SEARCH

SEARCH BY CITATION

Keywords:

  • prostate cancer;
  • radical prostatectomy;
  • biochemical recurrence;
  • nomogram

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

BACKGROUND

Men who undergo radical prostatectomy (RP) are at long-term risk of biochemical recurrence (BCR). In this report, the authors have described a model capable of predicting BCR up to at least 15 years after RP that can adjust predictions according to the disease-free interval.

METHODS

Cox regression was used to model the probability of BCR (a prostate-specific antigen level >0.1 ng/mL and rising) in 601 men who underwent RP with a median follow-up of 11.4 years. The statistical significance of nomogram predictors was confirmed with a competing-risks regression model. The model was validated internally with 200 bootstraps and externally at 5 years, 10 years, and 15 years in 2 independent cohorts of 2963 and 3178 contemporary RP patients from 2 institutions.

RESULTS

The 5-year, 10-year, 15-year, and 20-year actuarial rates of BCR-free survival were 84.8%, 71.2%, 61.1%, and 58.6%, respectively. Pathologic stage, surgical margin status, pathologic Gleason sum, type of RP, and adjuvant radiotherapy represented independent predictors of BCR in both Cox and competing-risks regression models and constituted the nomogram predictor variables. In internal validation, the nomogram accuracy was 79.3%, 77.2%, 79.7%, and 80.6% at 5 years, 10 years, 15 years, and 20 years, respectively, after RP. In external validation, the nomogram was 77.4% accurate at 5 years in the first cohort and 77.9%, 79.4%, and 86.3% accurate at 5 years, 10 years, and 15 years, respectively, in the second cohort.

CONCLUSIONS

Patients who undergo RP remain at risk of BCR beyond 10 years after RP. The nomogram described in this report distinguishes itself from other tools by its ability to accurately predict the conditional probability of BCR up to at least 15 years after surgery. Cancer 2008. © 2008 American Cancer Society.

Radical prostatectomy (RP) represents a treatment modality for patients with clinically localized prostate cancer (PCa).1 However, up to 33% of patients who undergo RP will demonstrate biochemical recurrence (BCR),2–4 and long-term data indicate that the BCR rates may be closer to 40%.5–7 Twenty-seven percent of BCRs occur ≥5 years after RP,2, 8 and the risk of BCR may not plateau until at least 15 years.5, 6, 9 In addition, 17% of patients with BCR die of PCa.10 Therefore, BCR is common: It may occur several years after RP, and it may be associated with PCa-specific mortality.

Several prognostic models help to identify men at risk of BCR.11–14 In 1999, Kattan et al. developed and internally validated a nomogram that predicted BCR after RP by using pathologic variables and preoperative serum prostate-specific antigen (PSA) levels.13 Their model demonstrated from 77% to 82% accuracy in external validation.13 Stephenson et al. updated that nomogram to 10 years of follow-up and demonstrated from 79% to 81% accuracy in independent validation sets.14 The limitations of the updated postoperative BCR nomogram relate to the use of the year of surgery (restricted to 1996-2004) and to the need of preoperative serum PSA for predictions. Only patients who were treated between 1996 and 2004 and only patients with available preoperative serum PSA levels can be included. In the early 1990s, preoperative serum PSA levels were not available invariably. In addition, the nomogram is limited to predicting BCR up to 10 years after RP. However, BCR does occur well beyond 10 years.

We decided to devise a prognostic nomogram that would predict BCR up to 20 years after RP without requiring the input of a preoperative serum PSA level. We relied on a surgical series of patients who underwent RP between 1960 and 1994 who had available PSA follow-up.15 Moreover, because the risk of BCR decreases with disease-free interval, we adjusted the nomogram predictions according to the BCR-free interval, an adjustment that also was used by Stephenson et al.14 Finally, we internally validated the nomogram at 5 years, 10 years, 15 years, and 20 years; and we externally validated the tool in a contemporary cohort at 5 years after RP to ensure its accuracy in patients who undergo RP in the new millennium.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

