Predictive model containing PI-RADS v2 score for postoperative seminal vesicle invasion among prostate cancer patients.

BACKGROUND: Seminal vesicle invasion (SVI) is considered to be one of most adverse prognostic findings in prostate cancer, affecting the biochemical progression-free survival and disease-specific survival. Multiparametric magnetic resonance imaging (mpMRI) has shown excellent specificity in diagnosis of SVI, but with poor sensitivity. The aim of this study is to create a model that includes the Prostate Imaging Reporting and Data System version 2 (PI-RADS v2) score to predict postoperative SVI in patients without SVI on preoperative mpMRI.
METHODS: A total of 262 prostate cancer patients without SVI on preoperative mpMRI who underwent radical prostatectomy (RP) at our institution from January 2012 to July 2019 were enrolled retrospectively. The prostate-specific antigen levels in all patients were <10 ng/mL. Univariate and multivariate logistic regression analyses were used to assess factors associated with SVI, including the PI-RADS v2 score. A regression coefficient-based model was built for predicting SVI. The receiver operating characteristic curve was used to assess the performance of the model.
RESULTS: SVI was reported on the RP specimens in 30 patients (11.5%). The univariate and multivariate analyses revealed that biopsy Gleason grade group (GGG) and the PI-RADS v2 score were significant independent predictors of SVI (all P<0.05). The area under the curve of the model was 0.746 (P<0.001). The PI-RADS v2 score <4 and Gleason grade <8 yielded only a 1.8% incidence of SVI with a high negative predictive value of 98.2% (95% CI, 93.0-99.6%).
CONCLUSIONS: The PI-RADS v2 score <4 in prostate cancer patients with prostate-specific antigen level <10 ng/mL is associated with a very low risk of SVI. A model based on biopsy Gleason grade and PI-RADS v2 score may help to predict SVI and serve as a tool for the urologists to make surgical plans. 2021 Translational Andrology and Urology. All rights reserved.

Introduction

The incidence of seminal vesicle invasion (SVI) has decreased over time because of the application of serum prostate specific antigen (PSA) for prostate cancer screening, and the prevalence of SVI in a contemporary surgical series was 3–17.6% (1-5). SVI is a well-established indicator of adverse prognosis. Compared to organ-confined disease or extra prostatic extension (EPE), SVI is associated with worse outcome, a higher rate of recurrence and mortality (6). In patients with suspected SVI, multimodal therapy should be considered, such as radical prostatectomy (RP) combined with radiotherapy. Due to poor prognosis, treatment and operation should think carefully for patients with SVI. Therefore, accurate presurgical diagnosis of SVI is critical for urologists to make a treatment decision.

Several models that rely on clinical information and biopsy data have been developed for the prediction of prostate cancer staging, such as the Partin tables (7) and the nomogram developed by Gallina et al. (8). These models show high sensitivity (90%) but low specificity (30–60%) for the detection of SVI (9). On the contrary, the prediction accuracy of multiparametric magnetic resonance imaging (mpMRI) for SVI reveals high specificity (>90%) but poor and heterogeneous sensitivity (30–70%) overall (10). The recent models that added mpMRI information to predictive models yield better accuracy, such as the combination of mpMRI and the Partin tables (11) and the novel nomogram developed by Martini et al. (12). However, these models rely on the SVI findings of mpMRI, and none of the models currently integrate the Prostate Imaging Reporting and Data System version 2 (PI-RADS v2) score to predict post-operative SVI in patients without SVI on preoperative mpMRI.

Therefore, we aimed to construct a new model based on the PI-RADS v2 for the prediction of SVI in patients without SVI on preoperative mpMRI. This predictive model may also serve as a tool for the urologists to make surgical plans. We present the following article in accordance with the TRIPOD reporting checklist (available at http://dx.doi.org/10.21037/tau-20-989).

