Reflections on the Recent Controversy Around Lithium and Suicide Risk
Reflections on the Recent Controversy Around Lithium and Suicide Risk
The weaponization of scientific uncertainty poses unique challenges
It has long been speculated that lithium, a gold standard treatment for bipolar disorder and commonly used adjunct medication in depression, has anti-suicidal properties. Over the years, proponents have amassed a significant body of evidence in favor of this claim, based on data from randomized controlled trials of lithium in mood disorders, observational data, clinical case studies, lithium discontinuation studies, and studies looking at concentrations of lithium in drinking water. In many corners of psychiatry, the claim is (erroneously) treated as an established fact. In recent years, two notable publications have highlighted the substantial uncertainty around this claim, and this has generated a debate that has taken a rather bitter turn.
The first is a negative trial by Katz et al (2022) in JAMA Psychiatry conducted within the VA system serving American military veterans. The second is a meta-analysis by Nabi et al (2022), with Joanna Moncrieff as the senior author. In this post, I will focus almost entirely on the Nabi et al meta-analysis and the subsequent controversy. (For a discussion of the Katz et al paper, see the accompanying editorial by Baldessarini and Tondo.)
Earlier this year Nassir Ghaemi published a strongly worded response to the meta-analysis, calling it “pseudoscience” and presenting his own analysis of the data showing a statistically significant result in favor of lithium. The difference between the results depends on whether certain older studies and placebo deaths are included or excluded, and whether studies with no suicide events are included or excluded. Ghaemi says that the exclusion of uninformative studies with no events is standard practice in meta-analysis while Nabi et al argue that it is more statistically rigorous to include studies with zero events. I don’t have the requisite statistical background to take a strong position on whether studies with zero events should be included or not, but it does seem to me that if statistical significance hinges on the inclusion/exclusion of studies with zero events, then there remains considerable statistical uncertainty and the matter cannot be considered settled.
(I also recommend
’s analysis of Ghaemi’s response: “Our blessed science (their barbarous pseudoscience)”)It is worth reflecting on why the academic debate has taken such an acrimonious turn, and why is it that we find Ghaemi accusing Nabi et al. of “social activism disguised as science,” using “scientific journals as a public relations tool, providing a patina of respectability for explicit opinion-based propaganda on the internet and in social media,” and being “pseudoscientific” in their employment of meta-analysis as a methodology.
Ghaemi is well-known for his view that psychiatric medications are by and large symptomatic and not disease-modifying, with one notable exception of lithium (Ghaemi, 2022).1 So it makes sense for him to prefer inclusion/exclusion criteria that would portray lithium in a more positive light. However, Ghaemi is not only one in this debate with strongly held views. Moncrieff is well-known for her distaste for lithium as a psychiatric treatment.
This presents an opportunity for me to highlight two related aspects of this controversy whose relevance generalizes beyond this specific debate:
The first is a tendency to weaponize scientific uncertainty in the service of an entrenched ideological position. That is, instances where uncertainty around statistical significance with regards to an issue (based only on RCTs or some other restricted set of data) is used as evidence that the phenomenon is false or mythical.
The second is the important role “ideological conflicts of interest” (so to speak) play in these debates; we will be blind to the vital drivers of such controversies unless these are made explicit.
The Nabi et al. meta-analysis concludes:
“The evidence from randomised trials of the new millennium is inconclusive and does not provide support for the belief that lithium reduces the risk of suicide or suicidal behaviour. More data are needed to estimate the effect of lithium with more precision in general, and in subgroups of patients, specifically.”
Note the careful wording around the inconclusive nature of evidence from RCTs and the need for more precise estimation. Now compare this to the language used by Moncrieff.
On publication of the meta-analysis in Sep 2022, Moncrieff tweeted:
“Our meta-analysis shows there is no evidence for the longstanding idea that lithium prevents suicide. Another myth that should be put to bed.”
