CYtCUPAVa

O

M

R

C

Pctnc

FnC

aS

©A

Adult Urology

ritical Assessment of Ideal Nodalield at Pelvic Lymphadenectomyo Accurately Diagnose Prostateancer Nodal Metastasis in Patientsndergoing Radical Retropubicrostatectomy

lberto Briganti, Felix K.-H. Chun, Andrea Salonia, Andrea Gallina, Giuseppe Zanni,incenzo Scattoni, Luc Valiquette, Patrizio Rigatti, Francesco Montorsi,nd Pierre I. Karakiewicz

BJECTIVES To study the relation between the number of removed and examined lymph nodes at pelviclymph node dissection and the rate of lymph node invasion (LNI).

ETHODS A total of 858 patients aged 45 to 85 years were predominantly treated with extended pelviclymph node dissection before radical retropubic prostatectomy. The pretreatment prostate-specific antigen level was 0.24 to 49.9 ng/mL (median 5.8). Most lesions were Stage T1c (55.2%)or T2 (40.7%), with a biopsy Gleason sum of 6 or less (62.2%) or 7 (25.1%). Receiver operatingcharacteristic curve coordinates were used to determine the probability of finding LNI accordingto the number of removed and examined lymph nodes. Moreover, the association between thenumber of removed lymph nodes and LNI was tested in univariate and multivariate logisticregression models.

ESULTS From 2 to 40 nodes (mean 15, median 14) were removed and examined, and 88 patients (10.3%)had LNI. The LNI rate increased with the number of removed nodes (P �0.001): 2 to 10 nodesremoved, 5.6% LNI rate; 10 to 14 nodes removed, 8.6% LNI rate; 15 to 19 removed, 10.2% LNIrate; and 20 to 40 removed, 17.6% LNI rate. On multivariate analysis, the number of examinednodes predicted for LNI (P �0.001), after accounting for prostate-specific antigen level, clinicalstage, and biopsy Gleason sum. The receiver operating characteristic coordinate plot indicatedthat the removal of 28 nodes yielded a 90% ability to detect LNI. Conversely, the assessment of10 or fewer nodes was associated with a virtually zero probability of finding LNI.

ONCLUSIONS We have provided a critical assessment of the concept that the nodal yield at pelvic lymph nodedissection is closely associated with the rate of LNI. UROLOGY 69: 147–151, 2007. © 2007 Elsevier

Inc.

beonsdntn

M

F

elvic lymphadenectomy (PLND) represents an es-sential staging procedure for patients undergoingradical prostatectomy for localized prostate can-

er.1 The rate of nodal metastases has ranged from 1.1%o 26%.2–8 Moreover, more extensive PLND might beecessary to detect occult lymph nodes metastases, be-ause the prevalence of lymph node invasion (LNI) could

rom the Department of Urology, Vita-Salute University, Milan, Italy; Cancer Prog-ostics and Health Outcomes Unit, University of Montreal, Montreal, Quebec,anada; Department of Urology, University of Hamburg, Hamburg, GermanyReprint requests: Pierre I. Karakiewicz, M.D., F.R.C.S.C., Cancer Prognostics

nd Health Outcome Unit, University of Montreal Heath Center (CHUM), 1058 rue

rt-Denis, Montreal, QC H2X 3J4, Canada. E-mail: pierre.karakiewicz@umontreal.caSubmitted: March 1, 2006, accepted (with revisions): September 4, 2006

2007 Elsevier Inc.ll Rights Reserved

e directly related to the extent of PLND.2–4,9,10 How-ver, no consensus has yet been reached about the extentf PLND or, more specifically, the number of lymphodes that should be removed to achieve optimal cancertaging.2–5 We hypothesized that it would be possible toefine the relationship between the number of lymphodes removed and the ability to predict LNI and testedhis hypothesis in a large cohort of men who predomi-antly underwent extended PLND (ePLND).

