Meta-Learning on Clinical Data for Diagnosis Support Systems: A Systematic Review

In recent years, the volume of data on people's health has increased, giving rise to numerous investigations related to health, such as for the prediction of diseases. These types of predictions help clinicians identify potential risk at an early stage and thereby provide better patient care. This represents significant practical challenges for the application of meta-learning for diagnosis. Meta-learning is the process of learning to learn where algorithms learn about other algorithms using experience. Recently, new researches in meta-learning uses deep neural networks. This study aims to review the literature on the use of meta-learning in the field of health for the detection, classification, or prediction of diseases in patients. A systematic review of the papers published according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guide of 2020 was carried out. The state of the art is very limited considering applied meta-learning in health. The manuscripts were identified by means of a search string in the Scopus database. All manuscripts that used meta-learning for the prediction, classification and detection of diseases were included. The search obtained a total of 31 papers. From this initial set, 13 publications were those that met the eligibility criteria and were included in the review. These studies are focused on the prediction of clinical risk from biomedical data, such as: disease prediction; epidemic prediction; adverse drug reaction prediction; and of patients under medical supervision.