Optimizing Laboratory Autoverification in Complete Blood Count Testing Using Machine Learning: A Performance...

Area of Science:

• Hematology
• Clinical Pathology
• Artificial Intelligence in Medicine

Background:

• Traditional rule-based autoverification systems for complete blood count (CBC) testing lack flexibility.
• Machine learning (ML) offers potential for enhanced pattern recognition in laboratory data.
• Evaluating ML-based autoverification systems is crucial for improving CBC testing efficiency.

Purpose of the Study:

• To compare the performance of ML-based autoverification systems against traditional rule-based methods for CBC results.
• To assess the ability of ML models to classify CBC results with high accuracy.
• To identify the best-performing ML model for autoverification in a clinical hematology setting.

Main Methods:

• A cross-sectional study analyzed 63,201 CBC results from a Sysmex XN-10 analyzer.
• Data were split into training (80%) and testing (20%) sets for Random Forest, Decision Tree, and Extreme Gradient Boost models.
• ML models were evaluated against the hospital's rule-based system using sensitivity, specificity, and predictive values.

Main Results:

• 30.9% of CBC results were non-verifiable due to WBC abnormalities, platelet clumps, or differential discrepancies.
• All ML models outperformed the rule-based system in the test set of 12,641 samples.
• XGBoost demonstrated optimal performance with 95%-98% sensitivity and 68%-81% specificity.

Conclusions:

• Machine learning models show superior performance for CBC autoverification compared to rule-based systems.
• ML can potentially reduce the need for manual reviews in laboratory testing.
• Further validation in diverse clinical settings is necessary before widespread adoption of ML for autoverification.