External validation of the Briganti 2019 nomogram to identify candidates for extended pelvic lymph node dissection among patients with high-risk clinically localized prostate cancer

Utility of Machine Learning Models to Predict Lymph Node Metastasis of Japanese Localized Prostate Cancer

BACKGROUND/OBJECTIVES

Extended pelvic lymph node dissection is a crucial surgical technique for managing intermediate to high-risk prostate cancer. Accurately predicting lymph node metastasis before surgery can minimize unnecessary lymph node dissections and their associated complications. This study assessed the efficacy of various machine learning models for predicting lymph node metastasis in a cohort of Japanese patients who underwent robot-assisted laparoscopic radical prostatectomy.

METHODS

Data from 625 patients who underwent extended pelvic lymph node dissection or standard dissection with lymph node metastasis between October 2010 and February 2023 were analyzed. Four machine learning models-Random Forest, Light Gradient-Boosting Machine, Logistic Regression, and Support Vector Machine-were used to predict lymph node metastasis. Their performance was assessed using receiver operating characteristic curves, a decision curve analysis, and predictive values at different thresholds.

RESULTS

Lymph node metastasis was observed in 34 patients (5.4%). The Light Gradient-Boosting Machine had the highest AUC of 0.924, followed by the Random Forest model with an AUC of 0.894. The decision curve analysis indicated substantial net benefits for both models, particularly at low threshold probabilities. The Light Gradient-Boosting Machine demonstrated superior accuracy, achieving 95.6% at the 0.05 threshold and 96.7% at the 0.10 threshold, outperforming other models and conventional nomograms in the validation dataset.

CONCLUSION

Machine learning models, especially Light Gradient-Boosting Machine and Random Forest, show significant potential for predicting lymph node metastasis in prostate cancer, thereby aiding in reducing unnecessary surgical interventions.

Predictors of Progression to Castration-resistant Prostate Cancer After Radical Prostatectomy in High-risk Prostate Cancer Patients

Background/Aim

To examine the specific time frame and identify associated risk factors from commencement of hormonal therapy to the onset of castration-resistant prostate cancer among patients who have developed biochemical recurrence following radical prostatectomy.

Patients and Methods

We retrospectively reviewed the records of 92 patients who developed biochemical recurrence and received hormonal therapy as initial salvage treatment after radical prostatectomy for high-risk localized prostate cancer from 2005 to 2021. The castration-resistant prostate cancer-free survival rates from the commencement of salvage hormonal therapy were analyzed using log-rank methods. Cox proportional hazard regression was performed to analyze the risk factors associated with acquiring castration resistance. The patients were stratified based on those risk factors.

Results

During a median follow-up duration of 57 months, 24 (26.1%) patients developed castration-resistant prostate cancer. The 5- and 10-year castration-resistant prostate cancer-free survival rates were 73.6% and 54.5%, respectively. A multivariate analysis showed that Grade Group of 5 and prostate-specific antigen doubling time at biochemical recurrence of ≤3 months were independent predictors of castration-resistant prostate cancer. The 5-year castration-resistant prostate cancer-free survival rates in the low- and high-risk groups, stratified according to the aforementioned factors, were 85.4% and 47.6%, respectively.

Conclusion

Patients in high Grade Group and short prostate-specific antigen doubling time after radical prostatectomy are more likely to develop resistance to salvage hormonal therapy.

Rational use of Ga-68 PSMA PET-CT according to nomograms and risk groups for the detection of lymph node metastasis in prostate cancer

OBJECTIVES

The aim was to ensure efficient utilization of PSMA PET-CT by examining the correlation of pathological lymph node metastasis with nomogram scores and risk classifications.

METHODS AND MATERIALS

Robot-assisted radical prostatectomy and bilateral pelvic lymph node dissections for pelvic lymph nodes were performed using the same template. Bilaterally pelvic lymph nodes were removed within the boundaries of genitofemoral nerves, psoas muscle and lateral pelvic wall laterally, ureteric crossing of the iliac vessels superiorly, lateral bladder wall medially, Cooper ligaments distally, and endopelvic fascia, neurovascular bundles and internal iliac arteries posteriorly. Clinical nomograms were used to calculate the probability of lymph node metastasis preoperatively. Using receiver operating characteristics analysis, discriminatory cut-offs were calculated. The diagnostic performance of PSMA PET-CT was determined for detecting lymph node metastasis.

RESULTS

For 81 patients, the median age was 64 years. The median PSA was 6.8 ng/ml. Most patients were in the D'Amico intermediate (56.8%) and high (37%) risk groups. Median Briganti 2017, MSKCC, and Partin scores were 35 (4-99), 37 (8-90), and 12 (2-38), respectively, in pN1 patients. The area under the curve for Briganti 2017, MSKCC, Partin nomograms and PSMA PET-CT scans were 0.852, 0.871, 0.862, and 0.588. Sensitivity, specificity, positive predictive value and negative predictive value for Ga-68 PSMA PET-CT for lymph node metastasis detection were 21.4%, 94%, 42.9%, and 85.1%, respectively, for the whole group. By using higher threshold values for clinical nomograms (Briganti 2017 >32, MSKCC >12, Partin >5), PSMA PET-CT had higher sensitivity (42.9, 30, 27.2) in detecting lymph node metastasis.

CONCLUSIONS

Patients in the D'Amico high-risk group and those with high nomogram scores are the best candidates who will benefit from preoperative PSMA PET-CT staging to estimate lymph node metastasis.

