18CSG MODELLING THE IMPACT OF COMPUTER AIDED DETECTION AND DOUBLE READING OF SCREENING MAMMOGRAMS ON BREAST CANCER DETECTION AND MORTALITY

Wednesday, October 22, 2008
Columbus A-C (Hyatt Regency Penns Landing)
Paul Taylor, PhD and Henry W.W. Potts, PhD, University College London, London, United Kingdom
Purpose

A Markov model was used to assess the impact of two interventions often proposed as enhancements to breast cancer screening: computer aided detection and double reading of mammograms.

Method

Bibliographic databases were searched for evaluations of CAD and of double reading. A meta-analysis identified a range of values for the impact of both CAD and double reading on both the cancer detection rate and recall rate of screening. A Markov model was then constructed to examine the implications for the effectiveness of screening.

Age-specific figures were taken from national data for: screening uptake, cancers detected at screening, non-screening detected cancer and other cause mortality. Audits were used to identify stage profiles for screen-detected and non-screen detected cancers. Different survival curves for the stages of cancer are calculated from published data. The model is used to assess the impact on various outcomes, including mortality, of changes in the screen detected rate by running the simulation and varying the figures for cancer detection based on the results of the meta-analysis described above.

Results

Eight studies were found comparing single reading to single reading with CAD and fifteen comparing single reading with double reading. Where double reading is performed with arbitration it shows a significant increase in detection rate (odds ratio: 1.08, 95% CI:1.02-1.15) and a significant decrease in recall rate (odds ratio:0.94, 95% CI:0.92-0.96). CAD studies do not show a significant increase in cancer detection rate (odds ratio: 1.05, 95% CI: 0.95-1.17) and show an increased recall rate (odds ratio: 1.14, 95% CI: 1.11-1.16).

The simulation was run with data from UK screening, where double reading is standard practice. The simulation was repeated varying the cancer detection rate across a range of 4 cancers per 1000 women screened. Double reading results in fewer women being recalled and leads to more screen detected cancers than CAD. CAD and double reading both lead to a small but detectable difference in the total of breast cancer deaths.

Conclusions

The model shows a clear benefit from double reading with arbitration, with lowered recall rates, higher cancer detection rates and fewer deaths from cancer. CAD also shows improved cancer detection rates and fewer deaths, but the improvement is smaller and at the cost of an increased recall rate.