of death rather than to prevent death itself. In such an instance of lead-time bias, the elderly person is likely to be subjected to the negative effects that arise from being labeled as diagnosed with cancer, as well as the likelihood of additional uncomfortable or painful instrumentation or surgery; these effects occur in the absence of any real postponement of death from the cancer. Also, one could conceive of a screening modality whose effect was to identify tumors that are declared to be cancer but that in fact are without lethal potential or that, as a group, take many years to kill the individual and are much less aggressive in comparison to cancers that are identified on the basis of symptoms or routine clinical exams. In such an instance of length bias sampling, the social and medical consequences are the same as in the prior example: adverse interventions without material prolongation of life.
There are few available data on the efficacy of screening for breast cancer in those over the age of 75, but screening efficacy is well supported in the years from ages 50 through 65. Although data on screening for cervical cancer in women over age 50 are sparse, the Clarke and Anderson case-control study16 and the ecologic studies noted earlier are consistent with a strong protective effect from cervical cancer screening through age 65. Taken together, the data on these two very different epithelial cancers support the conclusion that the predictive value of screening tests in the elderly is at least as good if not better than the positive predictive values characteristic of younger age groups.
From the standpoint of evidence of efficacy, it is difficult to support the notion of establishing an arbitrary age at which to terminate screening for cancer such as, for example, terminating screening for cancer of the cervix at age 65. Yet the committee notes that recent recommendations regarding cervical cancer are consistent with this practice.32 This may well be counterproductive: if either sensitivity or specificity are, indeed, less in a given screening modality (which seems not to be the case in screening for cervical and breast cancer), this deficit may be partly or entirely offset by much greater predictive value at any given level of sensitivity and specificity owing to the more common occurrence of cancer (see Tables 9-1 and 9-2).
The issue of competing causes of death becomes increasingly important in the context of cancer screening and aging in the very elderly; for example, it is recognized18 that average life expectancy is 6.9 years for an 85-year-old woman and that the average, untreated natural history between onset of detectable preclinical cancer and death from it may be nearly as long or longer. The importance of