Researchers Compare EHR Linkage Methods For MS

Saudi researchers from the Saudi Food and Drug Authority and King Saud University analyze deterministic, probabilistic, and machine-learning linkage methods on deidentified EHRs for 2,247 multiple sclerosis patients spanning 2016–2023. In simulations, deterministic and probabilistic approaches achieved F1 scores of 97.2% and 93.9%, respectively, while ML reached up to 99.8% but incurred substantially higher compute times. They conclude probabilistic linkage yields the best balance of recall, precision, and computational efficiency for real-world EHR integration when unique identifiers are absent.