Machine learning tools identify rare and undiagnosed immune diseases from patient EHRs
By HospiMedica International Staff Writer
Posted: May 2, 2024
Patients suffering from rare diseases often experience significant delays in receiving accurate diagnosis and treatment, which can lead to unnecessary testing, poor health, psychological burden, and significant financial costs. . To address these challenges, artificial intelligence (AI), including machine learning, is increasingly integrated into healthcare. Researchers now use AI to help diagnose undiagnosed individuals suffering from rare diseases by identifying patterns in electronic health records (EHRs) similar to those observed in patients with known diseases. We have developed a method to speed up the process.
UCLA Health (Los Angeles, California, USA) studies how machine learning tools have the potential to significantly speed up identification of patients with rare and undiagnosed diseases, improving outcomes while reducing healthcare costs and morbidity. They have proven it. Their research focused on a group of diseases known as common variable immunodeficiency disorders (CVID). CVID is often missed by medical diagnoses for years, even decades, because these diseases are rare, symptoms vary greatly from person to person, and they have commonalities. Symptoms of more common diseases. The complexity is compounded by the fact that each case can be caused by mutations in any of over 60 different genes, and there is no uniform genetic mutation that links them. This genetic diversity means that simple genetic testing cannot conclusively diagnose all CVID cases.
Image: Machine learning tools can identify patients with rare, undiagnosed diseases years in advance (Photo credit: 123RF)
A team at UCLA has developed a machine learning application named PheNet. The name comes from “phenotype,” which is an observable trait or characteristic of a patient's disease. PheNet is designed to learn the phenotypic patterns associated with confirmed cases of CVID and apply this knowledge to assess and rank patients according to their likelihood of developing CVID. Because CVID does not have a single clinical symptom, identifying her EHR's “signature” of this disease is a complex task. To address this, the researchers created a computational algorithm that could infer the characteristics of her EHR from her known CVID patient health records and disease patterns described in the medical literature. The system calculates a numerical score for each patient and prioritizes those with the highest likelihood of CVID. These high-scoring patients are what the researchers describe as “hidden in the health care system” and are recommended for referral to an immunology specialist. When the UCLA team applied his PheNet to the extensive UCLA electronic medical records database and conducted a blinded review of the top 100 patients identified by the system, they found that 74% of these patients suffered from her CVID. It turns out that there is a high possibility that
“We have shown that artificial intelligence algorithms such as PheNet can provide clinical benefit by speeding the diagnosis of CVID, and we hope this will be applied to other rare diseases as well. ,” Pasaniuk said. “Implementation at all five UC medical centers is already having an impact. We are now expanding to other diseases while improving the accuracy of our approach to better identify CVID.” We also plan to teach the system to read medical notes to gather more information about patients and their illnesses.”
Related Links:
UCLA Health
