Penn Medicine researchers develop an artificial intelligence algorithm to diagnose rare diseases

A team of researchers from Penn Medicine and the University of Florida Health is making developmental strides in rare disease diagnosis using medical artificial intelligence algorithms. 

Rare diseases, often referred to as "zebras" in the medical world because of their rare occurrence rates and “vague and perplexing” symptoms, create difficulty for doctors in diagnosis. Researchers developed a method to forecast rare disease development called the Predictive Analytics via Networked Distributed Algorithms for multi-system diseases, or PANDA, with $4.7 million of funding from the National Institutes of Health.

Rare diseases are classified as having fewer than 200,000 cases nationwide. Currently, there exists about 7,000 of them, and they impact approximately 10% of people in the United States. This results in rare diseases being hard to identify because they mimic many other common illnesses and because clinicians have little data on these cases and patients to work with in the first place. 

The data to drive this new algorithm will be pulled through the National Patient-Centered Clinical Research Network, or PCORnet. Anonymous patient data draws on records of labs, comorbidities, and prior treatments. This source contains information from 27 million patients globally.

PANDA will use electronic health records to scan for concerning symptoms and patterns that could “identify which patients are at risk for five different types of vasculitis and two different types of spondyloarthritis.”

Early diagnosis could reduce the fatality of some of these conditions and help clinicians “make better decisions for their patients,” Peter Merkel, Penn's chief of rheumatology and a medical professor, told UF Health.

PANDA's use to predict rare disease development is not the first or only application of algorithms in medicine. Algorithms developed by the Wisconsin Diagnostic Breast Cancer dataset have aided radiologists in breast cancer diagnosis with an accuracy upwards of 99%.

Another example of a diagnostic medical algorithm is one used to predict the probability that a baby with a high risk of autism, such as those who have autistic siblings, will be diagnosed with autism. The algorithm uses data points like brain volume and surface area to compute development risk with around an 80% accuracy.

PANDA has the potential to revolutionize diagnosis and save many patients, according to researchers. 

"To leverage these large collections of real-world data, which are often distributed across multiple sites, novel distributed algorithms like PANDA are much needed," chief data scientist at UF Health Jiang Bian told UF Health.