Yonsei nomogram to predict lymph node invasion in Asian men with prostate cancer during robotic era




  • To develop a novel nomogram to predict lymph node invasion (LNI) in Asian men undergoing radical prostatectomy (RP) and pelvic LN dissection (PLND) for localised prostate cancer.

Patients and Methods

  • The patient cohort included 541 patients who underwent robot-assisted RP and PLND by a single surgeon between January 2008 and December 2011.
  • Patients with dissection of <10 LNs, prostate-specific antigen (PSA) levels of >50 ng/mL, incomplete biopsy data, and treatment with neoadjuvant therapy were excluded.


  • The median (interquartile range) number of LNs removed was 17 (14–22) and 45 patients (8.3%) had LN metastases.
  • On multivariate logistic regression analysis, PSA level, clinical stage and Gleason score were independent predictors of LNI.
  • The bootstrap corrected area under curve of the model incorporating PSA level, clinical stage, and biopsy Gleason score was 0.883.
  • With a cutoff value of 4%, PLND could be omitted in 326 patients (60.2%), missing only two patients (4.4%) with LNI. The sensitivity, specificity, positive predictive value and negative predictive value were 95.6%, 65.3%, 20.0% and 99.4%, respectively.


  • We report a nomogram to predict LNI in Asian men with prostate cancer. The model demonstrated high accuracy and could be used for counselling patients and the selection of candidates for PLND.

area under curve


interquartile range


lymph node (invasion)


pelvic LN dissection


radical prostatectomy


prostate-specific antigen


With the advent of PSA testing, early detection of prostate cancer has resulted in stage migration and a decreased incidence of lymph node (LN) metastases. However, LN metastases still represent an adverse prognostic factor for disease progression and survival [1, 2]. Unfortunately, recent sophisticated imaging procedures have had limited success in accurately identifying LN involvement [3, 4] and pelvic LN dissection (PLND) is still the most accurate method for determining LN staging.

Nevertheless, the indication and optimal extent of PLND remains debatable. No consistent conclusion has been reached about the therapeutic benefit of PLND, and the impact of adjuvant hormonal treatment in patients with LN invasion (LNI) after radical prostatectomy (RP) is also controversial [5-7]. Moreover, PLND is associated with increased morbidity and longer operating times, and there has been a significant decline in the use of PLND following the introduction of minimally invasive surgery in prostate cancer [8].

Current guidelines recommend performing PLND based on the estimated risk of LNI [9, 10] and several nomograms have been created to better predict this risk [11, 12]. However, the accuracy of these predictive models may be influenced by several factors. The characteristics of the study population are crucial in the performance of nomograms and ethnic differences in the behaviour of prostate cancer have been well documented [13, 14]. There are also nomograms in Asian men to predict pathological staging of prostate cancer [15-17]. However, these nomograms were not constructed exclusively for the prediction of LN metastases and PLND was performed to a limited extent in most cases.

In the present study, we aimed to develop a nomogram to predict LNI after RP in Asian men with localised prostate cancer and evaluated the performance characteristics of the nomogram. We consistently performed PLND in all patients undergoing robot-assisted RP during the study period and the median number of LNs removed was 17.

Patients and Methods

From January 2008 to December 2011, 772 consecutive patients with prostate cancer underwent robot-assisted RP and PLND. All surgeries were performed by a single surgeon using the da Vinci Surgical System (Intuitive Surgical, Sunnyvale, CA, USA). All data were collected prospectively using an electronic database, and we analysed the database after Institutional Review Board approval (4-2012-0819). In all, 38 patients with neoadjuvant hormonal treatment or a preoperative PSA level of >50 ng/mL were excluded. We also excluded 134 patients with <10 LNs removed and 59 patients with incomplete biopsy information. Thus, the final study cohort included 541 patients. Preoperative variables, i.e. preoperative PSA level, clinical stage, biopsy information including Gleason score, the total numbers of biopsied cores and positive cores were evaluated. PSA level was measured by electrochemiluminescence immunoassay using the Cobas e601 analyzer (Roche Diagnostics, Indianapolis, IN, USA). Clinical stage was assessed using the 2009 American Joint Committee on Cancer TNM classification system [18].

All PLNDs were performed before RP via a transperitoneal approach. The region of PLND included the intrapelvic area in all patients, and extended up to the ureteric iliac crossing in 240 patients, at the surgeon's discretion. Dissection of the external iliac packet was limited by the lateral border of the external iliac artery and inferiorly by the node of Cloquet. Dissection was extended cephalad to the common iliac bifurcation. The internal iliac artery was identified and nodal tissues around the internal iliac were removed. Lymphatic tissues around the obturator fossa were also removed, sparing the obturator nerve. LN specimens were submitted for pathological analysis separately. The LN specimens were fixed in 10% neural buffered formalin and embedded in a paraffin block. Slides were then stained with haematoxylin and eosin and examined microscopically. All LN specimens were examined by a single genitourinary pathologist with >15 years' experience. The total number of LNs removed and the presence of LNI by prostate cancer were examined.

