[mikehall314]

The Placebo Effect is something of a misnomer, as there are actually many different placebo effects, which bias trial data in subtle and interesting ways. Effects like regression to the mean and the natural history of the disease, for example, will result in objective changes to a patient's condition even in cases where the patient is acutely aware they are only taking a placebo.

Analyses comparing placebo interventions to no treatment reveal that the apparent power of the placebo may be overstated. No placebo effects are observed, for example, when comparing placebo to no treatment for objective endpoints, or binary endpoints (Hróbjartsson et al, 2001). They are observed in subjective endpoints (e.g. pain, nausea) where the condition of the patient is filtered through the opinion and biases of the patient and/or the clinician - which makes it quite possible that this aspect of placebo action can be accounted for by the experimenter effect.

All of which leads to my primary problem with this paper. It is a comparison of open-label placebo to no treatment, with a relatively small number of participants (n = 80), studying only subjective end-points (hello, experimenter effect). The media coverage of this paper (c.f. the NPR article) makes the claim that an "honest placebo" was given, with the patients informed they were only taking placebo, which is true. But patients were also told the placebo could "present significant improvement in IBS symptoms through mind-body self-healing processes", which just as readily primes the patient for the experimenter effect as does telling them they're taking a drug.

On top of that, the clinical relevance of the IBS-GIS improvement seen in the placebo arm is questionable, improvement from "(4) no change" to "(5) slight improvement" on a seven point scale.

Small effect, small numbers, and potentially flawed methodology.

[disgruntledphd2]

OK, lets take this in order. Hrobjarrtsson and Goetzche's review actually undermines your point around there being many placebo effects (which is the approach taken by Benedetti), as they modelled wildly different clinical trials (the inclusion criteria were simply possessing both a placebo and no-treatment arm).

Furthermore, Meissner et al 2007 re-analysed those studies and found large effects where the outcome could be mediated by nervous system, and very low effects where there was no direct nervous system link.

Additionally, uncertain expectations (you may receive a drug) are given in clinical trials and Vase (2002) found that placebo effect sizes were much smaller in this context than they are when placebos are deceptively administered (this is a potent painkiller) (as they normally are in clinical practice and experimental research) (c.f. Kirsch & Wiexel 1998, Amanzio et al 2001). Also certain vs uncertain expectations are associated with differential amounts of dopamine release, which has been associated with response to placebo (Scott et al 2007, DeLa Fuente-Fernandez, 2002, 2004).

I agree with many of your comments around this paper, and essentially it was only published because Kaptchuk and Kirsch are two of the leading names in the non-clinical field.

And the papers from 2010, and has been getting this treatment for a while....

[mikehall314]

Indeed, all fair points.

Though I disagree that Hróbjartsson et al undermines the notion of many placebo effects.. or maybe I don't, depending on our definition of placebo effect.

If we define "placebo effect" as any clinically significant change observed in the placebo wing of a clinical trial, there are many placebo effects. As I mentioned before, regression to the mean, natural history of the disease, experimenter effect, and so on.

Alternatively, we could define it as any clinically significant change observed in the placebo wing of a clinical trial which is not also observed in a no treatment wing. This would eliminate things like regression to the mean and arguably leave behind only a "true" placebo effect. Though even here, we have to account for bias (and perhaps even classical conditioning?) I wonder how much placebo effect is left if these are controlled for?

Within the context of analysing trial data, it is the former definition we are interested in. But within the context of discussing "the placebo effect" as a standalone phenomenon, I'd argue the latter definition is more useful.

Unfortunately, as most trials don't have no treatment arms, any clinically significant changes observed in the placebo arm are chalked up to "the placebo effect", especially by the media (though sometimes by clinicians), even when there is good reason to think many of those same changes would have been observed under no treatment. Which IMHO leads to a distorted view of the clinical relevance of the placebo effect outside the context of a clinical trial. Especially as the placebo effect appears to have a reputation as a bizarre mind-over-matter affair.

[disgruntledphd2]

Ah, I see what the issue is here. I typically (used to) work in experimental placebo research, rather than clinical placebo research (I'm a psychologist by trade) so I would be more interested in those studies, rather than clinical trials.

In terms of what's left over after accounting for no treatment, Vase and H&G got into a big academic fight about this, and it appears that the effect size for placebo effects in pain is approximately (d=0.5), which is a relatively large effect (especially within psychology). This doesn't entirely account for experimenter effects, though with the use of a balanced-placebo design, those can be accounted for.

In terms of clinical trials, I would tend to agree with your second definition. Its not perfect, but its as good as it tends to get.

I think that one of the issues with placebo research is this notion of mind-over-matter, in that such a viewpoint is the reason that it is perceived as special, and also a reason why people disbelieve in it.

Based on old research (Levine, 1979) many (but not all) placebo effects in pain appear to be mediated by endogenous opioids,which I would take to mean that they are pretty naturally mediated by the brain, and so its a physical phenomenon. Many people do go a bit crazy with the woo around it though, I do agree.

Funny that this came up today, when I'm currently finalising a hopefully final draft of my thesis on the placebo (PROTIP: never, ever leave your university before submitting a PhD, it tends to go badly).

[garyrob]

Are you posting your thesis online? I'd like to have a look.

[disgruntledphd2]

Its on Github, in a private repository. When I'm finished, I plan to make the repository public (which is probably going to be embarrassing).

[ronaldx]

Suppose that by "placebo effect", we really are identifying a marker for something like "experimenter effect" or "physician effect" and only on subjective end-points.

I believe it would still be correct, and of practical value, to say that the placebo effect is real. Prescribing someone a placebo results in them receiving the experimenter/physician effect - they subjectively report feeling better - that's a good thing.

[gus_massa]

The problem with "they subjectively report feeling better" is that sometimes the people "lie" voluntary or involuntary. Perhaps they don't feel better but want to make the experimenter happy. Perhaps they don't feel better but wish to feel better. Too much possible sources of confusion.

Just using an exaggeration for comparison, if you pay $1000000 to the experiment subject in one arm to say that they fell better, then you will get a very big improvement, but it doesn't mean that they really feel better. This is an exaggeration, but the problem is that there are a lot of more subtle things that can change the self reported feeling.

I don't know how to measure the well feeling in a non subjective way. If I may just made up an inexistent medical device, perhaps I can put a 24hs endorphin measurer to the test subject and look for a difference in the mean concentration.