Placebos are a lot like the New York Yankees; everyone’s happy when they lose. And I can say that, because I’m a Yankees fan. But when placebos are just as good as drugs, when the effect of getting designer pharmaceuticals is the same as getting a few grams of glucose, there’s an open conversation as to why. Much has been made of the placebo effect – modern theories associated the simple action of taking a pill with activation of endogenous endorphin and cannabinoid receptors. Efforts to harness this effect, and to do so ethically, have consumed a lot of brain power and journal pages. But it’s not the only explanation.
Another possibility is that what we’re really seeing is regression to the mean, a well-described statistical phenomenon in which the natural tendency of measurements far from a mean to normalize over time gives the impression of an effect that isn’t there. Consider:
Image from Barnett et al., Int J Epidemiol (2005) 34 (1): 215-220.
If low HDL levels are one of your inclusion criteria, it’s reasonable to think that your patients are more likely to have higher HDL levels, that is to say levels closer to a population mean, on follow up. Maybe it’s not the placebo that’s making your control group get better – maybe it’s the fact that values far from a mean are naturally more likely to be closer to it when you repeat them, or rather to have been speciously far from it on your baseline measurement.
I tell you that story to tell you this one. Today, we’re taking a look in the journal Pain at a trial of open-label placebo for back pain. In what can only be described as a gutsy move, investigators gave 97 patients with back pain for at least 3 months treatment as usual versus addition of an open label placebo. Patients were told the pill contained no drug, but that placebo had been found to be helpful for back pain in numerous RCTs (the truth and ethics surrounding this statement are a topic for another day). They were excluded if they had prior surgery, had cancer, or had taken opioids for pain in the last 6 months. And what happened? Three weeks later, pain score on a standard 10-point scale decreased by an average of 1.5 points in the placebo group vs 0.2 points in the plac- errrr, control group. Non-placebo group? Whatever. Disability scores also decreased by 2.9 vs 0 in the placebo vs usual treatment groups, respectively.
So is this regression to the mean? Well, it’s well documented that back pain gets better over time. But, in a really cool and unconventional design element, there was a planned post-study crossover, in which the treatment as usual group got their chance at placebo. 31 of the 38 patients did try the sugary stuff, and over an additional three week period usual pain improved by a mean of 1.7 points, significant and roughly comparable to the initial open label placebo group. Hard to argue this is anything other than an effective placebo – if it’s regression to the mean, why is the treatment as usual group waiting until they get their placebos to regress?
At the end of the day, while regression to the mean is definitely a real effect, and may be responsible for some of the negative trials we have seen, modern trials manage to anticipate and adjust for this fairly well in their design and analysis. The scientific consensus is definitely on the side of a real and measurable placebo effect, and this trial and others are exploring an interesting and possibly ethical method of exploiting this. And I think we can all agree that any efforts to expand the number of effective interventions for low back pain is a noble effort.
Read the original study here. If you want to learn a little more about the most up-to-date thinking on the placebo effect, try this NEJM review, and try this article to read a bit about regression to the mean (Warning: Lots of great explanation up front and at the end, lots of stats that can be tough to follow in the middle).