Menu
We sat down with Prof. Jean-Christophe Bernhard, UroCCR Coordinator, and Dr. Gaëlle Margue, Urology Fellow, at University Hospital Bordeaux, to discuss the collaboration between UroCCR – the French Kidney Cancer Research Network - and SOPHiA GENETICS, and get their insights on the use of AI-powered multimodal approaches to improve patient care.
Watch the spotlight:
Gaëlle, Jean-Christophe, hello. It's a real pleasure to be with you today to discuss the collaboration between SOPHiA GENETICS and UroCCR.
Before we go into more details about this collaboration, I would like you to tell us a little about your background, your life as a surgeon at the hospital, and also introduce the UroCCR network and the role you play within this network.
Gaëlle Margue, I am a junior doctor in urology at the Bordeaux University Hospital. I arrived in 2018 to start my internship and now, I have been a PhD student for two years and my science thesis focuses on kidney cancer, surgical and oncological themes, within the framework of the UroCCR network, the French Kidney Cancer Research Network, and the I.CaRe team, a Kidney Cancer Research team in Bordeaux.
Jean-Christophe Bernard, I am a professor of urology at the Bordeaux University Hospital, and I coordinate the I.CaRe team, the Kidney Cancer Research and Innovation Program in Bordeaux. I am also the national coordinator of the French Kidney Cancer Research Network, UroCCR.
Gaëlle, we worked together on your medical thesis around UroPredict. Could you describe UroPredict, and what it does?
Yes, absolutely. That was the subject of my medical thesis. The initial idea was to better characterize the risk of recurrence after kidney cancer surgery. Therefore, for patients with localized kidney tumors that are operated and are considered cured or in remission after surgery, we want to better determine what factors lead to patient relapse or not. We have prognostic scores to try to determine that, but they are not very effective, and so we wanted to better characterize that, to propose a follow-up schedule tailored to the risk of recurrence, or additional treatments for patients who have a high risk of recurrence.
In UroCCR, we have a lot of data that relates to these kidney cancer surgeries - clinical, biological, imaging, surgical, and monitoring data -, which we were not able to leverage, with traditional statistics, to better characterize the risk of recurrence. So the goal was to create a new machine learning score, a tool to predict recurrence in these patients, based on all this data that we have in UroCCR.
It's quite fascinating, from an outside perspective, to see surgeons saying that we need Artificial Intelligence and Machine Learning, to be able to advance our care. Jean-Christophe, from a strategic point of view, how do you see the collaboration with SOPHiA GENETICS, in particular? And then more broadly, the role these precision medicine tools are expected to play in the future?
We see it as something that will be decisive, and I would say, necessary to maximize the value of all the data we gathered over time.
UroCCR, to go back a little into the history, is a project that dates back to 2006, which was certified by the National Cancer Institute in 2011. It was initially deployed as a multicenter project in 2013 across 11 centers. And today, we’ve grown from 11 centers to 54 and soon 58. And so, the positive excitement of the system means that we are constantly collecting highly qualified data on the pathology of patients who are diagnosed with a urological tumor. We collect this data regardless of the treatment method, whether the patient is in active surveillance, whether he is operated on, whether he is treated systemically with medication, or whether he is treated with interventional radiology.
In doing so, coupled with the increase in the number of centers, and the increase in the number of patients, we reached the 20,000 patients included in the UroCCR, this dataset became even more considerable since we are linked to the SNDS, the National Health Data System, and therefore we can also do medico-economic evaluation, representing nearly 10 million data points available on patients treated contemporaneously by French teams for kidney cancer.
It actually becomes a need, not so much a vision but ultimately a necessity to have new statistical methods to process this data, and derive insights to enhance our ability to provide patients with care that is as personalized and individualized as possible.
This is, I believe, the foundation of our collaboration with you and your team at SOPHiA GENETICS, to explore how we can take advantage of all the data patients have entrusted us with, regarding their illness, since everything is done with the patients’ consent - and I think it’s also important to point this out. Being able to produce new tools, and what Gaëlle said is that her thesis work, which initially is a scientific and fairly general work, I would say, has nevertheless led to the online publication of a calculator, which can now be used in a pragmatic way to answer a question that we may ask ourselves for a given patient.
Ok. So, beyond the research aspect that we were able to carry out together, which I think we are all very happy with, there is this role that an industrial company can play afterward, which will be the deployment of the tools, their validation, their improvement, the entire life cycle of the software.
Are these the things that are beginning to resonate in the minds of the medical profession, or are we still at the beginning and still at a time when everyone needs to find their role?