ATERIAL AND METHODS

rom November 2002 to August 2005, 944 patients underwent

adical retropubic prostatectomy and PLND for localized pros-

0090-4295/07/$32.00 147doi:10.1016/j.urology.2006.09.008

tw3awpaPaloflitaPapqnadilstw

S

sncwncfilowrcars

RT6T(l7T5n4

1

ate cancer at a single institution. Of these 944 patients, 31ere excluded because of incomplete pathologic information,4 because PLND was not performed, and 21 because they hadprostate-specific antigen (PSA) level greater than 50 ng/mL,hich is highly suggestive of metastatic disease.11 Thus, 858atients were included in this study. No patient had receivedny neoadjuvant therapy. Most PLNDs consisted of ePLND.atient inclusion to the ePLND group required the removal oft least 10 nodes within the confines of an ePLND. The ePLNDimits were the genitofemoral nerve laterally, the medial aspectf the obturator fossa up to the perivesical fat, the deep circum-ex vein and the femoral canal distally, and, finally, the prox-mal dissection included fibrofatty tissue at the bifurcation ofhe common iliac artery. Lymph nodes along the internal iliacrtery were also removed. In select patients, less-extensiveLND was performed according to the preference of the oper-ting surgeon. Pelvic lymph node specimens were submitted forathologic evaluation in multiple packages to improve theuality of specimen assessment.12 Frozen section analysis wasot used. Lymph node specimens were meticulously dissected,nd all identified lymph nodes were counted, sectioned, embed-ed in paraffin, and stained with hematoxylin-eosin. Four ded-cated genitourinary pathologists examined the PLND patho-ogic specimens. The patient data included the preoperativeerum PSA level, clinical stage, and biopsy Gleason sum. Fur-hermore, the number of removed and examined lymph nodesas recorded for each patient.Statistical tests were performed using Statistical Package for

ocial Sciences, version 13 (SPSS, Chicago, Ill). The relation-

Table 1. Characteristics and descriptive statistics

Variable All Patients

Age (yr)Mean (median) 65.8 (66.6)Range 45–85

Clinical stageT1c 474 (55.2)T2 349 (40.7)T3 34 (4)Tx 1 (0.1)

PSA (ng/mL)Mean (median) 9.1 (5.8)Range 0.2–49.90–4.0 107 (12.5)4.01–10.0 475 (55.4)10.01–20.0 186 (21.7)�20 48 (5.6)Missing 42 (4.9)

Biopsy Gleason sum�6 534 (62.2)7 215 (25.1)8–10 70 (8.2)Missing 39 (4.5)

No. of lymph nodes removedMean (median) 15 (14)Range 2–40

No. of positive lymph nodesMean (median) 3 (2)Range 1–19

Lymph node invasion 88 (10.3)Total 858

PSA � prostate-specific antigen; NA � not available.

Data presented as number of patients, with percentages in parenthes

48

hip between the number of removed and examined lymphodes and the rate of LNI was tested in receiver operatingharacteristic (ROC) analyses. The ROC curve coordinatesere used to define the probability of LNI, according to theumber of removed and examined lymph nodes. The ROCurve coordinates were then used to graph the probability ofnding LNI according to the number of removed and examinedymph nodes. The same analyses were repeated in a subcohortf 285 patients with biopsy Gleason sum 7 or worse to examinehether the probability of LNI relative to the number of

emoved nodes changes in men at greater risk of LNI. Toomplete the analyses, we performed univariate and multivari-te analyses to test the association between the number ofemoved lymph nodes and the rate of LNI. Two-sided tests withignificance level set at 0.05 were used.

ESULTShe patient age was 45 to 85 years (mean age � SD6.6 � 7.3). The clinical stage was T1c in 474 (55.2%),2 in 349 (40.7%), T3 in 34 (4.0%), and missing in 1

0.1%) of 858 patients. The biopsy Gleason sum was 6 oress in 534 patients (62.2%), 7 in 215 (25.1%), 8 to 10 in0 (8.2%), and missing in 39 patients (4.5%; Table 1).he PSA level was 0.2 to 49.9 ng/mL (mean 9.1, median.8). LNI was diagnosed in 88 patients (10.3%). Theumber of removed and examined nodes ranged from 2 to0 (median 14; mean 15), while the number of positive

Patients withegative Nodes

Patients withPositive Nodes P Value

0.465.7 (66.6) 66.5 (66.9)

45–85 49–82�0.001

440 (57.1) 34 (38.6)310 (40.3) 39 (44.3)19 (2.5) 15 (17.1)

1 (0.1) —�0.001

8.8 (6.9) 12.1 (9.8)0.24–49.9 0.6–39.9100 (13.0) 7 (8.0) �0.001444 (57.7) 31 (35.2)161 (20.9) 25 (28.4)37 (4.8) 11 (12.5)28 (3.6) 14 (15.9)

�0.001511 (66.4) 23 (26.1)174 (22.6) 41 (46.6)

49 (6.4) 21 (23.9)36 (4.6) 3 (3.4)

�0.00114.7 (14) 18 (18)

2–40 6–36NA

NA 3 (2)NA 1–19NA 88 (100) NA

770 (89.7) 88 (10.3) NA

N

es.

