EURURO Vol. 72 No. 6

911 1030

1. Introduction
Radical prostatectomy (RP) is associated with excellent
oncologic outcomes in patients with localized prostate
cancer (PCa), with approximately 75% of such patients being
free from recurrence at 10-yr follow-up . Following
surgery, prostate-specific antigen (PSA) is expected to
become undetectable at approximately 6 wk postopera-
tively. However, up to 20% of patients with adverse
pathologic characteristics fail to achieve an undetectable
PSA after RP . These individuals are at increased risks
of recurrence and mortality compared with patients with
initially undetectable postoperative PSA . Consider-
able heterogeneity has been noted in the clinical outcomes
of patients with PSA persistence after surgery . A
detectable PSA after RP has the potential to reflect
persistent local or distant PCa cells not removed by surgery
as well as benign prostatic tissue left behind during the
procedure. While in the former case, timely administration
of additional cancer therapies might improve oncologic
outcomes , in the latter scenario, additional postop-
erative treatments may represent overtreatment and, thus,
possibly expose these men to unnecessary side effects
. While subanalyses of prospective randomized
trials have found a benefit to postoperative radiotherapy
(RT) in men with PSA persistence , to date no study
identified the optimal candidate for this approach in order
to maximize oncologic benefit for those most likely to
experience disease progression, while sparing the use of RT
in those less likely to benefit from it.
We hypothesized that the impact of postoperative RT on
disease progression and mortality varies according to an
individual’s risk of cancer-specific mortality (CSM). As such,
we aimed at developing a novel predictive tool to identify
patients with PSA persistence at a higher risk of CSM. We
subsequently evaluated the impact of postoperative RT on
CSM according to the risk of dying from PCa. We relied on a
large contemporary cohort of patients with PSA persistence
after RP treated at two high-volume tertiary referral centers.
2. Patients and methods
2.1. Population source
After Institutional Review Board approval, 982 patients treated with RP
between 1994 and 2014 at two tertiary referral institutions (IRCCS
Ospedale San Raffaele, Milan, Italy, and Mayo Clinic, Rochester, NY, USA)
with available data on the first PSA value after surgery were identified. All
patients had PSA persistence, defined as a PSA level of 0.1 ng/ml after RP.
Among those, we selected patients who underwent a first PSA assessment
between 6 and 8 wk after surgery (n = 612). Due to their increased risk of
harboring distant metastases patients with PSA levels >2 ng/ml at
6–8wk after surgery (n = 100) were excluded fromour analyses. Moreover,
patients with incomplete pathologic data and pNx status were excluded
from our study (n = 16). This resulted in a final cohort of 496 patients.
2.2. Covariates
All patients had complete data, including age at surgery, year of surgery,
preoperative PSA, pathologic stage, pathologic grade group, surgical
margin status, and lymph node invasion. Prostatectomy specimens were
evaluated by high-volume, dedicated uropathologists. Postoperative RT
was delivered to the prostate and seminal vesicle bed using previously
described techniques . Whole pelvis RT was administered to 7%
and 80% of patients with pN0 and pN1 disease included in the
postoperative RT group, respectively. Immediate androgen deprivation
therapy (ADT) was defined as ADT administered within 90 d from
surgery. The decision to administer postoperative RT ADT was based on
the clinical judgment of each treating physician according to individual
patient and cancer characteristics.
2.3. End points
The primary outcome of the study was CSM, which was defined as death
from PCa. Other-cause mortality (OCM) was defined as death due to
other causes. Follow-up time was defined as the time elapsed between
surgery and CSM or last follow-up.
2.4. Statistical analyses
Our statistical analyses consisted of multiple steps. First, multivariable
Cox regression analyses assessed predictors of CSM. Covariates consisted
of pathologic stage, pathologic grade group, pN1 status, positive surgical
margin status, and immediate ADT. The regression coefficients were
then used to generate a model predicting 10-yr CSM. A leave-one-out
cross validation was used to construct the Harrell c-index to assess
discrimination of our novel model. The relationship between the
predicted probability and the observed fraction of patients experiencing
CSM at 10 yr was depicted using the calibration plot method.
Second, we assessed whether the impact of PSA level at 6–8 wk after
surgery on CSM-free survival differed according to the risk of CSM.
Locally weighted 10-yr Kaplan–Meier estimates by values of a
continuous covariate (locally weighted scatterplot smoothing) method
was used to graphically depict the relationship between PSA at 6–8 wk
and 10-yr CSM-free survival in the overall population and after
stratifying patients according to the median 10-yr CSM risk (<10 vs
10%) .
Third, we sought to assess whether the impact of postoperative RT
was different by CSM risk. A multivariable Cox regression model
predicting CSM was developed for patients who did not receive
postoperative RT. The same covariates adopted in the nomogram
developed for the overall population were used. The 10-yr CSM risk was
calculated for each patient using the multivariate coefficients. We then
tested an interaction with groups (postoperative RT vs no RT) and the
probability of dying from PCa according to the newly developed model.
The nonparametric curve fitting method was used to graphically explore
the relationship between the risk of CSM and actual 10-yr CSM rates
according to the administration of postoperative RT.
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 <0.05.
3. Results
3.1. Baseline characteristics
depicts clinical and pathologic characteristics of
patients included in our cohort. Median age at surgery was
64 yr. When patients were stratified according to receipt of
postoperative RT, significant differences were observed
with regard to the year of surgery, preoperative PSA and risk
group, pathologic grade group, pathologic stage, nodal
status, positive surgical margin status, and PSA level at
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 9 1 0 – 9 1 7 911