Separate Estimation of Primary and Secondary Cancer Preventive Impact: Analysis of a Case-Control Study of Skin Self-Examination and Melanoma
Begg, Colin B., Huang, Ying, Berwick, Marianne, Journal of the American Statistical Association
In this article we present the analysis of a population-based case-control study to evaluate the potential impact of skin self-examination (SSE) in reducing mortality from melanoma. The series of cases comprised 650 subjects diagnosed with primary cutaneous melanoma between January 1987 and May 1989, identified from pathology laboratories that report to the Connecticut Tumor Registry. A total of 549 population-based controls were identified by random-digit dialing. The cases were followed through 1994, and those who died or developed advanced disease were designated as "lethal" cases. Further details of the study design have been provided by Berwick, Begg, Fine, Roush, and Barnhill (1996). Typically, analyses of studies using this design involve comparisons of the proportions "exposed" to screening among the lethal cases and the controls. This results in an odds ratio of .37 (95% confidence interval of .16-.84), adjusted for risk factors for melanoma, indicating that skin self-examination, if adopted on a population basis, has the potential to reduce cancer mortality by more than half.
Unfortunately, the preceding analysis is open to serious criticism on the grounds that the onset of lethality will tend to occur later in screened cases due to lead-time bias, thereby inflating the apparent protective effect of the intervention (Gill and Horwitz 1995). Lead-time bias is the spurious apparent increase in survival in cases detected by screening due merely to the fact that the disease is diagnosed earlier than it would be otherwise (Prorok, Connor, and Baker 1981). Furthermore, although SSE is intended to reduce mortality via early detection and effective (surgical) treatment of the cancer, its use has the potential to encourage the removal of precancerous nevi (moles) and other premalignant tissue. That is, exposure to screening has the potential to actually reduce cancer incidence (primary prevention) in addition to reducing mortality among individuals who already have cancer (secondary prevention). These important and distinctive effects are not distinguished in the analysis just outlined. In recognition of these issues, we have attempted to develop a more complete conceptual framework for the evaluation of case-control studies of screening, permitting the separate estimation of the primary and secondary preventive effects.
The goal of cancer screening is to reduce cancer-specific mortality. Therefore, if screening is effective, then a series of cancer deaths should exhibit a history of less-intensive screening than the source population from which the deaths are derived (Morrison 1982). For compelling practical reasons, it is more common to ascertain consecutive incident cases rather than consecutive deaths when conducting a case-control study. There are two principal reasons for this. First, it is more difficult to ascertain a population-based series of deaths than a series of incident cases. Second, one cannot interview dead cases to obtain the necessary information on screening history, lifestyle, and other relevant cancer risk factors. So the hybrid design used in our melanoma study has become popular in this setting (Weiss and Lazovich 1996). In this design, incident cases are recruited and relevant information on screening and risk factors is obtained. The cases are then followed for survival, and the data are analyzed when enough cases have died to facilitate the comparison of the dead cases with the nondiseased control group.
Characterization of screening history can be a challenging task, unless there is reliable and exhaustive information in the available medical records of the subjects. In the melanoma study, the screening intervention under evaluation was "purposeful" SSE on at least one occasion. In fact, SSE is a behavioral attribute rather than a medical intervention, and our study results indicate that most subjects who engage in it appear to do so on a regular basis. …