sensitivity

Every test that looks for a disease can fail in two directions. It can miss someone who has the disease, or flag someone who does not. Sensitivity measures the first kind of failure: the proportion of people who genuinely have the condition that the test correctly picks up. The National Cancer Institute puts it as how well a test detects a disease in the people who actually have it, and notes that no test reaches 100 percent, because some people with the disease will always come back negative.

In a screening programme the figure answers a blunt question: of the women who had breast cancer at the time of their screen, how many did the screen find? A higher sensitivity means fewer cancers left to surface between screening rounds. In the MASAI trial, the first randomised controlled trial of artificial intelligence inside a national mammography programme, sensitivity was 80.5 percent in the group screened with AI support and 73.8 percent in the group read the standard way, a difference the trial reported as statistically significant. Specificity, the matching measure for the other kind of failure, was 98.5 percent in both groups.