Patient Population

Between December 1954 and May 1994, 787 patients underwent RP at Virginia Mason Medical Center (VMMC) by 13 surgeons. No patient received neoadjuvant therapy. Clinical and pathologic data were logged into a prospective database by one of the authors (R.P.G.) from 1969 to 1999. In later years, the records were maintained electronically with institutional review board approval. Preoperative comorbidity was defined according to the Charlson comorbidity index.16 For purposes of the current analyses, the single patient who had a weighted index of 4 was grouped with patients who had an index of 2. Of 787 men, 186 were excluded because of unavailable PSA follow-up (161 men), missing age (9 men), or missing pathology (26 men). This resulted in 601 evaluable patients who underwent RP between 1960 and 1994. Pathologic stage was assigned according to 1992 tumor, lymph node, metastasis (TNM) classification. The Whittmore-Jewett classification was converted to the 1992 TNM classification for patients who were diagnosed before 1992. Two hundred ninety-seven patients (49.4%) underwent perineal RP, and 304 patients (50.6%) underwent retropubic RP. An obturator fossa lymph node dissection (PLND) at RP was performed in 308 patients (51.2%) who were selected according to individual surgeon preference and surgical approach. PLND pathologies were not recorded in the database and could not be included in the current analyses. From 1992 to 1994, tumors routinely were classified according to the Gleason grading system.17 Before 1992, tumor grade was recorded as well differentiated (grade 1), moderately differentiated (grade 2), and poorly differentiated (grade 3) and was recoded according to Roehl et al.4 Positive surgical margins were recorded as the presence of cancer cells against the inked resection margin. Adjuvant radiotherapy was delivered within 1 to 3 months after RP. PSA follow-up was performed at least quarterly for 2 years, then at least biannually for 2 years, and at least annually thereafter. Systematic PSA follow-up was introduced in 1988; thus, 253 patients who underwent surgery before 1988 did not have complete PSA follow-up according to the protocol described above. These patients were included in the analyses as long as they did not demonstrate clinical recurrence before the first available postoperative PSA measurement. BCR was defined as PSA level >0.1 ng/mL.15

Cause of death was ascertained according to detailed chart review or was obtained from the VMMC cancer registry, which uses links with the Washington State Death Certificate Office. PCa must be the first listed cause of death on the certificate for a patient to be classified as having died from PCa.

The first independent external validation cohort consisted of 2963 European patients who were treated exclusively with retropubic RP between 1992 and 2005 (Table 1). Of these, 1876 patients (63.3%) underwent RP during or after the Year 2000. None of the patients received neoadjuvant androgen-ablation therapy, and none received adjuvant radiotherapy. The second independent validation cohort consisted of 3178 patients who were treated exclusively with retropubic RP between 1998 and 2005 in an American institution. The last data follow-up was obtained on December 1, 2004.

Table 1. Descriptive Characteristics for the Nomogram Modeling Cohort and for the External Validation Cohorts of Patients Who Underwent Radical Prostatectomy
VariableNo. of patients (%)
Development cohortValidation cohort I: HamburgValidation cohort II: Cleveland
  1. pT indicates pathologic tumor classification; TNM, tumor, lymph node, metastasis classification system; NA, not available; RP, radical prostatectomy; PSA: prostate-specific antigen; RFS, recurrence-free survival.

No. of patients60129633178
Study period1960–19941992–20051988–2005
Age at radical prostatectomy. y
 Mean [median]64 [64.8]62.3 [62.9]60.5 [60.7]
 Range45.3–77.937.6–78.8388.5–78.9
pT stage: 1992 TNM
 T2a130 (21.6)310 (10.5)NA
 T2b/c186 (30.9)1612 (54.4)NA
 T3285 (47.4)1041 (35.1)1103 (34.7)
Positive surgical margin(s)242 (40.3)632 (21)926 (29.1)
RP Gleason sum
 2–5187 (31.1)324 (11)97 (3.1)
 6204 (33.9)988 (33.3)1024 (32.2)
 7161 (26.8)1591 (53.7)1875 (59)
 8–1049 (8.1)60 (2)182 (5.7)
Surgical procedure
 Perineal RP297 (49.4)0 (0)0 (0)
 Retropubic RP304 (50.6)2963 (100)3178 (100)
Lymph node dissection308 (51.2)1295 (43.7)1672 (52.6)
Comorbidities
 0549 (91.3)NANA
 133 (5.5)NANA
 ≥219 (3.2)NANA
Adjuvant radiotherapy94 (15.6)0 (0)72 (2.3)
Year of surgery
  1960–1987, pre-PSA era253 (42.1)0 (0)0 (0)
  ≥1988, PSA era348 (57.9)2963 (100)3178 (100)
PSA recurrence189 (31.4)602 (20)514 (16.2)
 PSA RFS at 5 y84.870.379.1
 PSA RFS at 10 y71.2NA67.7
 PSA RFS at 15 y61.1NA50.4
 PSA RFS survival at 20 y58NANA
Prostate cancer-unrelated mortality97 (16.1)NANA
Follow-up between RP and PSA recurrence or last follow-up, y
 Mean [median]11.6 [11.4]2.7 [2.1]3.4 [2.2]
 Range0.1–40.50.1–10.80.1–15.3

