Patients suffering from rare diseases often endure extensive delays in receiving accurate diagnoses and treatments, which can lead to unnecessary tests, worsening health, psychological strain, and significant financial costs. Artificial intelligence (AI), including machine learning, is increasingly being integrated into healthcare to address these challenges. Researchers have now developed a method using AI to expedite the diagnosis process for undiagnosed individuals suffering from rare diseases by identifying patterns in their electronic health records (EHRs) that are similar to those observed in patients with known disorders.

At UCLA Health (Los Angeles, CA, USA), researchers have demonstrated that a machine learning tool can significantly speed up the identification of patients with rare, undiagnosed diseases, potentially improving their outcomes while reducing healthcare costs and morbidity. The focus of their study was on a group of disorders known as common variable immunodeficiency (CVID), which is frequently overlooked in medical diagnostics for years or even decades because these conditions are rare, vary widely in symptoms from one individual to another, and share symptoms with more common ailments. The complexity is further exacerbated because each case may be caused by mutations in any of over 60 different genes, with no uniform genetic mutation linking them. This genetic variability means that no straightforward genetic tests can conclusively diagnose all cases of CVID.

The team at UCLA developed a machine learning application named PheNet, a name derived from "phenotypes," which are the observable traits or characteristics of a disease in a patient. PheNet is designed to learn the phenotypic patterns associated with confirmed cases of CVID and apply this knowledge to evaluate and rank patients according to their likelihood of having the disease. Since there is no single clinical presentation for CVID, identifying an EHR "signature" for the disorder is a complex task. To tackle this, the researchers created a computational algorithm that could deduce EHR signatures from the health records of known CVID patients and the disease patterns documented in medical literature. The system calculates a numerical score for each patient, prioritizing those most likely to have CVID. These high-score patients are those the researchers describe as "hiding in the medical system," and they are recommended for referral to an immunology specialist. When the UCLA team applied PheNet to the extensive UCLA electronic health records database and conducted a blinded review of the top 100 patients identified by the system, they discovered that 74% of these patients were likely to have CVID.

“We show that artificial intelligence algorithms such as PheNet can offer clinical benefits by expediting the diagnosis of CVID, and we expect this to apply to other rare diseases, as well,” Pasaniuc said. “Our implementation across all five University of California medical centers is already making an impact. We are now improving the precision of our approach to better identify CVID while expanding to other diseases. We will also plan to teach the system to read medical notes to glean even more information about patients and their illnesses.” 

http://www.gzjiayumed.com/en/index.asp .