Methods

Patient population

We retrospectively reviewed a total of 787 patients who underwent RP for prostate cancer at our institution between January 2012 and July 2019. The inclusion criteria were as follows: patients with a preoperative serum PSA level <10 ng/mL who underwent preoperative mpMRI before prostate biopsies. Patients who underwent prior hormonal therapy or radiotherapy as well as patients with incomplete data were excluded from the cohort. Seven patients were excluded because of suspicious SVI on preoperative mpMRI. Finally, a total of 262 patients were enrolled in the study.

Clinicopathological characteristics

All patient data, including age, body mass index (BMI), preoperative PSA level, free/total PSA ratio (f/t PSA), prostate volume (PV) measured by trans rectal ultrasound (TRUS), the percentage of positive systematic biopsies, maximum cancer percentage per core, clinical stage, biopsy Gleason score (GS), and pathological characteristics of specimens following RP, were collected. PSA density (PSAD) was calculated by dividing total PSA by the PV. Cancer of the prostate risk assessment (CAPRA) score was calculated according to the UCSF-CAPRA scoring system.

Biopsy procedure and histopathology

All patients underwent TRUS-guided systematic biopsies. All biopsy specimens were evaluated by two dedicated genitourinary pathologists to determine the cancer diagnosis and the GS in positive cases. The patients were classified into the following five groups using the new GS grading system: grade group 1, GS 6; grade group 2, GS 3+4=7; grade group 3, GS 4+3=7; grade group 4, GS 8; and grade group 5, GS 9 and 10.

MpMRI

MRI was performed using a 3.0T Discovery MR750 HDx (GE Healthcare, Waukesha, WI, USA) without the use of an endorectal coil. The imaging protocol included axial T1-weighted images of the pelvis, axial T2-weighted fast spin-echo images centered on the prostate, and dynamic contrast-enhanced images. In addition, axial diffusion-weighted imaging was performed with b-values of 0, 800, and 1,400 s/mm2.

MpMRI interpretation.

MRI images were retrospectively interpreted by one of the two experienced radiologists with >5 years’ experience in reading prostate MRIs. Any disagreement in the process of interpretation was resolved by the senior adjudicating radiologist. According to PI-RADS v2 assessment categories, clinically significant cancer is highly unlikely or unlikely to be present in lesions of PI-RADS 1 or 2 (13,14). What’s more, lesions with PI-RADS >2 were defined as MRI-visible lesion, which could be considered for targeted biopsy. For patients with PI-RADS 3, it may be beneficial to perform follow-up rather than immediate biopsy, as most lesions can be reclassified after a manageable period of time (15). PI-RAD 4 or 5 means highly or very highly likely existence of clinically significant cancer, which biopsy should be considered (14). Targeted MR biopsy should be considered for PI-RADS assessment category 4 or 5 lesions but not for PI-RADS 1 or 2 (16). The 3 point “Likert” scale was associated directly with clinical decisions. So, the probability of cancer was evaluated and scored on a three-point scale based on the PI-RADS v2 score, where group “negative” (PI-RADS 1–2) = low probability, group “suspicious” (PI-RADS 3) = equivocal, and group “positive” (PI-RADS 4–5) = high or very high probability.

Statistical analysis

The endpoint of the study was the identification of the presence of SVI on the RP specimens. Univariate analysis was performed to investigate the associations between clinical and pathological risk factors and the presence of SVI in patients with negative SVI on mpMRI. The factors evaluated for the prediction of SVI were age, BMI, preoperative PSA level, f/t PSA, PV, PSAD, the percentage of positive systematic biopsies, maximum cancer percentage per core, CAPRA score, Gleason grade group (GGG), and PI-RADS v2 score. Continuous variables were compared using Student’s t-test and the Mann–Whitney U test, and categorical variables were compared using Chi-square test or Fisher’s exact text, as appropriate. Univariate and multivariate binary logistic regression analyses were conducted to identify independent predictors of SVI.

We constructed the logistic regression model for prediction of the diagnosis of SVI, by utilizing selected variables based on the results of multivariate logistic regression analysis. Discrimination was measured using the area under the curve (AUC) derived from the receiver operating characteristic (ROC) curves. With the optimal cutoff value according to Youden’s index, the performance of the model was assessed through analysis of sensitivity, specificity, positive predictive value, and negative predictive value.

