Prostate cancer is the most common cancer in men and the third most common cause of death among malignancies in male population.¹ For localized prostate cancer, definite treatments such as radical prostatectomy (RP) or radiation therapy could be primary therapeutic options. However, once prostate cancer invades lymph node or other organs, therapeutic effects of such treatment modalities become limited. Even if the therapeutic role of pelvic lymph node dissection (PLND) in patients with low-risk prostate cancer is still an open question, patients with intermediate- to high-risk disease are likely to benefit from PLND.² For this reason, early detection of lymph node invasion (LNI) is important for treatment of prostate cancer.

The Partin tables,³ the National Comprehensive Control Network (NCCN) practice guidelines nomogram,^4,5 the D'Amico risk-classification^6,7 are commonly used to assess the risk of prostate cancer. Such risk assessment tools use pre-operative variables including prostate-specific antigen (PSA), biopsy Gleason score, and clinical stage to predict prognosis of LNI and prostate cancer after treatments.

However, such preoperative assessing tools do not include preoperative image studies including magnetic resonance imaging (MRI). Recently, the multiparacentric MRI (mp-MRI), which is the combination of conventional T1-weighted MRI (T1WI) and T2-weighted MRI (T2WI) with various functional imaging modalities including diffusion-weighted MRI (DWI), dynamic contrast enhanced MRI (DCEI), and magnetic resonance spectroscopy, has been applied in clinical filed.⁸ The role of mp-MRI has been expanded broadly to prostate biopsy, active surveillance, advanced disease detection, and local recurrence detection after RP.⁹ However, the diagnostic value of imaging studies including MRI are still debatable. Wolf et al.¹⁰ reported that CT and MRI showed high specificity of 97%, but relatively low sensitivity of 36%. They concluded that nodal imaging studies should only be recommended for patients having a probability of 45% or higher for LNI. Min et al.¹¹ reported that a combined approach of T1W, T2W, and DCEI with DWI demonstrated an accurate detection rate of prostate cancer, showing high specificity but relatively lower sensitivity. Hö vels et al.¹² reported MRI demonstrated an equally poor performance in the detection of LNIs from prostate cancer through meta-analysis.

The aim of this study is demonstrating the predictive power of preoperative MRI for LNI in prostate cancer patients who underwent PLND.

With an approval from Institutional Review Board (B-1710-427-110), the medical records of 3,379 patients who underwent RP for prostate cancer at Seoul National University Bundang Hospital from November 2003 to May 2017 were analyzed retrospectively. After excluding patients who did not undergo PLND and those with missing data about lymph node state, 1,528 patients were included in our analyses. We evaluated the various clinicopathologic variables including pre-operative MRI and pathologic LNI.

In our institution, every patient diagnosed with prostate cancer underwent preoperative prostate MRI, including T1WI, T2WI, DWI, and DCEI. Positive lymph node enlargement (LNE) on MRI was defined when increased over 10 mm.

We stratified the cohorts into 3 groups according to the number of LNE: no significant LNE group, one LNE group, and more than 2 LNE group. Each group was graded as 0, 1, and 2, respectively.

Along with preoperative MRI, we used the NCCN risk stratification to predict diagnostic values for metastasis. In NCCN risk stratification system, there are 5 risk categories: very low (T1c and Gleason score≤6 and PSA<10 ng/mL and <3 positive core biopsies); low (T1-T2a and Gleason score≤6 and PSA<10 ng/mL); intermediate (T2b-T2c or Gleason score 7 or PSA=10-20 ng/mL); high (T3a or Gleason score 8-10 or PSA>20 ng/mL); and very high (T3b-T4 or Gleason primary pattern 5).¹³ The entire cohorts were classified into 3 categories as followings: low-risk group (very low-risk and low-risk group), intermediate-risk group, and high-risk group (high-risk and very high-risk group) according to NCCN risk stratification, with each group graded as 0, 1, 2, respectively. To assess pre-dictive power of combination of preoperative MRI and NCCN risk stratification, we added the numerical grades of MRI group (0, 1, and 2) and NCCN risk group (0, 1, and 2).

The prediction model for pathologic lymph node metastasis was evaluated using univariate and multivariable logistic regression analyses. Areas under receiver operating characteristic (ROC) curves (AUC) was used as an index of diagnostic accuracy.

IBM SPSS Statistics ver. 19.0 (IBM Co., Armonk, NY, USA) was used for statistical analysis, and ROC curve and the prediction factors were estimated using MEDCALC ver. 12 (MedCalc, Ostend, Belgium). All p-values were 2-sided, and p <0.05 was considered as a significant result.

Demographic characteristics of subjects were as shown in Table 1. The mean age of the cohort was 66.4±6.7 years. Median concentration of serum PSA was 11.34 ng/dL (interquartile range, 6.91-20.11 ng/dL). Positive preoperative MRI finding was observed in 9.4% (145 of 1,528) of patients. 5.3% (81 of 1,528) of patients had confirmed final pathologic lymph node metastases.

Multivariable analysis showed that preoperative MRI findings (p<0.001), clinical stage (p=0.001) and biopsy Gleason score≥8 (p<0.001) were significant independent predictors for pathologic LN metastasis (Table 2).

Preoperative mp-MRI (T1W, T2W, DWI, and DCEI) of prostate alone achieved a sensitivity of 30.86% and a specificity of 91.71%, respectively. NCCN risk alone and the combination of preoperative mp-MRI and NCCN risk stratification achieved sensitivities of 90.12% and 99.45% respectively, and specific-ities of 52.87% and 49.38% respectively (Table 3).

Among 145 patients with positive LN on preoperative MRI, only 25 patients had confirmed final pathologic lymph node metastases. Especially, Among 126 patients with one positive LN on preoperative MRI, only 21 patients (16.7%) had confirmed final pathologic lymph node metastasis (Table 4).

