A machine-learning capillary electrophoresis-based pattern to identify suspected sickle cell trait and α-thalassemia co-inheritance.

Objectives: Sickle cell trait (HbAS) and α-thalassemia are common conditions. Although their hematological laboratory findings have been extensively characterized individually, there is limited evidence on the hematological pattern when they are co-inherited. Early and simple diagnosis of HbAS/α-thalassemia allows for optimization of the clinical approach and provision of appropriate genetic counselling. The objective is to identify, using machine learning algorithms, the most relevant hematological quantities and to establish a useful cut-off value for screening patients with suspected HbAS/α-thalassemia. Methods: A database containing 235 patients carrying HbAS was obtained from the laboratory information system. The following variables were collected: hemoglobin and erythrocytes concentration, mean corpuscular volume (MCV), and the fractions of hemoglobin S (%HbS), A (%HbA), A2 and F. A multivariate clustering analysis was performed using the k-means algorithm to divide the dataset into two subgroups: HbAS and suspected HbAS/α-thalassemia. ROC curve was constructed for each variable to identify patients with suspected HbAS/α-thalassemia. Results: Patients with %HbS <37.6 % and %HbA >59 % were classified as suspected HbAS/α-thalassemia with 100 % sensitivity, 97.7 % specificity and AUC [95 % CI] of 0.989 [0.977-0.997]. Patients with suspected HbAS/α-thalassemia exhibited significantly lower hemoglobin concentrations (p=0.008), MCV (p<0.001) and %HbS (p<0.001). In contrast, significantly higher erythrocyte concentrations and %HbA (p<0.001) were observed, compared to HbAS alone. Conclusions: The proposed discriminant values enable suspected cases of HbAS/α-thalassemia to be identified. This facilitates the prioritization of genetic testing for α-thalassemia to be conducted, the provision of genetic and family counselling, and the risk to be accurately assessed for these patients.