Background: Poor reporting of harms in randomised trials is a common problem for Cochrane reviewers. Overdiagnosis is the most important harm of breast cancer screening, and increases in surgery were first quantified in a review of the randomised trials. However, overdiagnosis in trials may differ from that in population-based screening due to lack of life-long follow-up, opportunistic screening in the control groups, less rigorous quality standards and training of radiologists, and increased detection of harmless lesions due to technical developments or fear of litigation if cancers are missed.
Objectives: To quantify overdiagnosis in clinical practice using published incidence trends of breast cancer in regions with mammography screening.
Methods: Articles publishing data on the incidence of breast cancer before and after introduction of breast cancer screening were identified through a literature search in PubMed using the search string: 1: (("Mammography" [MeSH] OR "Mass Screening" [MeSH]) AND (("Breast Neoplasm/epidemiology" [MeSH] OR "Breast neoplasms" [MeSH] AND "Incidence*[ti]))) 2: Breast cancer AND screening AND trend*[ti] 3: Breast cancer AND screening AND overdiagnos*[ti] 4: (#1 OR #2 OR #3) One author scanned titles and abstracts and potentially eligible articles were obtained for evaluation. Data were extracted by one author and checked by the other. Differences were settled by discussion. Data were evaluated based on the size of the population, uptake, reliability of registration, length of follow-up, and length of the baseline period before screening. We considered large studies with long time intervals before and after the introduction of screening as the most reliable. We based our estimate of overdiagnosis on these studies and used remaining ones in sensitivity analyses.
Results: Breast cancer incidence increased by a mean of 40% in countries with good quality data, which was sustained over time. We found little evidence of a drop in incidence among older age groups no longer screened, as would be expected if the increase were mostly caused by earlier detection rather than overdiagnosis. Several factors may have influenced this increase, however: Factors potentially leading to an overestimate. The necessary advancement of the time of diagnosis (lead time), which is difficult to estimate, leads to inflated incidence rates. However, as we found little compensatory decrease in incidence among women no longer screened, this could explain only a minor part of the observed increase. Reduced fertility and increased obesity increase the underlying incidence, regardless of screening. This was compensated for by regression analyses of pre-screening trends in some of the estimates. Hormone Replacement Therapy (HRT) increases the incidence, but one study compensated by comparing the incidence in regions with and without screening and found the highest increase in incidence related to screening (50%). Factors potentially leading to an underestimate. Uptake rates varied from about 60% to over 90%, which influence estimates on population level. Some studies only included invasive cases, although in situ cancers increase considerably with screening, and most of these are likely overdiagnosed. In some areas, there was substantial opportunistic screening before an organised program was instituted.
Conclusions: Several factors influence incidence rates of breast cancer in regions with screening, but we were able to estimate overdiagnosis using good-quality data from countries with population based mammography screening. We estimate an overdiagnosis in practical use of mammography screening of 30% to 40%. This is in good agreement with our prior estimate of 30% based on the randomised trials. The benefit of screening in reducing breast cancer mortality is about 15%. This means that for every 2,000 women screened regularly for 10 years, one woman will have her life extended whereas 10 to 13 women will receive an unnecessary breast cancer diagnosis and treatment.