Recognizing the Limitations of Cancer Overdiagnosis Studies: A First Step Towards Overcoming Them

Numerous studies have attempted to quantify the number of breast cancers that would never have been diagnosed in the absence of screening. Unfortunately, results are highly variable across studies and there is considerable disagreement about both the frequency of overdiagnosis and the validity of different methodologic approaches. In this Commentary, we review limitations of the two major approaches used in existing studies. Studies that use excess incidence as a proxy for overdiagnosis require a valid estimate of incidence in the absence of screening and sufficient follow-up to ensure the excess excludes relevant (ie, nonoverdiagnosed) cancers detected early. The requirement of sufficient follow-up applies to both population studies and clinical trials, but only certain clinical trial designs have the potential to yield unbiased results. Studies that model disease natural history to infer overdiagnosis must, in addition, examine whether their models produce valid estimates in the presence of nonprogressive cases. In this setting, limited follow-up could lead to a lack of identifiability of the parameters needed to accurately infer overdiagnosis. In a polarized research community, the excess incidence and modeling approaches are generally viewed as competitors, but we argue that they are complementary, with models being more complex but having greater potential to inform about disease natural history and the outcomes of candidate screening policies. Rather than arguing why one approach should be preferred to another, investigators should focus on developing studies that generate reliable estimates of overdiagnosis. Recognizing that both approaches have limitations, which existing studies rarely overcome, is a first step towards reconciling methodologic perspectives and achieving consensus about the real magnitude of the overdiagnosis problem.