Should we use total mortality rather than cancer specific mortality to judge cancer screening programmes? YesBMJ 2011; 343 doi: http://dx.doi.org/10.1136/bmj.d6395 (Published 13 October 2011) Cite this as: BMJ 2011;343:d6395
Cancer screening is a source of much dispute—in the case of breast cancer, arguments have raged for more than a decade.1 One major concern is how the effects are to be measured. Disease specific mortality is used extensively in trials of cancer screening,2 3and as the aim of screening is to reduce deaths from the target disease,2 4 this might seem to be a suitable end point. But the arguments against using disease specific mortality weigh heavily, and all cause mortality is a better measure.
Uncertainties relating to cause of death
Clearly, the accuracy of disease specific mortality depends on correctly identifying the cause of death. However, this is often unreliable,2 and it entails decisions that can introduce biases, either for or against screening.2 3 Claims that bias favouring screening predominates have been disputed.2 4 Nevertheless, one thing is sure: the accuracy of all cause mortality depends solely on the number of deaths identified and is not subject to bias. It is therefore a more reliable end point.
Disease specific mortality also ignores the fact that screening for cancer causes harm. Invasive procedures may have fatal complications, while overdiagnosis—that is, the identification and treatment of tumours that otherwise would have caused no disease—may also result in death.1 2 3 A review of 12 trials of screening for breast, lung, and bowel cancer raised doubts about both the identification of screening related deaths and their inclusion in disease specific mortality.2 If screening related deaths are not included in the mortality figures, the results will be skewed in favour of screening. In contrast, all cause mortality balances the benefits and harms of screening in a single measure.
Although disease specific mortality is unproblematic when used descriptively, difficulties arise when it is used as an end point in randomised controlled trials. Should the figure include death occurring in a case of overdiagnosis? And what about someone who does not have colorectal cancer but who dies from a perforation due to screening colonoscopy? Although such deaths are not strictly linked with disease specific mortality, they are obviously relevant.
We can accommodate all screening related deaths in a randomised trial only by changing disease specific mortality into a vague and arbitrary end point. Alternatively, we could use all cause mortality which is untouched by these problems.
The unfeasibility argument
Advocates of disease specific mortality have a fall-back position.4 The target cancer, they argue, contributes little to total mortality; trials would have to recruit millions of people to show a statistically significant reduction in all cause mortality; such trials are not feasible; hence, we have to rely on disease specific mortality.
This argument does not show that disease specific mortality is better than all cause mortality; indeed, it seems to concede the opposite point. It also assumes that huge trials would show a reduction in all cause mortality, whereas this is precisely what is in question. And it ignores the existing data that strongly support an absence of any effect of screening on all cause mortality, as, for example, in the case of bowel cancer screening.
Specific case against disease specific mortality
The NHS Bowel Cancer Screening Programme was implemented on the basis of four large randomised trials.5 Meta-analysis showed a reduction in disease specific mortality of 15% in the screening group compared with controls (odds ratio 0.85, 95% confidence interval 0.78 to 0.92).5 The odds ratio for all cause mortality, however, was 1.0 (0.99 to 1.02). Thus, in more than 300 000 people included in the four trials, there was no difference in survival5 6; nor was there anything to indicate that a larger trial would be worthwhile.
Given that the absolute reduction in disease specific mortality was only 0.1% over 10 years,7 it would require an enormous trial to detect a difference in all cause mortality between the screening group and controls. But is this really necessary? Taking the above criticisms of disease specific mortality into account and considering the robust nature of all cause mortality in this example, surely we should accept that screening for colorectal cancer has no effect on overall survival?
All cause mortality is a hard end point that is free from bias, produces a robust estimate of the effect, and answers the crucial question of whether cancer screening improves overall survival. In contrast, disease specific mortality requires decisions which introduce bias and fails to deal effectively with deaths from screening. Unsurprisingly, the two measures often support opposing conclusions.2
Disease specific mortality is used in cancer screening trials primarily because it allows the identification of very small reductions in mortality from the target disease. Without it, there would be nothing to support current cancer screening programmes. But this is no reason to accept flawed data. On the contrary, we should prefer the evidence of all cause mortality, recognise that bowel and breast cancer screening do not improve overall survival, and question whether these programmes should continue.
Cite this as: BMJ 2011;343:d6395
Competing interests: The author has completed the ICJME unified disclosure form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declares 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.
Provenance and peer review: Commissioned; not externally peer reviewed.