Acquisition of patterns from medical records

In recent years, the volume of information available electronically has increased exponentially, and the field of primary health care has not been an exception. The increasing availability of this electronic data, represents an impact on the potential discovery of patterns to predict the risk of new...

Descripción completa

Detalles Bibliográficos
Autor: Borrell Roig, Oriol
Tipo de recurso: tesis de maestría
Fecha de publicación:2021
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/356873
Acceso en línea:https://hdl.handle.net/2117/356873
Access Level:acceso abierto
Palabra clave:Data mining
Medical records
Mineria de regles temporals
Ciència de dades
Mineria de dades
Registres mèdics
Temporal rule mining
Data Science
Medical Records
Àrees temàtiques de la UPC::Informàtica::Sistemes d'informació
Descripción
Sumario:In recent years, the volume of information available electronically has increased exponentially, and the field of primary health care has not been an exception. The increasing availability of this electronic data, represents an impact on the potential discovery of patterns to predict the risk of new diseases, helping the personalized care and increasing the quality of life. Extracting frequent patterns from medical records represents a huge challenge in Data Mining, knowing that in this context the analysis of the temporality between clinical instances is a must. In the TADIA-MED research project, data containing information on visits of patients at Primary Care Centers (CAP) throughout Catalonia was obtained. All annotations in the textbook that the doctor registers in the health system during visits follow what is called the MEAP structure (Motiu de la consulta, Exploració, Avaluació i Pla d'actuació, in Catalan). The information contained in these MEAPs was classified into Diagnostics, Signs or symptoms, Drugs, or Body parts. This information was represented as a graph and stored in a Neo4J server. In this thesis, a new formulation is presented which defines how to compute the temporal association rules in the explained context. The obtained rules are intended to be diagnostic aid patterns. We also have developed an algorithm that uses our formulation to extract the temporal rules. This algorithm makes it possible to parameterize the desired rules in various aspects with respect to the desired format or temporality. We are also capable of extracting rules at different levels of abstraction. Finally, we have defined a process for evaluating the rules obtained. The designed process will be the evaluation process of the entire TADIA-MED project. In spite of the small volume of available data, the evaluation of the rules obtained has been very promising and will help us to continue improving.