Design of a retrospective study of computer-aided detection in mammographic screening: Computer Aided Detection Evaluation Trial

The National Health Service Breast Screening Programme is an intensive consumer of radiology time. In principle, screening mammograms must be double read, and this puts considerable strain on human resources. The possibility of using single reading assisted by computer-aided detection rather than double reading is therefore an important research issue. Before expending considerable resources on a prospective randomised trial, we design a retrospective study involving re-reading of previously double-read mammograms (Computer Aided Detection Evaluation Trial). This was an equivalence, powered to assess whether the sensitivity of a single human reader with the R2 system of computer-aided detection was at least no worse than 10% less sensitive than the previous two human readers. Various complications had to be taken account of in the design:

• power calculations suggested we needed 15,000 mammograms to give the required number of cancers, but funding was only available to re-read 10,000;

• the ethical problem of potential de novo discovery of previously missed cancer;

• a limited period of free availability of the R2 system; and

• a need to avoid previous readers re-reading the same mammograms.

Design strategies to cope with these are described in this poster.