It depends on who’s asking the question.
“Current evidence does not substantiate the claim that common cancer screening tests save lives.” As scientific mic drops go, it’s a doozy.
The Australian government alone spends more than $400 million a year on screening programs for just three cancers – breast, bowel and cervical – and is about to add a fourth – lung. Globally, the investment in cancer screening likely runs into the hundreds of billions of dollars.
Yet a systematic review and meta-analysis published in JAMA Internal Medicine in August suggests that four commonly implemented screening programs for breast, bowel, prostate and lung cancer do not meaningfully shift the dial on longevity.
While controversial for a number of reasons, the paper exposes some tricky, almost philosophical conundrums at the heart of cancer screening as a public health intervention.
Clinical epidemiologist Associate Professor Katy Bell describes the analysis as powerful. “It refocused people’s attention onto what actually should be the goal of cancer screening or any population screening program, which is to either improve quality of life or quantity of life and preferably both,” says Professor Bell, from the University of Sydney.
The meta-analysis includes 18 randomised controlled trials of cancer screening programs, involving a total of more than 2.1 million participants, with at least nine years of follow-up including all-cause mortality and estimated lifetime gained. The trials studied are four FOBT and four sigmoidoscopy studies for colorectal cancer, one colonoscopy trial, four PSA screening trials, three on CT lung cancer screening in current and former smokers, and two mammography trials.
The analysis concludes that none of the screening trials – with the slight exception of the sigmoidoscopy study – show improvements in all-cause mortality. The sigmoidoscopy study barely tiptoes over the line of statistical significance. It is also the only one to show statistically significant lifetime gained, while the others either show no gains or have wildly large confidence intervals.
The choice of all-cause mortality as the primary outcome for the analysis isn’t very conventional, because most screening trials are not, and cannot be, powered to examine this, says epidemiologist and population health researcher Professor Karen Canfell, director of the Daffodil Centre, at the University of Sydney and Cancer Council NSW.
The authors of the analysis acknowledge that no individual cancer screening trial has yet been big enough to show an effect on all-cause mortality. But they nonetheless argue that cancer screening needs to be assessed for its impact on all-cause mortality in those who undergo screening, because the screening – and potential subsequent treatments in the event of a positive result – can have negative effects on all-cause mortality.
“A cancer screening test may reduce cancer-specific mortality but fail to increase longevity if the harms for some individuals outweigh the benefits for others or if cancer-specific deaths are replaced by deaths from competing causes,” the authors wrote.
Because the impact on all-cause mortality is averaged out across all the individuals randomised to screening, that incorporates both the winners and the losers of screening.
It also includes individuals who, for example, might have their lung cancer detected early, but succumb to smoking-related heart disease, or those whose bowel cancer is prevented, but die from obesity-related health conditions.
“A mortality shift from cancer to other causes of death without increased length of life is thus plausible,” they wrote.
But Professor Canfell says focusing only on all-cause mortality misses the strong evidence of cancer-specific benefit shown both in clinical trials and observational studies.
“If you have a series of trials that are pointing towards a significant reduction in disease-specific, cause-specific mortality and/or incidence of that disease – so it’s been prevented, in the case of cervical and bowel cancer – then, yes, there’s an inference that that will save lives, and that inference can be tested in some of the real-world observational data,” she says.
For example, in Australia, mortality rates from cervical cancer have halved since the introduction of that screening program. Screening programs for bowel cancer reduce bowel cancer mortality by 21-41%, according to a meta-analysis that also showed no impact on all-cause mortality.
Cancer epidemiologist Professor Mark Jenkins, director of the Centre for Epidemiology & Biostatistics, School of Population & Global Health at the University of Melbourne, comments that it would be difficult to observe an improvement in all-cause mortality by preventing deaths from a single disease such as bowel cancer, as it is only responsible for around 3% of the approximately 190,000 deaths in Australia each year.
