Predicting joint damage in non-severe hemophilia A using machine learning


Ana Marco-Rico, M.D., Ph.D.
University General Hospital Dr. Balmis
Alicante, Spain

Joint damage in patients with non-severe hemophilia A (PwnSHA) is often underdiagnosed due to the need for specialized ultrasound evaluation. In this presentation, Ana Marco-Rico reports findings from a multicenter study that developed and tested a machine learning model to identify joint damage in PwnSHA using routine clinical data. The study evaluated 84 patients and applied the HEAD-US scale to assess 502 joints. Among five machine learning models tested, the Random Forest algorithm performed best, achieving 92.0% accuracy. Age, target joint history, thrombin generation capacity and the FVIII-Coag/FVIII-Chrom ratio emerged as key predictors. These results suggest that data-driven models may help identify joint damage efficiently in clinical settings for individuals with non-severe hemophilia A.

Previous Article ASH and ISTH publish revised clinical practice guidelines for pediatric venous thromboembolism
Next Article Mechanistic insights into detrimental VWF activity during stroke