Statistical Analysis

The cohort of 601 evaluable men was used to develop and internally validate the postoperative BCR nomogram. Pathologic stage, surgical margin status, RP Gleason sum, surgical procedure type (perineal RP vs retropubic RP), PLND status (performed vs not performed), Charlson comorbidity index, and adjuvant radiotherapy status were used as predictors in univariate and multivariate Cox regression models that addressed BCR after RP. Adjuvant radiotherapy represented the only variable that was not known at the time of RP: Because it invariably was delivered within 1 to 3 months after RP, it was not modeled as a time-dependent variable.

All predictors were included in a multivariate model. Stepwise, backward variable removal was applied to the multivariate model with the intent of identifying the most accurate and the most parsimonious set of predictors. Proportional hazards assumptions were verified systematically for the final model by using the Grambsch-Therneau residual-based test.18 Because a proportion of patients who are at risk of BCR die as a result of other causes before they develop BCR, competing-risks regression was used to test the significance of BCR predictors after accounting for other-cause mortality, as described by Fine and Gray.19 Subsequently, the nomogram S-Plus function was used to derive the graphic representation of the Cox model, termed nomogram. To our knowledge, there are no commercially available statistical packages that allow the application of competing-risks regression within a nomogram setting. In consequence, the nomogram was based on Cox regression models. Because the risk of BCR decreases with increasing disease-free interval, we used the conditional survival approach14 to provide nomogram predictions on the basis of disease recurrence-free interval.

The nomogram was validated in 3 steps. Internal validation with 200 bootstrap resamples was applied to the multivariate Cox regression coefficients of the nomogram predictor variables to quantify their discriminant ability according to the area under the receiver operating characteristics curve (AUC). Because our data were censored, we relied on Harrell et al.'s modification of the AUC for censored data.20, 21 Under the condition of censoring, the AUC describes the probability that, given 2 randomly drawn patients, the patient who recurs first had a higher probability of disease recurrence.22 The second step of the nomogram validation consisted of a comparison between the nomogram-predicted and observed probability of BCR-free survival. This was plotted by using the val.surv S-Plus function devised for censored data (see Fig. 1B). An ideal plot corresponds to a 45-degree line, in which the predicted probability parallels the observed rate of BCR-free survival. In the third step of the validation process, we applied the nomogram to 2963 patients from Hamburg, Germany and to 3178 patients from Cleveland, Ohio who underwent retropubic RP between 1992 and 2005 and between 1988 and 2005, respectively. For both external validation cohorts, the nomogram-predicted probability was compared with actual follow-up, and the AUC was calculated for specific time points after RP. Because of follow-up time restrictions, the validation was limited to 5 years after RP in the Hamburg cohort and to 5 years, 10 years, and 15 years after RP in the Cleveland cohort. All analyses were performed using the S-Plus Professional software package (version 1; MathSoft Inc., Seattle, Wash), and statistical significance (P) was set at .05.