Yes, it is... It is very timely, because the UroCCR network has just been certified by the ANR, the National Research Agency, as a clinical investigation network on medical devices. These are also themes that we are addressing within the framework of the I.CaRe program, within the framework of the RHU (University Hospital Research in Health) where there is this desire for collaboration between academics and manufacturers for carrying out projects and arriving at outlets that are tools, products that can be used in everyday life and in the routine of patient care.
The UroPredict tool that we co-developed during your thesis, Gaëlle, it is now deployed and accessible. How is it being used by you?
It is a tool that is not yet a certified medical device, so it cannot yet be used to change patient care strategies.
However, it is very useful for discussing with the patient. We can present it as a research tool that helps us better assess their risk of recurrence and concretely show the patient their risk factors for recurrence and their personalized risk. And this allows us to develop the discussion around monitoring or potential complementary treatments to surgery, for the continuation of their care.
Do you see among your colleagues or among the pharmaceutical companies, a little reluctance to apply them, to trust them, at the level of clinical routine?
So, reluctance… I don’t think it’s really a reluctance. I think that at each time there is something new, that potentially can bring a change in practices. There is always an observation phase and we have to assimilate what this novelty can be and what it can bring. We experienced the same thing, for example, for the introduction of robotic assistance in surgery, which today everyone is convinced on its benefit, both for health professionals, for surgeons, but also for patients. There was a whole phase where the community asked itself the question of whether introducing robotics would actually make it possible to do better than conventional techniques that were, I would say, well known, and had been validated for many years.
So there is always this moment where we ask the question of whether it is a real innovation, whether it will really bring added value. There is always this observation phase. Afterwards, I think that more and more, the medical and surgical community is convinced that with technological progress, we are finally able to improve what our daily routine is, what our practice is and obtain additional information that will enrich our practice and patient care.
What do you think the next algorithm we should develop is?
We have several ideas. We could further enrich this one by perhaps incorporating radiomics or pathomics, as we discussed, adding imaging data to increase the precision of the current tool. Then, we could explore many other algorithms with different objectives, such as predicting kidney function loss after kidney cancer surgery, or predicting the risk of morbidity, and intra or post-operative complications.
Objectives and things that we are trying to better characterize and there are several that we could look into and probably we have all the necessary data from UroCCR, we need to be able to exploit it.
And I think if I can complete.
There is also a question that we should address. Every time, we evaluate outcomes that are very objective, focused rather on the practitioner, the surgeon, the quality of his surgery, the outcome, the evolution of the disease, etc.
But I think that we should also be able to work on outcomes that are reported to us by the patient.
Quality of life, satisfaction, anxiety, which are things that we capture in UroCCR and in particular thanks to the UroConnect application. Today, in UroCCR, we have what we call PROMS and PREMS, which are patient-reported outcomes, and I believe this is another field worth investigating, moving beyond purely scientific outcome predictions, but also to take into consideration what the patient's experience is, and to judge what we do or to predict how we can improve what we do and propose to patients, based on their feedback about the quality of their care. I think this is one of the objectives we should collectively aim for.
We thank Prof. Bernhard and Dr. Margue for their time and for sharing their experience. Visit the UroPredict page to learn more about this machine learning model on real-world data for the prediction of kidney cancer recurrence.
SOPHiA GENETICS products are for Research Use Only, not for use in diagnostic procedures unless otherwise specified.
Explore this infographic summary to learn more about the machine learning model developed by Margue et al. for the prediction of disease-free survival in patients undergoing surgery for renal cell carcinoma.
Margue G, et al. UroPredict: Machine learning model on real-world data for prediction of kidney cancer recurrence (UroCCR-120). NPJ Precis Oncol. 2024 Feb 23;8(1):45.
SOPHiA GENETICS multimodal analysis technology and concepts in development. May not be available for sale.
Do you want to further your understanding of overfitting, or are you interested to learn how we avoid overfitting when developing predictive machine learning models for clinical research? Browse our tech note, which includes a step-by-step, published example of how we developed a model that was able to successfully support the evaluation of kidney cancer tumor upstaging in individual patients.
Multimodal healthcare datasets synergistically integrate diverse data modalities such as genomic, clinical, radiomic, proteomic, and biological data, to provide comprehensive insights into human biology and medical conditions. Multimodal datasets have the potential to predict outcomes more accurately and informatively than the sum of their parts (Fig. 1).
Figure 1.Multimodal healthcare data integrated and analyzed by artificial intelligence (AI)/machine learning can provide useful information for healthcare professionals to use to improve patient care.
Genomics data.
Radiomics data include x-rays, CT scans, MRI scans, ultrasound images, and mammograms.
