Placebo response rate is ruining drug development in psychiatry: why is this happening and what can we do about it?

Psychopharmacology's 'Golden Era' of the 1990s and early 2000s is over. The soaring placebo response rate combined with the lack of novel validated drug targets has ended it. Big Pharma have now largely abandoned psychiatry. The first 'Golden Era' of Psychopharmacology in the 1960s and 1970s ended when the therapeutic targets of that era became saturated, and not because the placebo response rate skyrocketed. Innovation was rejuvenated into a second 'Golden Era' by agents with fewer side-effects, not really by novel targets, yet on the back of a clinical testing methodology that robustly and consistently distinguished active agents from placebo. Will there be a third 'Golden Era?' Indeed, there is hope that academia and small pharma will find the novel targets for the future, but how can innovation be rejuvenated into a third 'Golden Era' with novel targets if we cannot tell active agents from placebo? No one really knows the cause of escalating placebo response rates, hence this editorial. But feasible answers may lie both in how clinical practice and clinical trials have changed since the second 'Golden Era' and in how clinical trial design and implementation have also changed over this time. In those 'old days', we used to put patients into clinical trials from our own practices, and to rate the patients ourselves and then to continue to see them after the trial. Clinical trials were at the beginning of the second 'Golden Era' very much like clinical practice. Not today. Over the past few decades, clinical practice has diverged radically from clinical trials. In clinical practice today, patients – perhaps especially in the United States – have come to experience abrupt visits with rotating providers and no hospitalizations, fighting for coverage of their drugs and visits with frequent denial of both. Furthermore, these patients are often not candidates for clinical trials because their cases are too complex and their clinical responses to prior treatment so poor. On the other hand, patients in clinical trials today are exposed to a business-like situation where the principle investigators are no longer their personal physicians who enter them into the trial and do the ratings themselves, but instead to a more professional machine-like operation with professional patients recruited from advertisements into settings where much more time is spent at each visit, where more attention is given by professional raters who are not the principle investigator and may not be psychiatrists, and with free treatment and often monetary compensation for travel if not more. Changes in clinical trial design over recent years include tightening pressure from sponsors to enroll rapidly and financial incentives to do so that have occurred in parallel with longer visits, more ratings and much more paperwork. Many of us fear these developments may have distorted the incentives of both patients and raters to exaggerate symptoms prior to randomization, and to get better during the course of the trial no matter what their treatment. Another development is that as ethical review committees have gotten stricter and legalistic, consent forms have exploded from a paragraph to legal documents of thousands of words with details of how the trial is designed. It is worrisome that detailed knowledge of clinical design features by both the patient and the clinical raters, including the presence and timing of placebo and the chances of getting placebo or a known active agent versus study drug, could possibly distort expectations in a way that escalates placebo response rates 1. Many have speculated on additional demographic, psychopathological and administrative factors without consistent findings or a consensus emerging as to why placebo response rates are rising (e.g., 2-11). The consequence of high placebo response rate, small effect sizes and frequent failed trials has even led adversaries of psychiatry and medication treatment to argue that drugs such as antidepressants are no better than placebo and only cause additional expense and side-effects 12. Even if we do not really understand all the causes, or the principle cause, of the escalating placebo response rate, the question is can we fix it? Or in a larger sense, can we develop clinical trial methodology or new statistics that will tell us whether our drugs 'work' without cluttering the landscape with 'failed studies' where placebo is as good as drug and tell us nothing about efficacy? Attempts to identify clinical trial factors responsible for high placebo response have been many and the study of Kubo et al. 13 in this issue of Acta describes some important insights into this problem seen in antipsychotic trials. They report that a clinical feature of the schizophrenia population studied (i.e., disorganized thought scores) as well as lower PANSS (positive and negative syndrome scale) total score reduction at week 1 were associated with higher placebo response. Implementing these findings in future clinical trials in an attempt to reduce placebo response might mean excluding those with high disorganized thought as well as those with large reductions in PANSS at week one. The latter would be tricky to implement because it might mean a week of placebo lead in, a controversial way to lower placebo response rate 5-11. One potential problem with placebo lead in is that as soon as the rater and the patient know from the consent form and from reading the protocol that there is the possibility