Quantitative variables were compared using the Student's t-test and qualitative variables were compared using the chi-squared test or Fisher's exact test. Univariate and multivariate logistic regression analysis were conducted to identify preoperative variables that predict LNI after RP and PLND. Multivariate logistic model regression coefficients were used to generate a nomogram predicting the probability of LNI at PLND. The predictive accuracy of the model was quantified by the area under curve (AUC). The calibration plots were generated using 200 bootstrap samples for the internal validation and overfitting correction, and the average value of these 200 model performance indices was considered the bootstrap-corrected AUC. For each cutoff value, sensitivity, specificity, the number of patients below the cutoff value and the number of patients with false-negatives were calculated. Decision curve analysis was also performed to evaluate the net benefit of the predictive model.


The patients' characteristics are given in Table 1. Of 541 patients, 45 (8.3%) had LNI. Patients with LNI had higher PSA levels, more positive cores, and a higher clinical stage (all P < 0.001) than those without LNI. There were no significant between-group differences in age or the total number of biopsy cores. The mean number of LNs removed was 18.8 (median 17). The mean number of LNs removed was higher in patients with LNI than those without (21.0 vs 18.6, P = 0.012). The median (interquartile range, IQR) time for PLND was 36 (24–43) min. On final pathology, 84 patients (40.7%, 84/206) with biopsy Gleason score 6 and 23 patients (12.0%, 23/191) with biopsy Gleason score 7 were upgraded. Of 485 patients with clinically organ-confined disease, 142 (29.3%) were upstaged to non-organ confined prostate cancer. For complications presumably associated with PLND, clinically significant lymphoceles requiring percutaneous drainage were observed in 12 patients (2.2%) and six (1.1%) had neuropraxia of the obturator nerve postoperatively. Ureteric injury occurred in one patient.

Table 1. Perioperative characteristics of the study cohort
 Overall (n = 541)LN– (n = 496)LN+ (n = 45)P
Age, years   0.071
Mean (sd)64.5 (7.2)64.3 (7.3)66.4 (6.5) 
Median (IQR)65 (60–70)65 (60–70)67 (63–69) 
PSA, ng/mL   <0.001
Mean (sd)11.00 (9.11)10.07 (8.06)21.21 (13.12) 
Median (IQR)7.89 (5.27–13.52)7.41 (5.12–12.16)16.96 (9.87–33.01) 
N (%):    
Clinical stage:   <0.001
T1374 (69.1)365 (73.6)9 (20.0) 
T2111 (20.5)95 (19.1)16 (35.6) 
T356 (10.4)36 (7.3)20 (44.4) 
Biopsy Gleason score:   <0.001
≤6206 (38.1)203 (40.9)3 (6.7) 
7191 (35.3)180 (36.3)11 (24.4) 
≥8144 (26.6)113 (22.8)31 (68.9) 
No. total biopsied cores:   0.623
Mean (sd)11.4 (3.2)11.4 (3.2)11.2 (3.3) 
Median (IQR)12 (10–12)12 (10–12)12 (10–12) 
No. positive cores   <0.001
Mean (sd)3.7 (2.8)3.5 (2.7)5.9 (3.4) 
Median (IQR)3 (2–5)3 (1–5)5 (3–8) 
Percentage positive cores   <0.001
Mean (sd)33.8 (24.7)31.8 (23.2)55.2 (30.2) 
Median (IQR)25 (17–50)25 (17–46)50 (29–85) 
No. LN s removed   0.012
Mean (sd)18.8 (6.2)18.6 (6.1)21.0 (7.8) 
Median (IQR)17 (14–22)17 (14–22)20 (15–24) 

On univariate logistic regression analysis, PSA level, clinical stage, biopsy Gleason score and percentage of positive cores were significantly associated with LNI. On multivariate analysis, PSA level, clinical stage and biopsy Gleason score (all P < 0.001) were independent predictors of LNI, whereas percentage of positive cores lost its statistical significance after controlling for other variables (Table 2). The bootstrap corrected AUC of the model incorporating PSA level, clinical stage, and biopsy Gleason score was 0.883 and the nomogram was generated based on these variables (Fig. 1). The calibration plot is shown in Fig. 2 and the newly developed nomogram closely follows the ideal line.

Figure 1.

Nomogram to predict the probability of LNI.

Figure 2.