Through ROC analysis, prediction model using preoperative MRI findings alone showed AUC value of 0.613 (0.588-0.638). The addition of preoperative MRI findings to NCCN risk stratification showed fair accuracy, improving the AUC to 0.758 (0.735-0.779) from 0.613 with NCCN risk stratification alone (Fig. 1).

There are several pretreatment prostate cancer risk stratification systems or nomograms based on prognostic power of initial PSA, biopsy Gleason score and clinical T stage including D'Amico risk stratification, NCCN practice guideline risk-stratification, and the D'Amico risk-classification.¹⁴

The Partin tables are the first nomogram to predict rates of organ-confined disease, positive margins, the risk of seminal vesicle and lymph node positivity after RP, which uses commonly available preoperative factors including serum PSA level, clinical stage and biopsy Gleason score.¹⁵ The Partin tables were initially established based on the information of the patients who underwent RP between 1982 and 1991, and it was upgraded with newer data acquired between 2006 to 2011.¹⁶ The predictive accuracy of the Partin tables for LNI was from 76% to 84%.^17-20 Recently, the Partin tables are commonly used for making therapeutic decision to perform a lymphadenectomy at the time of RP. The NCCN guideline risk stratification contains the minimum of clinical stage, grade and PSA it is applied for the selection of optimal therapeutic options as well as for the prediction of biochemical failure rate after definitive local therapy.^4,21 NCCN practice guideline LNI nomogram is also used for predicting the prognosis of LN positive disease after lymphadenectomy. Moreover, the D'Amico risk-classification was originally developed to predict biochemical recurrence in prostate cancer patients after treatment.⁶

These 3 tools have been applied frequently in routine clinical practice. When comparing these 3 tools for prediction of LNI, Abdollah et al.²⁰ showed that NCCN guideline LNI nomogram (AUC 82%) outperformed the Partin tables (73%) and the D'Amico risk-classification (75%).

However, among these 3 nomograms, imaging studies including preoperative MRI has not been accepted as essential parameter to predict LNI. Therefore, we investigated whether preoperative MRI could improve predictive power of conventional nomograms. This study aimed to evaluate the accuracy of preoperative MRI in patients who underwent RP with PLND and determinate whether preoperative MRI could improve the diagnostic value of conventional NCCN.

The advent of mp-MRI is known to provide more accurate mean of diagnosing prostate cancer in men with elevated PSA level, achieving a specificity of 90%, a negative predictive valve of 85%.^22,23 To achieve the highest sensitivity and specificity in detecting and assessing suspicious lesion within prostate, mp-MRI combines conventional T1WI and T2WI with various functional imaging modalities, with a minimum of 2 modalities. DWI and magnetic resonance spectroscopic imaging add specificity for lesion characterization, and DCE adds sensitivity for cancer detection.⁸

There are several guidelines from the European Association of Urology (EAU), the American Urological Association (AUA), and the NCCN, mentioning the potential role of mp-MRI in prostate biopsy, active surveillance, and recurrent prostate cancer.⁹

The EAU guideline suggests that mp-MRI followed by transrectal ultrasonography (TRUS)-guided or MRI-guided biopsies might be useful to detect prostate cancer especially in anterior region, whereas conventional TRUS-guided biopsy is still performed more frequently instead of the corresponding newer modalities. However, mp-MRI is not recommended for the purpose of staging in case of low-risk prostate cancer, except when the results of MRI could influence management.²⁴ The AUA guideline suggests that mp-MRI could identify local recurrence lesion and improve salvage radiation targeting,²⁵ and the NCCN guideline suggests that mp-MRI could aid in deciding active surveillance, yet concluded mp-MRI is not recommended for routine use.²⁶

In detecting lymph node metastasis, mp-MRI showed a high specificity of 91%-97% but a low sensitivity of 14%-36%.^10,11,27 Hö vels et al.¹² mentioned poor performance of MRI in detection of lymph node metastasis through meta-analysis study. Because of such low sensitivity, mp-MRI alone was not recommended for lymph node workup in patients without suspicious lymph node lesions on CT.²⁸ To evaluate predictive power of preoperative MRI for LNI, we analyzed the medical records of 1,528 patients who underwent RP with PLND, which is the largest cohort as far as we know.

In this study, the mp-MRI showed a high specificity of 91.7% and relatively a low sensitivity of 30.8%, which are similar to the previous study results. The ROC curve of pre-operative MRI alone showed AUCs of 0.613 for lymph node metastasis. ROC curve of NCCN nomogram showed AUCs of 0.713 that was similar to the results of Abdollah et al.²⁰.

When combining risk grades of NCCN nomogram with the outcomes of preoperative MRI, the specificity and the sensitivity were 99.5% and 49.4%, respectively. The ROC curve showed AUCs of 0.782. When the corresponding 2 parameters were combined, the AUCs became smaller than 0.9, but the value was greater compared to 0.613 and 0.713 of single parameters.

There are several limitations in this study. At first, while the large number of cohort enabled us to detect significant differences, our database is derived from the retrospective records of patients who were treated at single institution. Therefore, multi-institutional studies might be needed to confirm our findings. Furthermore, we were unable to match and compare the locations of lymph node lesions shown in preoperative MRI with the actual pathologically confirmed LNI sites. Moreover, the lack of information on long-term survival including cancer-specific survival or overall survival could be another limitation to our study.

Our results show that preoperative MRI findings for pathologic LN metastasis alone showed limited prediction value. However, when combining with conventional NCCN risk nomogram, preoperative MRI could improve predictive power. A large-scale, multicenter, and prospective study is needed to fully evaluate the clinical significance of preoperative MRI.