Re: Dormancy of mammary carcinoma after mastectomy.

Karrison et al. (1) retrospectively studied 1547 patients with breast cancer operated on during the period from 1945 through mid-1987 to evaluate the limit of breast cancer dormancy. In the report, the authors stated that they “did not see evidence for a second peak in the hazard curve as reported by Demicheli et al.” (2) and drew evidence for tumor dormancy from the elevation of mortality rates out to 20 years after primary tumor removal and from the detection of new recurrences even after this time. While we agree with the authors’ conclusions about the occurrence of tumor dormancy, we wish to focus on the question of why they did not observe a twopeaked hazard curve, as has been confirmed by other reports (3–5). In our opinion, the fact that two peaks were not observed could be due to both the chosen end point and to the characteristics of their data base. While the events “first recurrence” and “death from breast cancer” reliably identify patients who are not cured, the combined event “first recurrence or death from breast cancer” is not a useful end point for study of the time-dependent structure of the hazard of each of the two events. First recurrence and death may have different time distributions, and the use of the combined end point may obscure the fine structure of the hazard curve of each event. Moreover, since Karrison et al. showed that an association existed between the time from mastectomy to first recurrence and the subsequent residual survival time, such an association may affect the hazard curve of “first recurrence or death from breast cancer.” Even more confounding noise, masking the time dependence of hazard function, may have resulted from the possible inadequate reliability of their event dating. Indeed, the examined time