EURURO Vol. 72 No. 4

633 656

1. Introduction
Up to 15% of prostate cancer (PCa) patients harbor lymph
node invasion (LNI) at radical prostatectomy (RP) . These
individuals are at a higher risk of recurrence after primary
treatment Correct nodal staging is crucial to identify
patients with poor prognosis who would benefit from
additional therapies . The implementation of novel
imaging modalities such as prostate-specific membrane
antigen positron emission tomography /computed tomog-
raphy scan prior to RP is limited by their poor performance
characteristics . Conversely, an anatomically defined
extended pelvic lymph node dissection (ePLND) represents
the most optimal method for nodal staging . Given the
prolonged operative time as well as the increased risk of
complications associated with an ePLND , this proce-
dure is indicated only in selected patients at a higher risk of
nodal involvement . The European Association of
Urology (EAU)–European Society for Radiotherapy &
Oncology (ESTRO)–International Society of Geriatric Oncol-
ogy (SIOG) and the National Comprehensive Cancer
Network (NCCN) clinical guidelines recommend the use
of models based on preoperative characteristics such as the
Briganti and Memorial Sloan Kettering Cancer Center
(MSKCC) nomograms to estimate the risk of LNI and, in
turn, to select men who should be considered for an ePLND
. However, these tools need to be periodically
updated Moreover, although the Briganti and MSKCC
nomograms achieve very good performance characteristics,
they can certainly be improved . Indeed, none of them
included the precise assessment of cancer involvement
within the biopsy cores or accounted for intraprostatic
heterogeneity in PCa grade. This might lead to a limited
accuracy in estimating the risk of LNI Under this
light, we sought to develop a novel nomogram predicting
LNI in a contemporary cohort of patients treated with RP
and ePLND, with detailed biopsy information available after
a centralized biopsy specimen review.
2. Patients and methods
2.1. Population source and surgical procedure
After Institutional Review Board approval, clinical and pathologic data
were prospectively collected for 2872 patients treated with open or
robot-assisted RP and ePLND for localized PCa between January 2011 and
July 2016 at a single tertiary referral center. Patients with complete data
who underwent centralized biopsy specimens review performed by two
high-volume dedicated uropathologists (either R.M. or M.F.) were
selected (n = 681). No patients received neoadjuvant hormonal therapy.
All cases were performed by six surgeons with at least 200 cases at the
beginning of data collection who were trained by the same surgeon and
applied the same anatomical template for ePLND. The fibrofatty tissue
along the external iliac vein was dissected, the lateral limit being the
genitofemoralis nerve. Proximally, an ePLND was performed up to and
included the crossing between the ureter and common iliac vessels.
Lymph nodes along as well as medially and laterally to the internal iliac
vessels were removed. All fibrofatty tissue within the obturator fossa
was removed, and the Marcille’s triangular lumbosacral fossa was
dissected free . All specimens were submitted for pathologic
evaluation in multiple packages according to their anatomical location
and were evaluated by dedicated uropathologists according to a
previously described methodology .
2.2. Covariates and end points
All patients were subjected to a detailed preoperative evaluation that
consisted of prostate-specific antigen (PSA), clinical stage obtained
according to the digital rectal examination performed by the attending
urologist, and transrectal ultrasound-guided prostate biopsy . All
patients had complete data, including the percentage of positive cores,
biopsy grade group for each positive core, percentage of PCa involve-
ment, and exact tumor length in each core. We calculated the percentage
of cores involved by the highest-grade disease by dividing the number of
cores with highest-grade PCa by the total number of cores. The
percentage of cores involved by lower-grade PCa was abstracted. We
calculated tumor length (ie, the sum of tumor length in each single core)
overall and stratified according to highest- and lower-grade PCa.
Similarly, the percentage of tumor in biopsy cores was abstracted.
The modified Gleason scoring system was adopted according to the
International Society of Urological Pathology 2005 and 2014 consensus
conferences The outcome of our study was LNI, defined as the
presence of positive lymph nodes at final pathology.
2.3. Statistical analyses
First, univariable logistic regression analyses assessed predictors of LNI.
Given the small number of events in the biopsy grade group 1 and
5 categories, we decided to categorize biopsy grade groups in 1–2 versus
3 versus 4–5. We then developed four different multivariable models
predicting LNI including variables that might be considered as a proxy of
tumor volume. Preoperative PSA, clinical stage, and biopsy grade group
were included in all these models. The discrimination accuracy of
multivariable models based on these variables in our cohort was
quantified using the receiver operating characteristic-derived area under
the curve (AUC). Since the inclusion of information on the maximum
percentage of single core involvement and the percentage of tumor in
biopsy cores overall and according to the highest- and lower-grade
disease did not improve the predictive accuracy, we relied on the most
parsimonious model to develop a nomogram predicting LNI. Preoperative
PSA, clinical stage, biopsy Gleason grade group, percentage of cores with
highest-grade PCa, and percentage of cores with lower-grade disease
represented the basis for our coefficient-based nomogram. The extent of
over- or underestimation of the histologically confirmed versus nomo-
gram-predicted LNI rates was graphically explored using a calibration
plot. A decision-curve analysis (DCA) was used to determine the clinical
net benefit associated with the use of the model . The discrimination,
calibration, and DCA were corrected for overfit using leave-one-out cross
validation. We then compared the predictive accuracy of two existing
models (Briganti and MSKCC nomograms) with the novel nomogram
using the predetermined regression coefficients . Calibration
plots were used to assess the extent of over- or underestimation
associated with their use. Finally, DCAs were used to determine the
clinical net benefit associated with the adoption of these models.
All statistical tests were performed using the R statistical package
v.3.0.2 (R Project for Statistical Computing, ). All tests
were two sided, with a significance level set at p < 0.05.
2.4. Sensitivity analyses
We compared baseline and pathologic characteristics of patients
included in our cohort and those excluded due to incomplete biopsy
information, to investigate whether patients with missing data were
different from those included in our analyses. We then repeated our
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 6 3 2 – 6 4 0 633