Should we use total mortality rather than cancer specific mortality to judge cancer screening programmes? No
BMJ 2011; 343 doi: https://doi.org/10.1136/bmj.d6397 (Published 13 October 2011) Cite this as: BMJ 2011;343:d6397
- Correspondence to: R J C Steele, Surgery and Molecular Oncology, Ninewells Hospital and Medical School Dundee, Dundee DD1 9SY, UK r.j.c.steele{at}dundee.ac.uk
All medical interventions have the potential to cause harm. This is particularly important in the case of cancer screening because the intervention is offered to people who are, or at least believe themselves to be, in good health, and the tolerance limit of harm must accordingly be low.
Screening may cause harm in several ways.1 If the screening test is not highly sensitive, false negative results may induce reassurance and create a “certificate of health effect.” In other words, people who have received a false negative test result may ignore symptoms or continue to engage in risky behaviour. Then the test itself can cause harm—for example, colonoscopy as a consequence of colorectal screening may lead to colonic perforation or other complications, and this must be monitored. Given that most people who are screened will not have disease, unnecessary psychological morbidity may also be created. Finally, screening inevitably leads to a degree of overdiagnosis —that is, people will be found to have disease that was not destined to become symptomatic in their lifetime. If they are harmed by the treatment of that disease, they will have been disadvantaged by participating in screening. Indeed, if a patient dies as a result of the treatment of screen detected disease, their life will have been shortened by screening.
The question of how we should judge screening programmes is therefore extremely important. We fully concur that any screening programme that causes a demonstrable increase in total mortality, regardless of its effect on cancer specific mortality, is unsustainable. However, stopping a screening programme that does not show a decrease in total mortality is not justifiable.
Because of the biases inherent in screening, the only robust method of proving efficacy is by population based randomised controlled trials. If such trials show that screening reduces disease specific mortality, we can be sure that early detection has a true effect on the natural course of the disease and that the effect is not solely due to lead time or self selection by a particularly healthy population. Randomised trials carried out for both breast2 3 4 5 6 7 and colorectal cancer8 9 10 11 screening have consistently shown reductions in disease specific mortality in the region of 20%.
Too stringent
Demonstrating a reduction in all cause or total mortality, however, is a different matter. As even common cancers account for only a small proportion of total deaths (for example, in the United Kingdom, colorectal cancer accounts for 3% of all deaths), to show a reduction in disease specific mortality being translated into a reduction in total mortality would require trials that are too large to be feasible. Furthermore, it is inappropriate to use disease specific mortality as a surrogate for all cause mortality; cancer screening is not designed to reduce all cause mortality but the number of people dying prematurely, in a particularly unpleasant manner.
Proponents of using all cause mortality as an outcome indicator argue that it avoids the bias inherent in the determination of causes of death. In a frequently quoted paper that examined all cause mortality in randomised trials of cancer screening,12 the point was made that the effect on all cause mortality was often in the opposite direction from the effect on disease specific mortality. However, close examination of the data shows that the confidence intervals around the differences in all cause mortality figures were much wider than those around the disease specific figures and did not reach anywhere near statistical significance, with the exception of a beneficial effect on all cause mortality in the Edinburgh mammography trial. In the correspondence that followed this article, even the original authors concede that a significant reduction in all cause mortality is too stringent a requirement for the determination of the efficacy of screening.13
Thus, if a trial shows a reduction in disease specific mortality, even though it has no demonstrable effect on total mortality, it has provided sufficient evidence to offer such screening to the population that has been studied in the trial. To insist that a trial should show a reduction in all cause mortality would deny society the opportunity to engage in screening that, on balance, is more likely to prevent cancer death than cause harm. On the other hand, it is reasonable to insist that potential participants are provided with adequate information and that risks are expressed in absolute terms, with the proviso that such information is understandable to the majority of the population. The Scottish bowel screening programme leaflet, for example, states explicitly that one bowel cancer death is prevented for roughly every 650 people invited for regular screening14; even so, uptake is currently 53%.15
We emphasise again that it is both appropriate and necessary for the effect of screening on all cause mortality to be assessed and that any screening programme that has a significant adverse effect on this measure must not be supported. However, it is unreasonable to single out screening, which is an intervention with a specific aim, as having to prove a reduction in all cause mortality at a population level. If all medical interventions were similarly constrained, then much of what health professionals do would be deemed inappropriate.
Notes
Cite this as: BMJ 2011;343:d6397
Footnotes
Competing interests: Both authors have completed the ICJME unified disclosure form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare no support from any organisation for the submitted work; no financial relationships with any organisation that might have an interest in the submitted work in the previous three years; and no other relationships or activities that could appear to have influenced the submitted work.