Calibration plot of the nomogram. The 45 ° line represents the ideal nomogram in which the predicted and actual probabilities are identical. The dotted line indicates the apparent accuracy of the nomogram. The solid line indicates nomogram performance with bootstrap-correction.

Table 2. Univariate and multivariate logistic regression analysis for prediction of LNI
OR (95% CI)POR (95% CI)P
PSA1.08 (1.06–1.12)<0.0011.06 (1.03–1.09)<0.001
Clinical stage (≥ T2 vs T1)11.14 (5.22–23.76)<0.0015.62 (2.51–12.58)<0.001
Biopsy Gleason score 0.001 0.001
≤61 1 
74.13 (1.13–15.05)0.0312.69 (0.70–10.32) 
≥818.56 (5.55–62.07)<0.0017.91 (2.22–28.11) 
Percentage positive cores1.03 (1.02–1.04)<0.001Not applicable

Using a 4% probability of LNI as the cutoff value for pursing PLND, 326 (60.2%) patients would have been spared PLND, missing two patients (0.6%) who harboured LN metastases. With the cutoff value of 4%, the sensitivity, specificity, positive predictive value and negative predictive value were 95.6%, 65.3%, 20.0% and 99.4%, respectively (Table 3). Figure 3 shows the decision analysis curve showing the net benefit associated with the use of the nomogram. At 4% of probability, the net benefit was 0.066. This net benefit value means that compared with assuming that all patients are negative for LNI, using a model with a 4% cutoff value results in a net 6.6 true-positive results per 100 patients and avoids performing PLND in patients without LNI. If we perform PLND in all patients, the calculated net benefit is 0.045 and the net reduction with this model was 50. This net reduction indicates that the model leads to the equivalent of 50% fewer patients without actual LNI undergoing unnecessary PLND, with no increase in the number of patients with LNI left untreated.

Figure 3.

Decision curve for a nomogram to predict LNI. Dotted line: prediction model. Red line: assuming all patients have LNI, Green line: assuming no patients have LNI. The graph gives the expected net benefit per patients relative to not performing PLND in any patients (‘treat none’).

Table 3. Analyses of nomogram-derived cutoff values to determine whether to perform PLND
Probability of LNI, cutoff %No. patients not recommendedMissing LN (+) with cutoffSensitivitySpecificity
 1161 (29.7)0 (0)10032.5
 2257 (47.5)0 (0)10051.6
 3294 (54.3)2 (4.4)95.658.7
 4326 (60.2)2 (4.4)95.665.3
 5353 (65.2)6 (13.3)86.770.0
 6366 (67.6)7 (15.6)84.472.2
 7379 (70.1)7 (15.6)84.475.0
 8397 (73.4)8 (17.8)82.278.4
 9420 (77.6)10 (22.2)80.081.3
10429 (79.3)11 (24.4)80.081.5


In the present study, we developed a nomogram for the prediction of LN metastases in patients undergoing robot-assisted RP for prostate cancer. The nomogram was based on preoperative PSA level, clinical stage and biopsy Gleason score, which demonstrated high accuracy when internally validated. Using a cutoff value of 4%, ≈60% of patients would be spared PLND.

Several studies have tried to predict LNI in patients with localised prostate cancer using preoperative variables [11, 12, 19, 20]. Godoy et al. [12] recently updated their nomogram to include only patients who had undergone extended PLND and showed improved calibration with high discriminative accuracy. Briganti et al. [21] showed that the information derived from biopsies, specifically the percentage of positive biopsy cores, improved the ability to predict LNI when incorporated as a covariate in their nomogram. The updated Briganti nomogram was recently validated in a different European cohort and showed good performance characteristics [11, 22]. However, the characteristics of study cohorts affect the performance of the predictive model. Considering that all of these nomograms were developed in Western populations, they might not be applicable in Asian populations.

Nomograms are most useful when patients closely resemble the population from which the nomogram was derived and clinician should decide whether the predictive model is applicable to their patient population, as differences in patient characteristics may undermine the accuracy of the predictive model [23, 24]. Thus, existing nomograms based on data from Western populations might show worse predictive accuracy when applied to Asian men. Previous studies have noted racial differences in clinical characteristics and prostate cancer mortalities between Whites and Blacks [25, 26]. Reviewing the literature, Asian men also seem to have unique characteristics in prostate cancer; a significant proportion of high-grade disease was noted in studies of Asian countries, i.e. Korea, China and Japan [13, 27, 28]. Even in patients who underwent radiotherapy, Asian men had twice the percentage of Gleason score ≥8 disease compared with non-Asian men [29]. We think that the present model might have a better predictive accuracy than pre-existing Western models in Asian populations or even in Asian men in the West.

