A Predictive Algorithm for Discriminating Myeloid Malignancies and Leukemoid Reactions

Area of Science:

• Hematology
• Medical Informatics
• Machine Learning

Background:

• Neutrophil-predominant leukocytosis (WBC >50,000/μL) in adults requires urgent management.
• Distinguishing between chronic myeloid malignancies and leukemoid reactions is clinically challenging.
• Existing diagnostic models lack accuracy for these conditions.

Purpose of the Study:

• To develop and validate a machine learning model for differentiating myeloid malignancies from leukemoid reactions.
• To improve diagnostic accuracy in cases of profound leukocytosis.
• To identify key demographic and laboratory predictors.

Main Methods:

• Retrospective analysis of adult patients with WBC >50,000/μL and >50% neutrophils (2000-2021).
• Extraction of demographic and laboratory data at initial presentation.
• Application of a supervised machine learning approach, including support vector machine algorithms.

Main Results:

• The best support vector machine model achieved 96% sensitivity and 95.9% specificity (AUC=0.982) for myeloid malignancy detection.
• Identified significant predictors for myeloid malignancies.
• Observed a 6-fold increase in 12-month mortality for leukemoid reactions compared to myeloid malignancies.

Conclusions:

• Machine learning models can accurately diagnose profound neutrophil-predominant leukocytosis.
• These models address an unmet need for timely and precise diagnosis.
• Validated predictive models can enhance patient outcomes and guide clinical management.