UROLOGY 69 (1), 2007

nmnsrtPp(l

pqt11rlbTpPpePns(m6ns1ie

gtRcceD

wnfiaorwVw7

CSPpnLlHne(

ergrrp

yl(itLfut(i

U

odes ranged from 1 to 19 (median 2; mean 3). Of 858en, 665 (77.5%) had undergone ePLND according toode count and anatomic considerations. The clinicaltage, biopsy Gleason sum, PSA level, and number ofemoved lymph nodes differed significantly betweenhose with positive and negative LNI findings (all�0.001; (Table 1). LNI was found in 38 (6.5%) of 582

atients with a PSA level of 0.2 to 10 ng/mL, in 234.3%) of 534 patients with a biopsy Gleason sum of 6 oress, and in 34 (7.2%) of 474 patients with Stage T1c.

The probability of LNI increased significantly in pro-ortion to the number of examined nodes (P �0.001):uartile 1 (2 to 10 nodes), 5.6% LNI rate; quartile 2 (10o 14 nodes), 8.6% LNI rate; quartile 3 (15 to 19 nodes),0.2% LNI rate; and quartile 4 (20 to 40 nodes), and7.6% LNI rate. Univariate and multivariate logisticegression models predicting for LNI according to PSAevel, clinical stage, biopsy Gleason sum, and total num-er of removed and examined lymph nodes are shown inable 2. A significant association between each clinicalredictor and LNI was present on univariate analysis (all

�0.001). All variables were statistically significantredictors of LNI on multivariate analysis (P �0.002),xcept for PSA level (P � 0.2). After accounting forSA level, clinical stage, and biopsy Gleason sum, theumber of removed and examined lymph nodes was atatistically significant independent predictor of LNIP � 0.002), and its odds ratio of 1.06 indicated that theultivariate probability of detecting LNI increased by

% for each additional removed and examined lymphode. This relationship was virtually the same as thateen on univariate analyses, in which an odds ratio of.07 (P �0.001) indicated that the probability of detect-ng LNI increased by 7% for each additional removed andxamined lymph node.

Finally, ROC analyses were used to examine andraphically explore the probability of finding LNI relativeo the number of removed and examined lymph nodes.OC coordinates were used to plot that probability ac-ording to the number of nodes removed (Fig. 1A). Theurve indicated that 28 nodes needed to be removed andxamined to achieve a 90% probability of detecting LNI.

Table 2. Univariate and multivariate logislymph node invasion at radical prostatecto

Predictor

PSAClinical stage

T2 vs. T1cT3 vs. T1c

Biopsy Gleason sum7 vs. 2–68–10 vs. 2–6

No. of nodes removed and examined

PSA � prostate-specific antigen.

ismal probability values (less than 10%) were associated m

ROLOGY 69 (1), 2007

ith the removal and examination of fewer than 10odes. However, a sharp increase in the probability ofnding LNI occurred when 10 to 30 nodes were removednd examined. The maximal increase in the probabilityf finding LNI occurred when 10 to 20 nodes wereemoved (from 10% to 65%, respectively). Minimal gainas achieved with the removal of more than 35 nodes.irtually the same results were achieved when the curveas fitted using data from men with biopsy Gleason sumor worse (Fig. 1B).

OMMENTeveral studies have suggested that more extensiveLND could be associated with a greater incidence ofositive nodes.2–5 ePLND, with a mean of 17.8 lymphodes removed, was associated with a threefold greaterNI rate compared with modified PLND (mean 9.3

ymph nodes removed; 23% versus 7%; P � 0.02).5owever, the link between PLND extent and LNI hasot always been confirmed. In one randomized trial,PLND did not yield a greater rate of positive nodesePLND 3.2% versus limited PLND 2.4%).13

Thus, no consensus has been reached regarding PLNDxtent or the number of lymph nodes that should beemoved and examined to maximize the yield of locore-ional lymph node staging. We addressed the questionegarding the ideal number of nodes that need to beemoved and assessed pathologically to maximize therobability of correctly predicting the probability of LNI.Our study results have confirmed the notion that nodal

ield, defined by the number of removed and examinedymph nodes, is directly proportional to the LNI rateTable 2). Moreover, we demonstrated that this increases statistically significant, regardless of the method used toest it. First, our stratified analyses demonstrated that theNI rate increased when the nodal yield was divided into

our quartiles (P �0.001). Second, when nodal yield wassed as a linear variable, it was highly significantly relatedo the rate of LNI in univariate logistic regression modelsP �0.001). Moreover, when the nodal yield was testedn multivariate logistic regression models with adjust-

egression models predicting presence of

Odds Ratio (P Value)

variate Logisticression Model

Multivariate LogisticRegression Model

.05 (�0.001) 0.02 (0.2)— (�0.001) — (0.001)

