In the last two decades, a number of widely varying estimates of overdiagnosis rates in breast cancer screening have been published. The estimates of overdiagnosis, in terms of an increase in incidence due to screening, range from less than 5% to more than 50%.1 2 A recent paper by Chaltiel and Hill goes some way to resolve the issue of widely varying estimates.3 Chaltiel and Hill observed that all estimates higher than 17% arose from studies with no individual data, that is only from ecological observation. They concluded that the larger figures are serious overestimates.3
In the past, researchers have generally used a definition of overdiagnosis that is not pathological but epidemiological: overdiagnosis is defined as diagnosis as a result of screening of histologically confirmed cancer which would never have arisen in the patient’s lifetime if screening had not occurred. Because the vast majority of cancers are treated, individual overdiagnosed cases cannot be identified: we do not know what would have happened in the absence of screen detection and consequent treatment. As a consequence, the counterfactual estimate of what would have occurred had screening not taken place is highly variable, depending on assumptions, leading to the wide variation in estimates referred to above.
While intuitively, overdiagnosed cancers would be expected to have favourable pathological and biological characteristics and to be diagnosed at an early stage, this has not featured in the formal definition. A paper by Dempsey et al in this issue of the BMJ Oncology goes some way to filling this gap.4 The authors do not formally estimate overdiagnosis, but define potentially overdiagnosed cancers as either low or medium grade ductal carcinoma in situ or stage 1 invasive cancer which is grade 1, smaller than 10 mm in maximum diameter, Her-2 negative and not triple negative. Correspondingly, all other cancers were defined as not overdiagnosed. While one might argue with exact boundaries between potentially overdiagnosed and not overdiagnosed cancers, the above definition seems reasonable.
Applying their definition to 5183 screen-detected breast cancers in Australia and New Zealand, the authors estimate that 15.8% of screen detected breast cancers were potentially overdiagnosed. If we applied this to a woman attending all seven screens within the age range (50–70) of the UK programme, this would corresponding to an absolute risk of a possibly overdiagnosed tumour of less than 1%.
Perhaps more importantly, only 28% of the screen detected cancers received chemotherapy, compared with 48% of the non-screen detected cancers, and only 4 (less than 1%) of the potentially overdiagnosed cancers received chemotherapy. In terms of surgery, 24% of the screen detected cancers were treated with mastectomy compared with 41% of the non-screen detected cancers. Only 14% of the potentially overdiagnosed cancers were treated with mastectomy.
These results seriously call into question the figures and textual material related to overdiagnosis in many information resources for breast screening. For example, the information provided to invitees in the UK’s NHS Breast Screening Programme includes an estimate of three overdiagnosed cases per 200 women regularly screened between ages 50 and 70 years. This figure was taken from the UK Independent Review.5 Since then, it has been revealed that there are serious issues with the two trials used by the Independent Review to estimate overdiagnosis, which indicate that the 3 per 200 is an overestimate.6–8 The results of Dempsey et al,4 with those of the review by Chaltiel and Hill,3 indicate that the true figure is less than 2 per 200.
Equally importantly, the results of Dempsey et al suggest that any discussion of overdiagnosis in information resources should note that overdiagnosed cancers are unlikely to receive radical treatment. As the authors state,4 ‘A reduction in breast cancer mortality is not the only screening outcome of importance. Reduced intensity of recommended treatment is also a key benefit.’