Statistical analyses were performed using SPSS (version 24.0; SPSS Inc., Chicago, IL, USA). All analyses were two-sided, with statistical significance set at P<0.05. All procedures performed in this study were in accordance with the Declaration of Helsinki (as revised in 2013). Due to no influence on therapeutic strategy or need for patients’ follow-up, the Institutional Review Board replied that there was no need for ethical approval. And because of the retrospective nature of the research, the requirement for informed consent was waived.

Results

Baseline clinicopathological characteristics of the study cohort

The baseline clinical and pathological characteristics of the total 262 patients are summarized in . The median age was 66 years, the interquartile range (IQR) was 62–71 years, the median preoperative PSA level was 7.51 ng/mL (IQR, 6.07–8.63 ng/mL), and 30 patients (11.5%) presented with SVI on RP specimens.

Table 1

Patients’ characteristics, N=262

Variable	Total (n=262)	Histological SVI
		No (n=232)	Yes (n=30)
Age at surgery, median (IQR), years	66 [62–71]	66 [62–71]	65 (60–70.25)
BMI, median (IQR), kg/m2	24.47 (23.18–26.13)	24.49 (23.21–26.13)	24.19 (23.15–26.19)
PSA, median (IQR), ng/mL	7.51 (6.07–8.63)	7.44 (5.91–8.61)	7.75 (6.49–9.34)
f/t PSA, median (IQR)	0.13 (0.09–0.16)	0.13 (0.10–0.17)	0.12 (0.09–0.14)
PV, median (IQR), mL	38.20 (28.55–52.25)	38.74 (28.38–55.15)	34.78 (28.55–46.83)
PSAD, median (IQR), ng/mL2	0.18 (0.12–0.25)	0.18 (0.12–0.25)	0.22 (0.16–0.27)
The percentage of positive systematic biopsies, median (IQR)	0.25 (0.10–0.42)	0.25 (0.08–0.42)	0.33 (0.17–0.56)
Maximum cancer percentage per core, median (IQR)	0.70 (0.30–0.70)	0.70 (030–0.70)	0.70 (0.70–0.73)
Clinical stage, No. (%)
T1c	37 (14.1)	33 (14.2)	4 (13.3)
T2a	64 (24.4)	58 (25.0)	6 (20.0)
T2b	60 (22.9)	53 (22.8)	7 (23.3)
T2c	92 (35.1)	81 (34.9)	11 (36.7)
T3a	9 (3.4)	7 (3.0)	2 (6.7)
GGG, No. (%)
1	79 (30.2)	75 (32.3)	4 (13.3)
2	108 (41.2)	96 (41.4)	12 (40.0)
3	37 (14.1)	35 (15.1)	2 (6.7)
4	24 (9.2)	17 (7.3)	7 (23.3)
5	14 (5.3)	9 (3.9)	5 (16.7)
CAPRA score, median (IQR)	3.00 (2.00–5.00)	3.00 (2.00–4.00)	4.00 (3.00–5.00)
PI-RADS, No. (%)
Negative	40 (15.3)	39 (16.8)	1 (3.3)
Suspicious	79 (30.2)	76 (32.8)	3 (10.0)
Positive	143 (54.6)	117 (50.4)	26 (86.7)

SVI, seminal vesicle invasion; IQR, interquartile range; BMI, body mass index; PSA, prostate-specific antigen; f/t PSA, free/total prostate-specific antigen ratio; PV, prostate volume; PSAD, PSA density; GGG, Gleason grade group; CAPRA, cancer of the prostate risk assessment; PI-RADS, Prostate Imaging Reporting and Data System.