thumbnail image

Figure 1. Nomogram (A), calibration plot (B), and conditional probability plots (C-F) for prediction of the individual probability of biochemical recurrence (BCR) after radical prostatectomy (RP). For the calibration plot (B), the 45-degree line represents the ideal predictions. The x-axis indicates the nomogram-predicted BCR-free probability, and the y-axis indicates the observed BCR-free proportion of patients. Nomogram instructions: Locate the patient's value for pathologic T classification (pT). Draw a line straight up to the point axis to determine how many points toward the probability of BCR the patient receives for the value of pathologic T classification. Repeat the process for each additional variable. Sum the points for each of the predictors. Locate the value corresponding to the sum on the total point axis (A). Subsequently, choose the time point of interest after RP (eg, 5 years, 10 years, 15 years, or 20 years) (C-F). Within the specific time point panel, draw a line straight up from the total point's axis. Then, draw a horizontal line from a value on the y-axis that corresponds to the number of months without recurrence that elapsed between RP and the current time. Use the intersection of both lines to identify a slanted line that crosses it or passes next to it. Follow the slanted line down to the x-axis to determine the probability of remaining free of BCR at the prespecified number of years after RP.5, 10, 15, 20 For example, for a total of 153 risk points calculated with the nomogram (A), start locating this value on the total point axis in E. Then, draw a vertical line upward. This line intersects the x-axis, corresponding to the probability of remaining free of BCR at 15 years after RP, which ranges from 2.5% to 93%. If the calculation is made immediately after RP, then the probability of being BCR-free is 0.5 (50%). At 18 months after RP, the probability is virtually the same (50%), as indicated by the intersection of a horizontal line drawn from the 18 months without recurrence value on the y-axis. At 54 months, provided no recurrence because of RP, the horizontal line drawn from the value of 54 months on the y-axis now intersects the slanted line that originates from the 0.6 (60%) probability point on the x-axis. This means that the 15-year BCR-free probability improved from 50% to 60% based on 54 months without recurrence.

Download figure to PowerPoint

RESULTS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

The pathologic and treatment characteristics of the 601 assessable patients are listed in Table 1. The mean age at RP was 64 years (median, 64.8 years; range, 45.3–77.9 years). At the time of pathologic evaluation, 285 patients (47.4%) had pathologic T3 (pT3) disease, 49 patients (8.1%) had Gleason sum ≥8, and 242 patients (40.3%) had positive surgical margin status. PLND was performed in 308 patients (51.2%). Adjuvant radiotherapy was administered to 94 patients (15.6%). The mean follow-up from RP to either BCR or last follow-up was 11.6 years (median, 11.4 years; range, 0.1–40.5 years). BCR was recorded in 189 patients (31.4%). The median time to BCR was 32.5 years. Among the 601 patients, none developed local or distant recurrence before BCR. Finally, hormone therapy or salvage radiotherapy was never administered before documented BCR.

Figure 2A illustrates BCR-free survival at 5 years (84.8%), 10 years (71.2%), 15 years (61.1%), and 20 years and (58.6%). It is noteworthy that, only after 17 years of follow-up, the slope of the Kaplan-Meier curve flattened. Figure 2B illustrates the rate of other-cause mortality at 5 years (2.4%), 10 years (6.4%), 15 years (20.8%), and 20 years (33.5%). Figure 2 also illustrates the rate of BCR-free survival stratified according to the tested variables (Fig. 2C-J), in which more advanced pathologic T (pT) classification (P < .001), positive surgical margin status (P < .001), unfavorable RP Gleason sum (P < .001), retropubic RP (P = .008), completed PLND (P = .03), a comorbidity index ≥1 (P = .009), and delivery of adjuvant radiotherapy (P = .01) all were associated with a higher risk of BCR. Conversely, study period (pre-PSA era [1960–1987] vs PSA era [1988–1994]) was not related to differences in the observed BCR rates (P = .1).

thumbnail image

Figure 2. Biochemical recurrence-free survival (A) and other-cause mortality (B) at 5 years, 10 years, 15 years, and 20 years. The rate of biochemical recurrence was stratified according to pathologic stage (C), surgical margins status (D), pathologic Gleason sum (E), surgical procedure (F), presence or absence of lymph node dissection at radical prostatectomy (G), comorbidity status (H), adjuvant radiotherapy (I), and year of surgery (J). 95%CI indicates 95%confidence interval; pT, pathologic tumor classification.

