First Nomogram Predicting the Probability of Lymph Node Involvement in Prostate Cancer Patients Undergoing Radioisotope Guided Sentinel Lymph Node Dissection

Pelvic lymph node dissection (PLND) is still the gold standard for lymph node (LN) staging in clinically localized prostate cancer (PCa). The diagnostic accuracy of available imaging procedures is quite inferior to the histological verification of LN metastases. The LN status is a crucial prognostic factor in PCa. Presence and extension of LN involvement (LNI) is associated with an increased risk of systemic dissemination and progression of the disease. A debate is currently underway on the positive therapeutic impact of PLND, especially in patients with minimal LNI [1,2].

Numerous nomograms based on preoperative variables have been developed to predict LNI in PCa and to select candidates for PLNDs. The goal is to identify low-risk LNI cases and hinder additional morbidity from a PLND. Without exception, these decision tools were based on conventional PLND techniques. Many of these algorithms and Partin tables [3] were based on series in which the patients underwent a limited degree of PLND (lPLND). The nomograms now available are based on ePLND [4,5,6], which means they account for the fact that LNI prevalence is directly related to the number of dissected LNs and extent of the PLND [7,8]. However, the rate of complications rises along with the number of LNs removed [9,10,11].

Similarly, for radioisotope guided sPLNDs, one can demonstrate a high staging accuracy accompanied by even lower morbidity [11,12,13]. The sentinel approach allows an individualized extension of LN dissection outside the boarders of ePLND too [14]. Presently, different tracers, such as the near-infrared fluorescent dye indocyanine green, are being tested to mark sentinel LNs (SLNs), especially in connection with robotic [15] and laparoscopic [16] radical prostatectomies (RPs). So far, there is no LNI nomogram based on a sPLND.

We hypothesized that preoperative parameters obtained in patients undergoing sPLNDs can accurately predict LNI in sPLND specimens. To test our hypothesis, we used data collected from men who had undergone a sPLND in combination with a radical retropubic prostatectomy (RRP). Multivariable logistic regression was used to calculate the probability of LNI.

A total of 1,325 consecutive patients with PCa were identified, who underwent sPLNDs in combination with RRPs carried out by 4 highly experienced surgeons, in a single center between January 2005 and April 2010. We excluded patients with incomplete clinical information for prostate specific antigen (PSA), clinical stage or biopsy Gleason score (n = 4, 0.3%). Furthermore, we also excluded patients who had undergone a transurethral resection or laser therapy of the prostate (n = 14, 1.1%) and cT4 tumors (n = 8, 0.6%). An additional 3 patients (0.2%) were also excluded, since no SLN could be detected by the gamma probe. The final sample comprised 1,296 patients.

The clinical stage was classified per the 2002 Union for International Cancer Control TNM staging system. PSA was measured using standard assays. The primary pretherapeutic PSA value was considered in patients who had undergone hormonal therapy prior to operative treatment (n = 12, 0.9%). Prostate biopsies were performed at our hospital, other hospitals and medical offices, which were then examined histopathologically by internal and external uropathology experienced pathologists. All patients were informed orally and in writing about a sPLND and RRP, and they signed a consent form.

Using ultrasound guidance, 99mTechnetium nanocolloid was transrectally injected into the prostate 24 h before the surgery [12]. Three injections were administered per prostate lobe. Activity attained about 100 MBq per lobe and total injection volume was about 1.2 ml. A few hours after injection, scintigraphy was carried out. The radioactivity of the LN was intraoperatively measured using 2 different gamma probe systems (C-Trak System, Care Wise, Morgan Hill, Calif., USA; Crystal Probe SG04, Crystal Photonics GmbH, Berlin, Germany). LNs identified as SLNs by the gamma probe were dissected. For surgical reasons, LNs other than SLNs directly adjoining and adhering to SLNs were also removed, if an in situ separation was not possible. Furthermore, if the SLNs are present in the obturator fossa area, the surrounding non-radioactive lymphatic tissue of the fossa was also dissected. However, the lymphatic tissue of the fossa was not resected, if no SLN existed in the fossa area.

All LNs were initially cut into 3-mm transverse sections, routinely processed and completely embedded in paraffin; sections of thickness 4-5 µm were stained with hematoxylin-eosin. Selected cases of serial sections were analyzed. An immunohistochemical study with a pancytokeratin antibody (AE1/AE3) was carried out to confirm or exclude metastatic spread in rare cases with inconclusive conventional histology.

