms the prior literature that these two aspects of suicidality should be separated.
Future randomized studies of this topic should focus on excluding parasuicidal behavior, such as self-cutting. Only suicide attempts that are fatal in intent or method should be included. Further, future randomized studies should be conducted only in bipolar illness, in one set of studies, and only in non-bipolar illness, in other sets of studies, so that the different patient populations can be effectively analyzed. Lastly, completed suicide is not amenable to randomized research, and no single study can address that topic. All future randomized research on lithium should be seen as relevant to non-completed suicidal behavior only. The completed suicide literature will remain as it is, showing benefit with lithium, until future meta-analyses can aggregate sufficient outcomes to update the current evidence base.