Interpretable machine learning model using peripheral blood for non-invasive detection of moderate-to-severe myelofibrosis in JAK2 V617F-positive MPNs: A multicentre pilot proof-of-concept study.
A blood test using machine learning identifies aggressive myelofibrosis early, revealing inflammation as a driver that some existing drugs might help reverse.
An interpretable SVM model trained on 336 JAK2 V617F-positive MPN patients and externally validated on 92 achieves AUC 0.916 for detecting moderate-to-severe myelofibrosis using only peripheral blood parameters. The SHAP-identified features — haemoglobin, IL-1β, and age — reveal an inflammatory mechanism of fibrosis progression, with IL-1β upregulation reversible by ruxolitinib or pegylated interferon-α but not hydroxyurea.
What the study was
- Study design
- Multicentre prospective ML model development and external validation study
- Population
- JAK2 V617F-positive MPN patients
- Sample size
- 428
- Category
- Diagnostics
- Maturity
- Validated
- Journal
- British Journal of Haematology
Why it surfaced
Multicentre blood-based ML model achieves AUC 0.916 for non-invasive myelofibrosis staging in MPN, directly addressing the clinical burden of invasive bone marrow biopsy for regular monitoring. IL-1β mechanistic validation enhances translational confidence. Pilot label limits immediate practice change.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.