Surgical treatment of prostate cancer (PCa) is one of the most effective methods of treatment. Critically, the rate of survival depends on the quality of the treatment given [1
]. The aim of radical prostatectomy (RP) is to achieve oncological radicality while preserving continence and sexual function. Preserving neurovascular bundles (NVBs) during RP is important for maintaining potency and continence in patients [3
]. Planning the RP and decisions concerning the preservation of NVBs must be based on the preoperative prognosis of the T3–T4 local stage in order to minimize the risk of positive surgical margins (SM) [5
]. It has been demonstrated that in the T3 stage, positive SM increase the risk of biochemical (BCR) and clinical recurrence [6
Proper T-staging, especially distinguishing between T2 and T3 stage, is key in deciding the extent of local treatment. Digital rectal examination (DRE) and transrectal ultrasound (TRUS) are useful only in diagnosing stage T3, which is associated with the presence of large changes in the architecture of the prostate [8
]. Local staging based solely on physical examination is unreliable and unclear. Nearly one-third of patients with T3 disease are understaged during rectal examination [9
]. Thus, accurate and precise surgical techniques preclude the use of these methods as preoperative T-staging.
Multiparametric MRI (mpMRI) of the prostate, when used to assess local staging, has a low sensitivity and high specificity. Recent studies estimated MRI confirmation of the T3 stage at 61% and 88% for sensitivity and specificity, respectively [10
]. For assessing extraprostatic extension (EPE/pT3a), the sensitivity is estimated to be even lower at 57% [10
On the other hand, various clinical factors such as the PSA baseline, Gleason score (GS) from biopsy result, can determine the local stage of the PCa [11
]. These clinical parameters are used and combined in “traditional” predictive nomograms for preoperative PCa patient assessment [11
]. Recent studies have shown that supplementing “traditional” predictive models with mpMRI information significantly improves the diagnostic value of such nomograms [5
Currently, patients with PCa are qualified for RP based on prostate biopsy (systematic biopsy combined with targeted or targeted only) after previous mpMRI. Therefore, novel nomograms, utilizing mpMRI results, targeted biopsy, and clinical data (e.g.PSA), have been developed [5
]. MRI-targeted biopsy in combination with systematic biopsy enables the diagnosis of a greater percentage of clinically significant cancers (csPCa, GS ≥ 3 + 4) and more precisely predicts the risk of stage pT3 in the final histopathology [15
Based on mpMRI data and targeted prostate biopsy, an accurate and reliable nomogram for side-specific EPE can be developed. This allows optimization and personalization of decisions on the extension of RP.
Therefore, the primary aim of this study is to establish what parameters (clinical, radiological, and biopsy data) can best predict the risk of EPE for a specific site of the prostate. The secondary aim was to assess and compare novel models including mpMRI results and biopsy results (the Martini et al. model) and traditional models, based only on clinical parameters and biopsy results (Memorial Sloan-Kettering Cancer Center (MSKCC) and Partin Tables), in predicting the risk of EPE in patients qualified for RP after targeted biopsy combined with systematic biopsy based on mpMRI.
The characteristics of the study group are presented in Table 1
. The median patient age was 66 years. The PSA and PSA density (PSAd) were 6.99 ng/mL and 0.20 ng/mL/mL, respectively. EPE in mpMRI was present in 24.6% of patients. CsPCa (ISUP Grade ≥ 2) was detected during the biopsy in 42.6% of patients.
After RP, 31.2% of patients had an pT3 stage in the RP specimen, and 29.5% had positive SM. Ten patients (52.6%) with EPE had positive SM in the RP specimen. The final histopathology report revealed csPCa in 59% of patients. Seven out of twenty-one (33.3%) patients with spared NVBs had positive SM.
