About

Bitfount is a UK-based startup that is revolutionising B2B data collaboration. Bitfount’s federated AI and data science platform enables organisations to run successful data collaborations without any data sharing taking place, safely unlocking the value of sensitive and previously unused data.

Healthcare data scientists, researchers, and analysts use Bitfount to train, validate and deploy their ML models on imaging, EHR and tabular data held within hospitals and clinics. In addition to AI applications, Bitfount also supports multiple additional privacy-enhancing technologies (PETs) out of the box, including techniques like differential privacy, secure aggregation, PrivateSQL and private set intersection. This novel approach to data collaboration has unlocked previously inaccessible data sets, creating many exciting new collaboration opportunities in the healthcare industry.

For example, the process of identifying eligible patients for participating in clinical trials is currently excruciatingly inefficient, time-consuming and expensive. As a result:

• 86% of trials fail or are significantly delayed due to not meeting recruitment timelines
• Over 50% of trial sites recruit only 0 or 1 patient
• Estimates of missed revenue are on the order of $8M per day that a trial is delayed

Bitfount offers a transformative solution to these challenges by securely and efficiently unlocking the latent value in siloed clinical data.

The Bitfount platform is used to deploy AI algorithms on the data held within the clinics and hospitals participating in the trial to automatically identify and flag patients who meet the trial’s eligibility criteria. The desktop application or SDK takes minutes to install at the trial site and is easily integrated with imaging and EHR databases. It is HIPAA and GDPR compliant, and no data is transferred away from the trial site to Bitfount, to the trial sponsor or to anyone else.

Bitfount automatically analyses all routinely collected images as well as EHR information using state-of-the-art AI models, with zero transfer of data from the clinic. This includes cutting down image analysis times from 30 minutes to seconds per image. Detailed reports are provided on patients who may be eligible for the trial based on its inclusion/exclusion criteria. This reduces clinician pre-screening time and increases trial site yield by several orders of magnitude, and can also be used to improve trial feasibility and site selection assessments.