What Is the Negative Predictive Value of Multiparametric Magnetic Resonance Imaging in Excluding Prostate Cancer at Biopsy? A Systematic Review and Meta-analysis from the European Association of Urology Prostate Cancer Guidelines Panel

2.1 Objective

Our primary aim was to systematically evaluate the performance of negative prebiopsy prostate mpMRI in predicting a negative biopsy result for overall PCa and csPCa in biopsy-naïve men and in men with previously negative biopsies. A further objective was to explore and define factors that may contribute to relevant thresholds in order to provide guidance for future studies.

2.2 Data acquisition and search strategy

The review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement [9] . The review protocol was published in PROSPERO database ( http://www.crd.york.ac.uk/PROSPERO; registration number CRD42015021929). Databases searched included the Embase and OVID Medline databases, the Cochrane database of systematic reviews, and the Cochrane Central Register for Clinical Trials, covering from January 1, 2000 to February 13, 2016. Systematic or standard prostate biopsies were used as reference standards, with positive or negative cases of PCa being determined by histopathological examination. The detailed search strategy is presented in Supplement 1.

2.3 Inclusion and exclusion criteria

Included studies focused on men who were assessed for suspected PCa by mpMRI before undergoing prostate biopsy. Studies enrolling both biopsy-naïve men and men who had undergone previous negative biopsies were included. Prebiopsy prostate mpMRI was considered the index test and comprised T2-weighted imaging (T2WI) and at least one functional imaging technique (diffusion-weighted imaging [DWI], dynamic contrast-enhanced imaging [DCEI], or magnetic resonance spectroscopic imaging [MRSI]). For inclusion, studies had to report on 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 characteristics 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 prevalence 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 interquartile 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-naïve 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-naïve versus previous negative biopsy; patients with positive versus negative digital rectal examination (DRE); mpMRI performed with an endorectal versus without an endorectal coil; transrectal ultrasound (TRUS) versus template transperineal (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.

3.1 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 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 .

 

Fig. 1
Preferred Reporting Items for Systematic Reviews and Meta-analysis flow chart. Bx = biopsy; CAD = computer-aided diagnosis; mpMRI = multiparametric magnetic resonance imaging; RP = radical prostatectomy; T2WI = T2-weighted imaging.

3.2 Quality of studies

Out of the 48 included studies, 42 were single-centre and six were multicentre studies. Thirty-four studies were prospective 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 A and 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.

 

Fig. 2
(A) Assessment of the risk of bias for included studies. (B) Risk of bias summary graph. RoB = risk of bias.

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-naïve 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 35 studies, respectively. Nineteen studies also added MRSI. An endorectal coil was used in 18 studies. The definition of positive mpMRI varied across studies. The PI-RADS v1 score was used in 12 studies, a five-level subjective (Likert) score was used in eight studies, and one study reported data based on the two scoring systems. In-house criteria were used in 13 studies for defining positive mpMRI, and five studies used a dichotomous definition. Nine studies did not report on the criteria for positive mpMRI. No study used the PI-RADS v2 score.

Regarding the reference standard, TRUS-guided biopsies were used in 39 studies, TTP biopsies in six studies, and mixed TRUS-guided and TTP biopsies in two studies. In one study, the biopsy approach was unclear. The number of cores per biopsy procedure was ≤16 in 30 studies, >16 in nine studies, and variable among patients in three studies. For six studies, the number of biopsy cores taken was unclear.

3.4 NPV of prebiopsy mpMRI

At patient level, the median biopsy positivity rate (ie, cancer prevalence) was 50.4% (IQR, 36.4–57.7%) for overall cancer and 32.9% (IQR, 28.1–37.2%) for csPCa ( Table 2 ). The median mpMRI NPV was 82.4% (IQR, 69.0–92.4%) for overall cancer and 88.1% (IQR, 85.7–92.3) for csPCa. NPV significantly decreased when cancer prevalence increased, both for overall cancer ( r = –0.64, p < 0.0001) and csPCa ( r = –0.75, p = 0.032; Fig. 3 ). In addition, NPV was highly dependent on the definition used for csPCa, with differences of up to 21% when several definitions were used in the same dataset [12 13 38 47 48] .

