Machine Learning Model Predicts Abnormal Lymphocytosis Associated With Chronic Lymphocytic Leukemia

Area of Science:

• Hematology
• Medical Informatics
• Machine Learning

Background:

• Chronic lymphocytic leukemia (CLL) diagnosis is often delayed, missing opportunities for early intervention.
• A significant care gap exists in identifying patients at risk for CLL before disease progression.
• No widely adopted machine learning (ML) models currently exist to predict CLL development.

Purpose of the Study:

• To develop and validate ML-based risk models for predicting abnormal lymphocytosis associated with CLL.
• To leverage readily available laboratory data for predicting CLL risk.
• To address the diagnostic delay in CLL and identify at-risk individuals.

Main Methods:

• An observational study utilized deidentified laboratory data from 1,090,707 adult patients (age 50-75) with initial absolute lymphocyte count (ALC) <5 × 10^9/L.
• Data spanning 7 years were split into 80% training and 20% testing sets.
• Random forest survival methods were employed to build ML models, with outcomes defined as ALC ≥5 × 10^9/L and ≥40% relative lymphocytosis.

Main Results:

• A 12-variable risk classifier accurately predicted ALC ≥5 × 10^9/L within 5 years (AUC=0.92).
• Key predictors included initial and slope of ALC, White Blood Cell (WBC) counts, platelet counts, age, and sex.
• The model demonstrated strong predictive performance on an independent test set.

Conclusions:

• The developed ML risk classifier accurately predicts abnormal lymphocytosis indicative of CLL using routine laboratory data.
• The findings support the clinical utility of this model for earlier recognition of CLL risk.
• Further prospective studies are recommended to confirm the model's real-world clinical impact.