Statistical aspects of evaluating treatment and prognostic factors for clinically localized prostate cancer.

This article reviews the fundamental assumptions for survival analysis and discusses some of the difficulties specific to the treatment evaluation and the analysis of prognostic factors in clinically localized prostate cancer. What makes clinically localized prostate cancer different from other forms of cancer is the chronic nature of the disease and the introduction of prostate-specific antigen (PSA) evaluation following a primary treatment. It is known that a study evaluating survival benefit for localized prostate cancer requires a long-term follow-up. This brings up issues of time varying treatment effect and the necessary use of cause-specific survival. In addition, biochemical failure following radiation therapy based on ASTRO consensus definition is another major topic. We question the appropriateness of the last observation approach that censors patients at their last observation and uses the Kaplan-Meier method. We show that the last observation approach can underestimate the biochemical failure rate for a treatment, especially when follow-up is short. The estimate of lower and upper bounds for biochemical failure is recommended. Examples based on Radiation Therapy Oncology Group prostate cancer trials are provided. This article concludes with a discussion of some novel statistical approaches to the design of prostate cancer studies and the analysis of trajectories of PSA values.