Vacek P M, Mickey R M, Worden J K
Department of Medical Biostatistics, University of Vermont, Burlington 05405, USA.
Prev Med. 1997 May-Jun;26(3):287-91. doi: 10.1006/pmed.1997.0138.
Self-reported behavior is widely used to estimate the prevalence of breast cancer screening and to evaluate programs for promoting screening, but detailed studies of reliability have not previously been performed.
Reliability was assessed by comparing responses to questions about screening behavior from repeat personal interviews of 382 women age 40 and older living in low-income census tracts of two Florida communities. Reliability was assessed using Pearson's correlation (r) and kappa (kappa) coefficients.
Estimated reliabilities were kappa = 0.38 for "ever had clinical breast examination," kappa = 0.82 for "ever had mammogram," kappa = 0.65 for "mammogram in past year," r = 0.54 for "date of last mammogram," and r = 0.72 for "number of mammograms." The dates of last mammogram reported at the two interviews agreed within 1 month for 64% of the women, while the dates of last clinical breast examination agreed within 1 month for 50% of the women. Reliability of "ever had mammogram" was significantly related to demographic variables.
Women reliably report ever having mammography, but information about timing and frequency has lower reliability. The results have implications for breast screening research because measurement error affects the precision of estimates and the sample sizes needed to detect program effects.