Download figure to PowerPoint

The univariate Cox regression analysis replicated the relations described with the Kaplan-Meier survival curves and are listed in Table 2. Table 2 also provides results from the multivariate Cox regression models predicting BCR after RP. In the full multivariate model, RP Gleason sum (P < .001), pT classification (P = .01), positive surgical margin status (P = .007), and adjuvant radiotherapy (P = .03) represented independent predictors of BCR. Conversely, the type of prostatectomy (retropubic vs perineal; P = .2), PLND status (P = 1.0), and the extent of comorbidities (P = .3) failed to achieve independent predictor status. After stepwise, backward variable selection, only RP Gleason sum (P < .001), pT classification (P = .01), positive surgical margin status (P = .003), surgical procedure type (P = .007), and adjuvant radiotherapy delivery status (P = .03) remained in the final model.

Table 2. Univariate and Multivariate Analyses Predicting Biochemical Recurrence After Radical Prostatectomy
PredictorUnivariateMultivariate
Full modelReduced model
RR (95% CI)PRR (95% CI)PRR (95% CI)P
  1. RR indicates relative risk; 95% CI, 95% confidence interval; pT, pathologic tumor classification; PLND, pelvic lymph node dissection.

Gleason sum<.001<.001<.001
 6 vs 2–52.3 (1.5–3.7)<.0012.3 (1.4–3.6)<.0012.3 (1.5–3.7)<.001
 7 vs 2–55.0 (3.2–7.9)<.0014.0 (2.5–6.4)<.0014.0 (2.5–6.4)<.001
 8–10 vs 2–55.5 (3.2–9.6)<.0015.0 (2.8–8.7)<.0015.0 (2.8–8.8)<.001
pT classification<.001.01.01
 T2b/c vs T2a1.6 (0.9–2.7).081.4 (0.8–2.4).21.4 (0.8–2.4).2
 T3 vs T2a3.5 (2.2–5.5)<.0012.2 (1.3–3.6).0042.1 (1.3–3.7).004
Positive surgical margin2.5 (1.9–3.3)<.0011.7 (1.1–2.4).0071.7 (1.2–2.5).003
Surgical procedure (retropubic vs perineal)1.5 (1.1–2.0).0081.5 (0.8–3.0).21.5 (1.1–2.0).007
Adjuvant radiotherapy (yes vs no)1.6 (1.1–2.2).010.7 (0.4–1.0).030.6 (0.4–0.9).03
PLND status (yes vs no)1.4 (1.0–1.8).031.0 (0.5–1.9)1.0  
Comorbidities.04.3  
 1 vs 01.8 (1.0–3.1).031.4 (0.8–2.5).2  
 ≥2 vs 01.8 (0.9–3.6).11.5 (0.7–3.1).3  

Table 3 shows the univariate and multivariate competing-risks regression models that were developed on the cohort of 601 patients. In the multivariate competing-risks regression model, all variables that were included in the model were statistically significant predictors of BCR after RP after accounting for other-cause mortality except for adjuvant radiotherapy (P = .07).

Table 3. Univariate and Multivariate Competing Risks Regression Models for Predicting Biochemical Recurrence After Radical Prostatectomy and Accounting for Other-cause Mortality
PredictorUnivariateMultivariate
Full modelReduced model
RR (95%CI)PRR (95%CI)PRR (95%CI)P
  1. RR indicates relative risk; 95% CI, 95% confidence interval; pT, pathologic tumor classification; PLND, pelvic lymph node dissection.

Gleason sum
 6 vs 2–52.4 (1.5–3.8)<.0012.3 (1.5–3.7)<.0012.4 (1.5–3.8)<.001
 7 vs 2–54.8 (3.0–7.5)<.0013.8 (2.4–6.2)<.0013.8 (2.4–6.2)<.001
 8–10 vs 2–55.8 (3.1–10.6)<.0015.1 (2.8–9.4)<.0015.2 (2.8–9.5)<.001
pT classification
 T2b/c vs T2a1.5 (0.9–2.4).11.3 (0.8–2.2).31.4 (0.8–2.2).2
 T3 vs T2a3.3 (2.1–5.1)<.0012.0 (1.2–3.4).0072.1 (1.2–3.5).006
Positive surgical margin2.4 (1.8–3.2)<.0011.6 (1.1–2.3).011.6 (1.1–2.3).008
Surgical procedure (retropubic vs perineal)1.4 (1.0–1.9).021.3 (0.7–3.5).41.4 (1.1–1.9).01
Adjuvant radiotherapy (yes vs no)1.6 (1.1–2.2).0090.7 (0.4–1.0).080.7 (0.5–1.0).07
PLND status (yes vs no)1.3 (1.0–1.8).041.1 (0.6–2.2).7  
Comorbidities  
 1 vs 01.7 (1.0–2.9).051.3 (0.7–2.3).3  
 ≥2 vs 01.5 (0.8–2.9).21.1 (0.5–2.6).7  