Univariate and multivariate analyses

Univariate and multivariate logistic analyses in patients with negative SVI on mpMRI are described in . In the univariate analysis, SVI (+) patients had a statistically significantly higher percentage of positive systematic biopsies [odds ratio (OR), 1.019; 95% confidence interval (CI), 1.002–1.036; P=0.028], a higher maximum cancer percentage per core (OR, 1.019; 95% CI, 1.003–1.035; P=0.018), higher GGGs (OR, 1.747; 95% CI, 1.284–2.378; P<0.001), higher CAPRA score (OR 1.461; 95% CI, 1.119–1.908; P=0.005), and a higher PI-RADS score (OR, 4.095; 95% CI, 1.688–9.938; P=0.002) compared with SVI (-) patients. No difference was observed in age, BMI, PSA, f/t PSA, PV, and PSAD between SVI (+) and SVI (-) patients.

Table 2

Univariate and multivariate analyses of predictors associated with SVI

Variable	Univariate			Multivariate
	OR (95% CI)	P value		OR (95% CI)	P value
Age (y)	0.990 (0.944–1.039)	0.694
BMI (kg/m2)	1.038 (0.903–1.193)	0.599
PSA (ng/mL)	1.211 (0.955–1.536)	0.114
f/t PSA	0.014 (0.000–25.202)	0.266
PV (mL)	0.984 (0.962–1.006)	0.150
PSAD (ng/mL2)	12.465 (0.477–325.419)	0.130
The percentage of positive systematic biopsies	1.019 (1.002–1.036)	0.028*			0.489
Maximum cancer percentage per core	1.019 (1.003–1.035)	0.018*			0.993
Clinical stage, No. (%)	1.113 (0.806–1.594)	0.473
T1c
T2a
T2b
T2C
T3a
CAPRA score	1.388 (1.063–1.814)	0.016*
PI-RADS	4.095 (1.688–9.938)	0.002*		3.359 (1.355–8.323)	0.009*
Negative
Suspicious
Positive
GGG	1.747 (1.284–2.378)	<0.001*		1.535 (1.111–2.120)	0.009*
1
2
3
4
5

*, P value with <0.05 significance. SVI, seminal vesicle invasion; OR, odd ratios; CI, confidence interval; BMI, body mass index; PSA, prostate-specific antigen; f/t PSA, free/total prostate-specific antigen ratio; PV, prostate volume; PSAD, PSA density; CAPRA, cancer of the prostate risk assessment; PI-RADS, Prostate Imaging Reporting and Data System; GGG, Gleason grade group.

CAPRA score, calculated from five variables including the GS and the maximum cancer percentage per core, was insignificantly associated with SVI (P=0.179) in the multivariate analysis which containing three variables: CAPRA score, the percentage of positive systematic biopsies and PIRADS (Table S1).

In the multivariate analysis, which excluding CAPRA score, GGG and PI-RADS remained significantly associated with SVI, suggesting that these variables were independent risk predictors for the diagnosis of SVI.

With the application of the coefficients of the logistic function, a predictive model for postoperative SVI was constructed using selected risk factors, including GGG and PI-RADS scores, as follows:

ROC analysis was performed to assess the accuracy of this model, as shown in . The predictive model applied for SVI yielded an AUC of 0.746, with a 95% CI of 0.662–0.831 (P<0.001). At the optimal cutoff predictive value of 0.100, it showed a sensitivity of 83.3% (95% CI, 64.5–93.7%), a specificity of 58.6% (95% CI, 52.0–65.0%), positive predictive value of 20.7% (95% CI, 14.1–29.2%), and a negative predictive value of 96.5% (95% CI, 91.5–98.7%).

Figure 1

Receiver operating characteristic curves (ROC) for SVI for PI-RADS score, GGG and predictive model. AUC, the area under the curve; GGG, Gleason grade group; PI-RADS, Prostate Imaging Reporting and Data System; ROC, receiver operating characteristic curves.

Discussion

In our cohort, SVI was reported on the RP specimens in 30 patients (11.5%). Using the selected risk factors, containing biopsy GGG and the PI-RADS v2 score, a predictive model for postoperative SVI was constructed, which revealed a high negative predictive value of 96.5% (95% CI, 91.5–98.7%) at the optimal cutoff predictive value.

