Personalized Medicine based on patient journals and family medical history records
Author
Solo, Stavris
Term
4. term
Publication year
2014
Submitted on
2014-01-10
Pages
89
Abstract
Dette speciale udvikler en praktisk procedure til at analysere store mængder data fra elektroniske patientjournaler for at støtte læger i diagnoseprocessen og for at pege på patienter med forhøjet risiko for at udvikle en bestemt sygdom. Tilgangen kombinerer datarensning, analyse og visualisering. Vi undersøger relationer inden for og på tværs af familier samt ligheder i diagnosticerede sygdomme for at anslå risiko. Vi bruger collaborative filtering—en teknik kendt fra anbefalingssystemer—til at beregne en lighedsscore mellem patienter på baggrund af diagnoser og familiebånd. Patienterne grupperes derefter efter lighed. For at analysere disse grupper anvendes klyngeanalyse baseret på Latent Semantic Indexing (LSI), der kan fremhæve underliggende mønstre i data. Visualisering af klynger præsenterer resultaterne på en måde, som læger kan udforske for at forstå, hvorfor en patient vurderes som højrisiko. Arbejdet resulterede i en systemprototype, der implementerer proceduren. Løsningen blev udviklet på baggrund af et interview med en læge og en gennemgang af relateret litteratur og internetkilder. Målet er at gøre komplekse data mere overskuelige, så læger kan træffe mere informerede beslutninger under diagnosticeringen.
This thesis develops a practical procedure for analyzing large volumes of electronic health record data to support physicians during diagnosis and to flag patients at high risk of developing a specific disease. The approach combines data cleaning, analysis, and visualization. It examines relationships within and across families, as well as similarities in diagnosed conditions, to estimate risk. We use collaborative filtering—a technique widely used in recommendation systems—to compute a similarity score between patients based on their diagnoses and family links. Patients are then grouped by similarity. To analyze these groups, we apply clustering based on Latent Semantic Indexing (LSI), which helps reveal underlying patterns in the data. Cluster visualizations present the results in a way doctors can explore to understand why a patient may be considered high risk. The work resulted in a system prototype that implements this procedure. The solution was informed by an interview with a physician and a review of related literature and online sources. The aim is to make complex data more understandable so that doctors can make more informed diagnostic decisions.
[This abstract was generated with the help of AI]
Keywords
Documents