Defining Biochemical Recurrence of Prostate Cancer After Radical Prostatectomy: A Proposal for a Standardized Definition

Clinically localized prostate cancer, relative to other solid tumors, is characterized by a protracted natural history from diagnosis to metastatic disease progression (MP) and prostate cancer–specific mortality (PCSM). The serum prostate-specific antigen (PSA) assay has revolutionized the diagnosis and treatment of prostate cancer, and PSA is routinely used to monitor disease recurrence after definitive therapy because biochemical recurrence (BCR) antedates MP and PCSM by an average of 7 and 15 years, respectively.^1,2 BCR is widely used as an early end point to assess treatment success and frequently prompts the initiation of secondary therapy. Despite this widespread use, uniform criteria to define BCR after radical prostatectomy (RP) are lacking, and numerous definitions are used in the medical literature.

The definition of BCR has a substantial effect on the outcome reporting of definitive local therapy for prostate cancer. Several investigators have demonstrated that disease progression rates may vary by up to 35% depending on the BCR definition used.^3-5 Likewise, the 3-year probability of continued PSA progression and/or secondary therapy has been reported to vary from 50% to 79% when recurrence is defined by a single PSA cut point of 0.2 versus 0.4 ng/mL.³

Although a rising PSA level is the first sign of ultimate progression to distant metastasis and PCSM, the natural history of BCR is highly variable, and it is not a surrogate for these end points.⁶ For many patients, prostate cancer recurrence heralded by a slowly rising PSA may pose little threat to one's longevity or quality of life.⁷ Within 15 years of BCR, patients are as likely to die from prostate cancer as they are from competing causes.⁸ It is imperative, when proposing a standardized BCR definition, to demonstrate that the disease state defined by this criterion is associated with a high probability of MP or PCSM.

A standardized definition is necessary to enable valid comparisons between local therapies or combined modality treatment strategies, to identify patients most suitable for entry into clinical trials for the rising PSA state,⁹ and to better select patients for salvage therapy. For translational research purposes, a BCR definition with a close association with MP is also important to establish a standardized end point for studies assessing the prognostic utility of biomarkers. To be clinically useful, the optimal BCR definition must consider both specificity (association with MP) and sensitivity (identification of recurrence at an early stage that may be amenable to local salvage therapy).

Hitherto, no study has analyzed the association of different BCR definitions with MP. Given that metastatic prostate cancer is the primary determinant of survival for this disease, is associated with serious morbidity, has no proven cure, and antedates PCSM by an average of 3 to 6 years,^1,10,11 we endeavored to identify the PSA criterion, among 10 BCR definitions, that best explains MP after RP after controlling for all prognostic variables and the use of secondary therapy.

Between August 1985 and February 2004, 3,792 patients with clinically localized prostate cancer underwent RP at our institution without prior radiotherapy. Because of the significant impact of androgen-deprivation therapy (ADT) on the time to BCR, 667 patients (18%) who received neoadjuvant ADT were excluded, leaving 3,125 patients for analysis. All prostates were totally embedded, and those from patients treated since 1998 were sectioned by the whole-mount technique. The preoperative and pathologic characteristics of the cohort are listed in Table 1.

In general, patients were followed for disease recurrence with serum PSA determinations and clinical assessment every 3 months for the first 3 years, every 6 months in years 4 and 5, and annually thereafter. A median of seven serum PSA values (interquartile range, 4 to 11) were available per patient from the time of RP until last follow-up. PSA assays in use at our institution included Hybritech Tandem-E (lower detection limit, 0.3 ng/mL; Beckman Coulter Inc, Fullerton, CA) before 1996, Tosoh AIA (lower detection limit, 0.05 ng/mL; Tosoh USA Inc, Grove City, OH) from 1996 to 1997, and Bayer Immuno-1 (lower detection limit, 0.05 ng/mL; Bayer Diagnostics, Tarrytown, NY) since 1997. Overall, 2,410 patients (77%) have undergone RP since the introduction of the more sensitive assays. Overall, 156 patients received postoperative radiotherapy and 225 received postoperative ADT as adjuvant or salvage therapy. Over a median follow-up of 49 months (range, 1 to 227 months), 75 patients have progressed to distant metastases (including 27 who started salvage ADT > 6 months before clinical detection), and 24 have died of disease.

