Machine learning predictive model to identify metabolic status in Mexican children, using homeostasis model assessment insulin resistance and amylase enzymatic activity

Karen E. Villagrana-Bañuelos, Unidad Académica de Ingeniería Eléctrica, Universidad Autónoma de Zacatecas, Zacatecas, Zacatecas, México
Carlos E. Galván-Tejada, Unidad Académica de Ingeniería Eléctrica, Universidad Autónoma de Zacatecas, Zacatecas, Zacatecas, México
Antonio García-Domínguez, Unidad Académica de Ingeniería Eléctrica, Universidad Autónoma de Zacatecas, Zacatecas, Zacatecas, México
Erika Acosta-Cruz, Departamento de Biotecnología, Universidad Autónoma de Coahuila, Saltillo, Coahuila, Mexico, Instituto Mexicano del Seguro Social, México
Miguel A. Vázquez-Moreno, Departamento de Biotecnología, Universidad Autónoma de Coahuila, Saltillo, Coahuila, Mexico, Instituto Mexicano del Seguro Social, México
Miguel Cruz-López, Hospital de Especialidades “Dr. Bernardo Sepúlveda Gutiérrez”, Unidad de Investigación Médica en Bioquímica, Centro Médico Nacional Siglo XXI, Mexico City. Mexico, Instituto Mexicano del Seguro Social, México

Background: Childhood obesity is a global health problem, as it is a risk factor for developing diseases such as metabolic syndrome and diabetes. At present, identifying these already established diseases is relatively easy for health professionals with the support of laboratory studies. The global trend in health involves acting before the disease is established. Objectives: The objective of this study is to identify whether total amylase activity is useful to predict which patients will develop metabolic syndrome or diabetes. Material and methods: Using a database with 101 Mexican patients, considering the value of the homeostasis model assessment insulin resistance as a diagnostic variable in three groups < 2 normal, between 2 and 5 with metabolic risk and > 5 as diabetes, as well as the value of the amylase enzymatic activity. Random forest (RF) was used as a machine learning method. Results: The RF model obtained the following results: area under the curve 0.7075, specificity 0.7619, sensitivity 0.7142, and accuracy 0.7500. Conclusions: It is concluded that with these variables and RF, it is feasible to have a prediction model that contributes to identifying this type of patients in the prepathogenic period.

Keywords: Diabetes. Homeostasis model assessment insulin resistance. Mexican children. Metabolic status. Machine learning.