Diagnosis analysis through graph decomposition and association rules in the context of Covid-19
Visualitza/Obre
Estadístiques de LA Referencia / Recolecta
Inclou dades d'ús des de 2022
Cita com:
hdl:2117/351424
Tipus de documentProjecte Final de Màster Oficial
Data2021-06-22
Condicions d'accésAccés obert
Tots els drets reservats. Aquesta obra està protegida pels drets de propietat intel·lectual i
industrial corresponents. Sense perjudici de les exempcions legals existents, queda prohibida la seva
reproducció, distribució, comunicació pública o transformació sense l'autorització del titular dels drets
Abstract
Rule miners are unsupervised learning methods used to detect associations between items. These algorithms have been traditionally used in transactional datasets to synthesise significance associations between items. Extrapolating this behaviour to EHR data, the algorithms should be able to detect associations between diagnoses in a certain segment of the population, therefore suggesting relations of conditions prone to interest by the medical community. This thesis provides an evaluation of a proposal of a rule mining algorithm to detect associations of diagnostses in medical trajectory databases of patients. The approach uses the notion of redundancy to solve the main issues of output size and validity traditionally suffered by rule miners by finding only the non-redundant significant associations. The yacaree program is able to use this approach reducing at the minimum level the needing of expertise by the end user. This thesis evaluates the validity of this technique in a high demanding medical dataset with respect to other rule miner approaches. The procedure aims to state an initial proposal for mining EHR databases to detect between and within associations of diagnoses in segments of patients based on confounding factors age ans sex, with promising results. By imposing high-demanding thresholds the procedure is able to retrieve associations of diagnoses that although being evident suggest correctness of the approach. By softening the thresholds, one should be able to detect non-obvious associations prone to research. The method is tested in a database of visits during the covid-19 outbreak period to bring to light possible associations with the pandemic. Using network visualizations, the ultimatre goal is making a primal formulation of a tool that can be easily interpreted by the medical community. Two final research proposals are adressed. First the suggestion of a basic algorithm to detect morbidity groups of diagnoses. Second, the detection of directionality between diagnoses in rules to improve the visualization and suggest temporality, which turns to be very interesting from the medical perspective.
TitulacióMÀSTER UNIVERSITARI EN INNOVACIÓ I RECERCA EN INFORMÀTICA (Pla 2012)
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
160965.pdf | 1,849Mb | Visualitza/Obre |