There appears to be increasing acceptance of the idea that lithium prevents suicide, and even that it can reduce mortality rates. For a toxic drug that makes most people feel rather depressed, this seems curious. I did wonder whether it might be having this effect on suicide by sapping people of the will to act, but the proposed effect on mortality seems completely inexplicable. A closer look at the evidence, however, suggests the idea is simply not justified.
When I looked into the claims about lithium and sucide for my book The Myth of the Chemical Cure, the evidence consisted of follow-up studies of people on long-term lithium; people attending lithium clinics or other mental health services where they would be having their lithium levels monitored and their moods checked. So firstly, these people represent an especially compliant group, and we know that people who are compliant with any therapy, including placebo, generally have better outcomes than those who are not. The large Women’s Health Initiative trial of hormone replacement therapy, for example, showed that adherence to placebo was associated with lower risks of hip fracture, myocardial infarction, death from cancer and death from any cause.
People who follow their lithium regime religiously are, in general, not likely to be the people who are chaotic, impulsive, desperate and most likely to commit suicide. In fact, one study showed just this. People who were highly compliant with lithium had a five times lower risk of suicidal acts than those who were judged to be ‘poorly adherent’ with it. Secondly, because even small overdoses of lithium can produce dangerously toxic blood levels, people who are thought to be at a high risk of committing suicide are usually not prescribed it. If they are, it is with a strong warning about the toxicity of the drug, and close monitoring by mental health service staff – both of which may, in themselves, help to stabilise someone regardless of the lithium.
More recent cohort studies showing reduced suicide and mortality in people on lithium are likely to be what we call ‘confounded’ by these same issues. People who show suicidal tendencies are less likely to be given lithium and the same applies to people with medical conditions. Not only does lithium cause direct damage to organs like the kidneys and thyroid, it interacts with many drugs commonly prescribed for physical health problems. These interactions can lead to dangerously raised lithium levels. Hence any decent clinician is naturally cautious about starting lithium in anyone who is physically sick or taking other sorts of medication.
In order to control for these factors, we really need evidence from randomised controlled trials, where lithium is compared with no treatment, or a placebo, in a similar population. Suicide is thankfully a rare event, however, and it is even more rare in trials, which usually screen and exclude anyone who is thought to be a high suicide risk. Two meta-analyses have therefore combined data from randomised controlled trials to look at suicide rates.
So what does the trial evidence suggest? Unfortunately, as I pointed out in my last blog, most trials of lithium are not trials of starting lithium, but trials of stopping lithium. They consist of a comparison between people who have been taken off lithium (or other medication) and put on placebo and people who have continued to take it. There is some evidence from cohort studies that suicide risk may be increased following lithium discontinuation, although this could also be an artefact, since people may stop, or be taken off their lithium, if they become suicidal. Nevertheless, comparing the effects of continuing on lithium with the effects of stopping it is clearly not the same as establishing the prophylactic effects of starting lithium in terms of suicide as well as relapse.
Bearing this in mind, let’s look at the results of the meta-analyses of randomised trials. One of the meta-analyses combined studies of different drugs used for bipolar disorder and asked whether the suicide rate was increased in people randomised to placebo compared to those taking an active drug of some sort. The analysis included four studies of bipolar relapse prevention, all of which lasted at least a year, and three of which included a lithium-treated group. Combined, the studies included 943 patients on active drugs, of which 258 were on lithium, and 418 patients on placebo. There were two suicides in these trials during the experimental comparison phase, and a further one three weeks after it finished. All involved patients taking active drugs rather than placebo. There were ten suicide attempts, eight in patients on active compounds, and two in patients randomised to placebo.
Unfortunately results are not presented according to the different active agents used, so we don’t know if any of the suicides occurred in patients randomised to take lithium. However, the suicide rate in placebo-treated patients in these studies was zero.
