EURURO Vol. 72 No. 6

All tests performed were two sided, with a significance level of 5%.

Statistical analyses were performed using R version 3.3.0 (packages

ModelGood and rms; R Foundation for Statistical Computing, Vienna,

Austria)

[19,20] .

Decision curve analysis (DCA) was performed utilizing

the DCA package

[18]

. Reporting followed Standards of Reporting of

Diagnostic Accuracy (Supplementary Table 2)

[21]

Results

In total, 1159 men underwent mpMRI and subsequent

fusion biopsy during the inclusion period. Patient demo-

graphics, MRI, and biopsy data are given in

Table 1 .

sPC was

detected in 489 men (42%).

For RM development and validation, men under active

surveillance and those with missing data were excluded

(Supplementary Fig. 1). Prior TRUS biopsy and age were

significant predictors of sPC in the combined RM cohort of

1015 men (

= 0.006 and

= 0.004, respectively). We

accounted for differences in predicted risks of sPC with

respect to prior TRUS biopsy by developing one RM for

biopsy-naı¨ve men and one for men after previous biopsy,

and included age in both RMs (and refitted ERSPCs). In the

multivariate logistic regression analysis to predict sPC for

biopsy-naı¨ve patients, logPSA (

0.001), PV (

0.001),

DRE (

0.001), and PI-RADSv1.0 (

0.001) contributed

significantly to the model

( Table 2

and

Fig. 1 A

). For

previously biopsied men, PI-RADSv1.0 (

0.001), logPSA

(

= 0.006), PV (

0.001), and DRE (

= 0.03) were included

in the RM

( Table 2

and

Fig. 1 B

The novel RMs were internally validated by bootstrap-

ping. The discrimination of the RMs was compared with

ERSPC-RC3/4, refitted RCs, PI-RADSv1.0, and ERSPC-RC3/4

combined with mpMRI using ROC analyses

( Fig. 2

and

Table 3 )

. For biopsy-naı¨ve men, the RM reached a higher

AUC (0.83), compared with ERSPC-RC3 (0.81), refitted RC3

(0.80), and PI-RADSv1.0 (0.76;

Fig. 2

A and

Table 3

). The RM

AUC was comparable with that of ERSPC-RC3 + PI-RADSv1.0

(0.84). In men with previous biopsy, the discrimination of

the RM (0.81) was superior to that of ERSPC-RC4 (0.66),

refitted ERSPC-RC4 (0.76), PI-RADSv1.0 (0.78), and ERSPC-

RC4 + PI-RADSv1.0 (0.78). LR test results also showed that

[(Fig._2)TD$FIG]

Fig. 2 – ROC curve analysis for the performance of mpMRI PI-RADSv1.0 (yellow line), ERSPC-RC3/4 (green line), refitted RC3/4 (pink line), ERSPC-RC3/

4 + mpMRI PI-RADSv1.0 (blue line), and novel risk model (orange line) to predict sPC for (A) biopsy-naı¨ve and (B) postbiopsy men. AUCs are given in

Table 3 .

AUC = area under the curve; ERSPC = European Randomised Study of Screening for Prostate Cancer; mpMRI = multiparametric magnetic

resonance imaging; PI-RADS = Prostate Imaging Reporting and Data System; RC = risk calculator; sPC = significant prostate cancer.

Table 3 – AUC of ROC curve analysis for the performance of mpMRI

PI-RADS, ERSPC-RC3, ERSPC-RC4, refitted ERSPC-RCs, combination

of ERSPC-RC3/4 and mpMRI PI-RADSv1.0, and novel RMs to predict

sPC for biopsy-naı¨ve men and men after previous biopsy, and

likelihood ratio tests for model comparison

Parameter

Subset of biopsy-naı¨ve men (n = 660 available for

RM development)

AUC in ROC

curve analysis

Risk model

0.83

ERSPC-RC3

0.81

ERSPC-RC3 refitted

0.80

ERSPC-RC3 plus mpMRI PI-RADSv1.0

0.84

mpMRI PI-RADSv1.0

0.76

Subset of men with previous biopsy sessions

(n = 355 available for RM development)

AUC in ROC

curve analysis

Risk model

0.81

ERSPC-RC4

0.66

ERSPC-RC4 refitted

0.76

ERSPC-RC4 plus mpMRI PI-RADSv1.0

0.78

mpMRI PI-RADSv1.0

0.78

Comparison of models for biopsy-naive men using LR test

value

Risk model versus ERSPC-RC3 refitted

0.001

Risk model versus mpMRI PI-RADSv1.0

0.001

Comparison of models for men after previous biopsy

using LR test

value

Risk model versus ERSPC-RC4 refitted

0.001

Risk model versus mpMRI PI-RADSv1.0

0.001

ROC = receiver operating characteristics; AUC = area under the curve;

ERSPC = European Randomised Study of Screening for Prostate Cancer;

RC = risk calculator; RM = risk model; LR = likelihood ratio; mpMRI =

multiparametric magnetic resonance imaging; PI-RADS = Prostate Imaging

Reporting and Data System; sPC = significant prostate cancer.

Note that the LR test is only defined for nested models. Thus, no LR test can be

given to compare the risk models with original ERSPCs or with original risk

models and additional mpMRI PI-RADSv1.0.

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

892