“The bowel cancer screening kit is only designed to prevent deaths from bowel cancer, not any other causes of death.” Almost all screening programs or public health measures only save a relatively small proportion of lives, but cumulatively, they have a substantial impact, he says. “It is only when their impacts are combined that they have a substantial reduction in all-cause mortality.”
However, Professor Bell argues that to invest the huge amount of money and resources required for a population-wide cancer screening program, “you’re really wanting to have, ideally, clear evidence that that there is benefit for the population”. And the meta-analysis approach was intended to achieve the statistical power needed to identify even a small shift in all-cause mortality across the screened population, she says.
The paper does include analysis of cancer-specific outcomes, which isn’t as resoundingly positive as might have been expected, at least for programs such as colorectal cancer. But here the answer may lie in another analytical choice made in the paper, to focus on the intention-to-treat outcomes rather than the per-protocol outcomes.
Gastroenterologist Professor Finlay Macrae, head of colorectal medicine and genetics at the Royal Melbourne Hospital, says this analytical choice reflects different philosophical attitudes towards the purpose of screening; “whether you’re interested in supporting the individual to get the best outcome for himself, regardless of anybody else in the population, or whether you’re interested in … supporting the total population to get the benefits for the total population”.
Professor Macrae says in the United States, the focus is on what can be gained for the individual – the person who actually turns up and gets screened – so there’s more interest in the per-protocol results; in Europe and Australia, the question is about how the broader population benefits from an intervention, including those who are eligible but don’t participate. “What we’re interested in is reducing the mortality in the whole population from introducing colonoscopy,” he says.
The authors of the analysis – who are European – say they chose intention-to-treat to get the most unbiased estimate for the outcomes associated with screening, even though this might underestimate its impact because of non-adherence.
Professor Paul Glasziou, director of the Institute for Evidence Based Healthcare at Bond University, supported the choice of intention-to-treat analysis to avoid the bias that comes with only analysing results from people who turn up for screening.
“If you take the HIP study – which was the first mammographic screening study – the women who got mammographic screening had half the rate of heart disease mortality,” he says. “From the public health point of view and for politicians, they’re interested in what actually happens with the population, not for an individual deciding whether they should be screened or not.”
However that does mean that screening for a cancer such as colorectal cancer, where the most effective option – colonoscopy – has lower adherence rates, is always going to struggle to make a meaningful impact at a population level. And it also means that the population-level benefits increase substantially with increased participation, Professor Jenkins says.
“We know that unfortunately, only 40% of people sent the bowel cancer kit return it, and that means that those 40% are protected in some way. But the 60% that don’t do it don’t benefit from this test,” he says.
According to the Daffodil Centre, increasing participation to 60% could save 84,000 lives by 2040.
Another notable absence from the analysis is any consideration of the impact of screening on the morbidity and quality of life issues associated with cancer. The authors say it’s difficult to measure and interpret in clinical trials.
But Professor Jenkins says this is a fundamental point of screening, especially in programs that can effectively prevent disease in the first place.
“If you detect bowel cancer early by screening, and the cancer is removed successfully – which can be done in over 90% of early-stage bowel cancers – there are no long-lasting health effects,” he said. “But if it’s detected late because screening was not done, and it’s already spread, it can result in lifelong hardship for the person.”
That’s part of the justification for introducing lung cancer screening in Australia by July 2025, which targets current and former smokers with two-yearly low-dose computed tomography screening.
Lung cancer is the leading cause of cancer death in Australia, says Dr Marianne Weber, senior research fellow at the Cancer Council NSW, but it’s also associated with enormous morbidity costs. “Most people (40%) are diagnosed with late-stage disease, when survival is poor. A large proportion – around a third – of people with lung cancer experience psychological distress and poor quality of life.”
For all the vociferous criticism of the paper, there is a good reason for its publication in such a high impact journal. Professor Glasziou says while it doesn’t argue for or justify stopping cancer screening programs, it does prompt important questions.
“It should be a warning alert that we have to continue to be cautious about introducing screening programs generally, particularly cancer screening programs,” he says. “You need evidence that there’s a benefit and it would be nice to have evidence that there’s an overall net benefit across the board.”