Comparison of Four Validated Nomograms (Memorial Sloan Kettering Cancer Center, Briganti 2012, 2017, and 2019) Predicting Lymph Node Invasion in Patients with High-Risk Prostate Cancer Candidates for Radical Prostatectomy and Extended Pelvic Lymph Node Dissection: Clinical Experience and Review of the Literature

Background: The indication for extended pelvic lymph node dissection (ePLND) at the time of radical prostatectomy (RP) is based on nomograms predicting the risk of lymph node invasion (LNI). However, limited data are available on the comparison of these predictive models in high-risk prostate cancer (PC) patients. Therefore, we compared the accuracy of the most used nomograms (MSKCC, Briganti 2012, 2017, and 2019) in the setting of high-risk PC patients submitted to ePLND. Methods: 150 patients with high-risk PC disease treated from 2019 to 2022 were included. Before RP + ePLND, we assessed the MSKCC, Briganti 2012, 2017, and 2019 nomograms for each patient, and we compared the prediction of LNI with the final histopathological analysis of the ePLND using pathologic results as a reference. Results: LNI was found in 39 patients (26%), and 71.3% were cT2. The percentage of patients with estimated LNI risk above the cut-off was significantly higher in pN+ cases than in pN0 for all Briganti nomograms. The percentage of patients at risk of LNI, according to Briganti Nomogram (2012, 2017, and 2019), was significantly higher in pN+ cases than in pN0 (p < 0.04), while MSKCC prediction didn’t vary significantly between pN0 and pN+ groups (p = 0.2). All nomograms showed high sensitivity (Se > 0.90), low specificity (Sp < 0.20), and similar AUC (range: 0.526–0.573) in predicting pN+. Particularly, 74% of cases patients with MSKCC estimated risk > 7% showed pN0 compared to 71% with Briganti 2012 > 5%, 69% with Briganti 2017 > 7%, and 70% with Briganti 2019 > 7%. Conclusions: Despite the high-risk disease, in our patients treated with ePLND emerges a still high number of pN0 cases and a similar low specificity of nomograms in predicting LNI.

Prediction of pelvic lymph node metastases and PSMA PET positive pelvic lymph nodes with multiparametric MRI and clinical information in primary staging of prostate cancer

Purpose

To compare the accuracy of multiparametric MRI (mpMRI), ⁶⁸Ga-PSMA PET and the Briganti 2019 nomogram in the prediction of metastatic pelvic lymph nodes (PLN) in prostate cancer, to assess the accuracy of mpMRI and the Briganti nomogram in prediction of PET positive PLN and to investigate the added value of quantitative mpMRI parameters to the Briganti nomogram.

Method

This retrospective IRB-approved study included 41 patients with prostate cancer undergoing mpMRI and ⁶⁸Ga-PSMA PET/CT or MR prior to prostatectomy and pelvic lymph node dissection. A board-certified radiologist assessed the index lesion on diffusion-weighted (Apparent Diffusion Coefficient, ADC; mean/volume), T2-weighted (capsular contact length, lesion volume/maximal diameters) and contrast-enhanced (iAUC, k_ep, K^trans, v_e) sequences. The probability for metastatic pelvic lymph nodes was calculated using the Briganti 2019 nomogram. PET examinations were evaluated by two board-certified nuclear medicine physicians.

Results

The Briganti 2019 nomogram performed superiorly (AUC: 0.89) compared to quantitative mpMRI parameters (AUCs: 0.47-0.73) and ⁶⁸Ga-PSMA-11 PET (AUC: 0.82) in the prediction of PLN metastases and superiorly (AUC: 0.77) in the prediction of PSMA PET positive PLN compared to MRI parameters (AUCs: 0.49-0.73). The addition of mean ADC and ADC volume from mpMRI improved the Briganti model by a fraction of new information of 0.21.

Conclusions

The Briganti 2019 nomogram performed superiorly in the prediction of metastatic and PSMA PET positive PLN, but the addition of parameters from mpMRI can further improve its accuracy. The combined model could be used to stratify patients requiring ePLND or PSMA PET.

Validation of models predicting lymph node involvement probability in patients with prostate cancer

OBJECTIVES

There are many models to predict lymph node involvement in patients with prostate cancer. We aimed to externally validate several models in a Japanese cohort.

METHODS

We considered patients who were treated with robotic-assisted radical prostatectomy with extended pelvic lymph node dissection for prostate cancer. The risk of lymph node involvement was calculated for each patient in several models. Model performance was assessed by calculating the receiver operating characteristic curve and the area under the curve, calibration plots, and decision curve analyses.

RESULTS

We identified lymph node involvement in 61 (18.4%) of the 331 considered patients. Patients with lymph node involvement had a higher prostate-specific antigen level, percentage of positive biopsy cores, primary Gleason grade, Gleason group grade, and clinical T-stage category. The Memorial Sloan Kettering Cancer Center web calculator presented the highest area under the curve (0.78) followed by the Yale formula area under the curve (0.77), the updated version of Briganti nomogram of 2017 area under the curve (0.76), and the updated version of the Partin table by Tosoian et al. had an area under the curve of 0.75. However, the 95% confidence interval for these models overlapped. The calibration plot showed that the Memorial Sloan Kettering Cancer Center web calculator and the updated version of the Briganti nomogram calibrated better. In the decision curve analyses, all models showed net benefit; however, it overlapped among them. However, the Memorial Sloan Kettering Cancer Center web calculator and the updated Briganti nomogram presented the highest net benefit for lymph node involvement risks <35%.

CONCLUSION

Models predicting lymph node involvement were externally validated in Japanese men. The Memorial Sloan Kettering Cancer Center web calculator and the updated Briganti nomogram of 2017 were the most accurate performing models.