Defining the indications for pelvic lymph node dissection (PLND) in prostate cancer (PCa) patients within a statewide quality improvement collaborative
Abdollah F, Betrus G, Cher M, Dalela D, Keeley J, Kim T, Lane B, Mansour S, Montie J, Schervish E, Sood A, Swama K, and Peabody J. Defining the indications for pelvic lymph node dissection (PLND)in prostate cancer (PCa)patients within a statewide quality improvement collaborative. Eur Urol Suppl 2019; 18(1):e2050.
Eur Urol Suppl
Introduction & Objectives: The National Comprehensive Cancer Network (NCCN)recommends PLND in PCa patients with a calculated pN1 risk of >2% based on Cagiannos model. The 2% cut-off was chosen to avoid PLND in 48% of patients, at a cost of missing 12% of pN1 disease. We test the performance of the NCCN model in Michigan Urological Surgery Improvement Collaborative (MUSIC)patients and compared it to a novel model tailored to the MUSIC cohort. Materials & Methods: MUSIC, a physician led statewide quality improvement consortium, collects data via online registry from diverse urology practices. The MUSIC registry identified 5023 patients treated with radical prostatectomy and PLND from March 2012 to September 2018, and randomly split the cohort: 70% development and 30% validation. The development cohort used logistic regression analysis to fit a pN1 predication model—including a priorselected variables: PSA, clinical stage, primary Gleason grade, secondary Gleason grade, and % of positive cores. The novel model performance was tested in the validation cohort, using accuracy and calibration, and compared to the NCCN guidelines model. Results: Median (interquartile range)age was 66 (61-71), PSA was 6.0 (4.6-8.7), and % of positive cores was 40% (30%-60%). Most patients had clinical T1 stage (68.3%), primary Gleason 3 (56.4%), and secondary Gleason 4 (54.8%). Multivariable analysis notes all variables in the methods above to be independent predictors of pN1 disease. In the validation cohort, the novel model accuracy was 80%—comparing favorably to the NCCN model accuracy of 76%. Similarly, the novel model had a better calibration (data not shown). The NCCN 2% cut-off would avoid PLND in only 12% of the validation cohort, while missing no pN1 disease. Conversely, the same cut-off calculated using the novel model would avoid PLND in 43% of the validation cohort, while missing 8.5% of pN1 disease. Conclusions: NCCN model application in MUSIC patients does not meaningfully decrease the number of PLND performed, and might not be so different from treating everyone with PLND. Conversely, the new MUSIC model can help selecting patients appropriately for PLND, which might significantly improve the overall quality of care.