Multiclass classification of metabolic conditions using fasting plasma levels of glucose and insulin

In clinical practice, plasma glucose concentrations in fasting and postprandial are measured to assess glucose metabolism and to diagnose diabetes. Plasma glucose and insulin concentrations in fasting and postprandial have been used to better characterize normal, prediabetic and diabetic conditions. In this paper, we seek to automatically recognize nine classes of metabolic conditions (three normal, three prediabetics, and three diabetics) by considering the age the patient and its fasting plasma glucose (FPG) and insulin (FPI) concentrations. Multinomial logistic regression (MLR), artificial neural network (ANN), support vector machine (SVM), decision tree (DT) and random forests (RF) were set for different attribute combinations (age, FPG and FPI). Accuracy, and macro-average and weighted-average measures of precision, recall and F1-score were employed to assess the performance of the classifiers. Accuracy and weighted-average of precision, recall and F1-score above 79% were obtained using an ANN and an RF with age, FPG and FPI as attributes. In terms of the weighted-average of F1, an ANN with FPG and FPI as attributes was the best classifier (weighted-average F1 = 81.50%). Age, FPG and FPI provided information to recognize the nine metabolic classes. Moreover, age helped to distinguish between two diabetic classes with overlapping glucose and insulin levels. Given the morbidity and mortality rate of metabolic diseases (Latin America counts for 26 million diabetic people and 10 million undiagnosed), the significance of this work lies in the conception of an automatic classifier for diagnosis support or preliminary screening in places with limited or non-existent health service delivery systems.