Upcoming event

Machine learning approach to predict pT3a upstaging of clinically localized renal cell carcinoma and oncological outcomes after surgery (UroCCR 15 study)

  • Astrid Boulenger de Hauteclocque,
  • Loïc Ferrer,
  • Damien Ambrosetti,
  • Solène Ricard,
  • Pierre Bigot,
  • Karim Bensalah,
  • Arnauld Villers,
  • François Henon,
  • Nicolas Doumerc,
  • Arnaud Méjean,
  • Virginie Verkarre,
  • Charles Dariane,
  • Stéphane Larré,
  • Cécile Champy,
  • Alexandre de La Taille,
  • Franck Bruyère,
  • Morgan Rouprêt,
  • Philippe Paparel,
  • Stéphane Droupy,
  • Alexis Fontenil,
  • Jean-Jacques Patard,
  • Romain Boissier,
  • Mokrane Yacoub,
  • Thierry Colin,
  • Jean-Christophe Bernhard

Introduction and objective

Pathological upstaging risk from clinically localized to locally advanced (pT3a) diagnosis increases with tumor size. The extension of partial nephrectomy indications to larger renal masses may question on the procedure safety in such situation. Moreover, in the era of personalized medicine and increasing interest for perioperative treatments with immune-checkpoints inhibitors, pT3a upstaging prediction gains interest. Using machine learning processes, we aimed to develop a clinically relevant model for individual preoperative prediction of this situation and comparatively assess the oncological outcomes of radical and conservative surgical approaches.


Clinical data from patients treated with either PN or RN for cT1/cT2a RCC between 2000 and 2019 and included into the French kidney cancer network database UroCCR (NCT03293563; CNIL DR2013-206) were retrospectively analyzed. Seven machine learning algorithms were applied to the cohort after a train/test split to develop a predictive model for upstaging to pT3a. DFS and OS were compared by survival curves after G-computation for pT3a tumors between PN vs RN and between laparoscopic PN vs open PN. Multivariate Cox regression models were applied to identify predictors of disease-recurrence or death in these patients.


In total, 4,395 patients were included and among them, 667 patients (15.2%, 337 PN and 330 RN) harbored a pT3a-upstaged RCC. The final prediction model based on tumor size, age, hilar location of the tumor, RENAL score, ASA score and symptoms at diagnosis, presented a prAUC of 0.41. An online calculator computing a single patient’s individual risk of upstaging was created. Survival analysis after adjustment on confounders showed no difference in DFS or OS for PN vs RN in pT3a tumors (DFS, HR 1.08, P=.717; OS, HR 1.03, P=.994); neither for laparoscopic PN vs open PN (DFS, HR 0.82, P=.429; OS, HR 1.04, P=.946). Type of surgery was not a predictor of disease-progression nor of all-cause death.


Our study suggests that machine learning technology has a great role to play in the evaluation and prognosis of upstaged RCC. In the setting of incidental upstaging, PN does not seem to compromise oncologic outcomes, even for large tumor sizes.

Source of Funding