The screening that found too much

A screening programme detected three times as many cases of a cancer as had been diagnosed before it began. Deaths from that cancer did not fall.

The explanation is overdiagnosis, and it is the hardest idea in screening to convey. The programme was finding real cancers, confirmed by pathology, in people who had them. What it was also finding were cancers that would never have caused symptoms in that person's lifetime, because some tumours grow so slowly, or regress, that the person dies of something else first. Before screening, those cases were never counted. After screening, they are counted, treated, and recorded as survivors, which makes survival statistics improve while mortality does not move.

The statistical signature is specific and worth knowing. Overdiagnosis produces a rise in incidence with no fall in mortality, and no corresponding fall in the rate of advanced disease. If a programme were catching dangerous cancers early, late-stage diagnoses would fall as early-stage ones rose. When early-stage diagnoses rise and late-stage ones stay flat, the extra cases are the kind that were not going to progress.

None of this means screening does not work; several programmes show exactly the pattern that indicates benefit. It means the benefit has to be demonstrated in mortality and in stage distribution rather than in survival rates, which are the figures most often quoted because they are the most flattering.

The practical consequence for a person offered a screening test is a question that clinicians increasingly welcome: for every thousand people screened, how many avoid a death, how many are treated for something that would never have harmed them, and how many have a false alarm. Good programmes publish all three.