Risk and Reasoning
(written in 2004, reviewed 2008)
A patient calls and says her stomach hurts. Is this because she has eaten something bad? Or because she has a new stomach ulcer? Or a new viral illness?
Or, could it be the medication that was just started? In this circumstance, we generally presume, as a physician friend of mine says, that "medications are guilty until proven innocent". This is an extension of the general principle we were all taught in medical school: Primum non Nocere, "First Do No Harm". Doctor, you may not be able to help this patient, but make sure that while you're trying, whatever you do, don't make things worse.
Another extension of this concept would be to have a different standard for judging whether a medication is helping someone, versus whether it could harm them. We use this sort of reasoning all the time: if a medication has clear risks, there better be some pretty darn good evidence that it is helpful before we recommend it. For example, clozapine (Clozaril) is a remarkable medication for schizophrenia and severe bipolar disorder -- but it took some overwhelming evidence of effectiveness before the FDA finally approved it for use in the U.S., because it can cause a dramatic decrease in white blood cells in about 1 in 100 patients (in the first 6 months of treatment, that is; it's safer after that).
Similarly, if a medication has good evidence for a lack of risk, we are more willing to try it even before the evidence is solid for effectiveness. For example, verapamil has been used for years for blood pressure treatment, and its risks are well known. Because there have been few known problems with pregnancies with this medication (at least to other mood stabilizer options), it was recently proposed as worth considering for women with bipolar disorder who become pregnant, even though the data on its effectiveness are limited.Wisner
This different standard for judging risk evidence, versus effectiveness evidence, seems to be the basis for the FDA warning about antidepressants and suicidality/homicidality. They take care to indicate that they have issued a call for increased vigilance about a possible causal relationship, not a conclusion about causality. Yet they are clearly saying there is somehow enough evidence out there to issue this warning, though the evidence is not good enough yet to reach a more certain conclusion. What would constitute such evidence?
The Perfect Study
If we wanted to design the perfect study to show than antidepressants do cause suicidality (which is a lot easier than showing that they don't), let's think about what that would require. First, it should be done with a group of people who need to take an antidepressant. We couldn't just study people with no such need, if we honestly think there could be some risk in taking this medication, right? And obviously we do think there may be such risk, or we wouldn't be doing the study.
Secondly, this group of people should not be depressed. Depression can cause people to be suicidal all by itself. We want to see if the medication can cause someone to become suicidal. So we'd better not have some other "built in" reason why suicidality could show up in our test group. Therefore, we'll have to study some group of people who have anxiety, or obsessive compulsive disorder, or pain, or some other reason to consider giving them an antidepressant.
Third, we would have to divide this group -- randomly, not by gender or age or something -- in half, giving one group an antidepressant medication, and the other group a placebo. As you have probably already concluded, "wait a minute, this has surely already been done". Correct, there are many such studies out there. But...
Fourth, the groups would be monitored for increased suicidality. How would we know if "suicidality" was going up? We would need some sort of monitoring tool, a questionnaire or something, that could recognize such an increase. And that, you see, is the bad news: there is no standardized way of monitoring for suicidality, so in all those previous such studies, we do not have systematic data about suicide. (The FDA struggled with this same issue in the data they received from the studies of people with depression).
And so, while we await a study that looks specifically for suicidality in randomized patients who are given an antidepressant for something other than depression, and are monitored closely for the new appearance of suicidal thinking.... we are stuck with case reports.
Testimonials and Case Reports
Usually we dismiss testimonials. They are not accepted as evidence. This is because a person may be completely convinced that some miracle drug or herb cured him, but we cannot even begin to sort out all the other potential explanations: at the same time as starting the pill, he stopped smoking; or he started a relationship with a great woman; or he quit the job he hated; and so forth. And we don't know about all these things, because the person himself doesn't know to include them as relevant in his explanation of his improvement. After all, he is convinced the pill was the cure, so he tells us up and down about the pill, and omits all the information we'd need to evaluate whether it was the pill or something else.
Worse yet, people really want to get well. So if some source (an advertisement, or someone selling a drug) quite convincingly suggests to a man that the pill or herb will get him well, and he really believes that, his own hope may influence the outcome. Somehow belief and hope can cause people to get well. This is known, of course, as the "placebo effect".
Any testimonial is limited, as evidence, by these two problems: "invisible" alternative explanations for a result; and possible placebo effects. Let's look at these two limitations in regard to the question of whether antidepressants cause suicide. First, the common presumed "invisible" explanation is that the depression got worse. After all, the antidepressants don't always work, and if the depression was getting worse, that could easily have been the reason for a person to commit suicide.