In Oct 2022, Moncrieff tweeted:
“I have had my sights on lithium for a long time. It is a toxic sedative with strong discontinuation effects…”
In Nov 2022, she tweeted:
“its a polarised debate because to admit that lithium is just a toxic sedative is to admit a huge and longstanding medical error - therefore most psychiatrists just can’t look at it clearly”
Moncrieff stated in an interview with Medical Minefield that prescribing lithium allows psychiatrists to “feel that they are proper doctors”2 because its use requires monitoring and various medical tests, and that “if psychiatrists admitted that it was just a sedative, they’d have to admit that they've been really using something that was quite toxic on people for many decades.”
Moncrieff was quoted in Mad in America’s coverage of the Nabi et al meta-analysis:
“The idea that lithium helps prevent suicide really should be put to bed now… The idea that lithium prevents suicide has added to its mystique and helped propagate the idea that lithium is a highly specific and effective treatment. But it never made any sense. Lithium is a highly toxic, sedative substance.”
Moncrieff is clear about her intended goals and her preferred interpretation, and has no qualms sharing that with confidence to the public, but she also has the option to hide behind the more cautious wording of Nabi et al. 2022 in case she comes under criticism.
If it seems odd that Ghaemi and others are reacting so strongly to a carefully worded paper that merely says the evidence is uncertain, this is because they aren’t responding only to the text of the paper. They are also responding to how this paper has been presented to the public by its senior author, its reception in circles hostile to psychopharmacology, and the manner in which it has been weaponized to dismiss lithium as a legitimate treatment to begin with.3
In my opinion, skepticism and moralism about the clinical use of lithium are pretty good indicators of dogmatism about psychopharmacology in general. There is no pharmaceutical lobby behind it. There is no ‘Big Lithium.’ Despite calls for increased utilization of lithium in clinical practice, only around 17-18% of the patients with bipolar disorder are prescribed lithium in the US. Lithium reduces the risk of manic relapse by 30-40% in clinical trials and outperforms other medications in many observational studies when it comes to outcomes like re-hospitalization. Nothing so far has produced superior outcomes in the maintenance treatment of bipolar disorder. Folks trying to convince the public that these efficacy outcomes aren’t good enough have nothing better to offer. To assert that lithium is “just a toxic sedative” is not just a dismissal of lithium, but a dismissal of all existing psychopharmacological treatments for bipolar disorder because nothing outperforms lithium in its prophylactic effects.
What the Nabi et al. meta-analysis reveals is that there is considerable uncertainty around lithium’s anti-suicidal effects based on randomized trials. It doesn’t mean that the effect reported in other meta-analyses is false, mythical, or based on junk science. In fact, all these meta-analyses show reduced number of suicides in the lithium group, with odds ratios around 0.2 to 0.4. What changes is whether this effect is statistically significant or not. Only someone red-pilled by null hypothesis significance testing or someone with a predetermined ideological position would use this as conclusive evidence that the matter is settled and that this is another myth that should be put to bed rather than as evidence of uncertainty around a potential effect that requires further empirical investigation.
Even when meta-analyses of RCTs show a significant p-value, we’d have to accept that there are considerable limitations with existing trials of lithium, and in order for RCTs to appropriately guide clinical practice in this regard, we’d have to conduct an adequately powered RCT that uses suicide as a primary outcome. There are serious practical challenges with regards to conducting such in a trial in a manner that would settle the disagreement.
In addition to RCTs, we also have to take into account non-randomized and observational data. The frequently-cited Baldessarini et al. (2006) meta-analysis included both randomized and non-randomized data, and found a robust effect: risks of completed and attempted suicide were lower by approximately 80% during treatment of bipolar and other major affective disorder patients with lithium over an average duration of 18 months.
Fitzgerald et al. (2022) report a lower rate of suicide (HR 0.40) with lithium versus no treatment in a large nationwide Danish cohort (as well as lower rates of self-harm and psychiatric hospital readmission). Chen et al. (2023) report a lower rate of suicide (HR 0.39) in a large nationwide Taiwanese cohort. Nabi et al. (2022) report a non-significantly lower rate of suicide (OR 0.42) in their meta-analysis of RCTs. This is a situation that calls for the triangulation of results from multiple methodologies, while taking into account the limitations of each. This is not a situation where a non-significant result in an underpowered sample from RCTs can be taken as conclusive evidence that there is no there there.
