Minimal data are available regarding selection of an optimal biochemical failure (BF) definition for patients treated with brachytherapy, external beam radiotherapy (EBRT), and combinations of these treatments with or without androgen deprivation (AD). We retrospectively analyzed our institution's experience treating localized prostate cancer in an attempt to determine a BF definition that could be applied for these various treatment modalities.
A total of 2376 patients with clinical stage T1-T3 N0 M0 prostate cancer were treated with conventional dose (median, 66.6 Gy) EBRT (n = 1201), high-dose (median, 75.6 Gy) adaptive radiation therapy (n = 465), EBRT + high-dose-rate brachytherapy boost (n = 416), or brachytherapy alone (n = 294) between 1987 and 2003. A total of 496 patients (21%) received neoadjuvant AD with radiation therapy. There were 21924 posttreatment prostate-specific antigen (PSA) measurements. Multiple BF definitions were tested for their sensitivity, specificity, positive predictive value (+PV), and negative PV (-PV) in predicting subsequent clinical failure (CF) (any local failure or distant metastasis), overall survival (OS), and cause-specific survival (CSS). Median follow-up was 4.5 years. The date of BF was the date BF criteria were met (e.g., date of third rise).
A total of 290 patients (12%) experienced CF at a median interval of 3.6 years (range, 0.2-15.2 years). The 5- and 10-year CF rates were 12% and 26%, respectively. Three consecutive rises yielded a 46% sensitivity and 84% specificity for predicting CF. The 10-year CF for those 475 patients who experienced three rises (BF) was 37% vs. 17% for those patients who did not meet these criteria (biochemically controlled [BC]). For all patients, the following definitions were superior to three rises for predicting CF for both +PV, and -PV: n + 1 (> or =1 ng/mL above nadir), n + 2, n + 3, threshold 2 (any PSA > or =2.0 ng/mL at or after nadir), threshold 3, threshold 4, and threshold 5. For the subset of patients treated with EBRT alone, the n + k definitions and threshold k definitions maintained superior predictive capacity. However, the threshold k definitions seemed to maintain a slightly greater separation in 10-year CF rates (43% for BF vs. 13% for BC = 30% difference for threshold 3). Surprisingly, all definitions generally had better predictive capacity for those patients who received brachytherapy or neoadjuvant AD vs. EBRT alone. The endpoints appeared similar for n + 1 vs. threshold 3 and n + 2 vs. threshold 4 in EBRT alone patients, but for brachytherapy or neoadjuvant AD patients, there were similarities for n + 2 vs. threshold 3 and n + 3 vs. threshold 4. This may be a reflection of the lower nadir levels in patients receiving AD (median <0.1 ng/mL vs. 0.2 ng/mL for brachytherapy vs. 0.8 ng/mL for EBRT alone, p < 0.01). When examining CF correlation for the various classes of BF definitions, the threshold k definitions clearly demonstrated the greatest area under the receiver operating characteristic curve, followed by the n + k definitions. For OS, the threshold k definitions again demonstrated the greatest area under the curve, followed by definitions based on specific nadir cutoffs (nadir > or =k ng/mL).
Biochemical failure definitions applying a PSA threshold at or after the nadir (e.g., threshold 3) demonstrated the highest association with CF, OS, and CSS for all assessed treatment modalities. Definitions incorporating a PSA increase above the nadir value (e.g., nadir + 2 ng/mL) were also superior for all modalities. In general, BF definitions have greater predictive capacity for clinical outcome with brachytherapy or neoadjuvant AD than EBRT alone, possibly because of less "noise" from production of background PSA.