In her first year at the David Geffen School of Medicine at UCLA, Katharina "Kat" Schmolly, MD, learned a popular medical adage: "When you hear hoofbeats, think of horses, not zebras." The saying encourages doctors to consider common diagnoses before rare ones. Initially, Dr. Schmolly, with her background in equine science, found it sensible. However, her perspective shifted during a hepatology lecture by Dr. Simon W. Beaven, who treats patients with acute hepatic porphyria (AHP), a group of rare genetic disorders.
AHP mainly affects women, often coinciding with menstrual cycles, and causes severe symptoms like abdominal pain, nausea, vomiting, limb weakness, and anxiety. "Women often get dismissed when they repeatedly visit emergency departments with these complaints," Dr. Schmolly noted. "It might look like menstrual pain, but it could be a serious liver disease."
This unfairness motivated her to found zebraMD, an AI-driven tool designed to help diagnose and manage rare and genetic diseases. ZebraMD uses a predictive algorithm to sift through electronic health records, identifying patterns and flagging patients at risk so physicians can conduct further tests and diagnoses. The tool aims to highlight rare diseases (the zebras) over common ones (the horses).
Addressing Diagnostic Delays
"Diagnostic delays for these diseases average 10 to 15 years because physicians rarely encounter them," Dr. Schmolly explained. "During this waiting period, the disease can progress and cause irreversible damage. Our goal is to diagnose patients earlier and manage their disease appropriately."
By definition, a rare disease affects fewer than 200,000 people. However, there are over 10,000 known rare conditions collectively affecting more than 30 million people in the U.S., similar to the prevalence of diabetes. While some rare diseases like multiple sclerosis are well-known, most are not.
AHP affects 1 in 100,000 people. In 2019, the FDA approved givosiran for treating recurrent AHP attacks, but it typically takes 15 years to diagnose the condition. To expedite diagnosis, Alnylam Pharmaceuticals, the drug's manufacturer, collaborated with UCSF Porphyria Center researchers to develop an algorithm for identifying potential patients.
Project Zebra
When Dr. Schmolly expressed her interest in rare diseases, Dr. Beaven connected her with Dr. Vivek Rudrapatna, an assistant professor at UCSF and director of The Real-World Evidence Lab, which applies data science techniques to health records. Together, they co-invented Project Zebra's predictive algorithm to analyze healthcare records and identify suspected porphyria patients.
Dr. Rudrapatna trained as a clinical data scientist during his UCSF fellowship, while Dr. Schmolly drew on her pre-medical experience at Medtronic, where she worked on a diagnostic algorithm for a cardiac monitoring device. They used de-identified patient records from UCSF and UCLA, part of the UC health network's roughly 10 million records.
Overcoming Challenges
The algorithm's first challenge was handling messy data from patient records, which include both structured data (vital signs, lab results, demographics) and unstructured data (physician notes). The researchers organized this information to train the machine learning algorithms.
Another challenge was preventing the algorithm from "cheating" by using data points that would reveal a disease diagnosis too directly, like referral orders for specialist visits. "We want algorithms that can discover patients way before clinicians suspect the disease," Dr. Rudrapatna emphasized.
Training the Algorithm
The algorithm was trained using three main resources: expert knowledge from Dr. Bruce Wang, director of the UCSF Porphyria Center; a rare and genetic disease database from the NIH; and the algorithm's own data analysis to identify signals. This approach is crucial because rare diseases often get misdiagnosed.
In their study published in the Journal of the American Medical Informatics Association, the algorithm predicted with 89% to 93% accuracy which patients would be referred for AHP testing. It identified 71% of AHP patients earlier than their actual diagnosis, saving an average of 1.2 years.
Dr. Schmolly and her team aim to expand the use of zebraMD to help more patients receive timely and accurate diagnoses for rare diseases, ultimately improving their quality of life.
More: https://medicalxpress.com/news/2024-07-ai-powered-tool-doctors-rare.html