For nomograms in Asian populations, several investigators have developed predictive models for pathological stage of clinically localised prostate cancer [15, 17, 30, 31]. Two of these nomograms were able to predict LN metastases with relatively high accuracy [15, 31]. However, these models were developed using patients undergoing limited PLND, and the information such as LN yield or the characteristics of patients with LN metastases was not available. Moreover, their study cohort did not consist exclusively of contemporary patients. In the present study, we included only contemporary patients who underwent RP between 2008 and 2011. Further, PLND including intrapelvic area LN dissection was performed consistently during our study period and patients with <10 LNs were excluded from the analysis. The median number of LNs removed was 17 and the rate of LN metastases was 8.3% in the present study, which were comparable to other nomogram series based on extended PLND [11, 12]. Thus, the present nomogram would predict the real incidence of LN metastases in a contemporary population more accurately than previous nomograms in Asian men.

Most existing nomograms use readily available preoperative clinical variables, and Briganti et al. [21] emphasised the importance of the percentage of positive biopsy cores for predicting LNI. In the present results, the percentage of positive biopsy core was significantly associated with LNI in univariate analysis. However, after adjusting for PSA level or biopsy Gleason score, the percentage of positive biopsy core lost its predictive significance. Also, we tested the predictive accuracy of the model with various coding methods of preoperative variables: PSA as categorical variables and clinical stage or biopsy Gleason score coded with various stratifications (results not shown). However, the presented nomogram showed the highest discriminative accuracy and better calibration than the other tested models and was adopted as the preferred predictive model. Although comparison between predictive models with a different dataset may cause problems with interpretation, the present nomogram had similarly high accuracy compared with other nomogram series [11, 12, 19].

No general consensus has been reached on how to decide the optimal nomogram cutoff value to recommend PLND. Existing guidelines and nomograms suggest a cutoff value which allows ≈50% of patients to be spared PLND while minimising missing patients with LNI [9, 11, 19]. In the present results, using a cutoff value of 4% showed similar results with sparing PLND in 60.2% of patients. The two patients who were false-negatives both had Gleason score 3+4 and a single positive LN. However, the decision to perform PLND according to the probability of LNI is actually considering PLND only as a staging procedure. Besides being a staging procedure, the therapeutic role of PLND has been also suggested. While results from the Mayo Clinic and Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) databases did not show any benefits of PLND on biochemical outcomes [32, 33], other large series showed improved biochemical or cancer-specific survivals with increased number of LNs removed [34, 35]. A growing body of evidence suggests the therapeutic benefits of PLND, possibly from the elimination of micrometastases. Indeed, a recent randomised study from China reported improved biochemical outcomes in patients with intermediate- or high-risk prostate cancers who were treated with extended PLND [36].

Recently, there has been a significant decline in the use of PLND [8]. Several reasons could be put forward to explain this trend. The increased use of robotic techniques in RP may have attributed to this trend; although technical feasibility and adequate safety have been shown in recent studies [37, 38], PLND was not typically performed in the early dissemination of robotic RP. In addition, there is a lack of evidence for a definite therapeutic benefit of PLND and the role of adjuvant hormonal treatment in patients with LN metastases is also controversial [5-7]. A low incidence of LN metastases due to stage migration may have also contributed. Nevertheless, it is necessary to evaluate the risk of LN metastases and it would be against the oncological principle to arbitrarily omit PLND.

The present study is not without limitations. The study is not a multi-institutional study, thus the cohort may not be representative of all Asian men. Also, this is not a prospective study and the templates for PLND were not the same in all patients, with more extended PLND templates in some patients. On the other hand, nodal tissues around the internal iliac area might not be completely dissected during the surgeon's learning curve or may be sent together as one package with the obturator LNs in some patients. Thus, we used LN counts as a surrogate for defining an adequate PLND and this might have skewed the results. Nevertheless, we think that the present results are valuable, as we consistently performed PLND during the study period and the number of LNs removed was similar to other nomogram series based on extended PLND [11, 12]. Although internal validation using bootstrap resampling demonstrated excellent performance, external validation is required for the application of the present nomogram to clinical practice. However, to our knowledge, our dataset is unique in Asian men regarding the performance of PLND. Most series are based on limited PLND, which might not be applicable for external validation of our nomogram.

In the present study, we developed a new highly accurate nomogram predicting the probability of LNI in contemporary Asian men who underwent PLND during robot-assisted RP. This model is easily applicable using routinely available preoperative variables and can be used for counselling patients and for the selection of candidates for PLND.


This study was supported by a Research Foundation of Korea (NRF) grant funded by the Korean government (MEST) (2011-0029348).

This study was supported by a faculty research grant of Yonsei University College of Medicine for 2012 (6-2012-0181).

Conflict of Interest

None declared.