of a placebo lead in, this would change expectations of response in a way that could alter the week one response 14. Alternatively, eliminating patients with big responses at week one after randomization might reduce placebo response at the end of the trial, but excluding patients already randomized is a big no-no for current statistics used in clinical trials required by government drug licensing authorities. What are some other ideas to 'fix' the placebo response problem? Another finding published in Acta is that placebo response rate may depend upon the investigated drug and that placebo response in one study of one drug may not be able to be compared with placebo response in another study with a different drug, complicating or possibly invalidating the process of doing meta-analyses 11. One approach to foiling the placebo response rate is to try to 'trick' the rater by using one rating scale for entry criteria to the study and another for efficacy rating outcomes; other approaches are to recognize and try to address the litany of factors that have been shown to contribute to placebo response rate such as rater variance in calibration and training/experience, placebo lead-in, trial duration, active controls versus placebo controls, academic versus commercial sites, percentage of subjects randomized to placebo, fixed versus flexible dosing, severity of illness, sample size, number of study centers and several others 1, 14. None has yet proven to be the 'silver bullet' to fix the escalating placebo response rate and number of failed trials. Nor is controlling any of these factors likely to solve the placebo response issue across psychopharmacology since the variables linked to placebo response differ in depression, versus psychosis, versus dementia, pain, anxiety and other conditions. So, what is the solution? In the short run, those who have run recently successful clinical trial programs that successfully distinguished drug from placebo (personal communication) suggest that selection of a limited number of experienced sites held accountable for their placebo response rate as well as for their clinical operations with protocols that are not too complicated is about all one can do. But this is not enough for the long run. Innovative solutions are needed such as the possibility of a paradigm shift from the randomized placebo-controlled design with classical statistics and 'purified' patients to consideration of other designs such as including real-world complicated patients and making comparisons of new agent augmentation to treatment as usual rather than to placebo; open-label treatment of new agents with randomization only of responders to double blind continuation or discontinuation of treatment; or here is a shocker: experimenting with statistics that would allow fair evaluation of drug versus placebo by making some adjustments AFTER randomization, such as those suggested by Kubo et al. 13, namely by dropping out early responders. A variation on this is already suggested by the novel Sequential Parallel design now being more commonly implemented in modern trials with novel statistics where the placebo non-responders are re-randomized to drug or placebo and the placebo responders are dropped out 14. Lead-in phases for Sequential Parallel design are long 1. Findings from Kubo et al. suggest placebo response as early as 1 week is predictive of later placebo response, so similar studies could be informative for providing empirical evidence to drive lead-in periods and criteria for future Sequential Parallel designs, which could save time and money. How about the possibility that drug levels are monitored after randomization and placebo patients found to be taking active drug or similar, prohibited drugs (amazingly common in clinical trials and a proposed factor for increased placebo response) as well as patients randomized to drug who have no drug in their system (also quite common) are not analyzed 15-18. Statisticians will not like this for violating randomization; but are statistics supposed to serve the feasible clinical trial design or are failing clinical trials supposed to serve the statistics just because that is the way we have already done it? Regulatory authorities will also not like this because it might tend to favor drug over placebo. But wait! That might be just what we are trying to accomplish. Whatever the solution, we need to think outside of the box, possibly develop and adapt a new 'gold standard' trial other than the randomized placebo-controlled study so that we can enter a new 'Golden Era' of psychopharmacology. Stephen M. Stahl, M.D., Ph.D is an Adjunct Professor of Psychiatry at the University of California San Diego, Honorary Visiting Senior Fellow at the University of Cambridge, UK and Director of Psychopharmacology for California Department of State Hospitals. Over the past 12 months Dr. Stahl has served as a consultant to Acadia, Adamas, Alkermes, Allergan, Arbor Pharmaceuticals, Avanir, Axovant, Concert, Clearview, Ferring, Intra-Cellular Therapies, Janssen, Lilly, Lundbeck, Neos, Otsuka, Pfizer, Servier, Shire, Sunovion, Takeda, Taliaz, Teva, Tonix, and Viforpharma; he is a board member of Genomind; he has served on speakers bureaus for Acadia, Lundbeck, Otsuka, Perrigo, Servier, Sunovion and Takeda and he has received research and/or grant support from Acadia, Avanir, Braeburn Pharmaceuticals, Eli Lilly, Intra-Cellular Therapies, Ironshore, ISSWSH, Neurocrine, Otsuka, Shire, Sunovion, and TMS NeuroHealth Centers. Gian D. Greenberg is an employee of Arbor Scientia and has no other disclosures.