So it is curious that the meta-analysis that focuses solely on lithium includes so little data from placebo-controlled trials. In fact it does not include a single placebo-controlled trial in which the placebo suicide rate is zero. Reading the paper again, I found that this is because the authors ‘excluded trials with no events in any treatment arm as uninformative’. This decision is totally unsound, however, as it reduces the denominator (the total number of participants) and thereby makes the events included appear more common than they actually were. How this passed the British Medical Journal’s referees is beyond me. This must be why well-known studies, such as the comparison between lithium, valproate and placebo and the two placebo-controlled studies of lithium and lamotrigine were not included in the analysis of suicides. I assume this means there were no suicides in these studies.
So the meta-analysis of suicide rates included only four placebo-controlled trials. There were 6 suicides in these studies, which all occurred among the 241 participants allocated to placebo and there were no suicides in the 244 participants on lithium. Thus the proportion of suicides in people on placebo as presented in the meta-analysis was 2.5%, and the proportion in the lithium group was 0%, a difference that is small, but not negligible. But if the studies in which there were no suicides had been included, the number of participants would have been much larger and the proportion of suicides in the placebo group much smaller. For example, if you add the valproate and lamotrigine trials, the total number of placebo treated patients reaches 524 and the proportion of suicides is then 1.1%. If you add in the large study of quetiapine, lithium and placebo (this study was included in the analysis of suicide attempts, but not completed suicides), then the total number of placebo-treated subjects is then 928, and the proportion of suicides in placebo-treated patients is only 0.6%.
Of the six suicides that occurred in the placebo-controlled studies of lithium, three occurred in one study. This study, conducted in Germany and published in 2008, is worth taking a more detailed look at. Unlike most of the other trials included in the meta-analysis, it is not a discontinuation study. The study explicitly set out to test whether or not lithium is effective in preventing suicide and attempted suicide. It involved randomising people with a variety of diagnoses, who had just made a suicide attempt, to have lithium or placebo. The authors don’t tell us whether any of the participants had been on lithium prior to the study, but I guess it would have been few, since most participants were diagnosed with depression, personality disorder and substance misuse rather than bipolar disorder. Other drugs people were taking were not stopped. The researchers found it difficult to recruit people to the study, and difficult to sustain lithium treatment. Importantly they say they did not maintain the double blind design in cases of ‘insufficient drug compliance’.
It is not clear exactly what this means, but it seems to suggest that people in the lithium group were unblinded in order to maintain what the researchers deemed to be adequate lithium blood levels. They don’t tell us how many people they broke the blind for in this way, but it is possible that it included a large proportion of the lithium group, since the average blood lithium level were frequently lower than the levels specified in the protocol, especially in the early months of the study. So these unblinded patients on lithium were aware they were taking a dangerous drug, and would have had extra visits and blood tests. Now, there is considerable evidence from trials conducted with people with depression that increased visits can improve outcome. So it is plausible that the extra attention received by unblinded lithium-treated participants in the German study prevented some suicides.
There were three suicides in the placebo group in this trial (incidentally far fewer than was predicted, which may indicate the placebo effect of being in a trial of a potentially toxic drug, with regular monitoring), and none in the lithium group. The difference was borderline statistically significant when analysed by the time to event technique (p=0.05). However, regardless of the statistics, the number of events is small, and as explained above, the conditions of treatment were not comparable across the groups because of the unblinding. There was also no difference in suicide attempts, with 7 occurring in each group.
A similar study was conducted recently in Italy in people diagnosed with ‘major depression’ who had just committed an act of deliberate self-harm. It was not blinded, and no placebo was used. Again, it proved difficult to recruit to the study. In the end 29 people were randomised to start lithium, and 27 to ‘usual care’. There was one suicide in the lithium group, and five suicide attempts. In the placebo group there were no suicides, but seven suicide attempts.
So there is the evidence on lithium and suicide. The meta-analysis that has been accepted as demonstrating that lithium prevents suicide spuriously inflated the suicide rate on placebo by excluding studies in which no suicides occurred. The only double blind, prospective study designed to test whether lithium reduces suicide in people at high risk, ended up unblinding many of its participants, and in any case suicidal events were low in both groups.
The fact that studies of suicide prevention have been so difficult to recruit to, suggests patients may have more sense than researchers in this field!