What is the work that you’re leading in harnessing real world evidence to develop novel therapies?
Over the past few years, we’ve worked collaboratively with hospitals, research centers and life sciences companies in public private consortia to develop solutions and platforms that enable federated data management. This data management approach uses privacy-preserving methods to query the patient level data, while the data does not leave the institution or the data custodian (e.g., the hospital) (Cowie et al, 2017).
These real-world data (RWD) platforms have now developed into mature commercial solutions. Recently, we are extending these query-based solutions with advanced analytics capabilities, e.g., artificial intelligence / machine learning (AI/ML). This novel approach, where we use federated or distributed machine learning, is also called federated learning. Conceptually, the analysis is brought to the data, rather than the data brought to the place where the analysis is conducted (Hallock et al, 2021).
Why is this approach so crucial for getting patients into trials?
Although the easiest approach is to analyze data which is pooled or centralized in one location, this is not always possible. Federated learning will allow for answers to more complex questions, e.g., physicians can gain insights in disease pathways or risk factor prediction to support clinical care. When using federated systems, a treating physician can have access to insights from similar cases outside of his or her hospital, in a privacy-preserving manner.
This can provide richer insights and optimize treatment in accordance to findings and experiences that go beyond one institution. For researchers in hospitals or healthcare organizations, this federated system is advantageous because it allows researchers to gain insights from a larger volume of complex data, while protecting data privacy and security.
A federated learning system can run advanced analytics on complex data that is stored in disparate sources. In addition to EHR data, which usually contains demographics, vital signs, diagnosis, there is more and more need to analyze genomics and imaging data to answer research questions in a personalized medicine setting.
One of the projects that we’re running, Project ATHENA (Augmenting THerapeutic Effectiveness through Novel Analytics), focuses on bladder cancer and multiple myeloma using these federated learning networks. Starting with a few academic and general hospitals in Belgium in collaboration with a strategic research institution in the Flanders region, a number of innovative start-up companies and Janssen, we are building an innovative ecosystem that enables several stakeholders to accelerate data and insight-enabled precision medicine.
How does clinical research and clinical trial matching fit into these projects?
If we go with the definition of personalized medicine as distinguishing the patients that will benefit from the treatments and those that will not benefit, then the approach of federated learning is also important in a clinical trial setting, where eligible patients need to be recruited applying more complex inclusion and exclusion criteria.
Currently, the feasibility analysis is often a manual process and many patients, who could have benefited from joining a clinical research study, are missed. We are only in the pilot phase for using federated learning, but this automated feasibility analysis is expected to become more important for clinical research and precision medicine in the coming years. Federated learning systems are also an area of interest within the emerging field of European Health Data Spaces (Digital Health Europe, 2020).
What is the role of federated learning in clinical trial matching and the optimization of precision medicine?
Federated learning can contribute to both clinical trial matching and precision medicine. It can also optimize recruitment for precision medicine studies. Data privacy is very important. The real-world initiatives that easily emerged in the US, like leveraging large volumes of claims data, are less common in Europe. We are investigating and piloting, with all stakeholders in the ecosystem, alternative approaches for using real world data.
Federated learning is one of these promising new approaches. There are already mature platforms that use federated data management to generate aggregated insights from patient level data. Federated learning, in turn, requires collaboration so you are now seeing demonstrations that this can exist, and we can layer these machine learning algorithms on top of it but it is still on a project by project basis.
What challenges exist in federated learning and using real world evidence to bring research and care together?
One of the main hurdles is the difficulty in generating high quality data at the source, e.g., high quality data in the EHR. The first step to making these systems and algorithms work in a reliable way is to have high quality, standardized data. Data harmonization is the first prerequisite to allow for interoperability.
The second challenge is that regulatory, legal, ethical frameworks are still in development, while the federated learning systems, technology-wise, are also still in the demonstration phase. In order to introduce these privacy-preserving methods at a larger scale, initiatives need to be launched to shape these frameworks.
A third challenge is the high degree of cross-functional collaboration, between different players in the health ecosystem, but also e.g., within a hospital, such systems are built in close cooperation with physicians and investigators, IT/ data management, legal and ethical departments, etc.
Besides pointing to the challenges, we should also emphasize the opportunities for engaging patients, physicians and life sciences researchers in new data and insight-driven precision medicine collaborative pilot projects.
How can other sponsors and hospitals implement these strategies?
These methods are currently developed in public- private partnerships, oftentimes based on open science principles. It’s very collaborative. Different types of hospitals, e.g., academic hospitals, general hospitals, large urban or smaller rural hospitals can potentially participate in similar programs.
What do you think precision medicine will look like in three to five years and how do we get there?
I would hope that privacy-preserving and secure methods in AI/ML, together with high quality data, can accomplish the broader benefits of precision medicine for patients. Nowadays we miss some of the insights to accomplish precision medicine at scale. Federated learning can be one of the analytic enablers to accomplish precision medicine, together with solid data privacy and security, legal and ethical frameworks to provide guidance on its applications for clinical care as well as for clinical research.
References:
Cowie, M.R.; Blomster J.L.; Curtis L.H.; ... Zalewski A. (2017). Electronic health records to facilitate clinical research. In: Clin Res Cardiol.; Jan;106(1):1-9.
Digital Health Europe (2020). Better Utilisation of Data Infrastructures to Support Secondary Uses of Health Data. White Paper
Hallock, H., Marshall, S., t Hoen, P. A., Nygård, J. F., Hoorne, B., Fox, C., & Alagaratnam, S. (2021). Federated networks for distributed analysis of health data. Frontiers in Public Health, 1316.
Hamburg, M. A. (2013). Paving the Way for Personalized Medicine. US Food and Drug Administration.
*Project ATHENA is a VLAIO Funded project, Project number: HBC.2019.2528