AI Model Could Cut Thousands of Unnecessary Blood Tests in Complex Diagnostics

A study conducted at the hematology laboratory of Ichilov Medical Center has developed a method using artificial intelligence (AI) to reduce the number of blood tests required in complex medical evaluations. The research focused on 7,454 diagnostic procedures for detecting lupus anticoagulant, an autoantibody linked to antiphospholipid syndrome, where nearly 90% of tests returned negative results. Typically, each evaluation involved about seven different coagulation tests.

The researchers aimed to identify early in the testing process which samples were likely negative, thereby avoiding further unnecessary tests. They utilized two routine coagulation assays that measure clotting time under different conditions, including varying phospholipid concentrations. An AI model analyzed the ratio between these two test results and identified a pattern that could predict the likelihood of a negative full diagnostic outcome.

When the AI classified a sample as highly likely to be negative, it was correct approximately 98% of the time. The researchers emphasized that the AI is not intended to replace existing tests or independently determine positive or negative status. Samples without a high-confidence negative classification continue through the full standard diagnostic process.

If validated in other laboratories, this approach could significantly reduce the number of complex coagulation tests performed, shorten result turnaround times, and save laboratory resources. The study was published in the journal Digital Medicine by Nature and was led by Dr. Bentzi Katz, head of the hematology lab at Ichilov, in collaboration with researchers from Ichilov and Tel Aviv University.