1.6 (0.048) 1.3 (0.3)0.2 (�0.001) 7.7 (�0.001)— (�0.001) — (�0.001)

5.2 (�0.001) 5.3 (�0.001)9.5 (�0.001) 9.5 (�0.001).07 (�0.001) 1.06 (0.002)

tic rmy

UniReg

1

1

1

ent for clinical stage, preoperative serum PSA level,

149

ap0tHei

ofbdtriw2Lfbttctpvf

oatetldb

FbwtntdrLtmL

clpibtftPetartsbapa

dL

Fcx

1

nd biopsy Gleason sum, the multivariate independentredictor status of nodal yield was maintained (P �.002). These analyses have confirmed the significance ofhe association between nodal yield and the LNI rate.owever, statistical significance or the magnitude of the

ffect of a risk variable on the targeted outcome could benterpreted as abstract, and may not appeal to everyone.

To address this potential limitation, we complementedur study with an ROC analysis, which provides a user-riendly, graphic display of the relation between the num-er of examined nodes and the rate of LNI. The ploterived from the ROC coordinates (Fig. 1A) indicatedhat the yield of PLND if fewer than 10 nodes wereemoved is inferior to 10% in detecting LNI. A markedncrease in the rate of LNI occurred when 10 to 30 nodesere removed. A plateau was reached when more than8 nodes were removed and the probability of detectingNI exceeded 90%. Virtually identical findings were

ound when the plot was restricted to patients with aiopsy Gleason sum of 7 to 10 (Fig. 1B). This represen-ation depicts more realistically than regression modelshe relation between nodal yield and the LNI rate, be-ause it does not force a linear or other predefined rela-ion between the risk factor and outcome. The data arelotted according to the unaltered distribution of theariables. Thus, the ROC-derived plot is slightly differentrom logistic regression analysis.

In the logistic models, the increase in the probabilityf diagnosing LNI corresponded to 6% to 7% for eachdditional node removed. This relation reinforces thathe increase in LNI for each additional removed andxamined node is important. However, it suggests thathe increase in the probability of diagnosing LNI behavesinearly. Moreover, it suggests that a high probability ofiagnosing LNI, according to the regression results, will

igure 1. Sensitivity curve for (A) entire cohort of patients aharacteristics (ROC)-derived sensitivity shown on y-axis a-axis.

e reached sooner than that suggested by the real data. p

50

inally, it provides no ceiling effect. However, it mighte suspected that the removal of a high number of nodesill result in a plateau effect, at which removing addi-

ional nodes will contribute little to the ability to diag-ose LNI. The ROC method circumvents these limita-ions and indicates that the increase in the ability toiagnose LNI is virtually linear when 10 to 25 nodes areemoved. However, the increase in the ability to diagnoseNI is very slow, indicated by the virtually flat portion ofhe curve, when 1 to 5 nodes were removed. After re-oving 30 nodes, the increase in the ability to diagnoseNI did plateau, as would be expected in real life.It can be argued that the number of removed nodes

annot be determined preoperatively, which certainly is aimitation of our analysis. This could be further com-ounded by the variability in nodal yield between spec-mens obtained from the same anatomic distributions,ecause some patients might have fewer lymph nodeshan others. Moreover, some pathologists might sampleewer nodes than others. Thus, surgical variables are nothe sole determinant of the nodal yield. In consequence,LND does not represent the only stage at which thextent of PLND can be optimized, and close collabora-ion with pathologists is crucial to optimize the ability toccurately stage LNI in prostate cancer. Finally, the trueate of false-negative findings represents another limita-ion of this study. However, this flaw biases all PLNDtudies, in which the true status of the disease is unknownecause of several uncontrolled sources of variability suchs the extent of PLND, patient anatomy, disease biology,athologic analysis, and other known or unknown vari-bility sources, all of which could bias the results.

