Background
Each year >150,000 patients in the United States will undergo radical prostatectomy (RP) for prostate cancer [
1]. Up to 30% of these patients will eventually experience biochemical failure (BF) as manifested by a rising prostate-specific antigen (PSA) [
2]. A rising PSA after RP can be indicative of local recurrence, distant disease, or both. Many patients who experience BF post-RP will go on to receive salvage external beam radiation therapy (SRT). Unfortunately, a suboptimal number of patients undergoing SRT obtain long-term recurrence free responses [
3]. It is likely that many of the patients who experience biochemical failure (BF) after SRT represent a subset of patients who harbor micrometastatic disease at the time of SRT. Improved prognostic factors are needed to better delineate the patients who have locally confined disease and are thus most likely to respond favorably to SRT.
A short PSA doubling time (PSADT) is known to predict both the development of metastasis (DM), prostate cancer-specific mortality (PCSM), and increased overall mortality (OM) in patients experiencing BF after RP [
4‐
10]. Similarly, a short PSADT after RP predicts worse clinical outcomes following SRT, including an increased likelihood of BF and development of distant metastasis [
3,
11‐
18]. To our knowledge, a direct association between a short pre-SRT PSADT and increased PCSM following SRT has yet to be reported; however, others have noted an association between a short pre-SRT PSADT and an improved overall benefit of SRT [
19,
20]. Furthermore, the most clinically useful PSADT threshold for defining high-risk patients has not been well established. Frequently proposed PSADT cut-points include 3, 6, 10, and 12 months [
3,
12‐
14,
19]. A well-known predictive tool for response to SRT, the Stephenson nomogram and decision tree, utilizes a PSADT threshold of 10 months [
3].
We sought to determine the most clinically prognostic pre-SRT PSADT cut-point for patients receiving SRT following RP.
Discussion
PSADT after BF following RP is prognostic for clinical outcomes as well as response to SRT [
3‐
18]. Boorjian et al. found that a PSADT <6 months strongly correlated with the development of metastatic disease following RP [
6]. Similarly, Pound et al. demonstrated that following RP, a PSADT <10 months was prognostic for the development of metastasis [
4]. Furthermore, a PSADT <3 months has been shown to be prognostic for increased prostate cancer-specific mortality and decreased overall survival following RP [
7,
10]. Similar to the results found in RP series, a short PSADT is prognostic for poor clinical outcomes following SRT, with multiple PSADT thresholds reported to be clinically relevant. Stephenson et al. identified that a post-RP PSADT <10 months resulted in increased risk for BF following SRT [
3,
12]. A number of studies have looked at PSADT prior to SRT and associations with BF or metastases with thresholds selected between 3 and 12 months, although, to date, no studies have evaluated an association between PSADT and PCSM following SRT [
11‐
18]. Interestingly, in our analysis using PSADT as a continuous variable, the strongest association as evaluated by c-index was between PSADT and PCSM (c-index 0.721) which was greater than that observed for more commonly reported end-points such as BF (0.623) and DM (0.670).
As a result, we confirm that a short PSADT post-RP is associated with poor clinical outcomes following SRT, including increased rates of BF and DM. Additionally, we establish that a rapid PSADT post-RP also increases the risk of PCSM. In our data set, analysis of hazard ratios and Kaplan-Meier curves demonstrated that a PSADT threshold of 6 months appears to be the ideal cut-point for predicting long-term response to SRT. This threshold strongly predicted BF, DM, and PCSM on multivariate analysis.
Predictive tools are needed to identify patients for whom SRT is appropriate. The Stephenson nomogram is one such tool, but only predicts for BF following SRT and has not been reported for DM, PCSM, or OM [
3]. Furthermore, while the Stephenson nomogram predicts BF, the overall discriminatory power of the model leaves significant room for improvement (c-index = 0.69) [
3]. In our cohort, PSADT alone had a c-index of 0.623 for predicting BF, resulting in only a minor reduction in discriminatory power compared to the Stephenson nomogram for predicting BF. This is not entirely surprising, given that PSADT is the single largest point contributor in the Stephenson nomogram, accounting for greater than 25% of the total points possible. Nevertheless, improved predictive tools are needed that can accurately and reliably predict which patients who experience BF following RP are appropriate candidates for SRT. Our data, which require validation, suggest that PSADT should be included in such nomograms, and that a threshold of 6 months would be most prognostic.
