252 320
false negatives and true negatives, in order to calculate NPV
(ie, results of systematic/standard prostate biopsies when
the mpMRI was negative). When available, false positive
and true positive findings were also noted to calculate the
positive predictive value (PPV) and the cancer prevalence.
There was restriction neither on the biopsy technique
(transrectal or transperineal) nor on the number of biopsy
cores. Studies using radical prostatectomy specimens as
reference standards were excluded, as were studies
evaluating men with histologically proven PCa. Studies
with less than 50 participants were excluded. No language
restrictions were applied.
2.4.
Data collection and data extraction
Two reviewers (P.C.M. and T.V.D.B.) independently
screened all abstracts and full-text articles for eligibility.
Disagreement was resolved by consensus or reference to an
independent third party (L.M.). All screening was performed
using a predefined eligibility form.
Using a data extraction form developed a priori, the same
two reviewers independently extracted data concerning
study methodology, patient characteristics, technical char-
acteristics of the MR scanners, mpMRI protocol, mpMRI
scoring system, definition of positive mpMRI, biopsy
protocol, and definition of csPCa. Any discrepancies
concerning data extraction were resolved by consensus or
reference to an independent arbiter (O.R. or T.B.L.).
2.5.
Assessment of risk of bias
To assess the risk of bias (RoB), all included reports were
reviewed using the Quality Assessment of Diagnostic
Accuracy Studies (QUADAS-2) tool for diagnostic accuracy
studies
[10] .2.6.
Data synthesis and analysis
Outcome data regarding false negative and true negative
values of mpMRI before prostate biopsy were recorded as
reported by authors. When not available, data were
indirectly derived from specificity, sensitivity, and preva-
lence values reported by authors using an online Bayesian
statistics calculator
( http://www.medcalc.com/bayes.html).
Descriptive statistics were used to summarise baseline
characteristics and outcomes, including median and inter-
quartile range (IQR) for estimates of NPV across studies. A
correlation between mpMRI NPV and a positive biopsy rate
was established using the Pearson’s correlation coefficient.
A meta-analysis was undertaken to calculate pooled NPV
and PPV. To ensure appropriate clinical homogeneity of the
studies included in the meta-analysis, we selected only the
studies enrolling biopsy-naı¨ve patients and/or patients with
a history of negative biopsy, and fulfilling the following
criteria that were defined a priori: (1) reference standard
consisting of prostate biopsy with at least 10 samples on all
patients; (2) mpMRI protocol comprising at least T2WI and
DWI; (3) mpMRI results presented as a five-level score,
using a subjective Likert scale or the Prostate Imaging
Reporting Data System (PI-RADS) score
[11]; (4) definition
of positive mpMRI as a score 3/5 or 4/5; and (5) results
reported on a per patient basis. In addition, only studies
defining csPCa as Gleason 7 cancers were selected for the
meta-analysis assessing the mpMRI NPV for csPCa. A
bivariate random-effects approach was employed using
the Midas package in Stata 12 (StataCorp LP, College Station,
TX, USA). Since the NPV decreases and the PPV increases as
the prevalence increases, post-test probability estimates of
NPV and PPV were reported for the given values of the
prevalence based on Bayes’ theorem.
For other studies not included in the meta-analysis based
on the criteria described above, a narrative synthesis of the
data was performed. To explore and define clinical
heterogeneity, subgroups were analysed at patient level
based on the following variables: biopsy-naı¨ve versus
previous negative biopsy; patients with positive versus
negative digital rectal examination (DRE); mpMRI per-
formed with an endorectal versus without an endorectal
coil; transrectal ultrasound (TRUS) versus template trans-
perineal (TTP) biopsy approach; and 16 cores versus
>
16 cores as the reference standard. Studies reporting
mpMRI NPV for patients with a prostate-specific antigen
(PSA) level of 10 ng/ml were also reported separately.
3.
Evidence synthesis
3.1.
Quantity of evidence identified
The study selection process is depicted in the PRISMA flow
diagram
( Fig. 1 ). A total of 2980 abstracts were retrieved.
After abstract screening and removal of duplicates,
240 articles were eligible for full text screening, of which
48 studies were eligible for inclusion
[12–59] .3.2.
Quality of studies
Out of the 48 included studies, 42 were single-centre and six
were multicentre studies. Thirty-four studies were pro-
spective and six were retrospective, whilst the design of the
rest was unclear. RoB assessment using QUADAS-2 was
performed for each of the individual studies
( Fig. 2 Aand B).
Overall, the RoB was highly heterogeneous across studies
for all criteria, except for the reference standard domain, in
which RoB was low in most studies.
3.3.
Characteristics of studies
The 48 studies comprised a total of 9613 men who
underwent prostate mpMRI followed by biopsy. The study
and patient baseline characteristics are presented in
Table 1 .The patient population consisted of biopsy-naive
men in nine studies, men with at least one previous negative
biopsy in 16 studies, and both biopsy-naı¨ve men and men
with a history of previous negative biopsy in nine studies. In
14 studies, the biopsy history of the patients was unclear.
The magnetic field strength was 1, 1.5, and 3 T in one, 28,
and 15 studies, respectively. Four studies used both 1.5 and
3 T MR systems. DWI and DCEI were used in 36 and
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 2 5 0 – 2 6 6
252