A total of 122 prostate lobes were analyzed. Final histopathology revealed EPE was found in 22.1% of the total prostate lobes in the final histopathology. Two comparative groups were established: the first with diagnosed EPE in a prostate lobe (pEPE+) and the second without EPE in the final histopathological results (pEPE-) (Table 2
The odds ratio (OR) was determined using the UVA results for each of the factors distinguishing the subgroups. Diagnosis or suspicion of EPE in mpMRI and a maximum lesion size of ≥15 mm were found to be independent predictors of EPE in the final histopathology. In addition, biopsy indicating the presence of csPCa (ISUP Grade ≥ 2) and an extent of PCa in a single core ≥ 50% were the factors that increase the probability of EPE in post-RP specimens (Table 3
). There were no differences between the groups in terms of PSA levels as a continuous variable. However, the use of a thresholded PSA level of ≥20 ng/mL significantly predicted the presence of pEPE (Table 3
). These results revealed that the combination of mpMRI indicating or suspecting EPE and PSA ≥ 20 ng/mL could significantly determine the risk of EPE for a specific side of the prostate.
The ROC curves for the MSKCC, Partin Tables, and Martini et al. nomograms and mpMRI are presented in Figure 2
and Figure 3
. The site-specific model by Martini et al. had the highest sensitivity (66.7%), while mpMRI had the highest specificity (93.7%). The model with the highest AUC for predicting the presence of site-specific EPE was the Martini et al. nomogram (AUC = 0.73, p
< 0.001); a cut-off point of 5% was proposed. The Hosmer–Lemeshow test for the Martini et al. model confirmed goodness of fit (p
= 0.71). However, the calibration plot showed an overestimation of the predicted risk (Figure 4
The AUC for mpMRI examination alone with regard to site-specific EPE assessment was 0.63 (p
= 0.005). Comparison of the AUC of Martini’s nomogram with that of mpMRI did not show a statistically significant difference (p
= 0.131, Figure 2
The AUC for the MSKCC nomogram for EPE prediction was estimated at 0.71 (p
= 0.007) with a cut-off point of 61%. The Partin Tables nomograms were found to be statistically insignificant, and the AUC was 0.63 (p
= 0.074). Comparison of the AUC of the MSKCC model with that of the Partin Tables was also not statistically significant (p
= 0.211, Figure 3
). The Hosmer–Lemeshow test for the MSKCC model confirmed goodness of fit (p
= 0.67). Using calibration curve showed that model underestimated the predicted risk.
The assessed sensitivities, specificities, PPV, and NPV of the models are reported in Table 4
Good planning of laparoscopic RP needs to take into account the achievement of oncological radicality and good functional results. The choice of the appropriate extension of RP depends primarily on the preoperative prediction of the pT3 stage. NVB sparing surgery in the presence of EPE is associated with an increased risk of positive SM in postoperative specimens [27
]. Positive SMs are associated with an increased risk of BCR [31
]. However, NVB sparing surgery helps to achieve good functional results, mainly in terms of continence and sexual function, which are crucial for patients’ quality of life [3
]. Accurate preoperative prediction of the EPE/pT3 stage aids clinical decision making and can minimize the risk of positive SM during RP [31
In the current study, the analysis of the discriminative value of mpMRI only in terms of stage pT3 assessment indicated a low sensitivity (33%) and high specificity (93.7%) of the examination. The problem of low MRI sensitivity has been repeatedly discussed in the literature. For example, a meta-analysis by de Rooij et al. reported sensitivity and specificity values of MRI for the detection of stage pT3a at 57% and 88%, respectively [10
]. Despite its rather poor sensitivity, MRI can enable the selection of the appropriate treatment range [32
]. The urologist’s decision on the extent of RP based on MRI findings has been found to change in 30–50% of cases, with the decision being judged appropriate in 75–90% [32
]. Moreover, patients so qualified have a reported lower rate of positive SM (12.4% vs. 24.1% p
≤ 0.01) [32
]. Conversely, the other studies have reported the limitations of MRI in detecting EPE [35
]. MRI alone cannot be used to precisely diagnose EPE [36
]. The lack of awareness of this limitation can result in wrong surgical decision making and a higher rate of positive SM [36
]. The aspect of extracapsular micro-infiltration, which is not detected by MRI but is recognized in histopathological examination and classified as stage pT3a, may be important in this respect. Therefore, due to the insufficient predictive value of MRI examination only, it is justified to consider clinical variables in order to achieve a model that can accurately and effectively define the risk of unfavorable histopathology (i.e., pT3).
