Radiation Therapy for Clinically Localized Prostate Cancer: A Multi-institutional Pooled Analysis

Context Prostate-specific antigen (PSA) evaluation leads to the early detection of both prostate cancer and recurrences following primary treatment. Prostate-specific antigen outcome information on patients 5 or more years following treatment is limited and available mainly as single-institution reports.

Objectives To assess the likelihood and durability of tumor control using PSA evaluation 5 or more years after radical external beam radiation therapy and to identify pretreatment prognostic factors in men with early prostate cancer treated since 1988, the PSA era.

Design and Setting Retrospective, nonrandomized, multi-institutional pooled analysis of patients treated with external beam radiation therapy alone between 1988 and 1995 at 6 US medical centers. Follow-up lasted up to a maximum of 9 years. Outcome data were analyzed using Cox regression and recursive partitioning techniques.

Patients A total of 1765 men with stage T1b, T1c, and T2 tumors treated between 1988 and 1995 with external beam radiation. The majority (58%) of patients were older than 70 years and 24.2% had initial PSA values of 20 ng/mL or higher. A minimum of 2 years of subsequent follow-up was required for participation.

Main Outcome Measure Actuarial estimates of freedom from biochemical failure.

Results The 5-year estimates of overall survival, disease-specific survival, and the freedom from biochemical failure are 85.0% (95% confidence interval [CI], 82.5%-87.6%), 95.1% (95% CI, 94.0%-96.2%), and 65.8% (95% CI, 62.8%-68.0%), respectively. The PSA failure-free rates 5 and 7 years after treatment for patients presenting with a PSA of less than 10 ng/mL were 77.8% (95% CI, 74.5%-81.3%), and 72.9% (95% CI, 67.9%-78.2%). Recursive partitioning analysis of initial PSA level, palpation stage, and the Gleason score groupings yielded 4 separate prognostic groups: group 1, included patients with a PSA level of less than 9.2 ng/mL; group 2, PSA level of at least 9.2 but less than 19.7 ng/mL; group 3, PSA level at least 19.7 ng/mL and a Gleason score of 2 to 6; and group 4, PSA level of at least 19.7 ng/mL and a Gleason score of 7 to 10. The estimated rates of survival free of biochemical failure at 5 years are 81% for group 1, 69% for group 2, 47% for group 3, and 29% for group 4. Of the 302 patients followed up beyond 5 years who were free of biochemical disease, 5.0% relapsed from the fifth to the eighth year.

Conclusions Estimated PSA control rates in this pooled analysis are similar to those of single institutions. These rates indicate the probability of success for subsets of patients with tumors of several prognostic category groupings. These results represent a multi-institutional benchmark for evidence-based counseling of prostate cancer patients about radiation treatment.

Counseling patients who have been recently diagnosed as having early clinically localized prostate cancer (stages T1 and T2) about the best curative treatment using updated evidence-based information has been vexing and controversial over the last decade. At the core of the debate is the efficacy of radical radiation therapy, currently delivered to more than 60,000 men each year, in achieving total tumor eradication in men presenting with early prostatic carcinoma.^1,2 Uncontrolled series are subject to selection biases; hence, interseries comparisons of results are usually flawed because of unequal proportions of patients with known (and unknown) significant adverse prognostic factors.^3-7 Most large single-institution series now report that pretreatment serum prostate-specific antigen (PSA) values, Gleason score (GS), and palpation stage are significant independent predictors of a successful outcome following either surgery or radiation therapy.^8-16 A successful outcome now requires that the patient has no PSA-based evidence of relapse with long-term follow-up.^1,2,7,8,17 Long-term follow-up for patients treated with external beam radiation in the PSA era (since 1988 when PSA values began to direct earlier diagnosis) is limited to fewer than 10 years; thus, most single-institution series have only a small number of treated patients with follow-up of more than 5 years. In an effort to define better the cancer control rates for these patients and to maximize follow-up, the American Society of Therapeutic Radiology and Oncology sponsored an analysis by an independent biostatistical unit. Eight centers known to have relatively long follow-up on their series of irradiated patients were invited to supply their outcome data on all patients treated from 1988 to 1995, and 6 medical centers agreed to participate.

Between 1988 and 1995, 1765 men with clinically localized prostatic cancer (stages T1b, T1c, T2, NX, M0) were treated with external beam radiation therapy alone (no androgen deprivation therapy) at 6 institutions. These institutions submitted follow-up results for all patients with these clinical stages treated at their respective centers during this period with external beam radiation alone. Also required were a known initial PSA value and at least 4 PSA measurements after radiation treatment. The mean pretreatment PSA level was 18.9 ng/mL, and the median was 10.1 ng/mL (range, 0.2-2028 ng/mL). Transurethral resection of the prostate was performed within 4 months preceding the initiation of radiation therapy in 7.5% of the patients.

