Try UroPredict

UroPredict

Leverage an innovative machine learning model to predict kidney cancer upstaging and recurrence

Overview Research Collaboration Publications Contact

Developing the UroPredict algorithms with UroCCR

The goal of UroCCR, the French Kidney Cancer Research Network (Réseau Français de Recherche sur le Cancer du Rein), is to connect a National, multidisciplinary network of medical and scientific professionals who focus on therapeutic management and applied research into kidney cancer.

One of the world’s largest collaborative kidney cancer databases

47

Multidisciplinary clinical teams

3

Countries – France, Belgium, French Guiana

>120

Publications

All clinical, biological, and radiological data from newly diagnosed kidney cancer patients in the UroCCR network are collected in a National, multidisciplinary, clinical and biological database which is shared and used to power collaborative research projects. It is labelled by the French national Cancer Institute (INCa) and referenced by the High Authority of Health (HAS) as a valuable source of real-world data.

Try the UroPredict calculators

Use predictors with your own data

For research use only – Not for use in diagnostic procedures.

At UroCCR, our research is centered on the idea that a shared database helps facilitate and expand the use of precision medicine for patients facing kidney cancer. Our work with SOPHiA GENETICS and the results of our latest study demonstrated the benefits of using AI to analyze real-world multimodal data. SOPHiA GENETICS’ technology and network have been key to the success of our research and will be imperative as we continue progressing with increased data inputs.
Prof. Jean-Christophe Bernhard
UroCCR Coordinator, Bordeaux University Hospital, France
Our work with UroCCR over the last three years has significantly advanced its research of RCC and shown the power of AI to provide insights from real-world multimodal data. We are extremely pleased with the findings from this study and look forward to continued collaboration with UroCCR.
Thierry Colin
VP, Multimodal Research and Development, SOPHiA GENETICS

Harnessing machine learning to advance kidney cancer research

UroCCR and SOPHiA GENETICS joined forces to develop machine learning models: First to predict prior to surgery whether kidney cancer would upstage, and second to predict whether kidney cancer would recur after surgery.

Read About the Collaboration

Research objectives

Preoperative prediction of pT3a upstaging in localized or locally advanced RCC

To develop a machine learning-based, contemporary, clinically relevant model for pre-operative prediction of renal cell carcinoma pT3a upstaging in patients undergoing nephrectomy for cT1/cT2a renal cell carcinoma.

In simple terms: Develop a model that can predict whether kidney cancer will progress from a localized tumor to a locally advanced tumor (associated with worse prognosis) before the patient undergoes surgery.

Prediction of prognosis after surgery in localized or locally advanced RCC

To develop a machine learning-based, contemporary, clinically relevant model for prediction of disease-free survival in patients undergoing surgery for localized or locally advanced renal cell carcinoma.

In simple terms: Develop a model that can predict whether a patient’s kidney cancer will recur after they undergo surgery.

Prediction of pT3a upstaging in localized renal cell carcinoma

UroCCR-15

Study type: Observational
Intervention: Retrospective analysis
Clinical Trials Identifier: NCT03293563
Publication: Boulenger de Hauteclocque et al. BJU Int. 2023
Cohort: 4395 patients treated surgically for renal cell carcinoma between 2000-2019, either by laparoscopic (pure or robot-assisted) partial nephrectomy (PN), open PN, or laparoscopic radical nephrectomy (RN)

Prediction of kidney cancer recurrence after surgery

UroCCR-120

Study type: Observational
Intervention: Retrospective analysis
Clinical Trials Identifier: NCT03293563
Publication: Margue et al. NPJ Precis Oncol. 2024
Cohort: 3372 patients treated surgically for localized or locally advanced renal cell carcinoma between 2000-2020

Why?

Knowing individuals’ risk of tumor progression/upstaging can guide treatment decisions, enabling selection of the strategy with the best chance of treating the kidney cancer while minimizing side effects and maximizing quality of life.

Research is needed to develop and test such a model.

How?

