Extended Data Fig. 3: Prevalent and incident screens per screening service, and overall.
Plots show cancer detection rate against recall rate curves for prevalent screens (orange), incident screens (green), and both prevalent and incident screens (blue). Individual panels represent: a) All screening services (N = 115,973 participants). b) Screening service 1 (N = 23,023 participants), c) Screening service 2 (N = 22,630 participants), d) Screening service 3 (N = 23,400 participants), e) Screening service 4 (N = 23,590 participants), f) Screening service 5 (N = 23,330 participants). The higher human reader recall rate for prevalent screens (first screens) vs incident screens (subsequent screens) is consistent across all services. AI appears to have potential benefits in the prevalent screen group for all sites. Error bars represent 95% confidence intervals.
