The findings mean that it was not really an RCT of the additive impact or benefit of PSA screening, since PSA was 'de facto' being screened in both arms. This shows one of the hazards of a rigorous "intention to treat" analysis when the control group actually veers to pragmatic treatment very similar to the treatment group.
[M]ore than 80% of the participants in the control group without baseline screening contamination (which for PSA wasdefined as ≥2 tests within 3 years before trial entry)reported having undergone at least 1 PSA test during the trial, with more than 50% undergoing testing within the past year and 70% within the past 2 years.
Overall, including the 10% of control participants with baseline PSA screening contamination, the proportion of control participants who reported having undergone at least1 PSA test before or during the trial was close to 90%.
___
To translate this to a more concrete example, imagine we could 200 persons who are 100 pounds overweight, and randomize them to bariatric surgery or a control group. Two years later, most people in both group lost 80 pounds. Then, you find out that almost all the people randomized on Day 1 to "control group" went and had bariatric surgery anyway. This is no longer a trial of the impact on medical health outcomes of bariatric surgery betweeen the two different arms -- it has degenerated to a study of what happens with human behavior and human health choices after you have set up an RCT for the particular purpose and design.
Intention to treat analyses can be very misleading in this situation. As we see in the NEJM letter today. If you randomize men in the 1990s and early 2000s to PSA screening or not, and, if most of the men in both arms get PSA screening, it became a study of human behavior under the study conditions, but not a study of the impact of PSA screening on cancer.
__
CMS proposed a PSA quality measure that violated even its own NCD on (for cause) PSA testing; original November 2015 blog here, follow-up February 2016 blog here.
