Machine Learning-Based Prediction of β-Thalassemia Trait Using Red Blood Cell Indices.

OBJECTIVES

Anemia is a significant global health concern, with hypochromic microcytic anemia being the most common type. Among its causes, β-thalassemia trait (β-TT) and iron deficiency anemia (IDA) share similar hematological features, making differentiation challenging. We aimed to develop machine learning (ML) models using routine red blood cell (RBC) indices to distinguish β-TT, IDA, and healthy cases.

METHODS

A total of 8106 individuals (3378 healthy, 2696 IDA, 2032 β-TT) were included in this study. Six RBC parameters-RBC count, hemoglobin (HGB), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), MCH concentration (MCHC), and RBC distribution width (RDW-CV)-were used to train eXtreme gradient boosting (XGB), random forest (RF), and neural network (NN) models. The dataset was split into training (70%) and testing (30%) sets, with feature importance assessed via the Boruta algorithm. All statistical analyses were performed using R version 4.3.1 Statistical Language.

RESULTS

All models achieved high accuracy (> 97%), with RF demonstrating superior performance (97.86% accuracy, 99.71% AUC). The most significant features contributing to the models are MCH for the XGB algorithm, MCV for the RF algorithm, and HGB for the NN algorithm.

CONCLUSIONS

Our findings demonstrate that ML-based models could offer a promising tool for improving β-TT detection, optimizing clinical workflows, and enhancing resource utilization in hematology.