Moncrieff’s strategy to use statistical uncertainty around lithium and suicide risk in RCTs to trash lithium as a treatment in her statements to the public (“just a toxic sedative”) parallels the strategy used by her in the umbrella analysis of the serotonin theory of depression, where lack of conclusive evidence of serotonin dysfunction in depression was used by Moncrieff in her public statements as a basis for questioning the efficacy of antidepressants and to challenge the status of depression as a medical disorder, issues that go well beyond the narrow scientific purview of the review in question. (See also: Serotonin strikes back)
What is a rational way of responding to such a state of affairs?
An important thing here is that the weaponization of scientific uncertainty by certain actors should not preclude the scientific and clinical communities from acknowledging the existence of empirical uncertainty. It is mistaken, IMO, to respond by insisting that the association between lithium and reduced suicide risk in RCTs is already a settled matter and assertions to the contrary are pseudoscience, or, in the case of the serotonin umbrella review, to insist that the evidence in favor of the existence of serotonin dysfunction is stronger than it actually is. We can accept the demonstration of uncertainty or lack of conclusive evidence while arguing for reasonable inferences based on the totality of evidence, advocating for the need for better data, and pushing back against the promotion of anti-medical and anti-pharmacological ideological positions.
Baldessarini and Tondo, two of the biggest proponents of the anti-suicidal effects of lithium, have not shied away from acknowledging this. In their editorial on the Katz et al paper, they say:
“An ironic final note is that recruiting participants to such trials may be made difficult by an evidently prevalent belief that the question of antisuicidal effects of lithium is already settled, which it certainly is not.”
And in their 2024 review (which I highly recommend):
“Findings summarized here, pertaining to research bearing on the hypothesis that lithium treatment may have a special role in reducing suicidal risk are encouraging, but it must be emphasized that such a role of lithium treatment is not securely demonstrated and has proved difficult to assess by use of randomized, controlled trials that support most clinically applied medicinal treatments. A major limitation of studies finding less suicidal risk during long-term treatment with lithium is that, to date, even in randomized controlled treatment trials, the outcomes involve incidental reporting of adverse events rather than testing for suicidal behavior as an explicit outcome.”
In my own clinical practice, I remain mindful of the potential effects of lithium on reducing suicide risk but I don’t use it purely for that indication. Given lithium’s status as a gold standard maintenance treatment in bipolar disorder and as a useful adjunct in depressive disorders, it is fortunate that most patients who might benefit from the potential anti-suicidal effects of lithium (if such effects exist) are also patients who stand to benefit from the direct effects of lithium on mood.
I won’t go into the symptomatic vs disease-modifying issue here, but I do find myself increasingly dissatisfied with this distinction.
Because, you know, psychiatrists are fake doctors treating fake disorders. *eye roll*
I am reminded of an analogy. Back when I wasn’t active on twitter, I read Jordan Peterson’s book 12 Rules for Life, and I was stumped as to why this book had generated so much anger and controversy. Then I became active on twitter and actually saw what Peterson was tweeting, and the backlash suddenly made sense.
I've never been on lithium, but Moncrioff has been adamant, too, that antipsychotics (and I was on Haldol for many years) are just sedatives, all they do is drug people down.
Making blanket statements like this isn't liberatory or respectful of madpeople! Because if you listen to US and what WE have to say, we'll report fairly different experiences with antipsychotics. For some people some medications are experienced as having a more targeted effect, not just a tranquilizing one.
Of course, people can be wrong about what effects a medication have on them, that's why we have RCTs. But we do have pro tanto reasons to trust people when they talk about their own experiences, we need some REASON to dismiss personal testimony. For instance, if people who claim more targeted effects from antipsychotics couldn't tell whether they got Haldol or Xanax (say) in a blinded trial, then we'd have reason to distrust their testimony. But in the absence of such blinded trials (which would probably be too complicated to do anyway - you'd have to make sure the study wasn't disturbed by discontinuation effects - and unethical to boot), we should trust people's testimony. Not go "well there's no objective evidence supporting what you say about your own experiences, so I'm gonna assume you're just wrong and confused".
This is such a well written article. Added to my link-list folder.