VI
Mark học AIThe Placebo Effect · Henry Beecher 1955, Ted Kaptchuk 2008, Hróbjartsson and Gøtzsche 2001–2010
Does a pill with no active ingredient make people healthier? Today's answer: yes, but far less than the legend says, mostly in what patients feel themselves (pain, discomfort), and most of the "effect" seen in placebo groups is really people getting better on their own.
The best-known "take it apart" trial: Kaptchuk and colleagues, Harvard, 2008. 262 people with irritable bowel syndrome (Kaptchuk et al., 2008).
10 things often said about placebos, checked against the original studies and reviews. Tap each line to see why.
True, but that is the "response", which includes natural recovery and regression to the mean. In Kaptchuk's trial, even the group that only waited had 28% reporting relief (Kaptchuk et al., 2008).
A review of 114 trials comparing placebo with no treatment: a small benefit on continuous outcomes (standardised mean difference −0.28), concentrated in pain and self-reported outcomes (Hróbjartsson and Gøtzsche, 2001). Naloxone removed placebo pain relief in responders, pointing to a biological pathway (Levine et al., 1978).
SAMSON trial: symptom score 15.4 in dummy-pill months, 16.3 on statins, 8.0 with no pills; nocebo ratio about 90% (Wood et al., 2020; Howard et al., 2021). Only 60 people, but double-blind and each person compared with themselves.
Sham surgery for knee osteoarthritis: arthroscopy relieved pain no better than sham at 1 and 2 years (Moseley et al., 2002, 180 people). Without the sham group, the operation looked very effective.
One trial, 262 people, one condition (irritable bowel), self-reported outcomes, and participants cannot be blind to how they are treated. The direction is plausible; the size needs more studies (Kaptchuk et al., 2008).
A short research letter, 82 healthy people, lab pain from electric shocks (Waber et al., 2008). The lead author called it a small, simplified experiment. Interesting, not yet something to apply to real patients.
Open-label placebo for 80 people with irritable bowel: about 60% reported adequate relief versus about 35–37% with no treatment (Kaptchuk et al., 2010; figures via secondary sources). Small sample, self-reports, no blinding possible, only 3 weeks of follow-up.
84 neuropathic-pain drug trials, 1990–2013: placebo responses rose, but only in US trials, which also grew bigger and longer (Tuttle et al., 2015). It may be about how trials are run rather than placebos getting "stronger".
Beecher's (1955) average of 35.2% came from 15 studies and 1,082 people. Kienle and Kiene (1997) re-read each study and found no evidence of a placebo effect in any of them: the improvement came from natural recovery, regression to the mean, parallel treatments and misquotation.
For "recovered or not" (binary) outcomes, placebo had no significant effect, whether reported by patients or measured objectively (Hróbjartsson and Gøtzsche, 2001: relative risk 0.95, confidence interval 0.88–1.02). The 2010 Cochrane review of 234 trials concluded that placebos have no important clinical effects in general.
Robustness 6/10. Placebos do not cure disease, but they change how people feel, especially pain. The bigger lesson lies elsewhere: much of the "improvement" we see would have happened anyway.
The often-told story: in World War II, out of morphine, a nurse injected a wounded soldier with salt water, and he still felt relief. Beecher is said to have seen this and turned to studying placebos.
Some critics say the story is hard to verify: Mark could not find Beecher's own original account of it, only retellings. Treat it as an anecdote.
What is certain is the 1955 JAMA paper "The Powerful Placebo", where the 35% figure was born and then quoted for decades.
Work skill: before praising an initiative, a tool or a training course, ask "compared with what?", and use the "warmth" part deliberately.
| Situation | Easy wrong conclusion | Check it like a trial |
|---|---|---|
| A new training course | "After the course everyone says they feel more confident. It works!" | Compare two months of errors or sales for the trained team versus an untrained one. |
| Changing a process after a bad month | "March had record errors, we changed the process in April, errors halved." | Look at the months before March: April may simply be back to normal. |
| Announcing a big change | Listing only risks: "it may be slow, may break, may lose data". | Give risks with real odds and what has been prepared, so worry doesn't create its own "symptoms". |
Every number in this article, the Facebook carousel and the captions comes from the sources below. Mark's working network blocks journal websites, so figures were checked against paper abstracts (PubMed, journal pages) and science news; anything seen only in secondary sources is labelled.