The primary end point in this study was MP after postprostatectomy BCR, and secondary end points included the use of secondary therapy, continued PSA progression (defined as a subsequent rise by any increment at any future time point), and the PSA doubling time (PSADT) after BCR. Although the BCR definition that best explains MP is likely to exhibit great specificity for recurrent disease, the BCR definition of optimal clinical utility must balance specificity and sensitivity. Currently, a rising serum PSA level is the most common indication for initiating salvage therapy,¹² and the majority of patients who receive salvage radiotherapy and ADT do so at PSA levels less than 1.0 ng/mL and less than 4 to 10 ng/mL, respectively.^7,13,14 Thus, the 10 BCR definitions analyzed were chosen on the basis of acceptable sensitivity without knowledge of their association with our primary and secondary end points. Table 2 summarizes these definitions along with the American Society of Therapeutic Radiation and Oncology definition (ASTRO). Except for ASTRO, the date of BCR was the date when the PSA criterion for each definition was met without back-dating. For ASTRO, the date of recurrence was the midpoint between the date of the last undetectable PSA level and the date of the first of the three consecutive PSA rises.¹⁵

Depending on the BCR definition used, the number of recurrences after RP ranged from 318 (PSA ≥ 0.4 and rising) to 557 (singe PSA ≥ 0.2), and the 10-year progression-free probability ranged from 63% (single PSA ≥ 0.2) to 79% (ASTRO) (Table 3). In eight of 10 definitions, the median time to recurrence ranged from 20 to 31 months. The two exceptions were ASTRO (15 months) and three consecutive rises (42 months). These two definitions are identical except for back-dating the date of recurrence in ASTRO; this large difference in the time to recurrence illustrates the substantial back-dating that occurs using ASTRO. The median PSA value at the time of recurrence ranged from 0.3 (single PSA ≥ 0.2) to 1.0 ng/mL (PSA ≥ 0.4 and rising).

The R² statistics from Cox models for MP by each BCR definition ranged from 0.14 to 0.21 (Table 4). BCR defined as PSA of at least 0.4 and rising had the highest calculated R² statistic (0.21). In general, BCR definitions that required rising PSA values at levels of at least 0.2 ng/mL, or single PSA values of at least 0.4 ng/mL tended to be the best predictors of MP (R² 0.17 to 0.21), and those that did not specify rising PSA values at levels of at least 0.2 ng/mL tended to be the poorest (R² 0.14 to 0.15).

The 4-year actuarial probability of subsequent PSA progression after BCR was 80% to 94% in eight of 10 definitions evaluated (Table 5). The median PSADT after BCR ranged from 12 (single PSA ≥ 0.6) to 21 months (PSA ≥ 0.1 and rising; Table 5). Recurrence defined as single PSA of at least 0.2 or PSA of at least 0.1 and rising is of questionable clinical significance in some patients, because an estimated one third was followed for 4 years without evidence of a continued rise in the serum PSA level, and the PSADT was greater than 20 months in half of the patients.

The 7-year probability of secondary therapy after BCR (or clinically evident recurrence, whichever occurred first) is listed in Table 6. Secondary therapy (as opposed to clinical recurrence) was the initial event in 67% to 72% of cases. In nine of 10 definitions, the probability of secondary therapy ranged from 58% to 66%, and the median time to treatment ranged from 41 to 65 months. As with PSA progression, the clinical course after BCR defined as single PSA of at least 0.2 was relatively indolent because an estimated 55% of patients were free of secondary therapy at 7 years. Interestingly, BCR defined as three successive incremental rises of at least 0.1 was associated with a high probability of secondary therapy but was among the definitions that predicted most poorly for MP.

