Estimating the sensitivity of breast cancer screening--experience with the Honolulu BCDDP data

The 'capture-recapture' models for estimating breast cancer screening sensitivity can be generalized to include factors that affect sensitivity such as cancer size. Including such factors can help reduce the covariance between mammographic and physical exam sensitivity, which will improve the estimates. One model relating sensitivity to cancer length is presented and tested using data from the Honolulu Breast Cancer Detection Demonstration Project (BCDDP). The model predicts sensitivity fairly well for large breast cancers, but underestimates sensitivity for small cancers. Using both mammography and physical examination, the total screening sensitivity is estimated as 0.91 +/- 0.03 for cancers longer than 12 mm (for the second through fifth screenings for Honolulu and Tucson screening data combined). Limited data suggest similar screening sensitivities for Orientals and Caucasians in Hawaii. Shortcomings of the BCDDP data include inaccurate measurements of cancer length. Suggestions to improve data for future work are given.