Cancer in the Elderly-Letter.

Abstract

We compliment Radkiewicz and colleagues for introducing extensive, high-quality data on cancer at old age (1), but we also wish to describe problems with their methods, including confusion between incidence and prevalence (2).Radkiewicz and colleagues find that cancer incidence rates in Sweden are highest at 75 to 90 years old, then reduce at older ages (1). These results agree with research from other countries (3, 4). However, Radkiewicz and colleagues argue for major corrections to address age-associated diagnostic bias. Their proposed corrections add incidence and autopsy prevalence together. This reverses the old-age reductions, which Radkiewicz and colleagues conclude were artifactual, and replaces them with the view that Sweden's medical system fails to diagnose or treat most cancers in the very old.Yet, Radkiewicz and colleagues’ conclusions are premised on interpreting autopsy prevalence as incidence, and prevalence is not incidence (2). Decades ago, similar misinterpretations of autopsy prevalence as incidence were used to conclude that increases in endometrial and lung cancer were also artifacts of diagnostic bias, rather than being due to the actual causes of estrogen replacement therapy and smoking (2). Unlike the historical precedents, Radkiewicz and colleagues’ methods are well intentioned and founded on reasonable concerns. However, the same mathematical problems arise (2).To explain briefly, we turn to a hypothetical example. Suppose cancer is diagnosed in 100 of 100,000 at-risk people this year (incidence rate: 100 per 100,000 person-years), with everyone receiving equal, appropriate diagnostic attention. Often, cancers have multi-year preclinical periods when they are still too small and early stage to cause symptoms or be diagnosed. So, for every person diagnosed this year, there can be several—say 4—who have not been diagnosed yet but already carry incipient cancer. If people who die are autopsied and autopsies detect even half of incipient cancers, then unexpected cancer prevalence at autopsy will be 200 per 100,000 this year—double the actual incidence. In this way, adding incidence and autopsy prevalence can greatly overstate true incidence. Radkiewicz and colleagues further multiply by all-cause mortality, which is highest at old ages, magnifying the problem there. Moreover, evidence shows the problem is large. For example, Karwinski and colleagues report 55% of cancers unexpectedly detected at autopsy were “incidental findings and thus of no importance to the individual concerned” (5).Another problem is that the proposed correction counts autopsy detections of nonprogressive, indolent cancers as though they should have been diagnosed clinically (overdiagnosed). In the unmodified data that are also presented by Radkiewicz and colleagues, clinical incidence declines at old ages, but autopsy prevalence usually rises further (1). From our perspective, these patterns actually tend to support the idea that cancers have biologically reduced proliferative potential at old ages (4), leading clinical incidence to decline while indolent prevalence (and thereby autopsy prevalence) rises.Finally, regardless of abovementioned disagreements, we agree more resources should be devoted to impending cancer increases as populations age. Cancer incidence ultimately declines with age, but not greatly until unusually old ages, meaning that population aging will increase incidence overall, requiring resources.See the Response, p. 1506C. Harding reports personal fees from Exergen, Corp., a manufacturer of thermometers, during the conduct of the study; personal fees from Exergen, Corp. outside the submitted work. No disclosures were reported by the other authors.