Clinical and biological data from electronic health records include patient histories, demographics, notes, diagnosis codes, procedure codes, laboratory results, and vital signs.
Proteomics data.
Digital pathology data.
Patient-reported data includes questionnaires and health journals, as well as data from wearable devices monitoring heart rate, sleep patterns, and activity levels, and implantable devices such as pacemakers, insulin pumps, and continuous blood glucose monitors.
Environmental data includes air quality and location data.
New data-driven technologies powered by novel ways of linking and analyzing patient data are set to transform the way that healthcare is delivered.1 Healthcare professionals routinely make use of multiple sources of data to arrive at a diagnosis and to decide on patient management.2 However, a significant level of expertise is required for an in-depth understanding of even a single data type (e.g. radiological images) such that it is unfeasible for individual healthcare professionals to master all areas. AI/machine learning technologies can be leveraged to bring together and analyze multimodal healthcare data, breaking data silos and creating robust and accurate predictive models.3 With the appropriate guidance around decision-making and communication, the valuable insights gained from these predictive models have the potential to support healthcare professionals to improve patient care.
Machine learning technologies can integrate data from disparate multimodal sources to provide a holistic understanding of patients’ health and medical conditions. Data are combined from multiple modalities with the aim of extracting complementary information to power predictive models that can find relationships between different variables/features that are not clearly visible or known by healthcare professionals. Indeed, multimodal data fusion models have consistently shown to provide increased accuracy (1.2-27.7% higher) and performance (AUC 0.02-0.16 higher) than models that utilize data from single modalities for the same task.4
Oncology is one of the medical specialties that most commonly leverages multimodal methods for clinical decision support.5 Machine learning technologies have the potential to explore complex and diverse data to support healthcare professionals from screening to treatment (including relapse).6 Identification of risk factors can support non-invasive patient screening and preventive care.3 Detection of patterns in easily accessible data can help identify diagnostic or prognostic biomarkers to improve patient risk stratification or selection for clinical trials. Identification of predictive signatures of risk factors, adverse treatment reactions, treatment responses, or treatment benefit, can guide decisions around patient management.
Figure 2. The number of PubMed articles published on multimodal oncology data has dramatically increased in recent years.
PubMed search for ((multimodal) AND (oncology)) OR ((multimodal) AND (cancer)).
*2023 analysis includes data available at time of writing (January-September).
With data privacy and security paramount, multimodal healthcare data can also be leveraged to accelerate advances in medical research, such as the discovery of novel biomarkers and therapeutic targets for drug development, as well as supporting population health management by providing a comprehensive view of health trends and outcomes. The rapid increase in peer-reviewed publications on the topic over the last 13 years demonstrates that the extraordinary value of multimodal oncology data is already recognized by the scientific and medical communities (Fig. 2). Leveraging machine learning to collate and analyze the vast diversity of multimodal data for data-driven precision medicine is on track to drive the next revolution in healthcare.
SOPHiA DDM™ multimodal healthcare analytics will have the potential to break data silos by streamlining the integration of longitudinal oncology data from multiple sources and modalities – including but not limited to genomic, radiomic, digital pathology, biological, and clinical data. The SOPHiA DDM™ Platform uses machine learning-powered analytics to assemble, standardize, and transform multimodal data into accessible data-driven insights, facilitating the identification of multimodal predictive signatures, as well as treatment response patterns and trends. To learn more and get in touch, visit the webpage.
Product in development – Technology and concepts in development. May not be available for sale.
Area under the ROC curve (AUC) – A ROC (receiver operating characteristic) curve is a graph that plots true and false positive rates to demonstrate the performance of a model. AUC measures the area underneath the ROC curve to provide an aggregate measure of performance. AUC values range between 0 and 1, with a score of 0 meaning that all predictions are wrong, and a score of 1 meaning that all predictions are 100% correct. Essentially, AUC represents the probability that a positive result is truly positive and a negative result is truly negative.
Omics data – Large-scale information related to the biology of organisms.
Digital pathology images – Scanned images of tissue samples on glass slides.
Precision medicine, also known as personalized medicine, aims to enhance healthcare quality by tailoring treatments to each person's unique genetic makeup, environment, and lifestyle. While a fully individualized approach to medicine is still a work in progress, the recognition that patient heterogeneity influences treatment effectiveness is not new1.
Historically, medicine has heavily relied on trial-and-error strategies for discovering, developing, and testing new treatments targeted at specific indications. This disease-centered approach resulted in predetermined standard therapies tailored to the “average patient.” While this one-size-fits-all approach has succeeded in many indications, it also carries significant drawbacks, particularly when dealing with complex diseases such as cancer or inherited disorders. In these cases, the risk of adverse side effects (e.g., toxicity) and reduced therapeutic response often result in poorer patient prognoses and quality of life1.
