AI Developers Need Real-World Data — and Atropos Health Is Stepping Up to Help – MedCity News

Real-world data is invaluable to healthcare tech developers. This type of data reflects actual patient experiences, treatments and outcomes in diverse, real-world environments — rather than controlled clinical trial settings, pointed out Atropos Health Brigham Hyde.

Yet, many of the developers creating new AI models for healthcare organizations struggle to access real-world data. An April paper analyzed more than 500 studies on large language models in healthcare and found that only 5% of them were conducted using real-world patient data.

To address this problem, Atropos recently announced that AI developers can now train their models on its real-world evidence network.

Atropos, which was founded in 2020 as a Stanford spinout, delivers real-world clinical data to physicians at the point of care. In 2023, the startup launched its evidence network — which is a federated healthcare data network consisting of more than 300 million patient records gathered from EHRs, claims data and patient registries

The network currently has “dozens” of members, including AI developers, practitioners, researchers, data holders and tech companies, Hyde said. With access to such a large amount of real-world patient data, members of the network gain a comprehensive and representative view of how diseases progress and treatments perform across varied populations, he explained.

Now that the evidence network offers AI model training, developers can seamlessly integrate their AI tools into the network’s infrastructure. This new capability is powered by Atropos’ GENEVA OS platform, which transforms real-world data into clinical evidence by providing physicians with quick, data-driven answers to complex medical questions.

“Using GENEVA OS, developers can train, test and validate predictive models on standardized, high-quality patient-level data. This eliminates the burdens of data acquisition and preparation, allowing for rapid model development while adhering to emerging AI assurance standards for transparency, bias detection and accuracy,” Hyde declared.

Overall, the data network’s infrastructure seeks to accelerate AI development, as well as boost AI reliability, with the overarching goal of driving innovation that improves patient care and outcomes, he added.

Hyde pointed out some use cases for AI tools that could be trained on the network — such as clinical trial simulation, patient journey mapping, cost of care estimation and outcome prediction. Developers can ultimately deploy validated models to Atropos’ channel partners, such as health systems or pharmaceutical companies, he said.

The CEO of one member of the evidence network — QuantHealth, a startup using AI to make it faster and cheaper for pharmaceutical companies to develop treatments — noted that Atropos’ data platform has allowed his company to quickly fine-tune its product.

“De-risking and optimizing clinical trials through robust patient-level simulations is no easy feat, which is why we have continued to evolve and mature our AI platform and underlying data frameworks,” QuantHealth CEO Orr Inbar said in a statement. “By doing this, we’ve been able to help 7 of the top 20 pharma companies simulate and optimize their trials and clinical programs to ensure clinical and operational excellence.”

QuantHealth can now perform real-time simulations and deploy its AI models to the point of care, “unlocking new opportunities and use cases” for its pharma customers, Inbar declared.

Photo: metamorworks, Getty Images