Introduction

BC is a common malignant tumor, which often occurs in the elderly and with an estimated 81 400 new cases and 17 980 deaths in 2020 in the USA.1 Evaluating the extent of bladder wall invasion should be prioritized when determining a therapeutic strategy because it can directly affect the prognosis. Non-MIBC is usually managed by TURBT and intravesical treatment.2 MIBC is an aggressive tumor with a poor prognosis and is mainly treated by RC with or without chemotherapy and radiotherapy.3 Muscularis propria layer invasion is primarily diagnosed based on the postoperative pathological findings of TURBT. In addition, TURBT might lead to tumor understaging and missed muscle infiltration detection in up to 25% of invasive cancers.4 Approximately 10% (range 6–14%) of T1 tumors are upgraded to MIBC at reTURBT.5

Traditionally, CT and MRI are used to identify extra-bladder diseases (upper tract disease, lymph node invasion, or metastatic tumors6). A recent meta-analysis showed a promising value of bladder MRI in T staging, with Sen and Spe for differentiating ≤T1 and ≥T2 of 0.87 and 0.79, respectively.7 Studies on mpMRI combining functional sequences, such as DCE and DWI, and anatomical sequences, such as T2W, showed high efficiency in distinguishing MIBC from NMIBC.8, 9

Therefore, VI-RADS was introduced in 2018 to define a standardized approach to imaging and reporting mpMRI for patients with BC.10 VI-RADS established a 5-point scoring system clarifying the probability of muscle invasion through comprehensive estimation of the separate score of T2W, DWI and DCE, with recommendations on patient preparations and technical parameters. Thus far, several studies were carried out on the diagnostic efficacy of VI-RADs, with a focus on differentiation between MIBC and NMIBC. Systematic reviews have shown that the pooled Sen and Spe of VI-RADS for predicting MIBC are 0.83–0.90 and 0.90–0.86, respectively.11, 12 However, considering that VI-RADS has only recently been introduced, few studies were included. Therefore, we carried out a meta-analysis to evaluate the diagnostic performance of VI-RADS for the local staging of BC and to update the results.

Methods

The current systematic review was carried out in line with the Preferred Reporting Items for Systematic Review and Meta-analysis Diagnostic Test Accuracy guidelines. A research question for this meta-analysis was drawn based on the PICOS criteria as follows: what is the diagnostic performance of VI-RADS in predicting muscle invasion of BC compared with the pathological results reported in studies since 2018.

Literature search

Embase, PubMed, and Web of Science were systematically searched from 1 September 2018, when the first article was published, to 30 July 2021 to include proper studies with a search query combining synonyms of BC, mpMRI, and VI-RADS as follows: (bladder OR urothelial OR papillary OR transitional) AND (cancer OR carcinoma OR tumor OR tumor OR neoplasm) AND (mpMRI OR mp-MR OR “multiparametric magnetic resonance imaging”) AND (“Vesical Imaging–Reporting and Data System” OR “VI-RADS” OR VIRADS). The articles cited in the papers were screened to increase the quantity of relevant literature.

Study selection Inclusion criteria

We included studies that met the following PICOS criteria (10): (i) P: patients with BC; (ii) I: VI-RADS based on mpMRI used as the index test for MIBC assessment; (iii) C: post-surgery (RC or TURBT) pathology for standard reference; (iv) O: sufficient material obtained for reconstruction of 2 × 2 tables to calculate Sen and Spe; and (v) S: publication or online publication type of original articles.

Exclusion criteria

Studies were excluded on the basis of the following criteria: (i) <10 patients were enrolled in the study; (ii) other publication types apart from original articles (such as review articles, letters, conference abstracts, and editorials); (iii) studies focused on other subjects (e.g. diagnostic accuracy of mpMRI for local staging of BC, but without VI-RADS or VI-RADS provided were not for assessing diagnostic performance in deciding muscle invasion); (iv) studies not using 1.5 T or 3 T MRI scanners; (v) studies with overlapped patient population; and (vi) studies that did not provide sufficient information to reconstruct 2 × 2 tables.

The literature search and study selection were carried out by two reviewers. Should there be any disagreements, an independent reviewer was consulted to reach a consensus.

Data extraction and quality assessment Raw data were extracted from the selected studies by two reviewers and checked for consensus. When studies provided the results of multiple readers, the data with the highest Youden Index were selected. When studies have multiple readers but do not provide separate results, pooled results were used. Data were extracted using a standard form: Patient characteristics: study population, mean age and range, number of tumors, percentage of muscle-invasive tumors, and histological subtypes; Study characteristics: study origin (nation and authors), study design (prospective or retrospective, single-center or multicenter, consecutive enrolment or not), reference standard (cystectomy or TURBT, interval between mpMRI and surgery) and pathological assessment (blind or not). mpMRI characteristics: MRI scanner magnet strength (1.5 or 3.0 T), technical parameters (including ST of T2W, b value of DWI, SI of DCE and image acquisition planes), bladder distension preparation, time of mpMRI (before or after TUR) and mpMRI interpretation (reader number and experience, blind or not).