The final model (Table 2) served as the basis for the multivariate nomogram (Fig. 1A). The nomogram predictions are shown for 5-year, 10-year, 15-year, and 20-years time points. At each time point, the prediction is accompanied by a graph that allows adjusting the BCR-free probability for the disease-free interval. The determination of the probability of BCR-free status is achieved in 2 steps. First, the nomogram axes are used to quantify the total number of risk points. Subsequently, the total points are applied to one of the time-specific scales (Fig. 1C-F). For example, 15-year predictions are shown on Figure 1E. When the predictions are calculated immediately after RP, a patient with 153 risk points has 50% probability of BCR-free survival at 15 years compared with 60% when the prediction is made 54 months after RP and 70% when the prediction is made 90 months after RP provided disease-free status is maintained at all time points.

After 200 bootstrap resamples of the nomogram regression coefficients, the accuracy of the BCR predictions was 79.3%, 77.2%, 79.7%, and 80.6% at 5 years, 10 years, 15 years, and 20 years, respectively. When the probability of BCR calculated by the nomogram was compared with the actual outcome of patients in the first validation set at 5 years after RP, the nomogram achieved 77.4% predictive accuracy. When the probability of BCR calculated by the nomogram was compared with the actual outcome of patients in the second validation set at 5 years, 10 years, and 15 years after RP, the nomogram achieved 77.9%, 79.4%, and 86.3% predictive accuracy, respectively.

Figure 1B shows the calibration plot of the newly developed nomogram predicting the individual probability of BCR after RP. The 45-degree line represents ideal predictions. The x-axis indicates the nomogram-predicted BCR-free probability, and the y-axis indicates the observed BCR-free proportion. It is noteworthy that the curve depicting the relation between predicted and observed BCR-free proportions closely approximates the ideal predictions, indicating excellent calibration.

DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES

One in 2 men who underwent RP before 1995 have developed or will develop BCR if follow-up is adequately long.9, 15 Pound et al. indicate that, among patients with BCR, 34% will progress to metastatic disease at 15 years of follow-up.9 Moreover, among men with BCR, 17% die of PCa if they are followed for up to 15 years.3, 10 Thus, BCR represents an important surrogate marker of progression to distant metastases and of PCa-specific mortality. Unfortunately, because of short follow-up, currently available studies may underestimate the risk of BCR.

To circumvent the lack of long-term BCR data, several investigators reported middle and long-term BCR data. In the series reported by Ward et al., 27% of patients with BCR demonstrated a rising PSA level ≥5 years after RP.8 The same trends were confirmed by others.5, 6, 9 These data indicate that patients are at continuous and non-negligible risk of long-term BCR, and a substantial proportion will recur biochemically beyond 10 years after RP.

On the basis of available data, late recurrences cannot be discounted as less important than early recurrences.8 However, recurrences that occur 10 years after RP cannot be predicted with any currently available prognostic tool. The nomograms provided by Kattan et al.13 and Stephenson et al.14 are limited to 7-year and 10-year predictions, respectively. Moreover, the limited follow-up of these models may lead to the belief that BCR risk is negligible beyond 10 years. Therefore, a tool capable of predicting long-term BCR, beyond 10 or 15 years, is needed.

To address this void, we decided to develop and internally validate a novel nomogram predicting the individual probability of BCR up to 20 years after RP. Our dataset consisted of 601 men who underwent either perineal RP or retropubic RP between 1960 and 1994. Because most of the men in our cohort were treated before the PSA era, preoperative serum PSA values were unavailable for a large proportion of these individuals. Therefore, we decided to omit preoperative PSA from consideration within the nomogram. The exclusion of preoperative PSA values prevented us from comparing our nomogram with the nomogram of either Kattan et al.13 or Stephenson et al.,14 both of which require preoperative PSA values.

