Researchers to Create AI Algorithms That Predict Patient Risk for Rare Diseases

Researchers from the Perelman School of Medicine at the University of Pennsylvania and the University of Florida College of Medicine are creating a set of artificial intelligence (AI) algorithms to determine patient risk for various rare diseases.

After receiving a $4.7 million grant from the National Institutes of Health (NIH), researchers are working to apply AI and machine learning (ML) to information from patient medical records to predict the risk of rare disease development.

With plans for development over the next four years, researchers aim to create a set of algorithms, driven by machine learning, to determine patients at risk for five types of vasculitis and two types of spondylarthritis. Vasculitis refers to a group of rare diseases that cause blood vessel inflammation, and spondyloarthritis is a type of arthritis that attacks the spine.

The set of AI algorithms, known as the Predictive Analytics via Networked Distributed Algorithms for multi-system diseases (PANDA) system, will scan data from patient EHRs to enable earlier diagnosis.

“This is an exciting step forward, building on our current PDA [Privacy-preserving Distributed Algorithms] framework, from clinical evidence generation toward AI-informed interventions in clinical decision-making,” said co-Principal Investigator Yong Chen, PhD, a professor of biostatistics at Penn Medicine, in a press release. “Despite the clear need to reduce the dangerous and costly delays in diagnosis, individual clinicians, especially in primary care, face important challenges.” 

To develop the AI method, researchers will use data from Patient-Centered Clinical Research Networks (PCORnet), which provides information on over 27 million patients. EHR data such as lab test results, comorbid conditions, and former treatments will be used to create the algorithms.

Initially, researchers plan on testing the algorithm in more than 10 health systems. The method may then be applied to conditions beyond vasculitis and spondyloarthritis, depending on the results.

“The increasing availability of real-world data, such as electronic health records collected through routine care, provides a golden opportunity to generate real-world evidence to inform clinical decision-making,” said co-Principal Investigator Jiang Bian, PhD, chief data scientist at the University of Florida Health and a professor in the health outcomes and biomedical informatics division at the University of Florida College of Medicine, in the press release. “Nevertheless, to leverage these large collections of real-world data, which are often distributed across multiple sites, novel distributed algorithms like PANDA are much needed.”

The use of predictive analytics to improve the diagnosis process is becoming increasingly common.

For example, in September, researchers at Mayo Clinic created an AI-based risk prediction model that used labor characteristics to define potential childbirth outcomes. Researchers tested the algorithm on thousands of delivery episodes, reviewing data on hundreds of variables. They found that the results provided by the model were accurate and applicable.