Machine-learning-based predictive classifier for bone marrow failure syndrome using complete blood count data

Accurate risk assessment of bone marrow failure syndrome (BMFS) is crucial for early diagnosis and intervention. Interpreting complete blood count (CBC) data is challenging without hematological expertise. To support primary physicians, we developed a predictive model using basic demographics and CBC data collected retrospectively from two major hospitals in South Korea. Binary classifiers for aplastic anemia and myelodysplastic syndrome were created and combined to form a BMFS classifier. The model demonstrated high performance in distinguishing BMFS, with consistent results across different CBC feature sets, confirmed by external validation. This algorithm provides a practical guide for primary physicians to identify BMFS based on initial CBC data, aiding in effective triage, timely referrals, and improved patient care.