Study summary: The present study was designed to examine the time-varying effects of tumor characteristics on breast cancer survival. While early detection and better treatments have improved survival, the risk of relapse remains long beyond diagnosis. Therefore, identifying prognostic factors for late breast cancer recurrence and metastasis is important. The study included data concerning breast-cancer-free survival for 3,088 breast cancer cases in four U.S. states. The subjects had participated in a randomized dietary intervention trial during 1995–2006. The median follow-up period was nine years; the maximum was 15 years. To allow for deviations from the proportional hazards assumption, a piecewise constant penalized spline approach incorporating time-varying coefficients was used. This method provides direct estimates of hazard ratios across time intervals. During the first 2.5 years after diagnosis, Stage II or III breast cancer at diagnosis was found to be associated with a three-fold higher risk of breast cancer compared to stage I breast cancer. While this hazard ratio decreased to 2.1 after 7.7 years, higher tumor stage persisted as an important risk factor. Similar diminishing effects over time were found for poorly differentiated tumors. Having estrogen receptor-positive (ER+) breast was protective up to four years after diagnosis. However, ER+ tumor status was found to be detrimental after 7.7 years (hazard ratio = 1.5). The authors comment that the results emphasize the importance of modeling that allows for time-dependent effects in long-term breast cancer survivorship studies.