Machine learning algorithms were designed, trained, and tested to predict kidney cancer upstaging and recurrence after surgery. As some data points might be missing in clinical routine, the algorithms were developed to handle incomplete data.

Read Tech Note on Avoiding Overfitting When Developing ML Models

Model for pT3a upstaging
(before surgery)

The participants were separated into two cohorts, training (n = 2636) and test (n = 1759). Seven pre-operative features were independently associated with pT3a upstaging and thus included in the predictive model:

Tumor size
Age
Hilar location
RENAL score
Sex
American Society of Anesthesiologists (ASA) score
Symptoms at diagnosis

Seven machine learning algorithms were tested and logistic regression was found to be the most effective, with a prAUC of 0.41 on the test dataset and an area under the ROC curve of 0.77.

Model for disease recurrence (after surgery)

The dataset was split into two cohorts, training (n = 2241) and test (n = 1131). 24 clinical, pathological, and biological features were included in the predictive model, including:

Tumor size
Histological subtype
Age
Fuhrman grade
Sex
Pathological primary tumor staging
Pathological regional lymph nodes staging

The final machine learning model surpassed the predictive performance of the most commonly used risk scores, with an integrated AUC of 0.79 (95% CI, 0.74–0.83) on the test dataset. The ML model had the further advantage of being able to handle incomplete data.

Applying the pT3a upstaging model

The model predicts individuals’ risk of renal cell carcinoma upstaging before surgery. As an example, the model predicted that one patient had a 32% probability of upstaging, higher than the average of tumor upstaging observed in this study (15%). Using SHAP values, we can interpret this elevated probability with risk factors (ASA score of 2, male sex, tumor size of 8 cm) counterbalanced by protective factors (young age, non-hilar location of tumor).

Applying the disease recurrence model

The model predicts the disease-free survival curve of each patient in the years following surgery, even in presence of incomplete values in predictors. It also assigns each patient to a risk group of recurrence or death within five years after surgery. Finally, using SHAP values, we are able to explain how each multimodal feature contributes to the patient-specific prediction.

What is the impact for patients?

The prediction model has the potential to support decision making around the treatment of clinically localized kidney tumors. Predicting risk of upstaging to pT3a could identify high-risk patients most likely to benefit from preoperative systemic therapy, or even low-risk patients for whom active surveillance could be sufficient.

What is the impact for patients?

The prediction model has the potential to support decision making around the management of patients undergoing localized or locally advanced kidney cancer. Predicting disease-free survival over the years following surgery could enhance the identification of patients candidate for adjuvant therapy or the identification of patients who could benefit from a less intensive post-operative follow-up.

Continuing to advance kidney cancer research using machine learning technology

UroCCR and SOPHiA GENETICS are continuing to make valuable contributions to the field of kidney cancer by training and testing machine learning technologies on the valuable UroCCR database.

Read our recently published research to discover more:

UroCCR-120: Margue G, et al. UroPredict: Machine learning model on real-world data for prediction of kidney cancer recurrence. NPJ Precis Oncol. 2024 Feb 23;8(1):45

Read Paper

UroCCR-120: Margue G, et al. Development of an individual postoperative prediction model for kidney cancer recurrence using machine learning. J Urol 2023 Vol. 209, No. 4S

Read Abstract

UroCCR-15: Boulenger de Hauteclocque A, et al. Machine-learning approach for prediction of pT3a upstaging and outcomes of localized renal cell carcinoma. BJU Int. 2023 Aug;132(2):160-169

Read Paper

UroCCR-15: Boulenger de Hautecloque A, et al. Individualized prediction of post-surgical pathologic T3a (pT3a) upstaging risk in localized renal tumors undergoing nephrectomy. JCO 2022 40:16_suppl, 4547-4547

Read Abstract

Margue G, et al. Disease-free survival in operated patients with nmRCC using machine learning. JCO 2023 41:16_suppl, 4539-4539

Read Abstract

UroPredict

Leverage an innovative machine learning model to predict kidney cancer upstaging and recurrence

Developing the UroPredict algorithms with UroCCR