NEAUA Main Site  |  NEAUA 2021  |  Past and Future Meetings
New England Section of the American Urological Association

Back to 2020 Abstracts


Computer Generated vs. Human Generated R.E.N.A.L Nephrometry Score to Predict Surgical Outcomes in Renal Cell Carcinoma
Zach Edgerton, MD.
University of Minnesota Medical School, Minneapolis, MN.

Background R.E.N.A.L. nephrometry score is associated with pathological outcomes, complication rates and survival. Despite its success, widespread uptake has been limited by interobserver variability and time investment to generate scores. We developed an algorithm to produce a computer generated (CG) RENAL score, and compared this with human generated (HG) scores to predict RCC, high grade (Fuhrman 3-4), high stage (pT3-4) and tumor necrosis. Methods Retrospective review of 544 patients undergoing nephrectomy following CT for suspected RCC from 2010-2018. After manually delineating tumors on CT using an internally-made application, we developed an algorithm to automatically generate each RENAL score component. Each tumor was also manually, independently scored by one of five medical professionals. We used ROC curve analysis to quantify the discriminative ability of HG and CG RENAL scores. Results CT imaging was available for 195 patients. 183 (94%) had malignant tumors. Interobserver agreement between CG and HG RENAL scores was significant, but slight (kappa=0.32, p<0.001). However, CG score had good discriminative ability for cancer (AUC 0.76), greater than HG (0.67). CG (0.59) and HG (0.62) scores were comparable for high grade, whilst HG score (0.80) outperformed CG (0.62) scores for high stage. HG (0.74) also outperformed CG (0.63) score for tumor necrosis. Conclusion CG RENAL scores demonstrate significant agreement with HG RENAL scores and have similar ability to predict clinically important pathologic outcomes. These are promising results and, with further refinement, automated RENAL scores may be more reliable, cheaper, faster and potentially supersede human RENAL scoring in predicting post-operative outcomes.


Back to 2020 Abstracts