AI Model Could Cut Unnecessary Lupus Anticoagulant Blood Tests by Predicting Negatives Early

A study conducted at Tel Aviv Sourasky Medical Center (Ichilov) analyzed 7,454 lupus anticoagulant (LA) test panels and found that nearly 90% ended with negative results after an average of seven coagulation tests per patient. Lupus anticoagulant is a key marker in diagnosing antiphospholipid syndrome (APS), an autoimmune disorder where antibodies disrupt blood clotting, increasing risks of thrombosis, stroke, and pregnancy complications. The current diagnostic process is complex, time-consuming, and costly, requiring multiple coagulation assays and sometimes repeated testing.

Researchers led by Dr. Bentzi Katz and Dr. Chen Harmesh sought to reduce unnecessary testing by leveraging existing laboratory data rather than introducing new tests. They focused on two routine coagulation assays that measure clotting time under different phospholipid concentrations. Using artificial intelligence, they developed a model that identifies a "fingerprint" in the ratio of these two test results, which correlates strongly with the presence or absence of pathological antibodies.

The AI model was able to predict negative LA results with approximately 98% accuracy, allowing early exclusion of samples unlikely to be positive. Importantly, samples not confidently classified as negative continued through the full testing panel, ensuring no positive cases were missed. This approach aims to streamline laboratory workflows, reduce costs, shorten wait times, and allocate resources more efficiently without compromising diagnostic accuracy.

The study, published in the journal Digital Medicine by Nature, highlights a novel application of AI in medicine by extracting new value from routine test data. The next steps include validating the model in other laboratories and testing systems before potential integration into standard clinical practice. The research team included experts from Ichilov and Tel Aviv University, with support from Prof. Irit Avivi and Hilik Avivi.