The utility of PSADT when addressing BF and PSCM after SRT is worth discussing so that SRT is not improperly selected only for patients based upon a long PSADT [
23]. In our cohort, where all patients received SRT, those with a PSADT <6 months are less likely to obtain a durable response to SRT compared to those with a longer PSADT, and are most likely to die from prostate cancer. However, this does not mean that patients with a short PSADT should not receive SRT. Two large retrospective studies assessing the survival benefit of SRT for patients experiencing BF following RP [
19,
20], where some received SRT and others did not, identified the greatest benefit in survival from SRT in those with the shortest PSADT. Trock et al. found that comparing those treated with or without SRT, prostate cancer-specific survival was improved with SRT, but only in patients with a PSADT <6 months [
20]. In contrast, Cotter et al. later demonstrated that SRT was effective in decreasing all-cause mortality both in patients with a PSADT <6 and >6 months [
19]. However, when analyzing patients without co-morbidities, the risk reduction offered by SRT was approximately 1.5 fold greater for patients with a PSADT <6 months [
19]. The primary outcome of interest in the study by Cotter et al. was all-cause mortality (ACM). Actuarial ACM was 49% for those receiving SRT with a PSADT <6 months, and 34% for those with a PSADT >6 months. In our cohort, median follow-up post SRT was 4 years shorter, which resulted in a lower actuarial ACM in both PSADT groups (33% for <6 months, and 20% for >6 months). However, the difference in actuarial ACM between those receiving SRT with a PSADT <6 months and >6 months was quite similar when comparing the Cotter cohort to our own (15% vs. 13%, respectively). As a result, the Trock and Cotter studies suggest that patients with a short PSADT are at greatest risk for PCSM but also most likely to benefit from SRT in terms of improved survival. Paradoxically, previous studies (11–14) and our data show that these patients are also the most likely to develop metastasis and experience a prostate cancer-specific death following SRT. These collective data are consistent with the hypothesis that following BF after RP, patients fall into two large categories: those with localized disease only and those with pre-existing metastasis. Those with a short PSADT are likely enriched for aggressive disease that could be either local or systemic. Patients with an aggressive recurrence (as represented by a short PSADT) without metastasis would have the most to gain from SRT and would be most likely to see an improvement in PCSM, whereas patients with micrometastatic spread outside of the treatment field would be least likely to benefit from SRT. In contrast, those with a long PSADT are likely to have less overall risk from their disease due to its more indolent nature and are less likely to see an improvement in PCSM with SRT (albeit more likely to have biochemical control following SRT). One can, therefore, conclude that PSADT alone is not sufficient to identify patients most likely to benefit from SRT in regard to all clinical end-points; although it does identify those at the greatest risk for PCSM who may have the most to gain if there is no evidence of metastatic disease. As a result, without improved means to identify metastasis, a short PSADT should not be used to select those not to receive SRT as this may remove this potentially curative therapy from those at greatest risk for PCSM who actually have the most to gain from SRT.
Limitations of this analysis include its retrospective design, which necessitates validation of these findings, and the fact that PSADTs could only be calculated for 48% of the SRT cohort. Unknown confounders may have been present and have the potential to bias our results if differences exist between the patients for whom PSADTs could be calculated for and those in whom we were unable to calculate PSADTs. However, no statistically significant difference existed between those with and without a determinable PSADT for any of our four primary outcomes of BF (p=0.3), DM (p=0.2), PCSM (p=0.1), and OM (p=0.1).
Conclusions
In summary, a short PSADT after BF post-RP predicts worse outcomes after SRT. A PSADT <6 months was the best predictor of outcomes in our data set, which was particularly true for BF and DM. We also establish that a post-RP PSADT <6 months is prognostic for PCSM following SRT on both univariate and multivariate analysis. The most common prognostic nomogram for SRT uses PSADT <10 months as the cut-off point for assigning points used to predict BF following SRT. Using a PSADT of <6 months may improve the nomograms functionality. However, since these data and the Stephenson nomogram only analyzed patients treated with SRT, they can provide information regarding the likelihood of biochemical control following SRT but, without data comparing outcomes without SRT, cannot reliably predict those most likely to benefit from SRT in regard to other end-points. Indeed, those at the greatest likelihood of BF following SRT are also those most likely to progress to DM and PCSM. If SRT can prevent these events, then these patients have the potential for greatest gain from SRT. Overall, improved prognostic and predictive models are needed to identify which patients are most likely to benefit from SRT, and our findings, once validated, suggest that PSADT should be included in such nomograms with a threshold of 6 months.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
WJ, BS, DL, CF, BF, YS, SJ, HS, DM, and FF made substantial contributions to acquisition of data. WJ, BS, DL, CF, BF, YS, MS, MS, HS, TM, GP, DH, and FF have all made substantial contributions to the conception and design of this manuscript, including analysis and interpretation of data, have been critically involved in the drafting and revising of this manuscript, and have given final approval for this final version of the manuscript to be published.