Placebos - now with added ethics!

Placebo effects can be induced in patients without deception, according to a new study in PLoS One. The study was an open label (no blinding for patients) single blind (the investigators did not know which treatment participants were given) controlled trial in 80 people suffering from Irritable Bowel Syndrome.

The patients were followed for one month, with a baseline, end and midpoint assessment. According to validated self report measures for the syndrome, the patients who were given the pill improved much more than the patients who received no treatment. Its important to note that the groups were randomised, and they were also matched in the amount of interaction they had with the medical providers.

An interesting point in the study (which doesn't appear to have been picked up by other science bloggers) is that some of the patients (N=17) also received counterbalanced provider interaction - they saw a male doctor once, and and also a female nurse. Contrary to some conceptions of the authority of the provider having an impact on the response to treatment, there were no differences in outcome which could be attributed to this difference. Given the small number of patients in this group, thats not very surprising. I really wish this trial had gone the whole hog and randomised everyone to see both practitioners, as that might have provided some very useful data.

Open-label placebo produced significantly higher mean (±SD) global improvement scores (IBS-GIS) at both 11-day midpoint (5.2±1.0 vs. 4.0±1.1, p<.001) and at 21-day endpoint (5.0±1.5 vs. 3.9±1.3, p = .002). Significant results were also observed at both time points for reduced symptom severity (IBS-SSS, p = .008 and p = .03) and adequate relief (IBS-AR, p = .02 and p = .03); and a trend favoring open-label placebo was observed for quality of life (IBS-QoL) at the 21-day endpoint (p = .08).

The above is from the finding section of the abstract, but it cogently sums up the major findings of the study. Now, there are a number of important caveats to the straightforward interpretation of the study. There are also a number of interesting implications arising from both the study, and the reactions of some of the better known science bloggers.

My first issue with this study is the number of statistical analyses which were carried out, with only 80 participants and at least ten significance tests reported in the article (and probably more which were not reported), the authors probably should have corrected for multiple tests (the most popular approach is to divided the required p value by the number of tests). That being said, this was a pilot study, so the results will require replication in a larger trial which would ideally have a protocol with details of analyses to be carried out published beforehand (what can i say, I'm an optimist).

Orac, (of respectful insolence fame) critiqued this study on a number of grounds. The first was (he claimed) a failure of randomisation. He based this on the numbers with each type of IBS (diarrhea or constipation primary) and argued that this could be responsible for the observed improvement. While i do take his point, i would suggest to my readers (all three of you) to look at the Table itself published in the article. Now, it can be seen from the table that indeed the groups did not appear matched on type of IBS.


However, if you look a little more closely in the table, it can be seen that the open label group had a longer mean duration of IBS,  a higher initial mean symptom severity score, and a lower initial quality of life. If anything, if there was no change over the course of the trial, then the no treatment group should have come out superior.  Given that only stable IBS patients were admitted into the study (look at the confidence intervals for the lenght of time with IBS) it seems unlikely that regression to the mean could account for these results.

The effect size for the mean difference between groups at the end of the study was d=.79, which is a large effect by anyone's standards (see the pun? you're a nerd if you do). To explain the effect size measure, its a difference in standard deviations, and one standard deviation is the difference between your high school teacher and Einstein, as measured by IQ (assuming an IQ for the teacher of 115 and giving Einstein 130).

This is not a small difference, and yet Ed Yong reports that Edward Ernst claims this is too small to be clinically significant, which makes me wonder what effect sizes he sees in everyday practice, cos thats a large effect to me (I'd kill for an effect size that large, i'd get t-shirts printed and everything).

Orac also takes the research to task for deception, as the placebo pills were described as empty sigar pills which have been proven in rigorous clinical testing to have an impact on self healing processes. He claims that this is the worst deception of all, far nastier than those involved in ordinary Randomised trials. Frankly, i don't agree. There have been a number of meta-analyses  conducted on the placebo, as well as re-analyses of data from many, many clinical trials, and what participants were told was not a deception, unless telling people about what clinical trials have shown about any medical treatment is a deception. So, i really dont see why this bothers him.

 Orac also uses guilt by association as he notes that the study was funded by the National Council for Complementary and Alternative Medicine, but i believe that argument is beneath any self respecting scientist so I'll ignore it.

PaiMD also takes note of the study, and claims that what was compared here was one non biological treatment against another. I would disagree, what this trial shows is that care and some kind of medical ritual (take these pills two times daily) are much better in combination than they are apart. That, to me, is perhaps the most interesting finding of this study.

Something which may also interests students of the placebo was the theoretical implications of this study. I've talked elsewhere about theories of placebo, and briefly i think that this study shows that the effects of expectancies are subservient to those of ritual. This is a clear mark in favour of Hyland's theory of motivational concordance, which basically holds that placebo effects arise from what we do that we find meaningful, rather than what we think (or expect) about a treatment. See the link to my previous post on theories if you'd like a more in depth discussion of the theoretical approaches in the field.

One brief detail that i would like to know more about in the study is how many pills the open label group took, and whether those who overdosed got more benefit. The authors note that a pill count was taken, but they do not report the results of this measure, which is a shame. This measure would have been especially great as it could be modelled as a Poisson variable (for count data) and then demographics and other measures collected could have been regressed against it to understand the causes and correlates of this interesting variable a little better.

