Abstract: SA-PO0631
Can Artificial Intelligence Reduce Racial and Ethnic Disparities in Fabry Disease Diagnosis?
Session Information
- Monogenic Kidney Diseases: Tubular and Other
November 08, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Genetic Diseases of the Kidneys
- 1201 Genetic Diseases of the Kidneys: Monogenic Kidney Diseases
Authors
- Kallish, Staci, University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Nelson, Matthew, Amicus Therapeutics, Inc., Princeton, New Jersey, United States
- Ostrovsky, Yuri, OM1, Inc., Boston, Massachusetts, United States
- Zabinski, Joseph, OM1, Inc., Boston, Massachusetts, United States
- Hennessy, Laura, University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Veleva-Rotse, Biliana O., Amicus Therapeutics, Inc., Princeton, New Jersey, United States
- Giuliano, Joseph D., Amicus Therapeutics, Inc., Princeton, New Jersey, United States
Background
Fabry disease (FD) is a rare, pan-ethnic, X-linked disorder that affects both males and females and is characterized by a deficiency in the α-galactosidase A enzyme. Its birth prevalence has been estimated at approximately 1 in 40,000 but the true rate is likely higher.
Methods
We previously developed and validated a machine learning model (MLM) that uses electronic health and claims records to identify patients at high risk of having undiagnosed FD. In the present study, medical records of adults from the Penn Medicine, University of Pennsylvania Healthcare System that met study inclusion criteria were classified based on the presence/absence of an FD diagnosis code or an FD medication code as the confirmed-FD cohort (at least one code, n=47) and the overall population (no codes, n=580,319). The MLM was applied to the overall population to identify the 100 patients with the highest risk scores for undiagnosed FD (high-FD-risk cohort). Demographic variables were analyzed, including age, sex, race, ethnicity and healthcare encounters.
Results
Patients in the confirmed-FD cohort were more likely to be male compared with the overall population (51.1% vs 37.5%, respectively) and were of similar age (mean 60.3 vs 58.7 years); Black patients were underrepresented (2.1% vs 18.5%) as were Hispanic patients (2.1% vs 4.5%), despite having higher numbers of healthcare system encounters. Application of the MLM bridged these disparities as evidenced in the demographics of the high-FD-risk cohort. Patients in the high-FD-risk cohort were more likely to be female (53.0%) than those in the confirmed-FD cohort and were slightly younger (mean 55.2 years); they were also 14 times more likely to be Black (31.0%) and 2.3 times more likely to be Hispanic (7.0%).
Conclusion
Sparse information exists on racial and ethnic disparities within the FD population and reasons for these disparities are only speculative at this time. We suggest that properly applied artificial intelligence may not only assist in identification of adults with undiagnosed FD in general (by flagging those with elevated risk for consideration for diagnostic confirmation) but may particularly benefit those in underrepresented groups.
Funding
- Commercial Support – Amicus Therapeutics, Inc.