We assessed the quality of the selected studies using the QUADAS-2 tool. Quality control and consensus view were also carried out by two independent reviewers.

Statistical analysis

In this work, 2 × 2 tables were reconstructed with raw data from the selected studies to compute their Sen and Spe. When studies had multiple readers but did not provide separate results, pooled results were extracted to evaluate the overall accuracy. Cut-off values of VI-RADS = 3 and VI-RADS = 4 were used to evaluate the diagnostic accuracy independently. Pooled Sen and Spe were evaluated using hierarchical logistic regression modeling, with both bivariate and HSROC modeling. The results were plotted using an HSROC curve and 95% confidence, and prediction region. The Deeks’ funnel plot and the Deeks’ asymmetry test were used to test publication bias.

The Higgins I2 test with an inconsistency index was used to assess heterogeneity. The results were interpreted as follows: 0–40%, heterogeneity might not be important; 30–60%, moderate heterogeneity, 50–90%, substantial heterogeneity and 75–100%, considerable heterogeneity.13 Meta-regression analysis was carried out for possible covariates to explore the possible causes of heterogeneity. We used the MIDAS module of Stata software (version 15.1; StataCorp, College Station, TX, USA) and Review manager 5 (Version 5.3; Nordic Cochrane Centre, Copenhagen, Denmark) for statistical analyses, with a P-value of <0.05 showing statistical significance.

Results Literature search

The systematic literature search initially retrieved 199 articles. After 161 duplicates were removed and the titles and abstracts of the 38 remaining articles were screened, 26 potential articles were obtained. Full-text reviews were considered, and 20 studies met the inclusion criteria,14-33 whereas the other six studies were excluded for the following reasons: original study involving VI-RADS (n = 2), non-English language (n = 1) and insufficient information provided (n = 3). Ultimately, the 20 included studies that reported on 2725 patients with more than 2861 lesions provided the raw data of this meta-analysis, as shown in Figure 1.

Flowchart of literature search.

Study characteristics

Patient and study characteristics are shown in Table 1. The number of patients ranged from 30 to 340. Among the study population, the mean/medium age ranged from 57.2 to 76.0 years, and the percentage of male patients ranged from 63.3% to 92.0%. The number of tumor lesions ranged from 30 to 340, in which the proportion of MIBC ranged from 10.3% to 57.1%. The characteristics of the included studies were as follows: 13 out of 20 included studies were retrospective, and the seven other studies were prospective. All but one32 study were carried out at a single center, and all studies recruited their patients consecutively. Seven studies used a combination of TURBT and reTURBT for high-risk NMIBC as the reference standard, and the remaining 13 used a combination of TURBT and radical or PC.

Table 1. Patient and study characteristics Author Patient characteristics Tumor characteristics Study design Reference standard No. Age (years)† Male (%) n MIBC (%) Prospective Single center Consecutive enrollment Surgery MRI-surgery interval Barchetti G 75 69.0 62 (80.3) 75 22 (29.3) Retrospective Yes Yes TUR, RC <6 weeks Makboul M 50 57.2 46 (92.0) 50 18 (36.0) Prospective Yes Yes TUR‡ NR Ueno Y 74 72.8 59 (79.7) 74 37 (50) Retrospective Yes Yes TUR‡ NR Wang H 340 64.0 296 (87.1) 340 85 (25) Retrospective Yes Yes TUR, PC, RC <2 weeks Del Giudice F 231 NR NR 231 62 (26.8) Prospective Yes Yes TUR, RC 2–6 weeks Hong SB 32 73.0 27 (84.4) 66 8 (25) Retrospective Yes Yes TUR, RC <2 weeks Kim SH 297 65.5 221 (74.5) 339 102 (30.1) Retrospective Yes Yes TUR, PC, RC NR Wang ZY 220 65.7 194 (88.2) 220 113 (51.4) Retrospective Yes Yes TUR, PC, RC <2 weeks Liu SH 126 67.6 104 (82.5) 126 50 (39.7) Retrospective Yes Yes TUR, RC <2 weeks Marchioni M 38 72.5 27 (71.1) 68 7 (10.3) Prospective Yes Yes TUR‡ NR Arita Y 66 74 60 (90.9) 66 17 (25.8) Retrospective Yes Yes TUR, RC NR Ahn H 82 68.6 73 (89.0) 82 19 (23.2) Retrospective Yes Yes TUR, RC NR Vaz A 30 68.0 19 (63.3) 30 8 (26.6) Retrospective Yes Yes TUR‡ 5–67 days Taguchi S 68 76.0 56 (82.0) 68 16 (26.0) Prospective Yes Yes TUR‡ 14 (7–22)§ Gmeiner J 51 68.2 48 (84.2)¶ 51 14 (27.5) Retrospective Yes Yes TUR, RC <1 week Kufukihara R 61 71.2 55 (90.2) 61 16 (26.2) Retrospective Yes Yes TUR‡ NR Wang X 179 67.1 147 (82.1) 179 65 (36.3) Retrospective Yes Yes TUR, RC <2 weeks Erkoc M 330 58.7 NR 330 94 (28.5) Prospective Yes Yes TUR‡ 2–4 weeks Metwally MI 331 61.9 274 (82.8) 331 189 (57.1) Prospective No Yes TUR, RC <1 week Delli Pizzi A 38 72.5 27 (71.1) 68 7 (18.4) Prospective Yes Yes TUR, RC NR

