Chances are, you’ve heard about the ‘NEJM colonoscopy trial’ published in October 2022 that had the internet all abuzz. In the same year that the USPSTF moved up colorectal cancer screening to age 45, this trial asserts that screening colonoscopies don’t make as much of a difference as we thought. [Enter the frantic headlines.]

Considering that colorectal cancer screening programs are operating based on presumed benefit demonstrated only by observational data, an RCT studying the “effect of colonoscopy screening on risks of colorectal cancer and related death” would be a landmark trial (and an easy acceptance into NEJM). The thing is, they didn’t exactly study what the title suggests.

Nearly 85,000 adults aged 55-64 in Europe were randomly assigned to receive either an invitation to undergo a single screening colonoscopy or no invitation (and no screening). (No colorectal cancer screening programs were in place at the trial outset in 2004, so this wasn’t unethical.) After ten years of follow up, the invited group saw no benefit in colorectal cancer-related mortality and only an 18 percent relative risk reduction in colorectal cancer incidence (0.98% vs 1.20%, risk ratio [RR] 0.82; 95% CI 0.70 to 0.93). But here’s what has everyone talking: only 42 percent of participants in the invited group actually got colonoscopies.

Statistically speaking, an intention-to-treat (ITT) analysis is overwhelmingly preferred in superiority trials like this one. That means all participant data are analyzed according to the group to which they were originally assigned, regardless of whether the individual got the assigned intervention, switched interventions, dropped out, etc. In this case, however, the ITT dramatically underestimates the effect of screening because so many people who were analyzed as being in the intervention group didn’t get the true intervention of interest (the colonoscopy). Interestingly, however, because of the compliance effect, the 42 percent who actually chose to get colonoscopies are statistically more likely to follow other health-related advice and thereby have a lower baseline risk of developing colorectal cancer compared to a population who doesn’t accept an invitation to screen. They are therefore less likely to realize the potential benefit that a screening colonoscopy would offer an average risk person with more precancerous polyps to find.

Here’s what has everyone talking: only 42 percent of participants in the invited group actually got colonoscopies.

In an effort to address this compliance issue, the investigators performed adjusted per-protocol analyses to estimate the effect if everyone in the invited group had gotten a colonoscopy. (Recall, the population who got colonoscopies from which they are extrapolating data are statistically not at average risk of colorectal cancer.) These post-hoc analyses demonstrated more substantial reductions in colorectal cancer incidence, and importantly, cancer-related death. However, the “50 percent reduction in mortality” you may have read about really comes down to an absolute risk reduction of 0.15 percent, from 0.30 percent without screening to 0.15 percent with colonoscopy screening. Ah, beware the risk of relative risk.

There are other major problems with this study that we could get into, such as 1) generalizability — colorectal cancer incidence is highly associated with geographic location; 2) inadequate length of follow-up; and 3) interval improvements in detection rates due to better technology and operator training. Suffice it to say that this was less a landmark study of the effects of colonoscopy screening and more a study of the efficacy of population screening programs. Regardless, the presumed benefit of colorectal cancer screening in general, and with colonoscopy specifically, remains intact, even if we didn’t end up with the landmark trial we were expecting.

Practice Point: Don’t trust the headlines. Colon cancer screening is still a good thing.

EBM Pearl: To best study the effects of an intervention, one should start by intervening. Low compliance yields low validity.

For more information, see the topic Colonoscopy for Colorectal Cancer Screening in DynaMed.