SVI is defined as pathologically invasion of the muscular wall of the extraprostatic seminal vesicle. The incidence of SVI seemed to be heterogeneous, ranging from 3% to 17.6% in recent studies (1-5). In our cohort, the rate of SVI was 11.5%, which is consistent with the rate reported in recent literature which the median PSA level of the patients (5.9–7.8 ng/mL) is similar with ours (7.51 ng/mL) (3-5).

SVI is considered to be one of the most adverse prognostic findings in prostate cancer, which affects the biochemical progression-free survival and disease-specific survival. The 5- and 10-year biochemical failure rate for SVI was reported as 60% and 72%, which was significantly higher than that for pT2 patients (17). In a study containing 31,415 patients, Kristiansen et al. showed that compared to EPE alone, patients with SVI had a higher risk of clinical progression and death after RP (6).

According to National Comprehensive Cancer Network (NCCN) guideline version 2. 2020, patients with SVI are defined as very high-risk group, which treatment should think carefully for them. Asymptomatic patients with <5 years life expectancy is only considered for androgen deprivation therapy (ADT), external beam radiotherapy or observation. Only selected patients are recommended to undergo surgery and more effective treatments. Besides, patients with SVI influenced the choice of nerve-sparing RP, which could improve urinary continence and erectile function and is recommend in men with localized prostate cancer. According to EAU-ESTRO-SIOG Guidelines on Prostate Cancer, high risk of extracapsular disease is contraindication for nerve-sparing RP (18). However, EPE and SVI did not always show up together (19,20). In our cohort, EPE and SVI coexisted in 46.7% (14/30) of patients, and the other 16 patients without EPE who are unsuitable for nerve-sparing RP. It’s of great importance to exclude these unsuitable patients before operating nerve-sparing RP. On the contrary, in the case of high likelihood of SVI, additional therapies should be discussed, for example, neoadjuvant chemotherapy (21). In terms of prognosis and therapy strategy, preoperative prediction of SVI is of great significance.

MpMRI examination plays an important role in clinical staging, which is recommended to all patients who are suspected of PCs in our institution. According to a contemporary critical meta-analysis that included a total 5,677 patients from 34 studies, MRI shows high specificity (0.96, 95% CI, 0.95–0.97) but poor and heterogeneous sensitivity (0.58, 95% CI, 0.47–0.68) (10), which means that more than half of the patients with SVI were underestimated in the worst cases. This is possible because radiologists have focused on high-specificity reading to minimize unnecessary exclusion of men from curative treatment.

Several models have been constructed in an effort to predict SVI in prostate cancer. Lughezzani et al. compared three different models that did not consider mpMRI results, including Partin tables, the European Society of Urological Oncology (ESUO) criteria, and the Gallina nomogram, for the prediction of SVI (9). This study showed that all these three tools had high sensitivity (92.7%, 89%, and 90.8%, respectively) but poor specificity (33.1%, 56.3%, and 47.6%, respectively), which confirms that these models overestimated the probability of SVI during application. MpMRI has been considered as an important part of these models for providing detailed anatomical information. Grivas et al. discovered that the AUC values of Partin with MRI predictive models were higher than those of Partin and MRI alone (0.929, 0.837, and 0.884), which showed that adding mpMRI findings to Partin tables could improve its predictive accuracy (11). Martini et al. developed a mpMRI and clinical data-based nomogram for the prediction of SVI, and this nomogram showed a relatively high AUC (0.847) (12). However, these results are limited by the relatively small number of SVI cases with respect to the variables included in the model. Besides, these models containing mpMRI findings highly rely on the negative or positive SVI results of mpMRI. To our knowledge, none of the models currently integrate the PI-RADS v2 score to predict postoperative SVI in SVI (-) patients on preoperative mpMRI.