The predominant method of treatment was a 4-field technique (anterior, posterior, and right and left laterals), with the tumor minimum dosage reported. In general, patients with T2 tumors and higher tumor grades were treated with fields encompassing the primary tumor target volume and the low pelvic lymph nodes to doses of 45 to 50 Gy (median dose, 46.5 Gy) with conventional daily fractionation. All patients were treated to the prostate target volume to total doses ranging from 63 to 79 Gy (median dose, 69.4 Gy), using a variety of techniques to cone-down the treatment field after 46 to 50 Gy. The daily radiation doses ranged from 1.8 to 2.1 Gy for 5 days a week; 51% of patients were treated using 3-dimensional conformal techniques. The median follow-up for patients treated with 3-dimensional conformal techniques was 3.1 years, and the median dose was 70.5 Gy. Because of the relatively short follow-up and the similarity of median doses, no comparisons can be safely made between conformal and more conventional techniques or between high- and conventional-dose therapy.

Treatment outcomes were measured in terms of overall survival, clinical recurrence-free survival, and survival free from biochemical recurrence. The American Society of Therapeutic Radiology and Oncology's consensus guidelines of biochemical recurrence¹⁷ were used, as modified: 3 consecutive rises in PSA values or any rise great enough to provoke androgen suppression with backdating of the failure time to the midpoint between the last nonrising and the first rising PSA value.¹⁷ Patients who died of other causes were censored at the time of death. Only patients whose follow-up was longer than 24 months and who had at least 4 posttreatment PSA measurements were included, except for those men who experienced a clinical failure or whose PSA level rose and met the criterion for biochemical recurrence prior to 2 years after treatment. The median number of posttreatment PSA values prior to failure or censoring was 7 (range, 4-27, with a total of 13,453 PSA value determinations evaluated in this population of 1765 men).

The freedom from biochemical recurrence (bNED control, biochemical, no evidence of disease) was calculated from the beginning of radiation therapy. Estimates of rates for bNED control were calculated using the Kaplan-Meier product limit method.¹⁸ Multivariate analysis was conducted using the Cox proportional hazards model to examine the effect of clinical factors on outcome.¹⁹ Comparisons of survival curves were carried out using the log-rank test or the Mantel-Haenszel test.²⁰ Recursive partitioning analysis was used to identify subgroups for which the risk of treatment failure is similar in terms of prognostic factors, such as the initial PSA level, GS, and palpation stage.²¹ This is achieved by identifying the cut points that minimize the P values between the different groups, thus giving the best separation between prognostic groups and checking their accuracy through cross-validation and bootstrapping. Recursive partitioning analysis was chosen instead of other approaches to classification problems because it handles categorical data in a natural fashion and gives easily understood and readily interpretable information regarding the predictive structure of the data. Calculations were carried out using realizations of these algorithms coded in S-Plus software.²² All P values are 2-tailed. Because of likely differences between the institutions in the way tumors were staged, graded, or treated, a preliminary analysis was performed to determine the degree of heterogeneity between the centers. This was done in 2 ways: (1) by comparing outcome of initial PSA value, stage of tumor, and GS with and without institution as covariate and (2) by analyzing the rankings of GSs by institution using the Kruskal-Wallis test.

Six institutions supplied information on all patients with T1b, T1c, or T2 tumors with a known pretreatment PSA value but unknown nodal status and who were treated with external beam radiation therapy alone between 1988 and 1995. The percentage of patients contributed by the following 6 institutions ranged from 8.2% to 26.1%: University of Michigan, Ann Arbor; Fox Chase Cancer Center, Philadelphia, Pa; Massachusetts General Hospital, Boston; Washington University, St Louis, Mo; Eastern Virginia Medical School, Norfolk; and Stanford University Medical Center, Palo Alto, Calif. The median patient age was 71 years and the median follow-up interval was 4.1 years. Patients with similar tumor GS groupings from the different centers were compared for bNED control. When stage, pretreatment PSA, and GS were accounted for, there was a significant institutional effect for 1 institution as judged by bNED control analyzed using the Cox proportional hazards model. A second analysis looked at the distribution of GS by institution, which showed that the institutions were significantly different in terms of the rankings of GS across institutions. The institutional effect as judged by bNED control was insignificant when 1 institution was omitted. The same institution stood out from the others by both of the above analyses. Subsequently, all analyses were carried out for all patients from the 6 institutions (n=1765) and for the statistically homogeneous group of patients from the 5 institutions (n=1607). The results were nearly identical whether 5 or 6 institutions were analyzed: the recursive partitioning analysis cut points were virtually the same, and the estimates of 5- and 7-year bNED control were within 2% absolute. However, for all of the analyses evaluating possible prognostic factor classifications and their possible interactions, only patients from 5 institutions were used.

