Do we Need to Perform Bone Marrow Examination in all Subjects Suspected of MDS? Evaluation and Validation of Non‐Invasive (Web‐Based) Diagnostic Algorithm

• Published in: European journal of haematology Vol. 114; no. 4; pp. 672 - 678
• Main Authors: Oster, Howard S., Polakow, Ariel M., Gat, Roi, Goldschmidt, Noa, Ben‐Ezra, Jonathan, Mittelman, Moshe
• Format: Journal Article
• Language: English
• Published: England Wiley Subscription Services, Inc 01.04.2025
• Subjects: Aged; Aged, 80 and over; Algorithms; Bone marrow; Bone Marrow - pathology; Bone Marrow Examination - methods; Creatinine; diagnostic model; Discordance; Female; gradient boosted model; Humans; Internet; Male; Middle Aged; Monocytes; Morphometry; Myelodysplastic syndrome; myelodysplastic syndromes; Myelodysplastic Syndromes - blood; Myelodysplastic Syndromes - diagnosis; Myelodysplastic Syndromes - etiology; Peripheral blood; Registries; validation; diagnostic model; gradient boosted model; validation; myelodysplastic syndromes
• ISSN: 0902-4441; 1600-0609; 1600-0609
• DOI: 10.1111/ejh.14379

ABSTRACT Background Bone marrow examination (BME) is the gold standard of diagnosing myelodysplastic syndromes (MDS). Problems: it is invasive, painful, causing possible bleeding, inaccurate (aspirate hemodilution), and subjective (inter‐observer interpretation discordance). We developed non‐invasive diagnostic tools: A logistic regression formula [LeukRes 2018], then a web algorithm using 10 variables (age, gender, Hb, MCV, WBC, ANC, monocytes, PLT, glucose, creatinine) to diagnose/exclude MDS [BldAdv 2021]. Here, we perform external validation of the model. Methods From the TASMC BM registry (2019–22) we identified and compared the model performance between MDS patients and controls (> 50 year with unexplained anemia, not MDS), all BME diagnosed, and not used in model building. Results The model was accurate and predicted MDS in 63% of 103 patients, and excluded (correctly) in 83% of 101 controls. It miss‐classified in 11%/7% respectively, and was indeterminate in 26%/10% respectively. The positive predictive value (PPV), NPV, sensitivity, and specificity (excluding the indeterminate group) were 90%, 88%, 86%, and 92%, respectively. Subgroup (Lower/higher risk, LR/HR) analysis results were similar. Conclusions The MDS diagnostic model was validated and can be used, mainly for MDS exclusion, especially in suspected LR‐MDS, avoiding BME in some patients. In the future incorporating peripheral blood genetics and morphometry can further improve the model.