The mpMRI characteristics are shown in Table 2. A total of 12 studies used scanners that provided a magnet strength of 3 T, whereas eight used scanners with a magnet strength of 1.5 T or the combination of 1.5 and 3 T. The bladder was distended in all 20 studies using multiple methods, such as holding back urine, oral hydration, saline injection using a catheter or drugs that inhibit bladder contraction. MpMRI was carried out before TURBT in all cases except for one study,26 in which 26.6% of patients underwent MRI after bladder biopsy or TURBT. Parameters differed among the included studies, and all studies met the requirement to obtain sufficient information that followed the recommendations of the VI-RADS guideline.10 In all but seven studies, mpMRI was interpreted by two readers, and each of the remaining studies had one, three, four and five readers.16, 19, 24, 27, 28, 32, 33 Readers were blind to the reference standard in all the studies. The cut-off value of VI-RADS ≥3 versus ≤2 was used in 18 studies, and the cut-off value of VI-RADS ≥4 versus ≤3 was used in all but three studies.15, 25, 31

Table 2. MpMRI characteristics Author Magnet strength Technical parameters Readers T2W DWI DCE Plane ST (mm) Plane b values (s/mm2) Plane SI (s) Barchetti G 3 T All 3–4 Axial 0–2000 Axial 5 2 Makboul M 1.5 T All 3 Axial 0–1000 Axial 50/110† 2 Ueno Y 1.5 T/3 T All 4 Axial 0, 1000 Sagittal, OA 40 5 Wang H 3 T All 4 Axial, OS 0, 1000 Axial/sagittal NR 2 Del Giudice F 3 T All 3–4 Axial 0–2000 Axial 5 2 Hong SB 3 T All 4 Axial 0–1000 Axial 5 3 Kim SH 3 T All 3 Axial 0, 1000 Axial/sagittal 30 2 Wang ZY 3 T All NR All NR All NR 2 Liu SH 3 T All NR Axial 800, 1000 All NR 2 Marchioni M 3 T All 3–4 All 0–2000 All 30 2 Arita Y 1.5 T All 2 Axial/sagittal 0–1000 All NR 4 Ahn H 3 T NR 4 NR 0, 1000 NR NR 2 Vaz A 1.5 T All NR Axial 50, 400, 800 Axial 60/300† 2 Taguchi S 3 T Axial/oblique 2/4 Axial 0–1000 Oblique 30 1 Gmeiner J 3 T All 3 Axial 50, 400, 800 Axial 6 2 Kufukihara R 1.5 T NR NR Axial/sagittal NR Axial or sagittal NR 2 Wang X 1.5 T/3 T All NR Axial/sagittal 0, 1000 Axial NR 3 Erkoc M 1.5 T All 3–4 Axial 0–2000 Axial 5 2 Metwally MI 1.5 T All 4–6 Axial 0, 800–1000 All NR 4 Delli Pizzi A 3 T All 4 Axial/sagittal 0–2000 Axial NR 4 All represents axial, sagittal, coronal. Quality assessment

Figure 2 shows the risks of bias and applicability concerns of the included studies using QUADAS-2. High risk of bias in patient selection was observed in one study, as it included post-TURBT cases with suspicious focal bladder wall thickening.26 Unclear risk was shown in another study because the patients were counted twice in the pooled estimation.25 In terms of the index test domain, one study discovered a substantial risk of bias, as VI-RADS scores were evaluated by a single radiologist.27 Six studies had an unclear risk of bias due to the extraction of data with the highest Youden Index to represent the pooled result in the absence of consensus and interpretation of VI-RADS by inexperienced physicians. Regarding the reference standard domain, one study had high risk of bias due to the omission of re-TURBT from the MRI evaluation when the VI-RADS cut-off value of ≥3 versus ≤2 was used.32 Eight studies had unclear risk of bias because not enough information on the pathologists was given in the article.15-17, 20-23,