Machine Learning Has High Accuracy, Little Prospective Validation in MPNs

AJMC's report covers a review titled "Artificial intelligence in myeloproliferative neoplasms: from digital bone marrow pathology to clinical decision support,"[1] which was published online in Blood Reviews on October 23, 2026[1].According to the review's abstract on ScienceDirect, the authors searched PubMed/MEDLINE through August 2026, supplemented by hand-searching and conference abstracts[2]. The abstract says AI has made progress in areas including automated bone marrow biopsy analysis, spleen volume segmentation, therapy response prediction and in silico drug discovery[2].The authors also write that "most published models to date remain confined to internal or single-cohort validation"[2]. They add that "prospective evaluation has been reported only for blood count–based fibrosis prediction."[2]AJMC reports that the tool in question, FIBOM-AI, uses 27 complete blood count variables plus age to flag grade 2 to 3 reticulin fibrosis[1]. The outlet says it scored an AUC of 0.92 in the external cohort and an accuracy of 85.2% when applied prospectively in routine practice[1].PubMed's record of the FIBOM-AI study in The Lancet Haematology says the prospective cohort included 493 patients from five centres[3] in 2024. According to AJMC, the developers see it as a triage tool rather than a substitute for biopsy[1].The review's authors concluded, as quoted by AJMC, that progress will require "prospective, outcome-oriented validation studies."[1]