top of page

How Federated Learning Can Improve Time to Value from Clinical Trials

Clinical trials are hugely expensive and time-consuming, with an estimated cost of $77M¹ for a single clinical trial and an average length of 6-7² years per trial. Shaving even a portion of this time and expense represents material savings to trial sponsors, and maybe more importantly, gives them a longer exclusivity period during which they can recoup their upfront investments. Federated learning, and specifically the end-to-end, highly reproducible, federated learning practices supported by the Rhino Health Platform, can help.

Federated learning is a powerful machine learning technique that allows for the training of models on multiple decentralized data sets. Federated learning (FL) has several applications for clinical trials:

  • FL can be used to analyze local data from electronic health records (EHRs) and other patient data collected by healthcare organizations, e.g., to screen for patients that may be eligible for a particular clinical trial. Due to its privacy-preserving properties, FL lowers the barriers for healthcare organizations to collaborate with industry on use cases like these, thus enabling broader access to data from more institutions and helping industry R&D teams enroll patients faster.

  • Variants of FL (e.g., vertical FL) can also be used to incorporate other types of data (e.g., data collected by wearable devices, such as fitness trackers) into models developed on data from clinical systems. This is valuable as it increasingly lets R&D teams incorporate ‘real-time’ information on patients' health and activity levels, such as when monitoring patients remotely during a clinical trial.

  • FL can also be used for prospective patient recruitment by deploying predictive models to run in real-time across varied data sets at multiple participating sites. By predicting who will be a good candidate for a clinical trial already ahead of time, clinical operations teams get more time to act, which becomes critical as the window for treatment is often limited.

One of the key advantages of FL is that it allows you to use a large and diverse data set for model development while letting that data remain at rest with the data custodians at all times. In turn, this helps protect patient privacy and reduce the risk of data breaches. Leaving data at rest behind institutional firewalls makes healthcare organizations' information security teams more comfortable with using applications such as the ones listed above to expedite clinical trials - broadening the potential network of sites AND expanding what data are available for use.

Additionally, because the data are processed where they live, there is less need for powerful centralized servers, which can reduce the cost of computing for the sponsor.

Currently, the use of federated learning in clinical trials is still in its early stages. Several real-world evidence (RWE) platform companies provide access to, or insights based, on anonymized federated data sets. These platforms connect to various data sources, such as EHRs and claims data, and apply a set of standardization and harmonization processes to link the data together to create a unified data model. This enables researchers to access the data in a consistent format and perform advanced analytics to generate insights and support clinical research. While these are critical first steps, the Rhino Health Platform can help RWE companies go one step further into true federated learning, by also supporting the training and validation of AI-models in a distributed fashion. Also supporting the execution of virtual containers ‘on the edge’, the Rhino Health Platform enables use cases like distributed ETL and data harmonization, which circumvents most of the need to centralize data.

Rhino Health is working with RWE platform companies and biopharma sponsors to realize the potential of AI in clinical trials via federated learning through use cases like the above. Rhino Health’s turnkey federated learning solution supports the full value chain of AI model development. The Rhino Health Platform specifically is totally extensible, allowing for work across any file format and system architecture, together with the tools and paradigms data scientists are already used to. Rhino Health is trusted by dozens of hospitals around the world, making implementation easy. These forward-thinking companies, powered by Rhino, are going to change the way clinical trials are run for the better - bringing drugs to market faster and less expensively.

  1. Tufts Center for the Study of Drug Development. “Research Milestones.”

  2. PhRMA. “Biopharmaceutical Research & Development: The Process Behind New Medicines.”

By Chris Laws, VP of Operations; Mathias Blom, VP of Partnerships

225 views0 comments


bottom of page