AI shows early promise for hemophilia diagnosis, care

Artificial intelligence (AI) shows promise for helping doctors diagnose hemophilia, detect joint complications, and predict disease severity and other clinical outcomes, a review of published studies suggested.

“However, methodological [diversity], limited external validation, and the scarcity of studies … remain important barriers to routine clinical implementation,” the researchers wrote. They said “the consolidation of these technologies in clinical practice” will require larger studies, validation in other populations, and evidence on their impact on time to diagnosis, prevention of joint damage, and quality of life.

The study, “Artificial Intelligence and Machine Learning in the Context of Hemophilia: A Scope Review,” was published in Haemophilia.

Hemophilia occurs when the blood cannot clot properly because certain clotting proteins — factor VIII in hemophilia A or factor IX in hemophilia B — are missing or deficient. This leads to hemophilia symptoms such as excessive bleeding. Repeated bleeds inside the joints can cause inflammation, pain, and lasting damage.

Advances in hemophilia treatment have improved care, but predicting complications and tailoring treatment to each person remain challenges. The growing availability of medical records, laboratory results, and imaging data has prompted interest in using AI to address these needs.

Recommended Reading

July 19, 2026 Columns by G Shellye Horowitz

Could AI allow for a faster medical diagnosis of rare conditions?

Machine learning

Machine learning, a branch of AI, trains computer programs to recognize patterns in data and make predictions. In hemophilia, researchers are exploring these tools to assess disease severity, monitor joint problems, predict outcomes, and estimate how clotting factor levels change after treatment to help guide dosing.

“Although initial results are promising, the available evidence remains heterogeneous and scattered across the literature, reflecting the still-emerging stage of these technologies in the field,” the researchers wrote.

The researchers, in Brazil, conducted a scoping review to identify available research on AI and machine learning applications in hemophilia A or B. Searching five databases for studies published between January 2014 and August 2024, they found four studies that met the inclusion criteria: two from the U.S., one from Japan, and one from India.

One U.S. study developed a model to identify people with hemophilia A in healthcare databases. Researchers screened insurance records to identify 2,252 potential cases, then reviewed medical records for 400 to confirm who had the disease. These findings helped them develop and evaluate the model, which correctly identified about 94% of confirmed cases. Nearly 95% of those it flagged had hemophilia A.

A study from Japan used 3,435 ultrasound images of elbows, knees, and ankles from people with hemophilia A or B to train and evaluate an AI tool for detecting synovitis, inflammation of the joint lining, and hemarthrosis, or joint bleeding. Compared with expert assessments, the tool performed well at distinguishing images with and without these problems, though performance varied depending on the joint examined.

In India, researchers explored whether machine learning could predict hemophilia A severity from mutations in the F8 gene, which cause factor VIII to be missing or deficient. They used 7,784 entries in a public database linking mutations to disease severity and compared different ways of presenting this information to the models. Including each mutation’s location reduced the time needed to train the models and generate predictions.

The other U.S. study tested whether AI could identify people with hemophilia who had poor clinical outcomes. Using data from 400 people, researchers trained and evaluated several models to predict outcomes including death, bleeding inside the skull, development of inhibitors — antibodies that can make factor replacement therapies less effective — and target joints, or joints repeatedly affected by bleeding.

When these outcomes were considered together, none of the models identified more than 53% of affected people. The researchers said the small dataset and limited clinical information may help explain this performance.

Together, the studies illustrate several possible uses of AI in hemophilia, though differences in their aims, data, algorithms, and training and testing strategies prevented direct comparisons. Most focused on statistical performance, with limited evidence that the tools improved clinical decisions or patient outcomes. Testing in populations beyond those used to develop the models was also limited.

“Overall, the findings indicate that AI and [machine learning] represent promising tools for precision medicine in hemophilia; however, their effective translation into clinical practice will depend on the development of models that are robust, interpretable, and externally validated, integrated into care workflows, and capable of supporting—rather than replacing—clinical judgment,” the researchers wrote.