The distribution of patient and tumor characteristics in the entire population is shown in Table 1. A 52.8% majority of the treated patients were older than 70 years, 24.2% had presenting initial PSA values of 20 ng/mL or greater, only 25.7% had stage T1b or T1c tumors, and 14.1% presented with initial PSA values of 4 ng/mL or less. For the 6-institution population, the 5-year estimates of overall survival, disease-specific survival, and the bNED control rates are 85.0% (95% confidence interval [CI], 82.5%-87.6%), 95.1% (95% CI, 94.0%-96.2%) and 65.8% (95% CI, 62.8%-68.0%), respectively. The PSA failure-free rates 5 and 7 years after treatment for patients presenting with a PSA of less than 10 ng/mL were 77.8% (95% CI, 74.5%-81.3%) and 72.9% (95% CI, 67.9%-78.2%). Recursive partitioning analysis of the data using only the initial PSA value as a continuous variable resulted in the definition of 4 prognostic subgroups for bNED control. These groups are patients whose PSA value was less than 9.2 ng/mL, from 9.2 to less than 19.7 ng/mL, from 19.7 to less than 31.7 ng/mL, and at least 31.7 ng/mL. The estimates of bNED control at 5 years for these groups were 81%, 69%, 50%, and 29%. Figure 1 shows the estimates of bNED control for the conventional or convenient PSA cut points of less than 10, at least 10 to less than 20, at least 20 to less than 30, and at least 30 ng/mL are nearly identical. There was no change in outcome at the level of 4 ng/mL, the clinical upper limit of normal for this test (Table 2). Recursive partitioning analysis of the data using initial PSA value as a continuous variable plus palpation stage and GS resulted in the definition of 4 significantly distinct prognostic subgroups for bNED control. Tumor stage by palpation did not play a role in the definition of the cut points. The recursive partitioning analysis defined 4 prognostic groups: group 1, patients with a PSA level of less than 9.2 ng/mL; group 2, patients with a PSA level of at least 9.2 and less than 19.7 ng/mL; group 3, patients with a PSA level of at least 19.7 ng/mL and a GS of 2 to 6; and group 4, patients with a PSA level of more than 19.7 ng/mL and a GS of 7 to 10. The estimated bNED rates for these groups at 5 years are 81% for group 1, 69% for group 2, 47% for group 3, and 29% for group 4 (Figure 2).

Table 2 compares by univariate analysis the significance of bNED status at 5 years according to pretreatment PSA, palpation stage, GS groupings, patient age, and posttreatment PSA nadir levels. The recursive partitioning analysis cut points are supported by these binary evaluations. The effect of tumor stage by palpation (T1b and T1c vs T2) is significant in univariate analysis. When patients were grouped by their specific GSs, a GS of 7 was not significantly different using the bNED end point when compared with a GSs of 8 to 10 (59% vs 49% at 5 years, P=.23). Similarly, the GSs of 6 and lower were all quite similar with 74% for patients with a GS from 2 to 4, 75% with a GS of 5, and 71% with a GS of 6. The bNED control at 5 years with a GS of 7 (59%) compared with these 3 groupings combined differed significantly (P<.001). The GS groupings of 2 to 4, of 5 alone, and of 6 alone are not significantly different even when subgrouped by initial PSA value. However, Table 2 and Figure 3 show that the GS groupings comparing 5 and 6 vs 7 to 10 are significant (P<.001). Patient age at presentation was not significant. The bNED control probabilities differ significantly by the nadir value that was reached following radiation. With the required postirradiation follow-up of at least 2 years, only 5.7% of the patients had PSA values that were still falling. There was no evidence that the achievement of a specific PSA nadir value was required to achieve durable biochemical control.

Figure 4 presents the bNED survival rates for 293 patients presenting with stage T1c tumors (those in whom the only indication for a diagnostic biopsy was an elevated and/or increasing PSA value). With a median follow-up of 3.4 years in this group, the 5-year bNED rate is 76.7% (95% CI, 71.5%-82.3%). For those with the pretreatment PSA of less than 20 ng/mL (chosen by recursive partitioning) compared with those of 20 ng/mL or greater, the 5-year bNED rates are 87% and 47%, respectively (P<.001).

Table 3 shows results of multivariate analyses for the group of 1443 patients for whom GS groupings were available and shows, as did the recursive partitions, that a biopsy on a GS of 6 or below compared with a GS of 7 or above was the Gleason cut point as an independent predictor of bNED control, using estimates of bNED control at 5 years as the end point. The multivariate analysis demonstrates that pretreatment PSA value (P≤.001) and GS(P≤.001) were highly significant independent predictors of bNED control. Palpation stage was insignificant on multivariate analysis (T1 vs T2, P=.17). However, the bNED control for patients with stage T1c tumors was significantly higher when compared with all other patients with T-stage tumors (T1c vs T1b+T2, P=.01).

