If a drug trial reveals that patients receiving the drug did not get any healthier than those who took a placebo, is this a failure? It depends what the alternative treatment is. Implicitly its a failure if we believe that doctors will prescribe a placebo rather than the drug. Of course they don’t do that (often) but we can think of the placebo as representing the next-best alternative treatment.
But the right question is not whether the drug does better than the next-best alternative, but instead whether the drug plus the alternatives does better than just the alternatives. It could happen that the drug by itself does no better than placebo because the placebo effect is strong, but the drug offers an independent effect that is just as strong.
If so, then the right way to do placebo trials is to give one group a placebo and another group the placebo plus the drug being tested. The problem here is that the placebo group would know they are getting placebo which presumably diminishes its effect. So instead we use four groups: drug only, placebo only, drug plus placebo, two placebos.
Maybe this is done already.
Followup: Thanks to some great commenters I thought a little more about this. Here is another way to see the problem. Conceivably there may be a complementarity between the placebo effect (whatever causes it) and the physiological effect of the drug. The more you believe the drug will be effective the more effective it is. Standard placebo controls limit how much of this complementarity can be studied.
In particular, let p be the probability you think you are taking an effective drug. Your treatment can be summarized by your belief p and whether or not you get the drug. Standard placebo controls compare the treatment (p=0.5, yes) vs. (p=0.5, no). But what we really want to know is the comparison of (p=1, yes) and the next-best alternative. If there is a complementarity between the placebo effect and the physiological effect then (p=1, yes) is better than (p=0.5, yes).
Cheap Talk 
6 comments
Comments feed for this article
September 14, 2009 at 1:00 am
Alan
If the placebo effect is generated by simply taking a pill, any pill, doesn’t taking just the drug already act like taking drug plus placebo?
September 14, 2009 at 7:35 am
billpetti
Alan, that’s exactly right. The power of the placebo is that psychologically it triggers the body’s self-healing processes. Just thinking that you are receiving an effective treatment is enough in certain disease states.
If you think that there may be some additional ‘combination’ effect (meaning, our body’s heal at a faster rate if they think they are receiving multiple treatments) than I think the idea makes sense. Right now, trials are often designed with combinations of the new treatment and the existing treatment to see if the new compound has greater efficacy on its own, or if taken in combination with the existing treatment. It would be interesting to see if the combination of the compounds has a significant impact or if the increased efficacy is an artifact of the placebo.
Your general point about whether placebo studies should be the final word on the efficacy of a treatment is valid, and drug companies have been looking at how to better design clinical trials to take into account that while a treatment may not have a general level of efficacy across all types of patients, it may have a significant impact amongst a sub-population with specific genetic characteristics (i.e. the personal medicine movement).
September 14, 2009 at 8:21 am
Pete
I don’t see this as valid simply because taking the drug only is effectively drug+placebo. So regardless if I’m taking a sugar pill or a “real drug” I think I’m taking something that will make me better, and that thinking is all that is required to get a placebo effect.
September 14, 2009 at 3:48 pm
Paul
billpetti – I think it would be hard for you to substantiate your claim that placebos work like that. Placebo controls exist to isolate variables related to the efficacy of the drug from unrelated variables. These include reporting effects and other statistical artifacts that don’t correspond to a real positive effect from treatment. One example that requires you not to be ANY better but still show a statistically significant “placebo effect” would be biased self-reporting. So imagine you are in the same pain today that you were in yesterday, but now you have taken a drug that, supposedly, makes you feel less pain. If you don’t have perfect recall about how much pain you were in yesterday, you will probably say that the pain was x +/- s. If you think the pill should be working, you might give more credence to a higher estimate of your pain for the previous day, in which case anyone asking you, “Do you feel less pain than yesterday” would get the erroneous answer “Yes” simply because you are mistaken about yesterday’s pain level. If effects like this drive placebo effects, you would expect that the more difficult something is to measure objectively, the larger the “placebo effect”, even in the absence of efficacy. I think this tracks with real studies (from what I’ve heard doctors say), but I don’t have a reference.
Here’s an article that goes into more detail about the nature of the placebo effect and why it likely has nothing to do with “triggering your body’s natural healing”: http://www.sciencebasedmedicine.org/?p=1248
I also wrote about the subject here: http://diversified.selocsg.com/?p=236
September 14, 2009 at 8:37 pm
jeff
Thanks for all your great comments. You led me to think more about this and I have added a followup to the post with some further thoughts.
September 15, 2009 at 2:27 am
Alan
While from a purely statistical point of view I’d agree with you. However, looking at the morals of achieving p=1 would suggest this isn’t terribly practical. You’d have to tell people they ARE getting the drug, when in fact there is a chance they won’t be.
One question, what is more important, the (perceived) probability of receiving the drug, or the strength of belief in the drug’s effects? Do the complementarities arise from simple belief in taking the drug, or from the belief the drug will be effective?
One impact of this question is in the design of the ‘perfect’ study. If your p is important than telling everyone they will be receiving the drug is all that matters. If instead belief in the effects of the drug matters then all that matters is trying to convince people this drug will work, and studies can still tell people they have p=.5 chance of getting the drug.