Katy Beckermann, Medical Director of GU Clinical Research at Tennessee Oncology, shared a post on X:
“Most kidney cancer risk tools score patients after surgery. This one runs before the first incision, and it’s fully interpretable.
In Nature Communications, Alessandro Larcher and Andrea Salonia built and externally validated a preoperative model for cancer-specific mortality in nonmetastatic RCC.
How it was built:
- Trained on 2,536 patients, externally validated on a separate 580
- Extremely randomized survival trees, kept as a transparent white-box model
- 8 routine preop inputs: tumor size, nodal status, performance status, hemoglobin, age, platelets, eGFR, BMI
What carries the weight:
- Tumor size dominates feature importance, ~3x the next variable
- By SHAP, nodal positivity and tumor size drive the largest risk shifts, then performance status and hemoglobin
Efficacy:
- Mean time-dependent AUC 0.895 vs 0.730 for the GRANT-based Cox model
- External C-index 0.88, Brier 0.02, cleanly splitting favorable / intermediate / unfavorable risk
- Free web app for point-of-care use
Why this matters for your practice: a transparent preop estimate, built on labs and exam you already have, can sharpen consent and surveillance before you operate.
Still single-program and retrospective, so prospective multi-site validation is the next bar.
Congratulations to Alessandro Larcher and the team!”
Title: A preoperative Artificial Intelligence model to estimate cancer-specific mortality in nonmetastatic kidney cancer patients
Authors: Alessandro Larcher, Alberto Traverso, Patrick Scuri, Umberto Capitanio, Giacomo Musso, Daniela Canibus, Simone Barbieri, Luca Caivano, Alan Zambello, Marco Denti, Riccardo Campi, Sergio Serni, Salvatore Granata, Anna Palmisano, Davide Vignale, Fracnesco Montorsi, Antonio Esposito, Carlo Tacchetti and Andrea Salonia.
Other articles about GU Oncology on OncoDaily.