 

Fig. 3
Negative predictive value of prebiopsy multiparametric MRI as a function of cancer prevalence (blue crosses: overall prostate cancer; red crosses: clinically significant prostate cancer). The blue line is the correlation line for overall prostate cancer; the red dotted line is the correlation line for clinically significant prostate cancer. mpMRI = multiparametric magnetic resonance imaging; MRI = magnetic resonance imaging.

Cancer prevalence tended to be higher and mpMRI NPV lower in the biopsy-naïve group as compared with the repeat biopsy group, in men with positive DRE as compared with men with negative DRE and when an endorectal coil was not used ( Table 3 ). There were no clear differences in the prevalence and NPV of the other analysed subgroups (TRUS-guided vs TTP biopsy, biopsy procedures with ≤16 cores vs >16 cores; Table 3 ). However, comparisons must be interpreted with care, due to the small number of studies in some subgroups. In patients with a PSA level of ≤10 ng/ml, the median NPV for overall PCa was 86.3% (IQR, 73.3–93.6%) for a median cancer prevalence of 35.4% (IQR, 27.6–42.5%).

3.5 Meta-analysis

3.5.1 NPV and PPV for overall PCa

Eight studies reported NPV at patient level for overall PCa and fulfilled the inclusion criteria for meta-analysis ( Table 4 ) [22 25 38 41 43 46 56 57] .

		Study	Prevalence a (%)	Neg MRI (%)	NPV (%)	PPV (%)	Spe (%)	Se (%)
All PCa	Score ≥ 3/5	Grenabo Bergdahl (2016) [22]	31.3	66.9	83.7	47.5	63.2	73.1
		Numao (2013) [38]	45	29.4	70.6	64.3	71.0	63.9
		Hauth (2015) [25]	45.7	14.3	85.7	48.3	11.8	97.7
		Vinet (2013) [57]	49.3	24.6	64.7	53.8	31.4	82.4
		Radtke (2015) [46]	51	27	73	73.2	71.5	74.7
		Thompson (2014) [56]	61.3	31.4	68.6	76.8	60.3	82.6
		Pokorny (2014) [43]	63.7	30.9	69.1	82.4	69.1	82.4
	Score ≥ 4/5	Pepe (2015) [41]	37	31	74.2	42	36.5	78.4%
		Vinet (2013) [57]	49.3	47.8	66.7	63.9	62.9	67.6%
		Radtke (2015) [46]	51	73.5	63.9	92.3	95.8	48%
Gleason ≥7 PCa	Score ≥ 3/5	Radtke (2015) [46]	29.3	27	87.9	45.1	59.6	80.2%

^a Prevalence of overall prostate cancers (10 first lines) or Gleason ≥7 cancers (last line).

PCa = prostate cancer; MRI = magnetic resonance imaging; Neg MRI = proportion of negative magnetic resonance imaging; NPV = negative predictive value; PPV = positive predictive value; Se = sensitivity; Spe = specificity.

Table 4Prebiopsy multiparametric MRI results in the series selected for the meta-analysis

 

Seven studies used a score of ≥3/5 for defining positive mpMRI ( Fig. 4 A and B) [22 25 38 43 46 56 57] . Fig. 4 C shows the conditional probability plot of 1 – NPV and PPV as a function of overall PCa prevalence. Table 5 shows NPV and PPV estimates for the given values of PCa prevalence.

 

Fig. 4
Forest plot showing the (A) NPV and (B) PPV of prebiopsy multiparametric MRI for overall prostate cancer in the seven studies selected for meta-analysis that used a cut-off score of ≥3/5 for defining positive MRI. Studies have been ranked according to cancer prevalence (left column). Intervals in the right column are 95% CIs of the (A) NPV or (B) PPV. As NPV and PPV vary with cancer prevalence, combined estimates of NPV and PPV have not been provided. (C) Conditional probability plot showing the estimation of the combined NPV and PPV in the seven studies, as a function of the prevalence of overall prostate cancer. The x axis (prior probability) indicates the overall prostate cancer prevalence. The y axis (posterior probability) indicates either PPV (dashed line, upper quadrant) or 1 – NPV (dotted line, lower quadrant). CI = confidence interval; MRI = magnetic resonance imaging; NPV = negative predictive value; PPV = positive predictive value.