Secondly, there is a possible "placebo effect" after a suicide. (I do not believe this is the best explanation for the suicides that have occurred on antidepressants; but family members of someone who committed suicide should brace themselves, this doesn't sound good). Consider a family whose daughter has committed suicide. Perhaps it is just too overwhelming for them to think that she really felt that badly, and somehow they missed it. Better to think that "the medication did it". Notice that they could really believe this, and that there is a "placebo" effect going on, only in this case it is an idea instead of a sugar pill. They want to believe the pill caused the suicide, because it can help them not hurt so much.
However, these two limitations may not completely doom testimonials as evidence in the particular case of antidepressants and suicide. There are some obvious logical criteria under which patient/family stories would be less subject to these limitations. First, there should be some aspect of the story that rules out or decreases the plausibility of "worsening depression" as the basis for the suicidal thinking or action. Secondly, there should be no evidence to suggest that some benefit to the patient or family led to their adopting the explanation that "the pill did it". That second one's tough: just sounding objective does not rule out the possibility of gross bias on the patient or family's part; however, obvious pain in the story-telling raises the possibility of the "placebo effect".
Can we find patient reports without background depression, and with sufficient objectivity? The FDA transcript include four examples meeting these criteria, in my opinion (granted, this judgement itself is subjective). In this case report table, their names appear in red.
From my own patients I hear much more detail, of course. After hearing accounts like those in the FDA transcripts for 2-3 years, at least one a month I would estimate, one came to me "in print" from a high-functioning, emailing client, Ms. B. Sensitized by previous, similar stories, I asked her if I might keep this one and publish it; i.e. this was not the only such story I've heard, nor necessarily the most striking, but rather the one that came most easily into my hands in original form.
I think Ms. B's story strongly meets the above criteria for a credible account implicating antidepressants. However, I also grant that skeptics would likely not find it as convincing as I do (perhaps rightly so; I could easily be biased from having met the patient herself).
In summary, on this issue of "testimonials": as just noted, both the patients' reports themselves and the process of selecting cases suggestive of causality are subjective. In that respect, they will never be adequate to establish firmly that antidepressants cause suicidality or homicidality. However, in my opinion, the more one listens, the more one becomes convinced of this relationship. On the other hand, because we do not yet have more objective data, maintaining some doubt about this relationship is also indicated. (Thanks to Dr. Rick Bingham for helping me maintain that perspective).
Can patients be good scientists?
We have looked at two reasons to dismiss testimonials: the "invisibility" of alternative explanations, and the possibility of a placebo-like value of placing blame on a medication. Now for some serious "opinion" on my part.
There may be a third reason we dismiss testimonials: they come from patients. Doctors can offer testimonials too, but their descriptions are more acceptable (we have been properly trained; we are more objective, less emotional; we aren't as susceptible to placebo effects because we know better, etc.) Accordingly, doctors' testimonials are given a different name: "case reports".
Fortunately medicine is slowly figuring out that patients can be scientists too, and can be enlisted as allies in the search for data. In fact, I think that's exactly what happened at the FDA on this issue of antidepressants and suicide: in 1991, the few case reports were presented only by doctors, but in 2004, by patients -- and especially, by a patients' advocacy organization, the Child and Adolescent Bipolar Foundation. I'll quote them again, though you may have seen this on a previous page:
Over the four years since CABF first launched its Web site, www.bpkids.org, numerous parents have reported on our Message Boards the results of their children using of antidepressants. Some report that their children first became suicidal immediately or shortly after being prescribed an antidepressant, and that these alarming symptoms stopped only when the medication was stopped. Others report that their child’s prior existing depression with suicidal ideation quickly escalated into rapid cycling/mixed states (the state with highest risk of suicide for any patient with bipolar disorder) soon after their child began treatment with an antidepressant, and that these increased symptoms subsided when the antidepressant was stopped. Conversely, other parents report that their children had suicidal thoughts or actions before taking any antidepressant, or which developed during treatment for bipolar disorder, and that the suicidal ideation and behavior improved when an antidepressant was added to the child’s treatment plan.
Because they are reporting multiple patients who received the same treatment, this testimony represents the conclusion from a large case series (granted that we have no access to the data from which the conclusion is derived; it might be wise to try to remedy that someday).
"Case series", or "testimonials"? The sound of those terms illustrates my point here. I think we are finally realizing that we should pay more attention to patients' accounts, and that strict reliance on "data" -- by which we mean our own productions, as opposed to anything patients might generate -- has been limiting our vision. Witness, for example, the benefit of the collaborations between Dr. Hirschfeld and colleagues and the Depressive and Bipolar Support Alliance (DBSA) on estimating how long it takes for someone with bipolar disorder to be diagnosed as such.Hirschfeld When we do not have randomized trial data to rely on, "testimonials" are worth listening to -- not to establish efficacy, but to identify risk.