Because patient variables and pathologic specimen han-ling could affect the yield of PLND and, indirectly, theNI rate, the information provided in our ROC-derived

) patients with Gleason sum 7 or more. Receiver operatingumber of removed and examined lymph nodes shown on

nd (Bnd n

robability plots cannot be directly converted into surgical

UROLOGY 69 (1), 2007

geoahlllttc

dantrntmlatt

CTygaoaTiab

R

1

1

1

1

1

U

uidelines about the required nodal yield at PLND. How-ver, our plot provides a valid approximation of the numberf nodes that the pathologist should assess to maximize theccuracy of pathologic LNI staging. For example, our plotsave indicated that limited PLNDs, such as described in a

arge North American series, in which an average of fiveymph nodes was removed, are associated with extremelyimited staging ability.14 Conversely, our plot has indicatedhat in some European series, in which a median of morehan 20 nodes was removed and examined, the ability toorrectly stage LNI was close to 80%.2,3

The novelty of our findings resides in their ability toirectly examine the relationship between PLND extentnd the LNI rate. It can be depicted using ROC coordi-ates and provides the clinician with tangible proof ofhe concept that the extent of PLND affects the LNIate. Moreover, our plots allow the determination of theumber of nodes that need to be examined to maximizehe accuracy of nodal staging. The proposed ROCethod is more user-friendly than the interpretation of

ogistic regression coefficients (Table 2), with which onlyrelative increase can be reported. Therefore, despite

heir limitations, our findings provide a practical approacho assess the relation between nodal yield and LNI rate.

ONCLUSIONSaken together, our data have indicated that the nodalield at PLND is directly related to the LNI rate. Areater nodal yield is associated with superior stagingccuracy. PLNDs in which fewer than 10 nodes arebtained should probably be omitted, because the prob-bility of finding positive lymph nodes is very limited.he practical application of more accurate LNI diagnosis

n patients with prostate cancer may allow more timelydministration of systemic therapy and may result inetter survival.10

eferences1. Aus G, Abbou CC, Bolla M, et al: EAU guidelines on prostate

cancer. Eur Urol 48: 546–551, 2005.

ROLOGY 69 (1), 2007

2. Bader P, Burkhard FC, Markwalder R, et al: Is a limited lymph nodedissection an adequate staging procedure for prostate cancer? J Urol168: 514–518, 2002.

3. Heidenreich A, Varga Z, and von Knobloch R: Extended pelviclymphadenectomy in patients undergoing radical prostatectomy:high incidence of lymph node metastasis. J Urol 167: 1681–1686,2002.

4. Allaf ME, Palapattu GS, Trock BJ, et al: Anatomical extent oflymph node dissection: impact on men with clinically localizedprostate cancer. J Urol 172: 1840–1844, 2004.

5. Stone NN, Stock R, and Unger P: Laparoscopic pelvic lymph nodedissection for prostate cancer: comparison of the extended andmodified technique. J Urol 158: 1891–1894, 1997.

6. Cheng L, Zincke H, Blute ML, et al: Risk of prostate carcinomadeath in patients with lymph node metastasis. Cancer 91: 66–73,2001.

7. Palapattu GS, Allaf ME, Trock BJ, et al: Prostate specific antigenprogression in men with lymph node metastases following radicalprostatectomy: results of long-term followup. J Urol 172: 1860–1864, 2004.

8. Bader P, Burkhard FC, Markwalder R, et al: Disease progressionand survival of patients with positive lymph nodes after radicalprostatectomy: is there a chance of cure? J Urol 169: 849 – 854,2003.

9. McLaughlin AP, Saltzstein SL, McCullough DL, et al: Prostaticcarcinoma: incidence and location of unsuspected lymphatic me-tastases. J Urol 115: 89–94, 1976.

0. Messing EM, Manola J, Sarosdy M, et al: Immediate hormonaltherapy compared with observation after radical prostatectomy andpelvic lymphadenectomy in men with node-positive prostate can-cer. N Engl J Med 341: 1781–1788, 1999.

1. Gerstenbluth RE, Seftel AD, Hampel N, et al: The accuracy of theincreased prostate specific antigen level (greater than or equal to 20ng./ml.) in predicting prostate cancer: is biopsy always required?J Urol 168: 1990–1993, 2002.

2. Bochner BH, Herr HW, and Reuter VE: Impact of separate versusen bloc pelvic lymph node dissection on the number of lymphnodes retrieved in cystectomy specimens. J Urol 166: 2295–2296,2001.

3. Clark T, Parekh DJ, Cookson MS, et al: Randomized prospectiveevaluation of extended versus limited lymph node dissection inpatients with clinically localized prostate cancer. J Urol 169:145–147, 2003.

4. DiMarco DS, Zincke H, Sebo TJ, et al: The extent of lymphade-nectomy for pTXN0 prostate cancer does not affect prostate canceroutcome in the prostate specific antigen era. J Urol 173: 1121–

1125, 2005.

151