Preoperative Nomogram Predicting the 10-Year Probability of Prostate Cancer Recurrence After Radical Prostatectomy

Abstract

An existing preoperative nomogram predicts the probability of prostate cancer recurrence, defined by prostate-specific antigen (PSA), at 5 years after radical prostatectomy based on clinical stage, serum PSA, and biopsy Gleason grade. In an updated and enhanced nomogram, we have extended the predictions to 10 years, added the prognostic information of systematic biopsy results, and enabled the predictions to be adjusted for the year of surgery. Cox regression analysis was used to model the clinical information for 1978 patients treated by two high-volume surgeons from our institution. The nomogram was externally validated on an independent cohort of 1545 patients with a concordance index of 0.79 and was well calibrated with respect to observed outcome. The inclusion of the number of positive and negative biopsy cores enhanced the predictive accuracy of the model. Thus, a new preoperative nomogram provides robust predictions of prostate cancer recurrence up to 10 years after radical prostatectomy.

Predictive models for recurrence of prostate cancer after definitive therapy are essential for patient counseling, the rational application of neoadjuvant therapy, and clinical trial design. In 1998, we developed a nomogram based on serum prostate-specific antigen (PSA), clinical stage, and biopsy Gleason grade that calculates the 5-year freedom from PSA-defined progression after radical prostatectomy (1). The nomogram has been independently validated in diverse patient populations (2-4). Currently, it is the most widely used disease-specific prediction tool in oncology (5). Nomograms also exist for prostate cancer recurrence after external-beam radiotherapy and brachytherapy (6, 7).

Despite the robust prognostic information contained within the nomogram, it has several limitations. The 5-year progression-free probability overestimates a man's probability of long-term cancer control because a substantial number of men will experience disease recurrence after maintaining an undetectable PSA level for 5 years or more after radical prostatectomy (8). Since 1998, an association between the extent of cancer in systematic prostate biopsy specimens and disease recurrence has been reported (9) . Last, widespread PSA screening has resulted in a stage migration such that patients of a similar grade and stage treated recently have an improved prognosis compared with longer ago (8, 10, 11). We have addressed these limitations in an updated and enhanced model by extending the predictions to 10 years, by including systematic biopsy results, and by enabling the predictions to be adjusted for the year of surgery. The 10-year progression-free probability more closely estimates the likelihood of cure because patients with an undetectable PSA level for 10 years after radical prostatectomy have a long-term progression rate of only 3% (8).

Multivariable Cox proportional hazards regression analysis was used to model the clinical information for patients treated by radical prostatectomy alone for tumor stage 1C–3 prostate cancer by two high-volume surgeons at Baylor College of Medicine (n = 1086) and Memorial Sloan-Kettering Cancer Center (n = 1134) between 1983 and 2002 (12). Complete information was available in 1978 patients pertaining to PSA levels, biopsy Gleason grade, 1992 American Joint Committee on Cancer clinical stage (12), and systemic biopsy results (Table 1). An external validation cohort consisted of 1545 patients who were treated by other surgeons at our institution during this time. The schedule by which patients were followed postoperatively has been described previously (8). Disease progression was defined as a serum PSA value of 0.4 ng/mL or greater (confirmed by a second PSA value higher than the first by any amount), secondary therapy, clinical recurrence, or aborted radical prostatectomy for lymph node metastases (13).

Table 1.

Clinical characteristics of patients in the modeling and validation cohorts^*

Patient characteristics	Modeling cohort, N = 1978	Validation cohort, N = 1545
Year of surgery, n (%)
1987–1992	294 (15)	40 (3)
1993–1998	649 (33)	794 (51)
1999–2003	1036 (52)	711 (46)
Median preoperative PSA, ng/mL (IQR)	6.1 (4.4–9.0)	6.6 (4.7–9.7)
1992 AJCC clinical stage, n (%)
T1C	803 (41)	809 (52)
T2A	509 (26)	299 (19)
T2B	335 (17)	307 (20)
T2C	244 (12)	113 (7)
T3	88 (4)	17 (1)
Biopsy Gleason grade, n (%)
Gleason 2–6	1348 (68)	1024 (66)
Gleason 7 (3+4)	397 (20)	295 (19)
Gleason 7 (4+3)	130 (7)	130 (8)
Gleason 8–10	104 (5)	96 (6)
Median positive biopsy cores (IQR)	2 (2–4)	2 (1–3)
Median total biopsy cores (IQR)	6 (6–9)	7 (6–10)
Median follow-up, mo (IQR)	54 (30–96)	68 (41–103)
Follow-up >5/>10 years, n (%)	976 (49)/391 (18)	1202 (78)/361 (23)
Lost to follow-up at 5/10 years, n (%)	252 (13)/321 (16)	297 (19)/307 (20)
Prostate cancer-specific death, n (%)	43 (2.2)	24 (1.6)

^*IQR = interquartile range; AJCC = American Joint Committee on Cancer.

