EURURO Vol. 72 No. 6

893 1030

both RMs performed significantly better compared with
refitted ERSPCs and PI-RADS alone ). Bootstrapped
calibration plots of the RMs and 3B) demonstrate
that there are no untoward deviations of predicted risk from
observed risk of sPC over the entire range.
In bootstrapped DCA for both biopsy-naı¨ve and post-
biopsy men, the RMs had a higher net benefit in terms of
accurately detecting patients with sPC, compared with PI-
RADS and the original and refitted ERSPC-RCs A and
C). The RMs showed a benefit for sPC threshold probabilities
larger than 10%. Net reduction curves, which show the
potential to reduce unnecessary biopsies B and D),
demonstrate that the RMs had excellent reduction rates of
unnecessary biopsies. ERSPC-RC3 + PI-RADSv1.0 had a
similar net benefit and reduction to the RM. Detailed TPR,
FPR, PPVs, and NPVs at exemplary probability cutoffs after
bootstrapping are given in Supplementary Table 3.
4. Discussion
Risk-based patient selection instead of PSA- or DRE-driven
indication for prostate biopsy can reduce unnecessary
biopsies by approximately 30% . Furthermore, the
ERSPC-RCs have been externally validated with comparable
discrimination results Alberts et al[3_TD$DIFF] recently adopted
the ERSPC-RC on an MRI-targeted biopsy cohort[9_TD$DIFF] . They
found that despite a systematic underestimation of PC risk,
the RC biopsy advice performed well. Thus, we tested
ERSPC-RC predictors on our large MRI/TRUS fusion cohort
and hypothesized further improvement by combination
with mpMRI . However, our approach was different from
that of Alberts et al[3_TD$DIFF], who used the RC as an upfront test to
avoid unnecessary MRI and biopsy[10_TD$DIFF] We aimed to
optimize noninvasive sPC-risk prediction by combining
mpMRI and clinical parameters. Our results show that the
novel models provide a statistically significant improve-
ment in the discrimination of men with a suspicion of PC.
ERSPC-RCs already showed a good prediction perfor-
mance for sPC detection in our cohort. However, the
prediction performance was inferior to that of Alberts et al
(AUC 0.84) and the original cohorts (AUC 0.86 for RC3
and 0.80 for RC4) . To the best of our knowledge, the RCs
have not yet been validated in a cohort of men who
underwent mpMRI followed by transperineal SB and FTB as
a reference test . Recently, Poyet et al[4_TD$DIFF] showed that the
prediction performance of RCs calibrated on sextant
biopsies decreases with an increasing number of biopsy
cores[11_TD$DIFF] This is confirmed in our study. In addition, the
relatively poor prediction performance of the original
ERSPC-RC4 in our postbiopsy cohort (AUC 0.66) is most
likely explained by including only men with a prior biopsy,
instead of including also men prescreened with PSA testing
alone. To attenuate these differences, clinical parameter
models, using the same parameters as the ERSPC-RCs
(adjusted for age and replacing linear PSA by logPSA), were
constructed and used as refitted ERSPC-RCs in our cohort.
However, as hypothesized, these models remained inferior
to approaches that combine clinical parameters with
imaging (our RMs and ERSPC-RC3 + PI-RADSv1.0).
The predictors within the novel RMs are known and
reproducible contributors. Improvement of risk stratification
by adding DRE to PSA has been demonstrated . Similarly,
the importance of PV in risk prediction has been analyzed in
a multicenter study, but is decreased in our cohort, since the
number of SB cores was adjusted to PV . Besides
potential harms driven by a lack of specificity, PSA-based
screening has shown increased detection of aggressive PC
and reduction of PC mortality . Thus, the contribution of
PSA in our RM is congruent with ERSPC results. Similarly, a
prior TRUS biopsy was also significant in our dataset despite
the stringent reference test.
The accuracy of mpMRI for sPC detection in our cohort is
consistent with a recent validation study using PI-RADSv1.0
. Regarding the predictive ability of mpMRI, [12_TD$DIFF]Meng et al
[(Fig._3)TD$FIG]
Fig. 3 – Calibration plots for the risk models to predict sPC. (A) Calibration plot for biopsy-naı¨ve men. (B) Calibration plot for postbiopsy men.
sPC = significant prostate cancer.
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 8 8 8 – 8 9 6 893