The actual study is (after some URL sanitization):<p><a href="https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2808648" rel="nofollow noreferrer">https://jamanetwork.com/journals/jamainternalmedicine/fullar...</a><p>and I do <i>not</i> like it. It suffers from what I consider an extremely common problem in statistics: if you define the question poorly, then your output is garbage no matter how fancy or careful your analysis is.<p>We can start with the beginning of the abstract:<p>> Importance Cancer screening tests are promoted to save life by increasing longevity, but it is unknown whether people will live longer with commonly used cancer screening tests.<p>I have never heard of a doctor suggesting a cancer screening by saying "this might save your life by increasing your longevity." What does that even mean?<p>So let's try to figure it out. The paper uses the terms "lifetime" and "longevity" somewhat interchangeably, and it does not define either term. The best I can figure out is that, <i>for an individual deceased person</i>, they have a certain lifetime in days from when they were screened to when they because deceased. (Or a certain lifetime in days from birth to death, and I'm not sure this distinction matters.)<p>Great, but this is only for one patient. What about for a sample of patients or for a population in general or for the probability distribution of lifetimes of a given patient conditioned on whether they do or do not get screened? The article <i>does not say</i>, and a single "lifetime" number is not a probability distribution. Is it an expected value or a mean? A median? A mode? No comment.<p>One of the headline conclusions is:<p>> Based on the observed relative risks for all-cause mortality and the reported follow-up time in the trials, the only screening test that significantly increased longevity was sigmoidoscopy, by 110 days (95% CI, 0-274 days) (Table 2, Figure 2)<p>Figure 2 is useless. Table 2 is somewhat informative, and it has a column for relative risk of all-cause mortality and a column for lifetime gained and its 95% CI. But WAIT A MOMENT! The only way you can know the lifetime of an individual patient is if they're dead. If they're dead, their risk of all-cause mortality by the time they died is 100%. That's not 100% plus or minus something with some relative risk thrown in -- they are dead enough to have a date of death so that someone could compute their lifetime! Or maybe "lifetime" means something else, and the authors didn't bother to figure it out and say what they meant. So what exactly is this paper even analyzing?<p>So I suspect this is a meta-analysis of studies, of which some may or may not have been high enough quality to define their terms, and probably several of which used "lifetime" to mean some estimated property of a distribution, and this meta-analysis completely failed to figure out what the included studies were talking about.<p>So I rate this meta-analysis as almost entirely useless, on account of it failing to actually analyze anything that makes sense.<p>So I don't think any conclusions can be drawn. Although... ACS puts the lifetime risk of colorectal cancer at 1:25 or so. So one might naively translate a 110-day lifetime extension for everyone to a 110 day · 25 = 2750 day = ~7.5 year expected lifetime extension for people who actually get colorectal cancer. Sign me up -- 7.5 more expected years of life and presumably more than that of quality life years in the event I contract a not-particularly-rare disease sounds like a pretty good deal. (Colorectal cancer screening is not all that unpleasant, and I apparently only have a 96% chance of the screening being unnecessary.)