Precision medicine represents a patient-centric paradigm shift, acknowledging each individual's uniqueness while using real-world data and advanced statistics to guide the discovery-to-treatment process. For instance, pharmacogenomics requires us to look at each patient genomic data individually and in the context of others, enabling stratification into cohorts for predicting treatment responses2.
Success in precision medicine hinges on the ability to derive meaningful insights from large patient datasets. Fostering data diversity has the potential to further advance progress in this area1,3.
Recent technological advances have made precision medicine more accessible and impactful than ever before. Next-generation sequencing (NGS) has become more affordable, transforming it from a research-focused technology into a tangible clinical reality. This progress was further propelled by the widespread adoption of electronic health records (EHRs) and laboratory information management systems (LIMS), which not only facilitate population-scale research but also enable the use of clinical decision support tools for the delivery of targeted therapies to individual patients4.
The ability to identify genetic biomarkers and assess variant pathogenicity has grown significantly in the past decade. This has not only revolutionized patient diagnosis but also transformed drug development. A pivotal moment was the approval of imatinib by the FDA in 2001, the first small molecule targeted therapy for chronic myeloid leukemia (CML)5. By inhibiting the BCR-ABL fusion protein, imatinib was shown to effectively slow the progression of CML from chronic phase to blast crisis, making it the first of its kind.
This groundbreaking milestone paved the way for the approval of many other targeted therapies, such as gefitinib targeting EGFR alterations associated with NSCLC (2003) and trastuzumab for HER2-positive breast cancer (2006). The pace of new targeted drug approvals continues to accelerate year after year, heralding a promising era of precision medicine6.
Timeline of FDA-approved targeted therapies in cancer. Grey bubbles represent the relative number of approvals per year. Data source: Waarts et al 2022.
To achieve a truly personalized approach to medicine, the harmonization of translational and precision medicine is paramount. This coordination between early mechanism-based drug development and late-stage patient-centric approaches gives rise to an end-to-end biomarker-guided process, allowing us to optimize treatment strategies for patient cohorts right from the outset7.
Known as translational precision medicine, this emerging concept brings a fresh perspective to the translational gap, calling for a broader scope beyond a purely genetic-based definition of biomarkers and introducing a multimodal approach by taking into account a wider range of healthcare variables. To make this new concept a reality, significant technological progress is required in several key areas8:
By addressing these critical areas of advancement, we can pave the way for a future where each patient receives personalized treatments tailored to her or his unique needs and characteristics. The pursuit of translational precision medicine promises to revolutionize healthcare, offering improved patient outcomes and transforming the landscape of medical research and development.
Powered by proprietary algorithms and enriched with data from 750+ institutions*, the SOPHiA DDM™ Platform accelerates advances in the field of precision medicine. Its core mission centers on empowering clinical researchers across healthcare and biopharma spheres alike.
To learn more about SOPHiA DDM™ BioPharma Solutions for biomarker-centric discovery, development, and application deployment, visit our page.
* The number of institutions represents active customers who have generated revenue through the SOPHiA DDM™ Platform usage or Alamut™ Visual Plus licenses as of September 30, 2022.
SOPHiA GENETICS products are for Research Use Only and not for use in diagnostic procedures unless specified otherwise.
SOPHiA DDM™ Dx Hereditary Cancer Solution, SOPHiA DDM™ Dx RNAtarget Oncology Solution and SOPHiA DDM™ Dx Homologous Recombination Deficiency Solution are available as CE-IVD products for In Vitro Diagnostic Use in the European Economic Area (EEA), the United Kingdom and Switzerland. SOPHiA DDM™ Dx Myeloid Solution and SOPHiA DDM™ Dx Solid Tumor Solution are available as CE-IVD products for In Vitro Diagnostic Use in the EEA, the United Kingdom, Switzerland, and Israel. Information about products that may or may not be available in different countries and if applicable, may or may not have received approval or market clearance by a governmental regulatory body for different indications for use. Please contact us to obtain the appropriate product information for your country of residence.
All third-party trademarks listed by SOPHiA GENETICS remain the property of their respective owners. Unless specifically identified as such, SOPHiA GENETICS’ use of third-party trademarks does not indicate any relationship, sponsorship, or endorsement between SOPHiA GENETICS and the owners of these trademarks. Any references by SOPHiA GENETICS to third-party trademarks is to identify the corresponding third-party goods and/or services and shall be considered nominative fair use under the trademark law.