Thread by @PloederlM

1/ Lithium & suicide: there is some (fierce) discussion about the suicide preventive potential of lithium. Some say it is a fact that lithium prevents suicides. However, an unbiased look at the evidence reveals that there IS uncertainty which should be acknowledged. A thread.

2/ There are meta-analyses concluding that lithium significantly reduced suicides - this one is perhaps the one most often cited

No other psychopharmacological tx has such effects. This is what I also told again and again in my talks, too.bmj.com/content/346/bm…

3/ Moreover, there is no big money in lithium, as it is a cheap medication. Thus it is often assumed that there is less bias in the research.

4/ In our recent MA, however, we concluded that the evidence from randomized controlled trials remains inconclusive, especially from the more recent trials, and that more data is needed. Confidence intervals for suicides are wide.
cambridge.org/core/journals/…

5/ Other meta-analyses came to similar conclusions. Statistical significance depends on the inclusion/exclusion of more recent data or different inclusion criteria (already a hint that the the effect is not robust).


cambridge.org/core/journals/…
cambridge.org/core/journals/…

6/ Moreover, a recent very good RCT was stopped early because it turned out that lithium had no protective effect with respect to suicide attempts.
jamanetwork.com/journals/jamap…

7/ Proponents argue that results of observational studies need to be considered. Fair point, but we know that observ stud can suffer from confounders. @joannamoncrieff argued that patients who stay on lithium have better compliance and thus better prognosis, indep of tx w lithium

8/ Interesting discussion about confounders also here recently on twitter here:

https://twitter.com/DebiecJacek/status/1625227416246296592?s=20&t=Clbyzp4cZjV2Of2OumzhVg

9/ Some proponents point out findings of ecologicial studies: fewer suicides in regions with higher levels of lithium in drinking water. However, after correction of the likely publication bias, the association almost nullifies.

https://twitter.com/PloederlM/status/1319254086554877954?s=20&t=Clbyzp4cZjV2Of2OumzhVg

10/ IMO, an under-acknowledged point in the discussion is that the findings for suicides point at a suicide protective effect (wih substantial uncertainty), the there was no such effect for suicide attempts, where statistical power is higher.

11/ This discrepancy is strange and a reason for concern, as fatal and nonfatal suicidal behavior are substantially correlated, and for this reason suicide attempts are often used as a proxy for suicides in many trials in suicidology.

12/ Another challenge is how to aggregate RCTs in meta-analysis when there were no suicides at all (small wonder, given the rarity of suicides). There is no consensus if such evidence should be included. Some specific stats methods are only available recently.

13/ We were accused of including these double-zero trials (leading to larger confidence intervals). It seems critiques are not aware that 1) there's no consensus to include/exclude such trials, and 2) recent statistical papers showing that excluding these trials leads to biases

13/ “To utilize all available information and reduce research waste and avoid overestimating the effect,
meta-analysts should incorporate DZS, rather than simply removing them”
link.springer.com/article/10.100…

14/ “To utilize all available information and reduce research waste and avoid overestimating the effect,
meta-analysts should incorporate DZS, rather than simply removing them”
link.springer.com/article/10.100…

15/ Thus, it is is not only unfair to accuse us of selectively stats methods bc there is no consensus about method is best. Moreover, we applied conventional stats methods in sensitivity analyses w similar conclusions. Seems that some critics are not aware about their own biases.

16/ Furthermore, we don’t know how strong the publication bias is in the trial data. As a rule of thumb, the efficacy of treatments is usually overestimated and the harms underestimated. The extend of this bias remains has still not been adequately addressed for lithium, AFAIK.

17/ Even proponents of lithium admit that there is uncertainty, as we point out in our reply to critics:
“Baldessarini and Tondo [] also recently acknowledged the uncertainty of lithium's anti-suicidal properties, describing how:cambridge.org/core/journals/…

18/ ‘recruiting participants to such trials [suicide prevention trials of lithium] 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’ "

19/ The uncertainty was also acknowledged in the excellent and IMO balanced discussion of the issue and the need for “An individualized risk/benefit analysis” by Hal Wortzel very recently
journals.lww.com/practicalpsych…

20/ “Conclusions that lithium unequivocally offers suicide prevention benefits do not appear warranted based on the strength of existing studies“
The paper gives good practical advice for prescribers.

21/ Thus, I think there is no way around to acknowledge the uncertainty in the evidence regarding the suicide preventive effect of lithium. Strong conclusions in either direction are not justified. More data is needed.

22/ Sorry for the typos and missing words - I was in a hurry.

23/ More thoughts: what's the necessary size of an RCT to prove that lithium prevents suicides. Some "optimistic" assumptions: lithium prevents 50% of suicides among a high-risk population (eg those after suic attempts, with ca 3% dying by suc in 1st year after the suic attempt)

24/ Such a trial needs ca 2500 patients. But compare this to the actual size of the existing trials.

25/ Using the actual rate of suicides in placebo arms of lithium (ca. 0.5%) as a basis for such a power analysis, you need nearly 15,000 people in a trial.
In ALL trials that were included in our meta-analysis, there were less than 4000 participants. We definitely need more data

@threadreaderapp unroll please