Mammograms may overdiagnose far less often than the long-quoted 30 to 50 percent

A reanalysis of all eight major randomized mammography trials concludes that their data are consistent with an overdiagnosis rate below 5%, rather than the estimates approaching 30% to 50% that have circulated for years. The paper, in the Journal of the National Cancer Institute, does not run a new trial. It argues that the old numbers were produced by how the trials were counted and followed up.

Sisse Helle Njor of the University of Southern Denmark and Lillebælt Hospital led the work with Casper Urth Pedersen of the same university, Elsebeth Lynge of the University of Copenhagen, Robert A. Smith and Matejka Rebolj of Queen Mary University of London. Pedersen’s contribution was supported by the Novo Nordisk Foundation and Rebolj’s by Cancer Research UK.

Three distortions in the older estimates

Overdiagnosis means detecting cancers that would never have caused symptoms or death. The National Cancer Institute’s breast screening summary gives a wide range, 20% to 50% of screen-detected cancers depending on age, life expectancy and tumor type, and notes that the available methods, long-term follow-up of randomized trials and excess incidence in screening programs, are imperfect. The reanalysis targets the first method, which supplied many of the highest figures.

According to the University of Southern Denmark’s release, the authors identified three factors that distorted earlier calculations: whether women in control groups were offered screening once the trials ended, whether the screened and control groups had the same number of screening rounds, and how long each group was followed. Each can inflate the apparent excess of cancers in the screened arm without any real overdiagnosis behind it. A control group that is screened after the trial closes, for example, catches up on cancers it would otherwise have recorded later, and a follow-up window that stops too early captures the surge of earlier diagnoses in screened women while missing the later dip when that group would otherwise have been diagnosed. Both errors push the apparent excess upward, and the reanalysis treats control arms that received exit screening as screened populations, so that the comparison matches what actually happened to them.

The eight trials and a Danish yardstick

The trials reanalysed were the New York Health Insurance Plan, Malmo, Two-County, Edinburgh, Canadian National Breast Screening Study, Stockholm, Gothenburg and UK Age studies. Instead of comparing each against its own control group alone, the team measured them against the real-world pattern of a population-based screening program in Funen, Denmark, as described on the JNCI abstract page. Denmark offered a natural experiment, because organized screening began in some regions 17 years earlier than in others.

The 17-year head start matters. It gives a long record of what cancer incidence does when screening starts and what it does afterwards, including the compensatory drop in incidence that follows once screening rounds end. Njor’s group compared that curve with what each trial produced, across 52 timepoints. Most confidence intervals overlapped the Danish expectation, meaning the observed numbers of extra cancers in the screened arms matched what organized screening in Denmark would have produced without invoking large-scale overdiagnosis, and the abstract says the trials “yield consistent excess-incidence patterns compatible with very low overdiagnosis (<5%)."

The statistical argument and what remains contested

The stakes of the number are practical. The National Cancer Institute’s summary stresses that no definitive method distinguishes life-threatening cancers from insignificant ones at the moment of detection, so surgery, radiation or chemotherapy is applied to all of them. Every percentage point of overdiagnosis is therefore a group of women treated for a cancer that would never have harmed them.

Lynge put the central intuition in one sentence in the release: when screening is introduced, diagnoses rise at first because cancers are found earlier, and unless that timing pattern is accounted for, researchers can mistake earlier detection for overdiagnosis. The authors add that some previous high estimates “were based on evidence before trial data had fully matured,” so the excess seen at the end of early follow-up had not yet been offset by the later dip.

The conclusion is conditional. The data are described as “consistent with” and “compatible with” an overdiagnosis rate below 5%, which is why the verdict is “may” and not “does not.” The estimate also leans on a single Danish program as the yardstick, and the abstract reports that earlier published estimates ranged from about 0% to 50%, so the reanalysis resolves a methodological dispute without ending the debate over how much unnecessary treatment screening causes in practice.

Njor framed the implication for patients in the release, as carried by ScienceDaily: women can now be reassured that the benefits of detecting breast cancer early and preventing premature death outweigh the small risk of unnecessary treatment. That is a reading of the benefit-harm balance, and it depends on the below-5% estimate surviving scrutiny from researchers who have published far higher ones. The JNCI paper, DOI 10.1093/jnci/djag302, is the document those critics will now have to answer.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

More from Morning Overview