PCaPrev	PPV	NPV
0.30	0.43 (0.34–0.53)	0.88 (0.77–0.99)
0.40	0.54 (0.45–0.64)	0.82 (0.70–0.94)
0.50	0.64 (0.55–0.73)	0.76 (0.64–0.88)
0.60	0.73 (0.65–0.80)	0.67 (0.56–0.79)
0.70	0.81 (0.75–0.87)	0.57 (0.47–0.67)
0.75	0.84 (0.79–0.89)	0.51 (0.42–0.59)

PCaPrev = prevalence of prostate cancer; PPV = positive predictive value; MRI = magnetic resonance imaging; NPV = negative predictive value.
Intervals in parenthesis are 95% confidence intervals. A score of ≥3/5 was used to define positive MRI.

Table 5Positive and negative predictive estimates for prebiopsy multiparametric MRI as a function of prostate cancer prevalence (meta-analysis)

 

Only three studies used a score of ≥4/5 for defining positive mpMRI ( Table 4 ) [41 46 57] , and a formal meta-analysis could not be performed.

3.6 Discussion

2.1 Objective

Our primary aim was to systematically evaluate the performance of negative prebiopsy prostate mpMRI in predicting a negative biopsy result for overall PCa and csPCa in biopsy-naïve men and in men with previously negative biopsies. A further objective was to explore and define factors that may contribute to relevant thresholds in order to provide guidance for future studies.

2.2 Data acquisition and search strategy

The review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement [9] . The review protocol was published in PROSPERO database ( http://www.crd.york.ac.uk/PROSPERO; registration number CRD42015021929). Databases searched included the Embase and OVID Medline databases, the Cochrane database of systematic reviews, and the Cochrane Central Register for Clinical Trials, covering from January 1, 2000 to February 13, 2016. Systematic or standard prostate biopsies were used as reference standards, with positive or negative cases of PCa being determined by histopathological examination. The detailed search strategy is presented in Supplement 1.

2.3 Inclusion and exclusion criteria

Included studies focused on men who were assessed for suspected PCa by mpMRI before undergoing prostate biopsy. Studies enrolling both biopsy-naïve men and men who had undergone previous negative biopsies were included. Prebiopsy prostate mpMRI was considered the index test and comprised T2-weighted imaging (T2WI) and at least one functional imaging technique (diffusion-weighted imaging [DWI], dynamic contrast-enhanced imaging [DCEI], or magnetic resonance spectroscopic imaging [MRSI]). For inclusion, studies had to report on 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 characteristics 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 prevalence 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 interquartile 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-naïve 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-naïve versus previous negative biopsy; patients with positive versus negative digital rectal examination (DRE); mpMRI performed with an endorectal versus without an endorectal coil; transrectal ultrasound (TRUS) versus template transperineal (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.

3.1 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 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 .

 

Fig. 1
Preferred Reporting Items for Systematic Reviews and Meta-analysis flow chart. Bx = biopsy; CAD = computer-aided diagnosis; mpMRI = multiparametric magnetic resonance imaging; RP = radical prostatectomy; T2WI = T2-weighted imaging.

3.2 Quality of studies

Out of the 48 included studies, 42 were single-centre and six were multicentre studies. Thirty-four studies were prospective 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 A and 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.

 

Fig. 2
(A) Assessment of the risk of bias for included studies. (B) Risk of bias summary graph. RoB = risk of bias.

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-naïve 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 35 studies, respectively. Nineteen studies also added MRSI. An endorectal coil was used in 18 studies. The definition of positive mpMRI varied across studies. The PI-RADS v1 score was used in 12 studies, a five-level subjective (Likert) score was used in eight studies, and one study reported data based on the two scoring systems. In-house criteria were used in 13 studies for defining positive mpMRI, and five studies used a dichotomous definition. Nine studies did not report on the criteria for positive mpMRI. No study used the PI-RADS v2 score.

Regarding the reference standard, TRUS-guided biopsies were used in 39 studies, TTP biopsies in six studies, and mixed TRUS-guided and TTP biopsies in two studies. In one study, the biopsy approach was unclear. The number of cores per biopsy procedure was ≤16 in 30 studies, >16 in nine studies, and variable among patients in three studies. For six studies, the number of biopsy cores taken was unclear.

