I’m really confused by this data. First of all, are they testing the efficacy of colonoscopy or the efficacy of inviting people to colonoscopy?
And then how is the former group’s reduction in deaths 50% and the latter group’s is about 0%?
This could also be true for the colonoscopy, that the more responsible/healthier people in the treatment group are more likely to get the colonoscopy and it's very possible responsibility/healthiness are driving the difference in health outcomes instead of the colonoscopy.
Basically in the treated group of 1103 of every 10,000 people died. And in the control group 1104 of every 10,000 people died.
So to summarize the study. Inviting someone to a colonoscopy reduces their risk of getting colon cancer by 22 basis points. Their risk of dying from colon cancer by 3 basis points, and their risk of dying of any cause by 1 basis point.
With the actual risk reduction being up to 5x this assuming it's a 20% difference in the rate of getting colonoscopies which is driving the difference.
But this makes metformin look good because it drives a much larger overall reduction in risk.
The measured intervention was not the colonoscopy, it was the invitation to screen. Only 42% of invited patients actually got a colonoscopy. This is far more persuasive to me:
> "When the investigators compared just the 42% of participants in the invited group who actually showed up for a colonoscopy to the control group, they saw about a 30% reduction in colon cancer risk and a 50% reduction in colon cancer death. “That adds to a bunch of observational study data that suggests exposing people to colonoscopy can reduce risk of developing and dying of colon cancer,” Gupta said."
As a member of the public, I don't really care about invitation to screen, but do care about the efficacy of colonoscopy. I can see invitation to screen being an important concern from a public health standpoint.
Inviting patients to undergo screening colonoscopy fails to reduce rate of cancer deaths
Imagine two cages filled with identical mice. One you drop some food into, and the other you don't. They starve to death at the same rate. Surprised?
What this appears to show is that you need to get a colonoscopy to avoid colon cancer; but that you don't need to get a colonoscopy to avoid death. I'd much rather avoid colon cancer entirely than have colon cancer and survive.
But as GP pointed out, maybe you need something else to avoid death: something that is correlated with responding to the invitation to get a colonoscopy. Maybe if you're willing and able to get a colonoscopy when invited, you're willing and able to more pro-actively go to the doctor when you notice other issues that are indicative of colon cancer, allowing you to get early treatment. And conversely, maybe if you're not willing or able to get a colonoscopy when invited, you're more likely to ignore symptoms until it's too late.
Again, avoiding colon cancer in the first place is better than successfully treating it; but it does point to the fact that other interventions might be more helpful in actually preventing deaths.
Yes. This is the argument against relying on the secondary analysis in this study. Although the invited and standard care groups were randomized such that differences in putative confounders were adjusted for, the rejection of the intervention itself may have reintroduced systematic differences that reduce the reliability of the hypothesis that intention to screen for colon cancer reduces mortality. Possibly those who accepted screening colonoscopy are more attentive to other health and lifestyle practices that reduce colon cancer mortality.
The study answers a question from health policy makers: Should we invite everyone (in some age group) for a colonoscopy? Based on this study, probably not.
It does not answer questions from individuals: Should I get a colonoscopy? If you have some good reason (symptoms, doctor advice, family history), probably yes (based on other studies, not this one).
If I then split by “had a colonoscopy between 2008 and 2012, inclusive” vs “didn’t” and look at 2012 through 2021 outcomes to draw conclusions, it’s possible that that filtering makes them unlike groups (I mean, it definitionally does in at least the primary selection criteria). Given that the effort is approximately that of a SQL query, I’d be interested to know if there’s a possible signal there, which would need to be corroborated with other data sets to determine the repeatability of the correlation and then if there’s any likely causal link.
To be fair, this is not two separate problems. Anyone who's willing to run the illegal study will not care whether they can get IRB approval.
The control group was 50% of the population who didn't get an invitation.
The experimental group was 50% of the population who got an invitation for colonoscopy, but turned out to be subdivided into ~40% "health conscious" who actually followed up on the procedure, and ~60% who ignored it.
Presumably there's a corresponding ~40% "health conscious" component of the control group, but this experiment had no method for identifying them.
If the study only looked at that ~40% subset of the experimental group, as opposed to the entire group who received invitations, then they could no longer compare them to the control group.
I'm not sure what gold standard you are referring to (or the article or the paper - https://www.nejm.org/doi/full/10.1056/NEJMoa2208375).
Double blind studies require there to be data. An invitation doesn't speak to the effects of Colonoscopy screening at all, while simultaneously adding a confounding variable about participation. The data is about the effects of offering screenings, not the effect of those screenings, per se.
Lifelong data is the gold standard for questions about mortality and most Colonoscopy randomized trials started around 2010 (hence this very early 10-year study, which I would say is premature).
Lifelong correlational data is not the gold standard for questions about mortality. It's intent to treat RCTs.
I believe that's what I said. That's certainly what was used. You can't compare the group subset that didn't participate, so it's a confounding variable.
> Lifelong correlational data is not the gold standard for questions about mortality.
AFAIK it is and has been over the last century. If you aren't tracking lifelong data, your mortality data is always skewed against hidden results because you didn't want to wait. When making a paper that isn't qualified (decade long effects vs effects), it's not expected to have short time-boxed data.
You're basicall saying "our randomization at the beginning of the trial is key to avoid biases, so we can't reassign people from the treatment group to the other group, even if they practically don't get the treatment". The reason is if you allow people to switch, your assignment is no longer random. People who avoid the treatment may have different health properties than the ones who don't.
In essence, you need your trial to be robust and large enough that a few people not getting the treatment don't matter.
Also I’m not sure if the article mentioned this, but the data seems to imply that the % of people who opted to get a colonoscopy was similar in the invited and control group.