Aprendizaje automático para predecir el diagnóstico temprano de personas con hipertensión arterial: Revisión sistemática

Abstract

Machine learning is widely used in the medical field and is increasing more and more because of the amount of data stored. The results obtained by the predictive models serve as support for good decision-making for medical personnel. The objective was to identify which methods, variables, and models are used for the prediction of arterial hypertension using machine learning. The systematic review was carried out in the PubMed, ScienceDirect, Redalyc and Scopus search engines, studies referring to the prediction of early diagnosis of arterial hypertension in people. For the selection process, Prisma was used, applying different exclusion criteria. 10,916 articles were found, 15 being included for the review. Several authors apply more than one model to compare the results in their research. The model most mentioned, used and with the best result was Random Forest, obtaining a Specificity (0.96), Precision (0.92) and AUC (0.95). Finally, it was possible to provide the models most mentioned and used in the investigations, as well as to identify the models with a high predictive performance. There are few studies that combine demographic, clinical, and pathological data to implement models to predict early diagnosis of people with arterial hypertension.