Given the frequently protracted time course from definitive local therapy to MP and PCSM, enormous emphasis has been placed on BCR as an early end point to enable assessments of treatment success, innovations, and identification of prognostic biomarkers to occur in a reasonable time frame. Currently, systemic therapy strategies are targeting the BCR population to enhance therapeutic efficacy. Thus, the need for uniform criteria to define BCR across all treatment modalities is evident. It is within this context that we endeavored to identify the BCR definition that best explains MP after RP.

Our results confirm the observations of others in terms of the varying probability of treatment success (63% to 79%), continued PSA progression (67% to 94%), and secondary therapy (45% to 66%) based on the BCR definition employed.^3,4,19 However, in an important extension of previous work, we analyzed the association of each BCR definition with MP in the context of multivariable Cox regression models using a goodness-of-fit statistic. PSA of at least 0.4 and rising was the BCR definition that best explained the MP, and it was also associated with a relatively high probability of PSA progression, secondary therapy, and a rapid PSADT. Thus, we propose adoption of this BCR definition as the standard for reporting outcomes of RP, as the biochemical end point for clinical trials of combined modality treatment strategies and for studies identifying prognostic biomarkers. Likewise, this definition should be used to define patient populations suitable for entry into clinical trials of systemic therapy for postprostatectomy BCR.

We believe that our approach to determining the optimal BCR definition is novel and valid. Distant metastasis was selected as the primary end point because it is the primary determinant of cancer-specific survival. Previous investigators have focused on secondary therapy and systemic or local recurrence as end points.^3,5 However, the timing of secondary therapy is highly controversial, and this decision may be dictated as much by the physician's treatment philosophy as by the patient's disease characteristics. In our study, we adjusted for the use of secondary ADT and radiotherapy in the predictions of distant metastasis by using them as time-dependent covariates. Likewise, the prognosis of patients with local versus systemic recurrence is vastly different.

We have also taken the unique approach to determining the optimal BCR definition by including pathologic stage, surgical margins, Gleason grade, and serum PSA in our multivariable Cox regression models for MP. This is an important consideration because the clinical significance of a postprostatectomy PSA of 0.23 ng/mL, for example, may be considerably different for a patient with lymph node metastasis compared to one with organ-confined prostate cancer. By correcting for the patients' disease characteristics, PSA of at least 0.4 and rising is a valid BCR definition to use for all patients, regardless of their pathologic stage, Gleason grade, or PSA.

There was considerable variation in the ability of each of the BCR definitions to explain MP. The top four definitions for MP were PSA of at least 0.4 and rising, PSA of at least 0.2 and rising, single PSA of at least 0.6, and single PSA of at least 0.4, and these were also associated with the highest probability of PSA progression, secondary therapy, and a rapid PSADT. However, the converse was not true in all cases, illustrating the potential disconnect between the use of salvage therapy for BCR and the actual risk of distant metastasis. It is not straightforward to conduct significance testing between differences in R² values across non-nested models, but PSA of at least 0.4 and rising had the highest R² value by a substantial margin. If alternative definitions are used, they should specify, at a minimum, two or more successive rising PSA values at levels of at least 0.2 ng/mL or single PSA values of at least 0.4 ng/mL.

PSA should reach undetectable levels within 4 weeks after RP. However, a detectable PSA level after this time does not necessarily represent clinically significant recurrent disease. Some patients with detectable PSA levels do not progress because of the presence of benign prostate glands at the margins of resection or from a dormant residual focus of prostate cancer at a local or distant site.²⁰ Similar to that reported by others,³ BCR defined as single PSA of at least 0.2 was associated with an indolent course in a substantial proportion of patients in our study. Thus, it would appear that this BCR definition is inadequate despite its widespread use in the literature.

