Randomized controlled trials are seen as the gold standard for guiding “evidence based medicine,” and up until recently, I didn’t pay much attention to the “kind” of evidence they provided. But after writing recently about two meta-analyses of antipsychotics and antidepressants that concluded these drugs were “effective,” I have come to think of RCTs in a new light.
The most important data in an RCT is not whether the drug provides a statistically significant benefit over placebo. The most important data is the “number needed to treat” calculation (NNT). This is the data that should be used to provide patients with informed consent about the likelihood that they will benefit from the treatment over the short term, or, conversely, be harmed by it. And it is this data—and not efficacy data—that should inform prescribing protocols.
In the past, I have written about the corruption in the RCTs of psychiatric drugs—the bias by design, the use of a placebo group composed of people withdrawn from their drugs, the spinning of published results, and so forth. But this post is about something different. Even if we take the RCTs of antipsychotics and antidepressants at face value, the NNT data from the meta-analyses reveals that the overwhelming majority of patients do not benefit from the treatment, and thus are harmed, at least to some degree, by their exposure to the drug.
In short, the NNT numbers provide evidence for utilizing the drugs in a selective way, and avoiding their immediate use in first-episode patients.
Efficacy vs. NNT Findings
The finding of efficacy in a trial of a psychiatric drug comes from calculating aggregate outcomes for the drug and placebo groups. If the reduction of symptoms, on average, is greater in the drug group than in the placebo group, and if this difference is statistically significant, then the trial is deemed to be positive for the drug. The treatment provides a benefit over placebo.
This is the evidence that is cited in psychiatry for using antipsychotics and antidepressants as first-line therapies for depression and psychotic disorders. These drugs—and this is how the findings are presented to the public—can be said to “work.”
However, the individuals in a trial of a psychiatric drug, in both the placebo and drug-treated groups, will have varying responses: some will worsen, some will stay the same, some will get a little better, and some may get a lot better. If you plot out the individual responses in the two groups, the bell curves for each group will overlap to some degree. The degree of overlap is reflective of the “effect size,” and that in turn leads to a calculation of the number of people who must be treated to produce one additional person who receives a benefit from the drug (NNT).
Thus, the NNT tells of the percentage of people who are being exposed to the adverse effects of the drug without any additional benefit, and the percentage of people who have a positive therapeutic response they otherwise wouldn’t have had. The first group could be said to have been harmed by the treatment, while the second group could be said to have benefitted from it.
For instance, an NNT of 10 means that 10 people must be treated to produce one additional person who has a positive outcome, which leads to this benefit-harm equation: nine people will be exposed to the adverse effects of the treatment without any additional benefit (and thus are harmed), while one will have a beneficial response that he or she would not have otherwise had.
With an NNT number in mind, a person contemplating taking a drug can assess whether the odds of having a better response that comes with use of the drug is worth the certain risk of being exposed to the drug’s adverse effects. That’s the equation patients should look to when considering whether to take a drug, as opposed to the “efficacy” finding we usually focus on.
It’s easy to appreciate the dramatic difference in understanding that comes from seeing the merits of a drug through the NNT lens, as opposed to the effective/not effective lens. The first provides a sophisticated understanding of the risks that come with taking the drug and reminds patient and prescriber alike that individual outcomes may vary widely. The latter leads to a simplistic “drug works” conclusion, which promotes the false notion that most patients can expect to benefit from the treatment.
In other words, the NNT illuminates the drug’s variable impact, while the efficacy finding leads to a hiding of this critical fact. The efficacy data, in fact, could be said to lead to a clinical delusion.
The NNT for Antidepressants
Irving Kirsch and others have calculated that antidepressants, in the RCTs, have an effect size of .30. Here is a visualization of a treatment with an effect size of .30:
In this graphic, there is an 88% overlap in the spectrum of outcomes between the two groups. This produces an NNT of 8. One in eight people treated with an antidepressant will have a positive response they otherwise wouldn’t have had; the other seven will have been exposed to the adverse effects of the treatment without receiving any additional benefit.
