The Theatre of Treatment

I think of myself as a skeptic, someone who is suspicious of much of what we do in medicine. The interventions in which I am confident are few; those that have been shown to have large treatment effects in robust clinical trials. When I read this essay, however, I found myself thinking, “Geez, Adam, you are an optimist, a true believer.”

I love this essay, mostly because Dr. Raudasoja has really made me think.

Adam Cifu

Modern clinical medicine rests on a quiet pretense: that the visit can read whether a clinical intervention worked. It cannot. The prescriber writes a prescription, the patient takes it, they meet again at three months, and a judgment is rendered. If the patient is better, the drug helped. If the patient is not better, either the dose is too low or the drug is wrong. Sometimes, if several drug treatments fail, the patient’s condition is considered treatment-resistant. This is the framework behind some of the most common conditions we treat — depression, dementia, and acute and chronic pain.

The framework rests on a biased assumption: that treatment effects are observable at the individual patient level. They mostly are not. Average drug effects in conditions like those named above are so small that they are rarely detectable. Set against the natural fluctuations of illness, detecting a drug response at the bedside becomes impossible.

Yet healthcare still leans on this framework. I call it the theatre of treatment. The performance is for the patient, for the chart, for the system that pays the clinician’s time. It is for the clinician’s own sense of doing something. The clinician follows a script: prescribe, wait, adjust the dose, switch the agent, and treat each step as a verdict on the medicine. No one is acting in bad faith; the performance is structural. It is the same performance medicine has staged for two thousand years, in which clinicians believe they are treating, while mostly they are watching the disease take its own course.

The numbers behind the prescription

Statistics support that we are acting in a theatre of treatment. Roughly one in eight patients with major depression responds to an antidepressant. The number is real and traceable to the literature. However, it does not say what it pretends to say. It is largely a statistical artifact of where the responder threshold is drawn and of the amount of variance in the study population, both treatment-related and not.

Consider a study in which all treatment group participants have 51% symptom improvement, and the control group participants 49%. Draw the response line at 50%, and you get a 100% response rate in the treatment group and 0% in the control group, number needed to treat of one. The figure does not tell us how much patients improve or whether some improve more than others; it is a single number that compresses many things together, and it is largely meaningless for treatment decisions.

What we should be looking at is the average benefit on the symptom scale itself. For antidepressants, the mean drug-placebo difference on the Hamilton scale (0–52 points) is about 1.8 points, against a minimal important change — the smallest difference patients themselves perceive as meaningful — of 3-5 points. A 1.8-point difference is unlikely to be noticed by a patient. Whether some subgroups of patients see larger effects, we usually do not know; what we do know is that responder analysis cannot answer the question. Even a patient who is worse at follow-up tells us nothing on their own: they may be even worse without the drug. In fact, because the absolute benefit of a drug tends to scale with symptom severity, the patients who look like the clearest non-responders could be the ones the drug is helping most.

The same logic applies elsewhere in the formulary. The new Alzheimer’s drug donanemab slowed cognitive decline by roughly half a point on the 30-point MMSE over 18 months. This compares with a placebo group whose scores declined by about 2 points. The minimal important difference is about 1.5 points. Compare these results to older acetylcholinesterase inhibitors that improve MMSE at 26 weeks by about 1 point.

Paracetamol improves the pain of osteoarthritis by about 4 points on a 0–100 scale, where the minimal important change is around 10.

The trials are likely well-designed. The benefits are real. They are also, for the average patient, small enough that the patient would likely call them no change.

The follow-up illusion

If the population-level signal is small, why do some patients seem to improve? Symptoms change for reasons that have nothing to do with the drug, and on a much larger scale. Depressive episodes lift on their own; chronic pain waxes and wanes; cognitive decline is non-linear . A Dutch primary-care cohort followed patients with major depression for three years: 43 percent recovered from their first episode, 40 percent had a fluctuating course, and only 17 percent remained continuously depressed. In clinical trials, about 43% of patients in the placebo arm improve by at least 50% at 6 to 8 weeks. That improvement is not the drug — it is the natural course of the illness, for example, regression to the mean (patients enroll when they are at their worst, and worst rarely lasts), the clinical contact itself, structured attention of being in a trial, and whatever else has shifted in their lives between visits.

Clinicians help maintain the illusion. In a 2021 study, 542 American primary-care physicians estimated how much routine prescriptions — warfarin, antihypertensives, bisphosphonates, statins — would help a specific patient. The median estimates ranged from 20% to 50% absolute risk reduction. The actual figures from the trials are around one percentage point. The doctors who overestimated the most were the ones most likely to prescribe.

The clinician walks into the visit, then, with a belief about drug benefit the trials do not support, meets a patient whose course is mostly determined by the illness and other non-treatment-related factors. The visiting doctor asks, “Is it better?” and then treats the answer as a verdict on the efficacy of the medication.

An old mistake in a new costume

Waiting for a response and reading the patient’s improvement as evidence that the treatment worked is the inferential structure medicine has always used. The history of medicine is full of treatments that did nothing, and clinicians and patients still waited for the “response.”

What has changed is that today’s drugs often do something, but the something is still not observable and not large enough to make the structure work. We are still in the same theatre, waiting for patients to “respond” and feeling vindicated when they do. When they do not, the answer is more medicine — a higher dose, a different drug, another chance at a response.

What should change look like

What should change? If we were honest about response, we would prescribe less. That would be a real gain, but probably not the main one. Think about the opportunity costs. A patient waiting for an antidepressant to work may be a patient not investing in the things that might do more for depression — exercise, sleep, the slow rebuilding of a life. A healthcare system organized around three-month follow-ups to read drug response is one whose clinicians spend their scarcest resource — time — on a reading they cannot reliably make. Drop the reading, and that time becomes available for more important conversations. Unlike our predecessors, we already have the evidence to see the theatre for what it is — and to act on something better.

Aleksi Raudasoja, MD, PhD, is an editor at Finnish Current Care Guidelines and is the responsible editor of Finnish Choosing Wisely Recommendations.

Photo Credit Giusi Borrasi