Should we expect all-cause mortality reductions in large screening studies?

Abstract

Randomised controlled trials of cancer screening are generally designed with disease-specific mortality (DSM) as the primary outcome. All-cause mortality (ACM) is often not reported, or reported only as a secondary outcome.1,2 Implicit in the choice of DSM as the primary outcome is that the screening intervention will not have an adverse effect on other causes of death, or at most that effect will be small in comparison with the DSM benefit. For example, the radiation exposure from mammography likely causes some cancers, but this number is small in comparison with the number of breast cancer deaths prevented.3 Harms due to overdiagnosis and overtreatment are more difficult to estimate, and may be substantial.4–6 Commentators have therefore argued that ACM is the preferred outcome for cancer screening trials, because DSM is a biased outcome due to incorrect assignment of the cause of death and failure to fully account for harms.7,8 If DSM decreases under screening, then there are three possible relationships between DSM and ACM in a cancer screening trial. First, both DSM and ACM may decrease by approximately the same absolute number of deaths, suggesting no important harms of screening. Second, DSM may decrease but ACM does not change significantly, suggesting that any benefits of screening are offset by harms. Third, ACM may increase while DSM decreases, suggesting that unintended harms of screening are greater than the benefits. An important limitation of determining which of these patterns has occurred in an individual screening trial is the difference in sample sizes needed to demonstrate a reduction in DSM as opposed to ACM. In a recent randomised controlled trial of screening for ovarian cancer (the United Kingdom Controlled Trial Ovarian Cancer Screening or UKCTOCS), a post-hoc analysis concluded that screening reduced DSM.1 …