3.4 NPV of prebiopsy mpMRI

At patient level, the median biopsy positivity rate (ie, cancer prevalence) was 50.4% (IQR, 36.4–57.7%) for overall cancer and 32.9% (IQR, 28.1–37.2%) for csPCa ( Table 2 ). The median mpMRI NPV was 82.4% (IQR, 69.0–92.4%) for overall cancer and 88.1% (IQR, 85.7–92.3) for csPCa. NPV significantly decreased when cancer prevalence increased, both for overall cancer ( r = –0.64, p < 0.0001) and csPCa ( r = –0.75, p = 0.032; Fig. 3 ). In addition, NPV was highly dependent on the definition used for csPCa, with differences of up to 21% when several definitions were used in the same dataset [12 13 38 47 48] .

 

Fig. 3
Negative predictive value of prebiopsy multiparametric MRI as a function of cancer prevalence (blue crosses: overall prostate cancer; red crosses: clinically significant prostate cancer). The blue line is the correlation line for overall prostate cancer; the red dotted line is the correlation line for clinically significant prostate cancer. mpMRI = multiparametric magnetic resonance imaging; MRI = magnetic resonance imaging.

Cancer prevalence tended to be higher and mpMRI NPV lower in the biopsy-naïve group as compared with the repeat biopsy group, in men with positive DRE as compared with men with negative DRE and when an endorectal coil was not used ( Table 3 ). There were no clear differences in the prevalence and NPV of the other analysed subgroups (TRUS-guided vs TTP biopsy, biopsy procedures with ≤16 cores vs >16 cores; Table 3 ). However, comparisons must be interpreted with care, due to the small number of studies in some subgroups. In patients with a PSA level of ≤10 ng/ml, the median NPV for overall PCa was 86.3% (IQR, 73.3–93.6%) for a median cancer prevalence of 35.4% (IQR, 27.6–42.5%).

3.5 Meta-analysis

3.5.1 NPV and PPV for overall PCa

Eight studies reported NPV at patient level for overall PCa and fulfilled the inclusion criteria for meta-analysis ( Table 4 ) [22 25 38 41 43 46 56 57] .

		Study	Prevalence a (%)	Neg MRI (%)	NPV (%)	PPV (%)	Spe (%)	Se (%)
All PCa	Score ≥ 3/5	Grenabo Bergdahl (2016) [22]	31.3	66.9	83.7	47.5	63.2	73.1
		Numao (2013) [38]	45	29.4	70.6	64.3	71.0	63.9
		Hauth (2015) [25]	45.7	14.3	85.7	48.3	11.8	97.7
		Vinet (2013) [57]	49.3	24.6	64.7	53.8	31.4	82.4
		Radtke (2015) [46]	51	27	73	73.2	71.5	74.7
		Thompson (2014) [56]	61.3	31.4	68.6	76.8	60.3	82.6
		Pokorny (2014) [43]	63.7	30.9	69.1	82.4	69.1	82.4
	Score ≥ 4/5	Pepe (2015) [41]	37	31	74.2	42	36.5	78.4%
		Vinet (2013) [57]	49.3	47.8	66.7	63.9	62.9	67.6%
		Radtke (2015) [46]	51	73.5	63.9	92.3	95.8	48%
Gleason ≥7 PCa	Score ≥ 3/5	Radtke (2015) [46]	29.3	27	87.9	45.1	59.6	80.2%

^a Prevalence of overall prostate cancers (10 first lines) or Gleason ≥7 cancers (last line).

PCa = prostate cancer; MRI = magnetic resonance imaging; Neg MRI = proportion of negative magnetic resonance imaging; NPV = negative predictive value; PPV = positive predictive value; Se = sensitivity; Spe = specificity.

Table 4Prebiopsy multiparametric MRI results in the series selected for the meta-analysis

 

Seven studies used a score of ≥3/5 for defining positive mpMRI ( Fig. 4 A and B) [22 25 38 43 46 56 57] . Fig. 4 C shows the conditional probability plot of 1 – NPV and PPV as a function of overall PCa prevalence. Table 5 shows NPV and PPV estimates for the given values of PCa prevalence.

