That is So Random: A Frontier Psychiatrists Guest Post!
Putting the plural in the Frontier Psychiatrist(s).
The Frontier Psychiatrists is a daily health-themed newsletter. It has historically been written by one somewhat lonely psychiatrist—Owen Muir, M.D. It was always envisioned as more than just me. Today, I’m thrilled to welcome
(a pen name, but a real Ph.D.) to the Frontier. Alex is a reader and writer and sent me this riff on the role of randomness in biasing psychiatric thinking. I would love to hear from other off-angle thinkers and writers…because I love collaboration. Alex is from the UK, so fair warning— color might end up spelled with a U.Is Poisson Clumping contributing to bias in psychiatric care?
Humans have the remarkable ability to recognize patterns. Unfortunately, it comes with side effects. We sometimes not only identify patterns where none exist, but we also draw conclusions from them.
Take astrology, for instance. From the earth’s point of view, stars are randomly distributed in the space above us. Yet, that has not stopped humans from drawing crude stick-figure pictures of animals in the sky and determining this means you will meet the love of your life next Wednesday.1
Perhaps the most surprising aspect about randomness is how “clumpy” it looks. I first came across this phenomenon in a maths lesson at school. My teacher asked the class to write their prediction for twenty “heads or tails” coin tosses. I wrote down a sequence I thought looked random, ensuring not to put too many heads or tails in a row. I still remember the astonishment of seeing the long string of “heads” in real life when my teacher flipped the coin. My mistake was very common, as the YouTube channel “Numberphile” quite humorously shows.
I was reminded of this maths lesson fifteen years later when my psychiatrist would not initially entertain even a discussion about a prescription of Pregabalin for my Generalised Anxiety Disorder (GAD). I found this outlook rather strange. According to UK NICE guidelines, Pregabalin was the next drug to trial after a Selective Serotonin Reuptake Inhibitor (SSRI) and Selective Noradrenaline Reuptake Inhibitor (SNRI). I had already trialed both. A severe reaction to an SSRI caused my GAD in the first place!
Further, the academic literature suggested that Pregabalin was effective and well tolerated. In terms of real-world experiences, the ratings on drugs.com showed that reviewers considered it the most effective medication for GAD2 (tied with escitalopram at the time of writing). I decided to look for a second opinion and emailed another prominent psychiatrist. In their reply, they wrote:
“Whoever prescribed the pregabalin needs to review that, as there are many better treatments available.”
Again, without a discussion. Pregabalin went on to become the most effective treatment out of the eight treatments I’ve tried so far. I couldn’t help but ponder how two highly experienced psychiatrists managed to form such a strong bias against Pregabalin despite the evidence of its effectiveness and tolerability. I wondered whether counterintuitive clusters within random distributions or “Poisson clumping” contributed to some bias against Pregabalin as an option?
This odd characteristic of randomness has often been seen as significant, whether it is a sudden spate of shark attacks, pedestrian deaths, or even crime. Events initially considered to represent a fast-growing trend by news organizations were merely statistical ”blips.” The immediate refusal of Pregabalin by two experienced psychiatrists made me speculate whether they had both previously encountered a Poisson clump of patients with severe adverse reactions.
Psychologically, one severe reaction would probably be seen as unfortunate; two in a row is a dangerous pattern! Coincidence? But three could be seen as causal! We humans are hungry for those patterns!
In other fields that require impartiality (e.g., asylum court decisions), it has been shown that professionals are prone to underestimation the likelihood of sequential streaks. This is called the “gambler’s fallacy.”
Whether or not Poisson clumping caused significant bias in my psychiatrists against Pregabalin, there is an argument it could influence prescribing patterns. Consider the following simplified hypothetical scenario.
The Doctors, A+B, MD.
Say we have two psychiatrists, whom we will call Psychiatrist A and Psychiatrist B.
They have both just started prescribing a new “antidepressant” for depression. Let us assume that the medication has a patient population with a remission rate of around 15 percent, with adverse drug reactions serious enough to affect their quality of life (e.g., a bad case of antidepressant discontinuation syndrome) occurring at a rate of about 5 percent. I assume the sequence of clinical responses (remission, complications, no response, etc.) for patients entering each psychiatrist’s clinic will be random3.
After a few months, Psychiatrist A is very fortunate. Three patients in a row achieved remission after the prescription of the new antidepressant. They have just encountered a Poisson clump. An opinion might form that this is a “good” drug.
Psychiatrist B, on the other hand, is extremely unlucky. They observe a string of three patients who suffer severe complications. An opinion may form that this is a “bad” drug. Importantly, whichever clump comes first may very well influence the future behavior of both psychiatrists. Psychiatrist A may eventually encounter a string of severe complications, but previous observations could reinforce the opinion that it was simply bad luck. This means they are likely to continue prescribing the drug.
Psychiatrist B, misinterpreting the clump as a pattern, could very well stop prescribing the medication. Therefore, they never encounter the strings of patients who achieve remission with minor or no side effects. Without the response rates of psychotropic medications and the number of new patients each psychiatrist sees every year, it is impossible to know how much Poisson clumping affects bias within psychiatry. Nevertheless, conveying the true characteristics of randomness in psychiatric education might help clinicians better identify when a trend in their clinical experience is causal or due to random fluctuations.
Thanks, Alex, for joining The Frontier Psychiatrists movement! Science writing can be fun and weird and help us think in new ways about our lives. His blog can be found here.
Alex has this feature on my friend
epic movement-unto-itself also!While not exactly a scientifically rigorous way of determining efficacy at a population level, it is data and the best we have at this point.
For example, the order of the first 10 patients with associated clinical responses (Remission [R] = 15%, Adverse Reaction [A] = 5%, No response [N] = 80 %) for one of the psychiatrists could be: N,N,A,N,N,R,N,N,R,R. Note: the website random.org was used to produce a random number between 1-100. If the number was from 1-15, this was allocated “R”. From 16-20, allocated “A”. And from 21-100 allocated “N”
Good post, especially for those of us who have, first-hand, observed buses coming in threes!