The current study has several limitations worth noting. The study cohort was composed of patients managed over a 20-year period who were followed for disease recurrence using PSA assays with lower detection limits between 0.05 and 0.3 ng/mL. The fact that BCR defined as PSA of at least 0.4 and rising was most closely associated MP may be driven by the fact that 31 (41%) of 75 events occurred in patients who were initially monitored for recurrence using the older PSA assay. BCR defined as PSA of at least 0.2 and rising potentially may be appropriate using more sensitive PSA assays; however, the data are not mature to test this hypothesis in patients monitored exclusively using these assays. Other limitations of the study include the fact that the follow-up of patients was not standardized, and thus the timing of PSA testing, the use of secondary therapy and imaging studies may have varied considerably among patients. During the study period, the use of secondary therapy for BCR may have delayed the progression to distant metastasis and influenced our results despite our best efforts to control for it.

The 318 patients who recurred on the basis of PSA of at least 0.4 and rising presumably had more advanced disease at the time of recurrence by virtue of having the highest median PSA level at recurrence. Although BCR definitions specifying higher PSA thresholds may have yielded better results in terms of the association with MP, the definitions selected for analysis all had acceptable levels of sensitivity given that the administration of salvage radiotherapy at low PSA levels leads to improved cancer control.¹³ BCR definitions associated with substantially higher PSA levels at recurrence, such as the recently proposed PSA nadir plus 2.0 ng/mL for radiotherapy,⁵ are not appropriate for the RP population given that a large proportion of patients will receive secondary therapy before fulfilling the criteria for recurrence.

For local salvage therapy considerations, PSA of at least 0.4 and rising may not be a sufficiently sensitive BCR definition as 75% of patients who recurred by this criterion had a PSA of more than 0.67 ng/mL at BCR and accumulating evidence suggests salvage radiotherapy should be administered at PSA levels less than 0.6 ng/mL to optimize outcome.¹³ PSA of at least 0.2 and rising may be a better BCR definition for patients with a rising PSA who are presumed to have local recurrence as the patients who recurred by this criterion had a substantially lower PSA at the time of BCR, and this definition was among the best predictors of MP.

The quest for a universal BCR definition that can be applied across all treatment modalities remains elusive. Our proposed BCR definition has not been tested in radiation therapy populations and is likely to suffer from inadequate specificity given that the 10% to 30% of patients who exhibit the “PSA bounce” phenomenon would be falsely declared as having recurrence.^21,22 The limitations of the recently proposed PSA nadir plus 2 ng/mL definition for the RP population has been discussed.

Even if a common BCR definition were adopted for both populations, the considerable differences in the PSA kinetics after radiotherapy and RP may not permit valid comparisons. For example, most BCR definitions for radiotherapy require that a patient first reach his nadir PSA. The nadir PSA is reached within 4 weeks of RP, but the median time to PSA nadir is 18 to 36 months after external-beam radiotherapy, and it may be considerably longer after brachytherapy.^23,24 Thus, few recurrences after radiotherapy are declared within the first 3 years after treatment,²³ whereas 50% of the recurrences after RP in our study were observed within 31 months in nine of the 10 BCR definitions tested. Back-dating also may impart a bias because the amount of time that a recurrence is back-dated may not be the same for radiotherapy and RP. Lastly, 40% to 50% of patients treated with radiotherapy will receive neoadjuvant or adjuvant ADT (often for the presence of adverse features) which may significantly influence the time to BCR; neoadjuvant ADT is not commonly administered to RP patients.^25,26 For these reasons, it is unlikely that any uniform BCR definition based on current PSA assays would be appropriate for radiation and surgically treated patients.

Interestingly, the BCR definition that best explains MP is the same definition that was proposed by the Prostate-Specific Antigen Working Group in 2004 on the basis of the available evidence.⁹ Their recommendations further support our proposal that PSA of at least 0.4 ng/mL followed by another higher value should be adopted as the standard for reporting postprostatectomy PSA outcomes. Use of alternative definitions should be based on a demonstrated association with MP or PCSM. More sensitive definitions, such as PSA of at least 0.2 and rising, may be more appropriate to use when selecting patients for salvage radiotherapy.

The authors indicated no potential conflicts of interest.