Objective
To determine the increased cancer detection rate, if any, of programmes in the UK National Health Service breast screening programme (NHSBSP) using more than single reading of mammograms.
Design
Information on the detection of cancers by individual screening programmes from annual (KC62) returns, supplemented by questionnaire information about the number of readers.
Setting
The 87 NHSBSP programmes from England and Wales for the screening year 1 April 1996 to 31 March 1997. The study includes all programmes for prevalent screens where two views are mandatory, but excludes the four programmes using two view mammography for incident screening.
Main outcome measures
Cancer detection, invasive cancer detection, and small (<15 mm) invasive cancer detection by mammographic reading protocol using single reading as the reference level.
Results
Programmes collectively using single reading detected the lowest rate of cancers at both prevalent (first) and incident (subsequent) screening. The highest rate of age standardised cancer detection was achieved by programmes using double reading with arbitration. At prevalent screens, where all programmes used two views, those programmes using double reading with arbitration detected 32% (95% confidence interval (CI) 3% to 69%) more small (<15 mm) invasive cancers than programmes using single reading. At incident screens, where all programmes analysed used one view this increased to 73% (95% CI 40% to 113%). Recall rates showed no obvious difference between single reading and the double reading protocols, being around 7% for prevalent screens and 3.5% for incident screens.
Discussion
The results suggest that the increase in cancer detection resulting from increasing the number of readers depends on the number of views, and is higher for one view than two views. Single reading of one view results in a low detection rate of small invasive cancers for most individual programmes. It is, however, recognised that a small number of individual readers may achieve high detection rates with such a protocol. All groups of programmes using different reader/view protocols are on average close to or above target cancer detection rates, except those using single reading of one view (mediolateral oblique) at incident screens.
