Thread by @ColinWestMDPhD

1/ The colonoscopy trial reported in @NEJM is being widely reported as suggesting that colonoscopy is not effective for colon cancer screening. This is not the correct interpretation of these results. #EBM #evidencebasedmedicine #epitwitter
nejm.org/doi/full/10.10…

2/ This study’s results do not demonstrate that colonoscopy is effective, but also do not demonstrate that it is ineffective. Why? Because only 42% of those in the screening arm actually had a colonoscopy!

3/ What *does* this study show? Only that a screening program involving an invitation to get a one-time colonoscopy modestly reduced colon cancer cases but did not reduce colon cancer deaths or overall mortality at a median of 10 years of follow-up.

4/ So this is a trial of the *invitation* to get a colonoscopy, NOT the colonoscopy itself. What intervention study would be accepted with an actual receipt of the intervention by only 42% of those randomized to that arm of the trial?

5/ Let’s consider a few ludicrous examples to make this point. These are intended to be absurd, so don’t @ me if you miss the point. First, I might report that smoking doesn’t cause lung cancer – cough cough, are you sure, you might say?

6/ What if I randomized people to smoke or not (yeah yeah), and then followed them to see if they developed lung cancer? The lung cancer rate was 0.56% in the smoking group and 0.06% in the non-smoking group, p = NS. But only 42% of those in the smoking group actually smoked.

7/ Does this provide evidence that smoking doesn’t meaningfully increase lung cancer rates? Of course not, and I hope that’s obvious! More than half of the smoking group DIDN’T ACTUALLY SMOKE, so the study doesn’t assess the effect of smoking.

8/ Second, I might report that super-loud rock concerts don’t affect hearing – say what now, you might say (literally)?

9/ What if I randomized people to attend rock concerts or not, and then followed them to see if they developed hearing loss? The hearing loss rate was 3% in the concert group and 2% in the control group, p = NS. But only 42% of those in the concert group actually went to concerts

10/ Does this provide evidence that rock concerts don’t cause hearing loss? Of course not, and I hope again that’s obvious! More than half of the rock concert group DIDN’T ACTUALLY GO TO A CONCERT, so the study doesn’t assess the effect of attending rock concerts.

11/ Third, let’s get a little more risque. I might report that sex doesn’t cause pregnancy – come again, you might say?

12/ What if I randomized people to have sex or not, and then followed them to see if they became pregnant? The pregnancy rate was 2% in the sex group and 0.5% in the no-sex group, p = NS. But only 42% of those in the sex group actually had sex.

13/ Does this provide evidence that sex doesn’t cause pregnancy? Of course not, and I hope for the final time that’s obvious! More than half of the sex group DIDN’T ACTUALLY HAVE SEX, so the study doesn’t assess the effect of having sex. Anti-climactic, huh?

14/ Back to the current study: this trial doesn’t tell us if colonoscopy is effective as a screening tool for colon cancer, especially to reduce mortality, ONE WAY OR THE OTHER! There is actually some signal of benefit in my assessment, but this is far from definitive.

15/ We still need better data to establish colonoscopy’s effect, unfortunately, but in the meantime I hope this trial does not push people away from colonoscopy because it really should have little to no effect on attitudes toward colonoscopy as a screening tool. /fin