Now, my own thoughts on the future of this line of research are as follows: it will be difficult to replicate this effect when a drug is being used in the study. This seems intuitive to me, given that the meaning of getting placebo or drug is very different from the meaning of getting placebo or no treatment. Mind you, I hope I'm wrong here. An interesting line of research which bears upon this study is the work using the balanced placebo design. Essentially, this work combines the drug and placebo arms of your standard clinical trial with two deceptive conditions, where the participant is told they get a drug but get placebo and vice versa.

I personally would find a replication of this study using that design to be far more interesting (though ethically challenging) as the effects of placebo-placebo condition could be contrasted with the others. Again, this is probably just a pipe dream, but if ever have enough funding, i'd like to make this happen.

One way in which these findings can be explained is as follows. I've spoken before about the effects of self monitoring of internal processes on the placebo effect (essentially, it increases them). It may be that the open label group, while taking the pills paid more attention to their bodies, and this attention increased their self healing processes (which is all the placebo effect really is). This work on somatic awareness also ties into the results of a recent meta-analysis on in what conditions the placebo is effective Just a thought, probably needs a little more refining into a useful predictive theory.

Anyway, thanks for reading. You may have noted that I haven't really talked that much about the paper itself. Its is PLos One, which is free to everyone, and its very clearly and engagingly written, so go read it yourselves, rather than relying on me or any other blogger to give you their perspective.


Kaptchuk, T., Friedlander, E., Kelley, J., Sanchez, M., Kokkotou, E., Singer, J., Kowalczykowski, M., Miller, F., Kirsch, I., & Lembo, A. (2010). Placebos without Deception: A Randomized Controlled Trial in Irritable Bowel Syndrome PLoS ONE, 5 (12) DOI: 10.1371/journal.pone.0015591

Placebo commentary

placTaking a little break from my placebo run down, as I found these old blog posts on placebos, both inspired by a wired article which notes the problem of decreasing drug-placebo differences in recent clinical trials.

The link to White Coat Underground is here, and Greg Laden's rather more thoughtful peice is here .

The first point I would like to make about PaiMD's article is that he seems to be unaware of all the experimental work done on placebos, focusing instead on the clinical trial use of placebos.

Recent work has shown that the nature of clinical trials actually decreases the placebo effect, due to whats known as the expectancy theory. It goes like this: in a trial, participants are told that they may get placebo or they may get the drug. This is a conditional expectancy. In experimental research, they are told that it is a powerful painkiller (uncondtional expectancy). The UCE's have been shown to produce much better results, with less requirements for painkillers following surgery, with stronger coffee effects and indeed even when measuring sleep patterns. 

So, what does this tell us? To me, it seems to indicate that placebo effects are underestimated in clinical trials, and if they appear to be getting stronger here then something weird is going on. I actually agree with Greg Laden's idea that it may be down to stronger cultural associations between pills and healing, which creates stronger expectancies that then interfere with the testing of new drugs. This could be tested properly by a meta-analysis comparing drug advertsing and placebo response across countries, and hopefully I'll get a chance to do this review after i finish my doctorate.

Something else that struck me, as I read through the comments, was that they were recapitualting the history of studies of placebo. We had people claiming that placebos were useless, that they were only a control and an artifact, while others claiming that modern medicine was a lie and only the mind had power (the burt dude at the end of Greg Laden's comments was hilarious, especially the way he made so many people angry).

Anyway, a few issues came up and I thought i should post some more recent research which hopefully, can illuminate the debate.

Firstly, the meta-analysis by Hrobjarrtson and Goetzsche (god those names are hard to spell). Now, this meta-analysis claimed that placebo was not significantly different from no treatment across 114 clinical conditions.

Personally, i can't see how they were able to do this meta-analysis, considering the extreme heterogenity of the data base (which is supposed to mean that you avoid the meta-analysis). Anyway, this review drove a lot of placebo people into studying pain, as it was the only part which they didnt slate.

Now, more recently, another meta-analysis was done by Meissner, Distal and Colleagues. This meta-analysis actually seperated the trials they found into different components, and found that there was a large placebo effect (d=.5 approx) when the outcome variable was a physical parameter, but none when the outcome variable was a hormone level. They re-analysed H&G's sample and replicated their results. For some reason, this study didnt get nearly as much attention as the negative one.

So, essentially, we can see placebo effects in some areas but not in others, which makes a lot of sense, if you study the literature. An alternative explanation is that the clinical trials only activated the expectancy pathways which affect these outcomes, while the conditioning pathways were not activated. This makes sense if you look at some of the work by Benedetti et al.

Another canard that was raised in the comments was the notion of response bias. This was popularised by Allan and Siegel in their Signal Detection Theory of the Placebo Effect. While I like SDT,  i dont think that it can account for all of the observed placebo effects. Referring to the surgery paper above, what the results suggest is that most of the response variability to opioids is the result of placebo effects, which is the result (mostly) of the endoegnous opioid system. Response bias cannot account for these well documented effects.

Anyway, thats probably enough for now. Hopefully I'll get my second part of the placebo review done today, if not my cousin's wedding will intervene and it'll be next Monday or Tuesday.