In the course of this study, it was demonstrated that mpMRI indicating EPE or its suspicion and lesion size of ≥15 mm are independent predictors of pathological EPE for a specific side of the prostate (OR 7.42, p < 0.001 and OR 3.67, p < 0.001, respectively). Furthermore, the MVA findings in our study demonstrated that mpMRI results indicating EPE or its suspicion with a PSA level of ≥20 ng/mL are the strongest predictors of EPE for a specific side of the prostate.
Subsequently, our comparison of constructed and validated models such as Martini, MSKCC, and Partin Tables nomograms aimed to assess whether the prediction of preoperative risk based on the combination of clinical data and mpMRI is better than prediction based solely on clinical data or imaging examination. AUC analysis indicated that “traditional” models based only on clinical variables, such Partin Tables or MSKCC, have a lower predictive value than models that also take mpMRI data into account [9
]. MRI improved the model’s discrimination value [9
]. A study by Feng et al. showed that the AUC for the MSKCC and Partin Tables nomograms increased after adding MRI results, changing from 0.86 to 0.92 and from 0.86 to 0.94, respectively [9
]. Morlacco et al. demonstrated that Partin Tables with MRI outperformed Partin Tables alone [38
]. Similarly, Ryan et al. reported that adding MRI data to the MSKCC nomogram increased the AUC by 0.10 for predicting EPE risk [13
]. Additionally, the authors showed that the GS obtained by targeted biopsy, compared to the results of systematic biopsy, significantly improved the discrimination value of the clinical MSKCC model (it increased the AUC by 0.07) [13
Our analysis of the MSKCC model, which only took clinical data into account, had better sensitivity and AUC than mpMRI alone. However, it should be mentioned that the biopsy results used in the MSKCC nomogram indirectly account for the mpMRI results, as systematic biopsy is performed in combination with targeted biopsy based on mpMRI.
However, we concluded that the nomogram by Martini et al. is the best tool for assessing the risk of cancer extending beyond the prostate, as well as for planning surgical treatment and making decisions to preserve NVBs. The AUC we assessed for the Martini model was 73%, which turned out to be the most sensitive of the compared models and was better than mpMRI alone for estimating EPE. In their own study of the model, Martini et al. estimated an AUC of 82.1% [5
]. Although the AUC for the Martini model in the current study was lower than in the original paper, the test of goodness of fit indicated good concordance between observed and predicted risk of EPE with p
= 0.71. However, the calibration plot showed that the nomogram overpredicted the EPE risk. It should be mentioned that our limited group of patients could have resulted in miscalibration of the model. The other measure of the external validation of the Martini nomogram performed by Soeterik et al. showed good discriminative ability but poor calibration and underestimation of the predicted risk [39
Martini’s analysis of the number of patients in stage pT3 and with positive SMs in relation to the proposed cut-off point of 20% indicates the highest utility for making decisions to spare NVBs. In the current study, the proposed cut-off point is lower and only differentiates the possibility of EPE or its absence.
Our study has several limitations. This was a preliminary study conducted in one center with a relatively small final group of patients (n
= 61). We used 1.5T or 3T MRI scanners. A systematic biopsy combined with a targeted biopsy was performed based on mpMRI results. Thus, the biopsy data used in the MSKCC and Partin Tables models differ from the original assumptions of the MSKCC and Partin Tables models [11
], where only systematic biopsy results are considered. However, despite the above limitations, the conclusions of our study are consistent with the currently available literature on the importance of combining predictive models with mpMRI in the planning of surgical treatment of patients with PCa. We believe that multicenter studies involving larger groups should confirm our conclusions.