I made a glaring error. I described the results of Carlesso et al like so:
The authors found that increased neck pain is 25% more likely with SMT than if you did nothing or stuck to safe and neutral treatments.
No. Wrong. Bzzz! Thank you for playing.
This morning I received a note from Lisa Carlesso, PT, MSc, first author of the paper, letting me know that I got it wrong: although her data showed that 25% number, it was not a statistically significant number. And that’s significant.
If there were any noteworthy increases in neck pain following this kind of neck treatment, presumably clearer data would have emerged. Thus the paper concluded that there is “strong evidence that neck manipulation or mobilization does not result in an increase in neck pain.” I’m not sure if I quite agree that a statistically insignificant number constitutes “strong evidence” so much as just generally low confidence in the results (and Carlesso acknowledges this in the paper as well, practically in the next sentence: “However, the limitations of the Strunk study and the low GRADE rating remain, affecting confidence in the estimate.”)
But I am guilty of doing something I’ve accused others of doing: emphasizing a statistically insignificant number to make my point. When the numbers lean your way but fail to reach statistical significance, it’s called a “trend.” A trend in favour of a therapy is often trotted out as if it were supportive evidence. I did the opposite, and so I am doing the walk of shame now. Bad science writer, bad! I erroneously thought the number was statistically significant, probably because I’m a debunker by nature, and I basically saw in the data what I wanted to see. Funny how that works.
(Gosh, I wonder what system of knowledge-seeking could possibly compensate for that aspect of human nature? You get a gold star if you guessed “science.”)
What does it mean?
Not much, really. “Here be statistical dragons.” There are so many ways that all this stuff about treatment harms can be wrong that I can’t really walk away from this feeling like I’ve learned anything terribly important one way or the other. The hard statistical bottom line is that statistically insignificant means that no conclusion can actually be drawn — the data was no more than suggestive. Interestingly, Lisa Carlesso pointed out that she has written another paper about how difficult it is to study adverse effects. Indeed!