Our data confirmed the importance of BCR beyond 10 years, because 27% of patients developed BCR ≥10 years after RP. At 5 years, 10 years, 15 years, and 20 years after RP, the respective BCR-free survival rates were 84.8%, 71.2%, 61.1%, and 58.6%. It is noteworthy that we did not observe any difference (log-rank P = .1) in the BCR rate according to the year of surgery when patients from the pre-PSA era (before 1988) were compared with patients from the PSA era (1988 or later).

Because many patients may die of other causes before BCR, we confirmed the statistical significance of nomogram predictors in competing-risks regression models. Except for adjuvant radiotherapy (P = .07), all variables that were included in the final model maintained their independent predictor status after accounting for other-cause mortality. Because the risk of BCR decreases with time, we complemented the nomogram predictions with the ability to adjust the BCR-free probability according to the disease-free interval.

After 200-bootstrap internal validation, the predictive accuracy of the nomogram was 79.3%, 77.2%, 79.7%, and 80.6% at 5 years, 10 years, 15 years, and 20 years, respectively. When the nomogram was applied to the first external validation cohort, 77.4% accuracy was recorded at 5 years. In the second validation cohort, the predictive accuracy of the nomogram was 77.9%, 79.4%, and 86.3% at 5 years, 10 years, and 15 years after RP, respectively. External validity could not be tested at 20 years because of insufficient follow-up in either cohort. It is interesting to note that the contemporary external validation cohorts differed drastically from the original modeling set. The main differences between the modeling and the external validation datasets related to years of surgery, pathologic grade and stage, positive surgical margin rate, and use of adjuvant radiotherapy. Moreover, the validation datasets exclusively relied on retropubic RPs. It also is worth noting that, in the multivariate model that was used for the nomogram, retropubic RP was associated with a higher rate of BCR (relative risk, 1.5; P = .007) than the perineal RP. It is conceivable that the retropubic approach was reserved for higher risk patients. However, the retrospective nature of the current study precluded valid conclusions regarding the true effect of the surgical approach. Despite all the considerable differences, thenomogram achieved very adequate accuracy at 5 years of follow-up, which validated its applicability in contemporary patients.

The performance of our nomogram is roughly comparable to the performance of the nomogram reported by Stephenson et al., which predicts with 79% to 81% accuracy up to 10 years after RP.14 The advantage of our nomogram relative to that of Stephenson et al.14 resides its the ability to predict BCR up to at least 15 years after RP instead of 10 years. The advantage of the Stephenson et al. model relies on superior sample size.14 Therefore, the nomogram reported by Stephenson et al.14 should remain the standard for 10-year predictions. Conversely, our model should be used when predictions are required beyond 10 years, or if predictions need to be made for men without an available preoperative PSA level, or for men who underwent RP before 1996. In consequence, both tools are complementary to one another.

Several limitations apply to our model. First, the sample size of our dataset was smaller than in the competing nomograms.13, 14 Second, the Gleason grading used in the current series has not been subjected to a review.23 However, the nomogram performed well in 2 contemporary cohorts in which contemporary criteria for Gleason grade assignment were used. It is interesting to note that the pathologic Gleason sums were not reviewed in the nomogram reported by Stephenson et al.14 Third, preoperative serum PSA levels were not available in the great majority of patients. Therefore, this variable could not be included. Fourth, our data did not allow distinguishing between pT3a and pT3b/pT3c disease: The latter is associated with a substantially worse prognosis and may help in distinguishing between patients who fail and those who do not. Even more important, we did not have information regarding lymph node involvement, which confers an even more unfavorable prognosis than pT3b/pT3c disease and may help further in accurately predicting BCR rates.24 Despite the omission of these important variables, our model's prognostic accuracy paralleled that of the nomogram reported by Stephenson et al. and of several other tools for predicting BCR after RP.13, 14, 22, 25, 26

In summary, patients who undergo RP remain at risk of BCR beyond 10 years after RP. Our nomogram distinguishes itself from other tools by its ability to accurately predict BCR up to at least 15 years after surgery. We recommend its use for risk stratification in men with ≥10 years of follow-up.

REFERENCES

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS
  5. DISCUSSION
  6. REFERENCES