Univariable and multivariable logistic regression models were carried out to test the association between preoperative tumor characteristics and the probability of LNI. The predictor variables were the preoperative PSA level categorized as 4, 4.1-10, 10.1-20 and >20 ng/ml; clinical T-category as T1, T2 and T3 and biopsy Gleason sum as 5-6, 7 and 8-10.

Regression coefficients were used to develop the nomogram that predicts the probability of LNI at a sPLND. Bootstrapping (9,999 replications) was applied to generate reliable 95% confidence intervals for the predicted probabilities and for internal validation. Predictive accuracy was quantified using the receiver operator characteristics of the AUC. The performance characteristics were evaluated using a calibration plot of predicted probabilities against observed LNI rates.

Statistical analyses were performed using the generalized linear model function of the open-source statistical software R (R Development Core Team 2008) [17].

Table 1 lists the summary of patient characteristics. By definition, details of the Gleason score in the surgical specimen could not be given for 12 patients who previously underwent hormonal treatment. The median number of LNs removed was 10 (interquartile range (IQR) 7-13), encompassing a median of 6 (IQR 4-8) SLNs. Overall, 17.8% of patients (n = 231) had LNI. The number of positive LNs ranged from 1 to 15 (median 2; IQR 1-3).

In the multivariate logistic regression analysis, all variables (pretherapeutic PSA, clinical T-category and biopsy Gleason sum) were significantly associated (p < 0.001) with LNI. The multivariate predictive accuracy (AUC) was 82%, under consideration of the 3 predictors. Univariate analysis also showed a significant (p < 0.001) association between each predictor and LNI. In the univariate predictive accuracy analysis, the biopsy Gleason sum was the most accurate predictor of LNI (74.5%), followed by the clinical T-category (69.3%) and the preoperative PSA value (68.9%). The results of the multivariate and univariate logistic regression analyses are detailed in table 2.

Figure 1 illustrates the nomogram tool in a graphical form as generated by the multivariate analysis. The probabilities for LNI, predicted by the multivariate regression analysis, ranged from 3% in low-risk to 88% in high-risk PCa patients. For example, the probability of LNI is 24% for patients with a cT1c tumor, a PSA value of 10 ≤ 20 and a Gleason sum of 7.

The calibration plot of predicted probabilities against observed LNI rates showed a high level of consistency between predicted and actual probabilities in low- and intermediate-predicted probability ranges. Variances from the ideal nomogram are shown in the high-predicted probability ranges (fig. 2).

There is general consensus that an ePLND performed on PCa patients achieves the highest staging accuracy. For sPLND, a high staging accuracy has been demonstrated too [12,13,18]. LNI predictor ePLND-based nomograms provide PCa patients a crucial basis to decide for or against a PLND [4,5,6]. Other predictive models are based on series of lPLNDs, and thereby, these models most likely underestimate the risk of a LNI [3,19]. It was possible to demonstrate that for a sPLND, the LNI rate was higher in a sentinel cohort than was expected from the European Association of Urology (EAU) guideline nomogram [20]. The validation of a corresponding sentinel-based nomogram is still pending. This study presents the first sPLND-based nomogram.

With an AUC of 82%, the sentinel nomogram presents a comparably accurate model for predicting LNI in patients with PCa. In various lPLND- or ePLND-based nomograms that use the same preoperative parameters as those of predictors of a LNI the reliabilities are 76-86% [3,4,6,19]. Despite the extended approach in these studies, the proportion of LN+ patients was significantly lower than in that of the sentinel series (table 3).

