Prediction of bone marrow fibrosis from complete blood count in myeloproliferative neoplasms (FIBOM-AI): a multicentre machine learning model development and validation study
A blood test algorithm may spare many patients with blood cancers from needing bone marrow biopsies by accurately predicting fibrosis without invasive procedures.
FIBOM-AI uses 27 routine CBC parameters plus patient age to predict high-grade bone marrow fibrosis in MPN patients across 18 European and Canadian centres, achieving near-clinical accuracy without invasive biopsy. The confident prediction mode achieved 96.9% accuracy in retrospective validation and 98.6% in prospective real-world use, positioning it as a practical triage and biopsy-deferral tool.
What the study was
- Study design
- Multicentre retrospective ML model development + prospective validation
- Population
- MPN patients undergoing bone marrow biopsy at 13 French + 1 Canadian centres
- Sample size
- 2488
- Category
- Diagnostics
- Maturity
- Validated
- Journal
- Lancet Haematol
Why it surfaced
Lancet Haematol publication; multicentre ML model (n=2488 retrospective, n=493 prospective) predicts bone marrow fibrosis in MPN from CBC alone with AUC 0.92, prospective rule-out sensitivity 98.6% — directly implementable as biopsy triage tool. Validated across 18 centres including Canadian external site.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.