Of the 302 patients available for follow-up beyond 5 years and who had bNED control, 5.0% relapsed from their bNED status from the fifth to eighth year (Table 4). Of the 57 patients available for follow-up beyond 7 years, only 1 had recurred.

Table 5 summarizes the 5-year bNED status by prognostic factor categories separately based on this pooled analysis data for the 413 patients with T1 tumors and for the 1194 patients with T2 tumors.

Disease-specific end points such as disease-specific survival and metastasis-free survival^3-6,23 have been studied in other pooled analyses. Our pooled analysis is the first to use the earlier PSA end point. A rising PSA level precedes clinical failure with a lead time of years²⁴ and may never result in clinical metastases in men with a limited life span. It is, however, justified as an end point for this study because it rigorously assesses the true ability of external radiation to cure the patient. Although a rising PSA value is a valid indicator of disease activity and is the most important marker for evaluating disease relapse following treatment, as yet, it has not been validated as an early surrogate for progression to clinically detectable metastatic disease nor has it been validated for death due to prostate cancer.⁷ Nevertheless, a rising PSA value is now frequently used by itself as an indication for salvage therapy and, thus, may have as much clinical relevance to patients as an overt clinical relapse.

Until the results of well-controlled, prospective, randomized studies are available, caution should be used in comparing nonrandomized series such as the present one with others since selection biases may affect outcomes more strongly than the difference in efficacy between treatment modalities.^5-7

The absence of substantial long-term follow-up in this pooled analysis compromises the evaluation of the durability of bNED status beyond 5 years. In this series, 448 (28%) of the group of 1607 patients have been evaluated beyond 5 years. The PSA failure-free rates beyond 5 years out to 8 years in our report suggest that late recurrences only occur in a small percentage of patients. This rate of loss of bNED status beyond 5 years is similar to those that have been reported following radical prostatectomy.⁸ This series and others, however, will have to mature an additional 3 to 4 years before a complete evaluation of the durability of the bNED status can be determined for 10 years following primary treatment.

In the PSA era, stage T1c, impalpable, PSA-detected tumor is becoming the most common presentation for this disease and thus deserves special mention. These patients are presumed to have the earliest detectable and therefore the most curable disease. This is the disease stage that any screening program aims to detect. The work of Catalona et al²³ on a serially screened population has shown that once the more advanced disease that is prevalent within the population has been detected, T1c tumors in patients with a PSA value of less than 10 ng/mL and a GS of 6 or less are the most common subgroup found. When treated by external radiation, our analysis shows that 87% of these men with initial PSA values of less than 20 ng/mL are free from a serially rising PSA for at least 5 years after treatment (Figure 4). Also interesting is that while palpation stage was insignificant in a multivariate analysis, stage T1c became significant when compared with all other T stages combined (Table 3).

There was a significant effect in 1 of the 6 institutions on both treatment outcome and the distribution of GSs, which may have resulted from a variety of factors including differences in patient material, treatment technique, and pathologic scoring. Institutional variation in histologic grading and Gleason scoring has been well documented recently^25,26 and yet, together with other known and unknown forces of heterogeneity, has influenced the bNED outcome in our pooled analysis very little.

Although the pretreatment PSA cut points of 4, 10, and 20 ng/mL are used in most single-institution reports of treatment outcome, the results of recursive partitioning analysis of this multiple-institution series show cut points (9.2, 19.7, and 31.7 ng/mL) for bNED failure-free curves to be better separated from each other than with the PSA cut points usually reported. Our analysis showed no separation for patients with PSA values of less than 4 ng/mL but did for those with PSA values ranging from 19.7 to less than 31.7 ng/mL compared with those with PSA values of more than 31.7 ng/mL. We carried out a bootstrapping calculation,²¹ which showed that the cut points of 9.2, 19.7, and 31.7 ng/mL can be replaced with 10, 20, and 30 ng/mL without any important change.

Table 5 gives an evidence-based prediction for 5-year bNED rates by prognostic factor categories for patients treated with external beam radiation alone in the PSA era. Similar prognostic factor categories have been used recently to predict 5-year freedom for treatment failure following surgery at a single institution.²⁷ Our analysis following radiation gives physicians new broad evidence-based information to advise better their patients who are requesting their counsel on the best curative treatment available. The results of this multi-institutional pooled analysis also provide clinicians a benchmark against which, in the absence of results from randomized trials, newer and less well-studied radiation treatments in patients with T1 and T2 tumors, such as brachytherapy or radiation combined with neoadjuvant and/or concomitant androgen deprivation therapy, may be compared.