Purpose: The optimal definition of biochemical recurrence of prostate cancer after definitive radiotherapy remains elusive. Different institutions have developed their own definitions, and a consensus conference (CC) sponsored by the American Society for Therapeutic Radiology and Oncology has recently proposed another definition. This study compares the definition previously used at our institution with the definition proposed by the CC.
Methods: Two hundred and eight patients were treated for localized prostate cancer with conformal external-beam radiotherapy between 1989-1993 at our institution and followed for at least 24 months. Patients were categorized as failures according to our institutional definition and the CC definition. Our definition (CPMC) required two increases in serum prostate specific antigen (PSA) over at least a 3-month period with a final value of at least 1 ng/ml or a single value resulting in clinical intervention. The CC definition required three consecutive increases in PSA. This was modified to also consider those patients with one or two increases leading to clinical intervention as failures. Differences in the failure rates between the two definitions were evaluated and factors influencing these differences were explored. In an additional analysis, CC was modified such that patients with one or two PSA increases were censored at the time of the PSA prior to the increases (CC-II), rather than at the last PSA (CC). The median follow-up time was 31 months.
Results: There were 36 fewer failures according to CC (n = 96) compared with CPMC (n = 132) (p < 0.001). Twenty cases called failures by CPMC subsequently had a decrease in PSA ("false failures"). The other 16 patients have had two increases in PSA, but are awaiting their next follow-up visit to obtain a third PSA ("pending failures"). Analysis of factors predicting "pending failures" showed Gleason score to be the sole predictor of this change in status in multivariate analysis (p = 0.03) with patients with lower-grade tumors being more likely to change status (Gleason 2-6: 15% vs. Gleason 7-10: 1%). On the other hand, "false failures," compared to true failures, had a lower mean PSA nadir (1.7 ng/ml vs. 7.0 ng/ml, p < 0.001) and significantly smaller mean increases in PSA (1st increase: 0.6 ng/ml vs. 3.4 ng/ml, p = 0.006; 2nd increase: 0.4 ng/ml vs. 4.8 ng/ml, p = 0.002). In 85% (17 of 20) of these patients, at least one of the increases was < or = 0.3 ng/ml compared with 44% (42 of 96) of the true failures (p = 0.0008). CC-II resulted in a small decrease in BDFS rates compared with CC, but did not affect the overall difference between CC and CPMC. A modified definition that defines failure as two consecutive increases in PSA over 3 months, with a final value greater than 1.0 ng/ml and each increase being at least 0.3 ng/ml, or three consecutive increases would result in a "false" failure rate of only 3% (3 of 99) and identify 56% (54 of 96) of the true failures after only two PSA increases.
Conclusion: The CPMC definition of two PSA increases can falsely identify patients as failures, particularly if the increases in PSA are small (i.e., < or = 0.3 ng/ml). The CC definition requiring three increases in PSA can falsely identify patients as disease-free when the time to failure is long relative to the follow-up time. We propose a that a definition that combines aspects of both definitions (two consecutive increases in PSA over 3 months, with a final value greater than 1.0 ng/ml and each increase being at least 0.3 ng/ml, or three consecutive increases) may be a better definition of biochemical failure.