Thus, for the person considering taking an antidepressant, the NNT data provides the “math” needed to weigh the potential benefit of taking the drug against the potential harm of doing so. The patient will know they have a one-in-eight chance of doing better than they would without the antidepressant, and thus the question for the patient: Is this possible benefit worth the negative aspects of being exposed to the drug?
To complete that informed consent, the patient would need to have an understanding of the possible negative effects of exposure to an antidepressant. The side effects are many, starting with sexual dysfunction and an increased risk of having a manic reaction, which may lead to a bipolar diagnosis. Other possible negative effects include suffering withdrawal symptoms when trying to quit the antidepressant and possibly ending up on the antidepressant long-term, which in turn often leads to a host of physical and emotional difficulties.
Many depressed people, presented with this information, might still choose to take an antidepressant. They would see it as a worthwhile risk. At the same time, many would likely choose to forgo taking the drug and seek out a non-drug alternative.
My own guess is that the vast majority of people, when confronted with the NNT of 8 finding, would take the second option, particularly when suffering a first episode of depression. Unfortunately, very few patients are presented with the NNT data when they are considering taking an antidepressant, and such data rarely informs the thinking of prescribers.
It should be noted too that this NNT of 8 is derived from industry-conducted trials of antidepressants, and that in studies of “real-world” patients, such as the STAR*D trial, drug-response rates have been much lower. It may be that in “real-world” patients, antidepressants provide no “efficacy” benefit at all over placebo in the short term. The NNT-of-8 calculation is a best possible scenario for assessing the benefit-harm ratio for short-term use of an antidepressant.
With Antipsychotics the NNT is 6
In a 2009 report, Leucht published a meta-analysis of 38 trials of second-generation antipsychotics and reported a response rate of 41% for the drug-treated patients versus 24% for the placebo group. These percentages, they noted, produce an effect size of .50, which translates into an NNT of six.
Six people must be treated to produce one more person who has a favorable response. The other five (80%) can be said to have received no additional benefit from the treatment but were exposed to the adverse effects of antipsychotic treatment, which, as is well known, are many.
The responder percentages cited by Leucht make it easy to see the benefit vs. harm equation in the use of an antipsychotic. Those harmed are the non-responders to the drug (59%) and those who would have responded without the treatment (24%), which equals 83% (or roughly five of every six patients.)
The adverse effects of antipsychotics are almost too numerous to list. Second-generation antipsychotics may cause weight gain, diabetes, metabolic dysfunction, Parkinsonian symptoms, cognitive slowing, emotional numbing, and brain shrinkage. It can be difficult to withdraw from an antipsychotic, and the risk of long-term use includes tardive dyskinesia and a host of other negative effects.
Thus, the benefit versus harm equation that comes from Leucht’s meta-analysis: Is it worth it for a patient who is “psychotic” to take an antipsychotic to gain this one-in-six chance of having a favorable response they otherwise wouldn’t have had, with this possible benefit coming at the expense of exposure to the adverse effects of the drug?
NNT-informed Prescribing Practices
The NNT data reveals that the majority of patients will be harmed to some degree by taking an antidepressant or an antipsychotic, even over the short term. With antidepressants, 88% will fall into this category of being exposed to the drug without any additional benefit; with an antipsychotic, 80% will fall into this category.
Given this fact, it is easy to understand that RCTs in psychiatry do not provide an “evidence base” for “one size fits all” prescribing practices. Instead, they provide compelling evidence that prescribers need to develop “selective use” protocols, which would seek to identify those who could get well without the drug treatment, and seek to halt treatment in non-responders to the drug.
Placebo response vs. drug response
There is ample evidence that recovery from a depressive or psychotic episode without exposure to drug treatment puts the patient onto a much better long-term path. Relapse rates are lower for those that recover without medication, and their long-term functional outcomes are better too.
There is a fairly easy way to identify those who might get well without drug treatment, particularly with first-episode depression or first-episode psychosis. Prescribers, while providing psychosocial care, could utilize a watch-and-wait practice. Wait for a week or two to see if the patient begins to improve without the drug, and this would help identify those who could get well without the drug.
