A new model based on data from the Breast Cancer Surveillance Consortium (BCSC) suggests that overdiagnosis of screen-detected breast cancer is less frequent than estimates from excess-incidence studies, but the model also takes into account indolent tumors and produced a higher estimate than previous models that didn’t consider this factor.
“There is a pronounced lack of consensus of the true rate of overdiagnosis in the contemporary U.S. mammography practice. This uncertainty about the extent of overdiagnosis is a problem for the development of guidelines and policies. By overcoming shortcomings of previous studies, we produced a defensible estimate of overdiagnosis in contemporary U.S. mammography practice. About one in seven screen-detected cancers in women (between 50 and 74 years) undergoing biennial screening will be overdiagnosed, and about one in three overdiagnosed cancers are attributed to the detection of nonprogressive cancers,” said Marc D. Ryser, PhD, in an interview. Dr. Ryser is an expert in mathematical and statistical modeling in population health science at Duke University, Durham, N.C. He presented the results of the model at the 2021 San Antonio Breast Cancer Symposium.
Previous models have come up with estimates ranging from 0% to 54%, but the heterogeneity makes them difficult to compare. “They differ in study populations, estimation methods and their definitions of overdiagnosis,” Dr. Ryser said.
There are two general ways to estimate overdiagnosis. One is a model-based approach that works out the tumor latency using models of disease natural history and clinical data, and then uses that to predict overdiagnosis. But these models may not account or indolent tumors, or tumors that would not likely cause death during the patient’s lifetime, and the assumptions behind the models can be opaque. On the other hand, the excess-incidence strategy compares incidence in screened versus unscreened populations and assumes that excess cancers in the screened group is caused by overdiagnosis, but this can be affected by bias.
To get around these limitations, Dr. Ryser’s group used a model-based approach, but also allowed for indolent tumors. They ensured transparency of the underlying assumptions of the model, and took advantage of a contemporary, high-quality data source in the BCSC.
They used individual mammography screening and breast cancer diagnosis records from 35,986 women aged 50-74 years, who were first screened between 2000 and 2018. To estimate overdiagnosis caused by indolent tumors, they used the risk of non–breast cancer mortality from age cohort–adjusted annual mortality risks. There were a total 82,677 screens and 718 cases of breast cancer diagnosed. 3.6% of detected tumor were indolent (95% credible interval, 0.2%-13.8%). The predicted overdiagnosis rate for a biennial screening program was 15.3% (95% prediction interval, 9.7%-25.2%). 6.0% of overdiagnosis was projected to be caused by indolent tumors (95% PI, 0.2%-19.0%) that don’t progress at all, and 9.3% to tumors that would progress, but not fast enough to cause mortality during the individual’s lifetime. An annual screening program had a predicted overdiagnosis rate of 14.6% (95% PI, 9.4%-23.9%).
Dr. Ryser identified some specific studies that used the same definition of overdiagnosis as his group used, and compared them with the 15.3% incidence that his group determined. Excess-incidence studies produced higher estimates, while modeling studies produced lower estimates.
The model did not distinguish between ductal carcinoma in situ and invasive cancers, and it did not account for patient race and breast density.
The study was funded by the National Institutes of Health. Dr. Ryser has no relevant financial disclosures.