There is no consensus on the risk level of a LNI that would be the ideal cutoff for choosing a PLND in patients with PCa. For instance, the National Comprehensive Cancer Network deems a cutoff acceptable if it leads to waiving about 50% of the PLNDs prior to RP at the expense of proof or removal of LNMs in 12% of the cases with LNI [21]. The EAU guidelines advise the usage of an LNI nomogram-calculated probability of 5% as a cutoff to perform ePLND, which would allow the avoidance of unnecessary PLND in about 65% of patients at the cost of missing 12% of patients with LNI [5,22]. Table 4 presents a systematic analysis of a range of nomogram thresholds from 1 to 10% to help in the correct discrimination of patients with or without histologically confirmed LNI taking into account the sentinel model. The number of avoidable sPLNDs versus the number of potentially missed patients with LNI was quantified. Accordingly, a 7% threshold would be regarded as the most favorable cutoff. In our population of 1,296 patients, 406 patients (31.3%) were classified below this threshold. A voidance of sPLND in those 406 cases would have resulted in missing LNI in 7 patients or in 3% of all patients with histologically confirmed LNI. Therefore, approximately one-third of patients could be spared from sPLND. Considerable costs and patient discomfort could be saved.

In view of the low morbidity of sPLNDs in combination with the high sensitivity of proof of metastases [11,13], we question the ability to define a cutoff. One should also note that patients with minimal LNI especially appear to benefit from removal of LN metastases [23]. Finally, the sPLND nomogram offers PCa patients the first-ever opportunity to make an informative decision about the probability of the sPLND detecting LN metastases, and thereby allows them to weigh the pros and cons of going for a sPLND for themselves.

On the other hand, a high risk of positive LNs may discourage urologists offering a RP. However, increasing evidence suggests that RP and PLND improve survival in LN+ PCa [24]. Besides being a staging procedure, PLND may be curative, or at least beneficial, in a subset of patients with limited LNI [23,25]. A retrospective observational study has shown a dramatic improvement in cancer-specific survival and overall survival in favor of completed RP versus abandoned RP in patients who were found to be LN+ at the time of surgery [26]. These results suggest that RP may have a survival benefit and the discontinuation of RP in LN+ patients may not be justified [22] or that it is useful to perform RP in such cases. Furthermore, RP and PLND are important components of multimodal strategies for patients with LN+ PCa [22]. Due to these results, we consider the definition of an upper cutoff for PLND as not useful.

No consideration has yet been given to the percentage of positive cores as a predictor, as in other nomograms [3,5,6,19,27,28]. In the Update Nomogram of Briganti et al. [5], the percentage of positive cores is the most accurate predictor of LNI. This was also confirmed by external validation studies [29,30]. On the flip side, the sentinel nomogram reflects the reality of care. One expects better predictability on inclusion of the percentage of positive cores. Yet, the requirements, such as compliance with standards for biopsies and histopathological preparation, have not yet been established fully in most regions. Another limitation of the study arises from the limitations inherent to unicentric analysis. However, the staging accuracy and the rates of LNI patients detected by sPLNDs in the monitored sample compare well with data from other sPLND-experienced centers [13]. Ideally, one should also externally validate the reliability of the sentinel nomogram [29,30,31].

No clear statement can be made about the sensitivity of a sPLND, since no additional ePLNDs were performed. However, this was not the aim of our research. In a meta-analysis [18], the pooled detection rate of sPLND was 93.8% with a pooled sensitivity rate of 94%. In the largest study [13] conducted, falsely detected negative results (non-SLNMs found in the absence of SLNMs) were found in <6% of the cases.

One fundamental problem with this technique is that when LNs are fully metastasized or when the lymph pathways blocked, the afferent lymph will be directed to other LNs/non-SLNs [32]. These nodes will not be positive on SLN imaging, resulting in false-negative findings. The false-negative rate was shown to correlate with the Gleason score [13]. Patients with a high-risk disease could thus have both positive SLNs and positive non-SLNs [33]. If the goal in such cases is to remove all pelvic LN metastases high-risk patients have the option of undergoing a combination of sPLND and ePLND. As such, the possibility of an ePLND overlooking a part of the LN metastases, possibly in the pre-sacral region, is overcome by being able to detect it through the sPLND. Reportedly, Joniau et al. [14] did not detect 13% of metastatic LNs by applying only an ePLND.

For radioisotope guided sPLNDs, one can demonstrate a high staging accuracy accompanied by even lower morbidity. We have developed the first nomogram to predict the probability of LNI in patients undergoing a sPLND at a RP. The first sentinel nomogram demonstrates a high degree of accuracy. This means that a nomogram can, for the first time, support clinicians and patients in making a key decision on whether to go for a sPLND. Compared with the ePLND-based nomograms, the higher rate of LN+ patients detected underpins the sensitivity of the sPLND. An external validation of the sentinel nomogram is still pending.

None.