Objectives. To examine pretreatment variables that influence biochemical failure, and to describe and test a new definition of prostate-specific antigen (PSA)-based biochemical failure.

Methods. We introduce and describe a new definition of biochemical failure, which is based on quadratic fitting of the logarithm of the follow-up PSA profile curve. From a data base of 449 patients with prostate cancer treated with definitive radiation therapy, 230 patients who had at least five follow-up PSA observations were chosen for analysis. The new definition of failure was applied to this cohort, as was the conventional definition of two consecutive PSA rises. Univariate and multivariate analyses were performed using established pretreatment prognostic factors as covariates. Also, the association of both definitions of failure with clinical outcome (local recurrence and any recurrence) was examined.

Results. Application of the new definition of biochemical failure resulted in smoothing of the “noise” that is inherent in using definitions based on successive PSA rises. This smoothing was verified by smaller P values for the statistically significant covariates in the univariate analysis. Furthermore, the new definition correlated better with clinical outcome, as demonstrated by the statistically significant P values on regression analysis when using the quadratic fitted nadir compared with using the observed nadir.

Conclusions. We devised a new criterion based on quadratic curve fitting for PSA-based biochemical failure. This definition is based on all available PSA information, correlates with both pretreatment factors and post-treatment clinical outcome, is relatively insensitive to noise, and allows for prediction of time of failure.