EURURO Vol. 72 No. 2

179 320

1. Study characteristics—authors, year of publication,
country and institution of origin, duration of patient
recruitment, and study design (prospective vs retrospec-
tive and consecutive or not)
2. Demographic and clinical characteristics—sample size,
number of patients with PCa, patient age, prostate-
specific antigen (PSA) level and Gleason score, number of
previous biopsies, and PCa diagnosis prior to mpMRI
3. Technical characteristics of mpMRI—scanner model and
manufacturer, magnetic field strength (1.5 vs 3 T), coil
type (endorectal vs pelvic phased array), and specific
sequences used (T2WI, DWI, DCE, or MR spectroscopy)
4. Interpretation of mpMRI—number of reviewers and
experience in prostate mpMRI, independent or consensus
reading, and blinding to clinicopathological information
5. Reference standard—type of reference standard (radical
prostatectomy, targeted biopsy, or systematic biopsy),
interval between MRI and pathology, outcomes assessed
(any PCa vs csPCa), definition of csPCa (studies assessing
‘‘clinically significant’’, ‘‘aggressive’’, or ‘‘high-grade’’ PCa
were all considered to assess csPCa; however, only
studies that used the definition as provided by the PI-
RADSv2 guideline [Gleason score >7 [3 + 4], volume
> 0.5 ml, or extraprostatic extension] were considered
not to have concern for applicability), separate analysis
for the PZ and TZ, and type of analysis (per patient vs per
lesion)
6. Diagnostic performance of PI-RADSv2 including criteria
or cutoff values (in case of multiple readers, the results of
the most experienced reader were extracted for this
meta-analysis)
The methodological quality of the included studies was
assessed using tailored questionnaires and criteria provided
by Quality Assessment of Diagnostic Accuracy Studies-2
. Data extraction and quality assessment were per-
formed independently by two reviewers (S.W. and C.H.S.).
All disagreements were resolved by consensus through
discussion with the third reviewer (S.Y.K.).
2.4. Data synthesis and analysis
The diagnostic performance of PI-RADSv2 for the detection
of PCa was the primary outcome for this meta-analysis. In
addition, a comparison between the diagnostic perfor-
mance of PI-RADSv2 and that of PI-RADSv1 using studies
that reported head-to-head comparison data of the two PI-
RADS versions was considered a secondary outcome.
Pooled estimates of sensitivity and specificity were
calculated using hierarchical logistic regression modeling
including bivariate modeling and hierarchical summary
receiver operating characteristic (HSROC) modeling
. For graphical presentation of the results, an HSROC
curve with 95% confidence region and prediction region was
plotted. Publication bias was evaluated using the Deeks’
funnel plot, and statistical significance was tested with the
Deeks’ asymmetry test
We performed [3_TD$DIFF]meta-regression analyses to investigate
the cause of heterogeneity. The following covariates were
considered for the bivariate model: (1) proportion of
patients with PCa (>50% vs 50%), (2) magnet strength
of MRI (3 vs 1.5 T), (3) use of endorectal coil, (4) cutoff value
( 4 vs 3), (5) reference standard (radical prostatectomy vs
biopsy), and (6) type of analysis (per patient vs per lesion).
In addition, multiple subgroup analyses were performed for
cutoff value, outcome, and previous biopsy history to assess
various clinical settings: (1) a cutoff value of 4 for all
studies, (2) a cutoff value of 3 for all studies, (3) a cutoff
value of 4 for determining any PCa, (4) a cutoff value of 3
for determining any PCa, (5) a cutoff value of 4 for
determining csPCa, (6) a cutoff value of 3 for determining
csPCa, (7) a cutoff value of 4 in studies using per-patient
analysis, (8) a cutoff value of 4 in studies using per-lesion
analysis, (9) studies analyzing PZ PCa, (10) studies analyzing
TZ PCa, (11) patients without previous biopsies, and (12)
patients [4_TD$DIFF]with previous biopsies. The ‘‘metandi’’ and ‘‘midas’’
modules in Stata 10.0 (StataCorp LP, College Station, TX,
USA) and ‘‘mada’’ package in R software version 3.2.1 (R
Foundation for Statistical Computing, Vienna, Austria) were
used for statistical analyses, with p < 0.05 signifying
statistical significance.
3. Evidence synthesis
3.1. Literature search
A systematic literature search initially identified 287 arti-
cles. After removing 46 duplicates, screening of the
241 titles and abstracts yielded 105 potentially eligible
articles. Full-text reviews were performed, and 84 studies
were excluded for the following reasons: not in the field of
interest (n = 80, including 68 studies that used only PI-
RADSv1), insufficient data to reconstruct 2 2 tables
(n = 2), and shared study population with other studies
(n = 2). Ultimately, 21 original articles including a total of
3857 patients assessing the diagnostic performance of PI-
RADSv2 were included in the meta-analysis No
additional studies were identified via screening the
bibliographies of these 21 studies. Among them, 15 studies
including 3099 patients dealt with PI-RADSv2 alone,
whereas six studies including 758 patients provided a
head-to-head comparison between PI-RADSv1 and PI-
RADSv2 The detailed study selection
process is described in
3.2. Characteristics of included studies
Patient characteristics are shown in . The size of the
study population ranged from 49 to 456 patients, with the
percentage of those with PCa ranging from 37% to 100%. The
patients had a median age of 62–69.6 yr, median PSA of
3.97–15 ng/ml, and a Gleason score ranging from 5 to
10. Patients had already been diagnosed with PCa prior to
MRI in all or some of the study populations in seven studies
. Biopsy was performed before MRI in
seven studies all patients were biopsy-
naı¨ve in three studies both patient types were
included in three studies , and data regarding
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 1 7 7 – 1 8 8 179