Machine Learning Classifier Using Blood Count Parameters and Erythropoietin to Predict JAK2 Mutations in Patients...

Area of Science:

• Hematology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Differentiating polycythemia vera from other causes of erythrocytosis presents a diagnostic challenge.
• While Janus kinase 2 (JAK2) mutations are common in polycythemia vera, widespread testing is impractical due to the condition's rarity.
• Identifying patients most likely to benefit from JAK2 testing is crucial for efficient resource allocation.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) / machine learning (ML) classifier.
• The classifier aims to predict JAK2 mutation status in erythrocytosis patients using blood count parameters and erythropoietin levels.

Main Methods:

• Utilized data from the Veterans Affairs data warehouse for model training and validation.
• Included cases with JAK2 results and elevated hemoglobin levels (≥15 g/dL in females, ≥17 g/dL in males).
• Evaluated 31 models, with Light Gradient Boosted Trees Classifier identified as the highest performing.

Main Results:

• The best-performing ML classifier achieved 100% sensitivity and 92.8% specificity on an out-of-sample dataset.
• The model demonstrated a potential reduction in JAK2 testing by 89%, significantly outperforming rule-based systems.
• Platelet count, relative distribution width, and erythropoietin were key predictors in the model.

Conclusions:

• An AI/ML classifier can serve as an effective decision support tool for JAK2 testing in polycythemic patients.
• This approach enhances diagnostic accuracy and optimizes the utilization of genetic testing resources.
• Machine learning offers a promising avenue for improving the management of erythrocytosis.