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

Area of Science:

• Hematology
• Medical Informatics
• Predictive Analytics

Background:

• Accurate risk assessment for bone marrow failure syndrome (BMFS) is vital for timely diagnosis and treatment.
• Interpreting complete blood count (CBC) data for BMFS requires specialized hematological expertise, posing a challenge for primary care physicians.
• Existing diagnostic pathways may delay critical interventions due to the complexity of initial CBC interpretation.

Purpose of the Study:

• To develop and validate a predictive model for identifying BMFS using readily available demographic and CBC data.
• To create a practical tool that assists primary care physicians in recognizing potential BMFS cases early.
• To improve patient triage and referral processes for suspected bone marrow failure syndrome.

Main Methods:

• Retrospective collection of demographic and CBC data from two major South Korean hospitals.
• Development of binary classifiers for aplastic anemia and myelodysplastic syndrome.
• Integration of classifiers into a combined BMFS predictive model, validated across different CBC feature sets and externally.

Main Results:

• The developed BMFS predictive model demonstrated high performance in distinguishing cases.
• Consistent accuracy was observed across various CBC data feature sets.
• External validation confirmed the model's robust and reliable performance in identifying BMFS.

Conclusions:

• The algorithm offers a practical, data-driven approach for primary physicians to assess BMFS risk from initial CBC results.
• This tool can significantly aid in the early identification, effective triage, and timely referral of patients with potential BMFS.
• Implementation of this model has the potential to improve patient outcomes through earlier intervention in bone marrow failure syndrome.