The PI-RADS scoring system has shown a great value in predicting biopsy outcome (22), biochemical recurrence (23), infiltration of the neurovascular bundles (24), and EPE (25). Our previous studies have shown similar results, confirming the value of PI-RADS in predicting prostate cancer and clinically significant prostate cancer in men undergoing repeat prostate biopsy and in predicting pelvic lymph node metastasis at RP (26,27). After adjusting for the other confounding factors, the GS and PI-RADS score were found to be independent risk factors for the prediction of SVI. On that basis, a model including PI-RADS to predict postoperative SVI in patients without SVI on mpMRI was constructed in this study. The AUC value of this model was 0.746 (P<0.001), which is better than that of mpMRI alone. This model also showed high sensitivity and negative predictive value, which indicates that it can precisely distinguish the real SVI (-) patients and help the urologists for the formulation of treatment plans.

Koh et al. found that among 275 patients with PSA ≤10 ng/mL and no cancer at the base in systematic biopsy results, none had SVI (28). Through creating probability plot graphs, Zlotta et al. pointed out that patients with PSA <10 ng/mL have a risk of SVI <5% when GS on biopsy is <7 or when the percentage of biopsies affected by cancer is <50% (29). In our series, we also observed that among the 111 patients with PI-RADS v2 score <4 and Gleason grade <8, extremely few patients (n=2) had SVI, which yielded only a 1.8% incidence of SVI with a high negative predictive value of 98.2% (95% CI, 93.0–99.6%).

PSA, a widely used serum marker for prostate cancer screening, plays an important role in other predictive models for SVI. In our study, the PSA level in SVI (+) patients was higher than that in SVI (-) patients, but without any significant difference (Median: 7.75 vs. 7.44 ng/mL, P=0.114). This is most probably because we selected patients with PSA <10 ng/mL, which narrowed the difference between the two groups. PSAD and f/t PSA did not show a significant difference as well as.

CAPRA score, which is calculated from the PSA level, the GS, the clinical T stage, the percentage of positive prostate biopsies and the patient age at diagnosis, has a great capacity in predicting prostate cancer outcomes, such as biochemical recurrence (30), metastatic potential (31), and prostate cancer-specific death (32). In our study, CAPRA wasn’t an independent risk predictor for the diagnosis of SVI (P=0.070) after adjusting other confounding factors. And further research is required for the value of CAPRA score for predicting SVI.

Emerging technologies and prostate cancer biomarkers are playing a vital role in prostate cancer diagnosis and treatment (33,34). Micro-ultrasound is a novel high-resolution imaging technology for diagnosing prostate cancer which is complementary for mpMRI (35-38). Compare to mpMRI, micro-ultrasound, which promises real-time visualization of suspicious lesions and targeting of biopsies, has shown same or superior sensitivity (37). For detecting clinically significant prostate cancer, micro-ultrasound biopsy has shown a higher rate with fewer biopsied cores (39), and could found prostate cancer missed by all other techniques (34). However, additional studies are needed to explore the application of micro-ultrasound for prostate cancer staging and predicting SVI.

Several limitations should be considered. First, these results were obtained from a retrospective cohort, thus bringing a certain risk of selection bias. Second, the predictive model was constructed on the basis on a small sample size from a single institution, and the accuracy of this model requires internal and external validation in a multicenter study to assess its wider applicability. Third, due to improvement in data quality over time and different MRI protocols performed in patients, there might be a certain difference in the outcome of MRI. Finally, the MRI performed previously at our institution could not meet the PI-RADS v2.1 technical stander (40), which would aggravate transferability for clinical usage, and the role of the PI-RADS v2.1 score for predicting SVI need further investigation. However, PI-RADS v2 has showed satisfactory inter-reader variability in previous study (41).

Conclusions

In the present study, PI-RADS assessment has been proved to be one of the valuable predictors of SVI in SVI (-) patients on mpMRI. The PI-RADS v2 score <4 in prostate cancer patients with PSA <10 ng/mL is associated with a very low risk of SVI. A model based on biopsy Gleason grade and PI-RADS v2 score may help to predict SVI and serve as a tool for the urologists to make surgical plans.

The article’s supplementary files as