 

Fig. 4
Forest plot showing the (A) NPV and (B) PPV of prebiopsy multiparametric MRI for overall prostate cancer in the seven studies selected for meta-analysis that used a cut-off score of ≥3/5 for defining positive MRI. Studies have been ranked according to cancer prevalence (left column). Intervals in the right column are 95% CIs of the (A) NPV or (B) PPV. As NPV and PPV vary with cancer prevalence, combined estimates of NPV and PPV have not been provided. (C) Conditional probability plot showing the estimation of the combined NPV and PPV in the seven studies, as a function of the prevalence of overall prostate cancer. The x axis (prior probability) indicates the overall prostate cancer prevalence. The y axis (posterior probability) indicates either PPV (dashed line, upper quadrant) or 1 – NPV (dotted line, lower quadrant). CI = confidence interval; MRI = magnetic resonance imaging; NPV = negative predictive value; PPV = positive predictive value.

PCaPrev	PPV	NPV
0.30	0.43 (0.34–0.53)	0.88 (0.77–0.99)
0.40	0.54 (0.45–0.64)	0.82 (0.70–0.94)
0.50	0.64 (0.55–0.73)	0.76 (0.64–0.88)
0.60	0.73 (0.65–0.80)	0.67 (0.56–0.79)
0.70	0.81 (0.75–0.87)	0.57 (0.47–0.67)
0.75	0.84 (0.79–0.89)	0.51 (0.42–0.59)

PCaPrev = prevalence of prostate cancer; PPV = positive predictive value; MRI = magnetic resonance imaging; NPV = negative predictive value.
Intervals in parenthesis are 95% confidence intervals. A score of ≥3/5 was used to define positive MRI.

Table 5Positive and negative predictive estimates for prebiopsy multiparametric MRI as a function of prostate cancer prevalence (meta-analysis)

 

Only three studies used a score of ≥4/5 for defining positive mpMRI ( Table 4 ) [41 46 57] , and a formal meta-analysis could not be performed.

3.6 Discussion

2.1 Objective

Our primary aim was to systematically evaluate the performance of negative prebiopsy prostate mpMRI in predicting a negative biopsy result for overall PCa and csPCa in biopsy-naïve men and in men with previously negative biopsies. A further objective was to explore and define factors that may contribute to relevant thresholds in order to provide guidance for future studies.

2.2 Data acquisition and search strategy

The review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement [9] . The review protocol was published in PROSPERO database ( http://www.crd.york.ac.uk/PROSPERO; registration number CRD42015021929). Databases searched included the Embase and OVID Medline databases, the Cochrane database of systematic reviews, and the Cochrane Central Register for Clinical Trials, covering from January 1, 2000 to February 13, 2016. Systematic or standard prostate biopsies were used as reference standards, with positive or negative cases of PCa being determined by histopathological examination. The detailed search strategy is presented in Supplement 1.

2.3 Inclusion and exclusion criteria

Included studies focused on men who were assessed for suspected PCa by mpMRI before undergoing prostate biopsy. Studies enrolling both biopsy-naïve men and men who had undergone previous negative biopsies were included. Prebiopsy prostate mpMRI was considered the index test and comprised T2-weighted imaging (T2WI) and at least one functional imaging technique (diffusion-weighted imaging [DWI], dynamic contrast-enhanced imaging [DCEI], or magnetic resonance spectroscopic imaging [MRSI]). For inclusion, studies had to report on 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 characteristics 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 prevalence 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 interquartile 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-naïve 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-naïve versus previous negative biopsy; patients with positive versus negative digital rectal examination (DRE); mpMRI performed with an endorectal versus without an endorectal coil; transrectal ultrasound (TRUS) versus template transperineal (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.

3.1 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 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 .

 

Fig. 1
Preferred Reporting Items for Systematic Reviews and Meta-analysis flow chart. Bx = biopsy; CAD = computer-aided diagnosis; mpMRI = multiparametric magnetic resonance imaging; RP = radical prostatectomy; T2WI = T2-weighted imaging.

3.2 Quality of studies

Out of the 48 included studies, 42 were single-centre and six were multicentre studies. Thirty-four studies were prospective 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 A and 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.

 

Fig. 2
(A) Assessment of the risk of bias for included studies. (B) Risk of bias summary graph. RoB = risk of bias.

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-naïve men and men with a history of previous negative biopsy in nine studies. In 14 studies, the biopsy history of the patients was unclear.