Internal and external validation of the nomogram was performed using methods described previously (1). Predictive accuracy was assessed using the concordance (c-) index (discrimination) (14) and a visual inspection of the plots comparing the predicted probability of recurrence with the observed outcome (calibration). Internal validation was performed using bootstrapping analysis (15). External validation was performed by applying the nomogram to patients in the validation cohort. All statistical analyses were two-sided and were performed using S-Plus software (PC version 3.3, Redmond, WA) with additional functions (called Design) (15).

Disease progression was observed in 220 patients in the modeling cohort, and the 10-year progression-free probability was 77% (95% confidence interval = 73% to 80%). A nomogram containing PSA (P≤.001), clinical stage (P≤.001), primary (P≤.001) and secondary Gleason grade (P≤.001), year of surgery (P≤.001), and number of positive (P = .25) and negative (P = .094) biopsy cores had c-indices of 0.76 and 0.79 in internal and external validation, respectively. The nomogram estimates the 1- to 10-year progression-free probability after radical prostatectomy alone (Fig. 1, A). A model that did not include systematic biopsy results had an inferior predictive accuracy in external validation (c-index = 0.77). The c-index value, which is similar to the area under the receiver operating characteristic curve, indicates that the nomogram performs slightly better than midway between a perfect model and a coin flip (c-indices = 1.0 and 0.5, respectively).

Fig. 1.

A) Preoperative nomogram estimating the 1- to 10-year progression-free probability after radical prostatectomy alone. B) Calibration plot of the nomogram in external validation. The 45° line represents an ideal model in which estimates of recurrence are perfectly calibrated with outcome. Vertical bars are 95% confidence intervals for quintiles in the validation set.

Instructions for physician: Locate patient's preoperative prostate-specific antigen (PSA) on the Preoperative PSA axis. Draw a straight line down to the Points axis to determine how many points toward disease recurrence that patient receives. Repeat this process for each of the remaining variable axes. Sum the points for each predictor and locate this sum on the Total Points axis. Draw a line straight up from the Total Points axis until it intersects with the horizontal line drawn from Months from Surgery, corresponding to the time point in the future within which the risk of recurrence is wished to be calculated. The progression-free probability can be estimated at 12 to 120 months after radical prostatectomy. The slanted vertical line that crosses this intersection point corresponds to the calculated progression-free probability within that point in time after radical prostatectomy.

Instructions for patient: Mr. X, if we had 100 men exactly like you, we would predict <predicted probability from nomogram×100> percent to remain free of disease progression at <specified months from surgery> following radical prostatectomy alone.

The model exhibited good calibration across the spectrum of predictions in internal validation (data not shown) but exhibited some optimism in external validation; patients with an 8-year progression-free probability between 70% and 85% had an observed rate of freedom from progression of 57%–72% (Fig. 1, B). Part of the optimism may be explained by the fact that the model was developed on patients treated by high-volume surgeons and that the validation cohort was managed by both high- and low-volume surgeons. We have previously reported that positive surgical margins, which are associated with a threefold increased risk of recurrence, are influenced by surgeon volume and individual technique (16, 17). Surgeon volume may also be an independent predictor of recurrence (18).

An added feature of the nomogram is the ability to estimate the probability of recurrence at any point in time from 1 to 10 years after radical prostatectomy. The ability to predict the risk of early recurrence may be important for neoadjuvant treatment strategies, because disease recurrence within 2–3 years of radical prostatectomy is associated with an increased risk of metastasis progression and cancer-specific mortality (19, 20).

The nomogram predictions are adjusted for year of surgery, enabling the model to account for the improved prognosis of patients treated more recently. Of note, the nomogram assumes the patient is treated in 2003 and thus, theoretically, its accuracy for patients treated after 2003 is uncertain. However, we previously demonstrated that the improved prognosis associated with year of surgery is most pronounced for the years prior to 1998 and the hazard ratio is similar for the years 2000–2003 (10). Thus, the nomogram predictions are likely to be valid for patients treated after 2003. The current nomogram is most appropriate to use in regions where PSA screening is widespread (21). The original nomogram may be better suited for use in non-PSA-screened patients given that it has been extensively validated in these populations (1, 2).

In a previous study, the percentage of positive biopsy cores did not improve the accuracy of the original nomogram (22). The addition of the number of positive and negative biopsy cores to the current nomogram modestly increased its predictive accuracy. Neither parameter was statistically significantly associated with recurrence in the model, but statistical significance is not a prerequisite for improved outcome prediction (23). We previously demonstrated that the length of cancer and noncancer in biopsy cores substantially enhanced the predictive accuracy of a nomogram for indolent prostate cancer (24). Incorporation of these parameters into future nomograms is anticipated.