.jpeg)
We’re excited to share details of funding awarded to Bitfount by Innovate UK, the UK’s innovation agency, to expand our distributed data science platform to address one of the most broken parts of global healthcare R&D.
It’s fairly well known that, on average, bringing a new drug to market costs around $1.5Bn and takes anywhere between 10-15 years. What’s less well known is that at present over 80% of clinical trials are either cancelled or significantly delayed due to inadequate patient identification and screening. New regulations and increasingly complex drug candidates have also contributed to a tightening of clinical trial eligibility criteria, making patient identification harder, driving up costs and stretching timelines. This urgent issue significantly delays new life-changing treatments getting to market and prolongs patients’ suffering.
Currently, eligible patients are identified through a variety of inefficient ways, ranging from posters hung in clinic and hospital waiting rooms, to expecting clinicians to remember all the ongoing trials and discuss them with patients during appointments, to advertising on dedicated boards, forums and apps. In the most advanced cases, a rudimentary search e.g. based on age and gender, is done on the clinic/hospital’s patient database in order to rule out clearly ineligible patients. These approaches are all slow, manual, error-prone and expensive, typically taking months or years to identify and recruit enough participants. if at all.
In recent years, however, two disruptive AI-based technologies have emerged which promise a paradigm shift in the efficiency and scale at which eligible participants can be identified for clinical trials:
- Firstly, the development of cutting-edge AI-biomarkers which enable fast, cheap and accurate analysis of large high-dimensional datasets such as 3D medical images, with accuracies matching or exceeding those of trained clinicians. However, getting these models to where they’re needed most remains challenging, in large part due to data privacy concerns.
- This AI-delivery problem is addressed by the second technology innovation: distributed (or ‘federated’) data science. In this paradigm, sensitive data remains with its custodian and doesn’t need to be shared with any external parties. Instead, analysis algorithms are sent to the location of the data, allowing insights to be extracted from it while protecting the privacy of data subjects. One example of distributed data science is Federated Learning, in which AI models are trained on remote datasets without requiring the data to be centralised. Another example is Federated Execution (or ‘inference’), in which models aren’t trained, but rather just deployed to remote datasets in order to analyse them.
Despite these technical advances, deploying these new tools in real-world settings has been challenging. Earlier this year, Bitfount was awarded a grant by Innovate UK as part of the Biomedical Catalyst programme* to demonstrate the combination of these innovative ideas in a real-world clinical setting - interfacing with real-world infrastructure and datasets in the context of an ongoing clinical trial.
We’re excited to be working with two great partners on this project: Moorfields Eye Hospital NHS Foundation Trust, and the University of Surrey. Moorfields is the largest centre for ophthalmic treatment, teaching and research in Europe, and the University of Surrey is home to one of the leading Clinical Trials Units in the UK. The project combines Bitfount’s platform for distributed data science with Moorfields’ cutting-edge AI-biomarkers for Age-related Macular Degeneration (AMD) to boost recruitment in an existing clinical trial run by a major pharmaceutical company. The AMD AI-biomarker model will be deployed against against routinely-collected imaging data within Moorfields in order to identify eligible patients, without requiring any data to be transferred outside the hospital. The new methodology will be independently evaluated by the University of Surrey under a Study Within A Trial (SWAT) protocol.
AMD is the most common cause of severe loss of eyesight among people 50 and older. The estimated worldwide prevalence of AMD among individuals aged 45 to 85 years is 8.7%, with a projected number of affected people of 196 million in 2020, increasing to 288 million in 2040. Crucially, some forms of the disease still do not have any approved treatment, while quantification of its progression still requires time-consuming manual analysis by clinicians.
The platform being developed as part of this project could lead to an immediate step-change in our ability to tackle this widespread and disruptive condition, and more broadly to transform patient recruitment processes across a range of other clinical trials, especially those with an imaging component due to their time-consuming nature. It will furthermore pave the way for real-world applications of distributed data science and federated machine learning both within and outside healthcare and biomedical research.
Bitfount’s distributed data science platform is already operational and has attracted interest from a range of industries and applications. You can try the Bitfount platform out for yourself for free by signing up at bitfount.com.
If you’d like to discuss patient identification using AI, or any aspect of privacy-preserving data science, don’t hesitate to drop us a line at info@bitfount.com.
* You can learn more about the Biomedical Catalyst here: https://www.gov.uk/guidance/biomedical-catalyst-what-it-is-and-how-to-apply-for-funding
.png)
.png)
.png)
