Extrapolation of pre-screening trends: Impact of assumptions on overdiagnosis estimates by mammographic screening. (June 2016)

• Record Type: Journal Article
• Authors: Ripping, T.M.; Verbeek, A.L.M.; ten Haaf, K.; van Ravesteyn, N.T.; Broeders, M.J.M.
• Is Part Of: Cancer epidemiology. Volume 42(2016:Jun.)
• Page Start: 147
• Page End: 153
• Publication Date: 2016-06
• Subjects: Breast cancer -- Excess incidence -- Mammographic screening -- Netherlands -- Overdiagnosis -- Pre-screening trend -- Screening; Cancer -- Epidemiology -- Periodicals; Cancer -- Prevention -- Periodicals; Cancer -- Diagnosis -- Periodicals; Carcinogenesis -- Periodicals; 616.994005
• DOI: 10.1016/j.canep.2016.04.015
• Languages: English
• ISSNs: 1877-7821

Abstract

Highlights: Extrapolation of pre-screening trends is a common method to estimate overdiagnosis. Overdiagnosis estimates using this method range from 0 to 57% for mammographic screening. Assumptions made to extrapolate the trend can fully explain the range of estimates. Choice of pre-screening period has the largest impact on overdiagnosis estimates. Researchers should acknowledge limitations of extrapolating pre-screening trends. Abstract: Background: Overdiagnosis by mammographic screening is defined as the excess in breast cancer incidence in the presence of screening compared to the incidence in the absence of screening. The latter is often estimated by extrapolating the pre-screening incidence trend. The aim of this theoretical study is to investigate the impact of assumptions in extrapolating the pre-screening incidence trend of invasive breast cancer on the estimated percentage of overdiagnosis. Methods: We extracted data on invasive breast cancer incidence and person-years by calendar year (1975–2009) and 5-year age groups (0–85 years) from Dutch databases. Different combinations of assumptions for extrapolating the pre-screening period were investigated, such as variations in the type of regression model, end of the pre-screening period, screened age range, post-screening age range and adjustment for a trend in women <45. This resulted in 69, 120 estimates of the percentage of overdiagnosis, i.e. excess cancer incidence in the presence of screening as a proportion of the number of screen-detected and interval cancers. Results: Most overdiagnosis percentages are overestimated because of inadequate adjustment for lead time. The overdiagnosis estimates range between −7.1% and 65.1%, with a median of 33.6%. The choice of pre-screening period has the largest influence on the estimated percentage of overdiagnosis: the median estimate is 17.1% for extrapolations using 1975–1986 as the pre-screening period and 44.7% for extrapolations using 1975–1988 as the pre-screening period. Conclusion: The results of this theoretical study most likely cover the true overdiagnosis estimate, which is unknown, and may not necessarily represent the median overdiagnosis estimate. This study shows that overdiagnosis estimates heavily depend on the assumptions made in extrapolating the incidence in the pre-screening period, especially on the choice of the pre-screening period. These limitations should be acknowledged when adopting this approach to estimate overdiagnosis.