specificity

A test that looks for a disease can fail in two directions, and specificity measures the second one. It is the proportion of people who do not have the disease who correctly test negative. The National Cancer Institute defines it as the percentage of people who test negative for a disease among a group of people who do not have it, and notes that no test is 100 percent specific, because some people without the disease will test positive anyway.

In breast screening, the people a test wrongly flags are recalled for further imaging and sometimes a biopsy, and most of them turn out not to have cancer. That is the cost side of any change that finds more cancers: a method that catches more disease by being readier to call something suspicious will also clear fewer healthy women, and specificity is where that shows up. It is the number to check whenever a screening result claims an improvement. In the MASAI trial, specificity was 98.5 percent both with artificial intelligence support and with standard double reading, so the extra cancers were not bought with extra false alarms.

The measure is not the same as the false positive rate, but it is its mirror: a specificity of 98.5 percent means 1.5 percent of the women without cancer were flagged anyway.