Machine Learning-Based Detection of HbS and HbC Carriers in the UK General Population.

Background: Haemoglobin S (HbS) and C (HbC) are the most important sickling variants on the African continent, imposing major health burdens. Early detection of carrier status is crucial but often hindered by resource limitations. Objectives: To develop machine learning (ML) models to accurately classify HbS and HbC carriers using readily available routine blood tests, facilitating cost-effective mass screening. Methods: We utilised demographic and routine blood parameters from 469,248 individuals from the UK general population, including 1635 individuals with HbS and/or HbC variants identified by whole exome sequencing, to develop ML models for carrier detection based on standard blood tests. Three ML models (Logistic Regression [LR], Random Forest [RF] and XGBoost [XGB]) were trained using 32 different standard blood test results. Results: All models demonstrated high discriminatory ability (ROC-AUC: LR 0.951; RF 0.943; XGB 0.956) in the UK general population. At a sensitivity of 95%, specificities were 77% (LR), 76% (RF) and 78% (XGB). SHAP analysis revealed consistent key features across models. When use was restricted to black individuals, performance fell considerably. Conclusions: ML models based on routine blood tests effectively identify HbS and HbC carriers in a mixed general population. This approach has the potential to enhance screening efficiency by reducing reliance on specialised techniques.