This is the antipsychotic protocol used in the Open Dialogue program developed in western Lapland in Finland, starting in the 1990s. They have reported that two-thirds of their first-episode psychotic patients recover without the use of antipsychotics and are well at the end of five years. Their results tell of the extraordinary public health benefit that can come from a watch-and-wait practice.
As for first-episode depression, prior to the marketing of the SSRIs, it was understood that depression was an episodic illness, and that most people could be expected to recover in time without any somatic intervention. Indeed, in the early years of the antidepressant era, a common thought was that the drugs were useful because they could speed up this natural healing process.
Thus, the benefit that would come with a watch-and-wait protocol is not simply that some patients would avoid the drug’s adverse effects over the short therm. Recovery without drug treatment also puts patients onto a better long-term course, as can be seen most clearly in the Open Dialogue outcomes.
The RCTs of antipsychotics and antidepressants tell of a significant percentage of patients who do not get better on the drug. Given this fact, the second part of a selective use protocol would be to stop prescribing the drug to non-responders, and to try other therapeutic methods instead.
This seems rather obvious—why prescribe a drug that doesn’t work—and yet the current practice is to double down on the drug treatment. Doctors put non-responders on additional drugs, and soon these patients are on a polypharmacy regimen, and struggling with a mounting burden of side effects.
The NNT numbers warn prescribers that there will be many non-responders. If prescribers were to adopt a selective-use protocol, they would try to identify these non-responders quickly, and switch them to a non-drug therapy. This would be an essential aspect of a prescribing protocol that sought to determine “for whom” the drug works, and “for how long,” which is understood to be the question that needs to be answered in order to make “best use” of a drug.
A Unified Body of Evidence
In Anatomy of an Epidemic, I presented the evidence base for psychiatric drugs in this way: I wrote of how RCTs provided evidence of their short-term benefits (at least to some small degree), and then examined evidence of many other types—epidemiological studies, longitudinal studies, MRI findings, and so forth—that told of how psychiatric drugs worsened long-term outcomes.
In other words, in that book, I told of a paradox: drugs that “worked” over the short term, but caused harm over the long-term.
However, if you focus on the NNT data from the RCTs, the paradox disappears. The RCTs tell of a majority of patients who do not benefit from the drugs, and thus are harmed by one-size-fits-all prescribing practices. Specifically:
- They tell of a significant percentage of patients who would have recovered over the short term without the treatment but have now been exposed to the drug, and this may put them onto a path of long-term use of an antidepressant and other psychiatric drugs, which is fraught with possible negative effects.
- They tell of a significant percentage of medicated patients who are non-responders and soon head down the polypharmacy rabbit hole.
Indeed, the RCTs and long-term outcomes literature now come together to tell of a paradigm of care that, as currently practiced, leads to more harm than good for patients right from the first moment. The biggest risk for first-episode patients is that first use of a drug will turn into long-term use, which means that in return for a small chance—1 in 8 for antidepressants, 1 in 6 for antipsychotics—that a person will have a better response on drug than on on placebo, that person is exposed to a significant risk that they will become a long-time user of a psychiatric drug.
There are hundreds of personal blogs that have been published on MIA that tell of this short-term to long-term pathway of harm. The authors tell of either not responding to the drugs in the first place or gradually worsening on a psychiatric drug, of falling into polypharmacy torment, and eventually of lives that were diminished or even ruined. Their personal stories attest to the very outcomes that are visible in the RCTs and long-term outcomes literature, if only the profession would take a close look at its own findings.
In the solutions section of Anatomy of an Epidemic, I wrote that psychiatry needed to adopt selective-use protocols, based on two precepts: avoid immediate use of drugs to identify those who can get better “naturally,” and try to minimize long-term use, which includes stopping treatment in non-responders.
The NNT numbers from the short-term RCTs argue for this same selective-use protocol, and this was the aha moment for me. There is a consistent evidence base, starting with the RCTs that assess short-term outcomes, that calls for psychiatry to dramatically rethink its use of these drugs and to adopt selective-use protocols. As long as it fails to do so and clings to its one-size-fits-all protocols, psychiatry will be using the drugs in a way that does—and there is no other way to put it—great harm.