izpis_h1_title_alt

Statistične metode za analizo intenzivnih longitudinalnih podatkov v medicini
ID Ahlin, Črt (Author), ID Lusa, Lara (Mentor) More about this mentor... This link opens in a new window

.pdfPDF - Presentation file, Download (6,05 MB)
MD5: 855B6AE9585AAE77A1F32E157E933D04

Abstract
Naloga obravnava metode za modeliranje intenzivnih longitudinalnih podatkov. Nosljive naprave s senzorji (na primer pametne ure) omogočajo olajšan in stalen zajem fizioloških podatkov. Longitudinalni podatki imajo posebnosti, ki izhajajo iz upoštevanja časa v analizah -- posamezniki so merjeni večkrat. Intenzivni longitudinalni podatki imajo zelo pogoste meritve in omogočajo modeliranje procesov na ravni posameznika. S tem se odpirajo nove možnosti, tudi za tako imenovano na posameznika osredotočeno medicino. V prvem delu naloge je predstavljeno interaktivno orodje za delo z longitudinalnimi podatki, {\tt medplot}, ki smo ga razvili, da bi omogočili ne-statistikom dostop do naprednih metod za analizo longitudinalnih podatkov. Z aplikacijo in interpretacijo rezultatov smo biomedicinskim uporabnikom ponudili orodje za napredne analize longitudinalnih podatkov in tako prispevali k razvoju področja uporabe metod v praksi. V drugem delu naloge je obravnavan problem periodičnih longitudinalnih podatkov. Gre za podatke, katerih vrednost se periodično ponavlja, na primer vsako leto. Predstavljena je nadgradnja uveljavljene metode modeliranja s kubičnimi zlepki z dodanimi omejitvami (RCS), da je upoštevana periodičnost podatkov. Z izpeljavo periodične RCS ter njeno implementacijo v paketu {\tt peRiodiCS} smo prispevali k področju modeliranja periodičnih podatkov, saj smo ponudili dodatno orodje, ki ima v nekaterih primerih boljše lastnosti kot obstoječa, in sicer zaradi manjšega števila ocenjenih parametrov. V tretjem delu naloge so obravnavani intenzivno longitudinalni podatki srčnega utripa z namenom, da bi napovedali dogodke atrijske fibrilacije. Uporabljena je metoda vreče vzorcev za prikaz podatkov in v postopku navzkrižnega preverjanja so bile preverjene napovedne lastnosti ob različnih vrednostih parametrov napovedovanja. Pomanjkljivost podatkov je bil razmeroma majhen vzorec. Menimo, da so ugotovitve glede parametrov metode vreče vzorcev in intervalov opazovanja lahko izhodišče oziroma usmeritev ob nadaljnjih raziskavah, ko bodo na voljo ustrezni večji podatkovni nizi frekvenc srčnega utripa na ravni posameznika.

Language:Slovenian
Keywords:intenzivno longitudinalni podatki, periodičnost, vreča vzorcev
Work type:Doctoral dissertation
Organization:MF - Faculty of Medicine
Year:2019
PID:20.500.12556/RUL-107397 This link opens in a new window
COBISS.SI-ID:34270937 This link opens in a new window
Publication date in RUL:11.04.2019
Views:1833
Downloads:248
Metadata:XML DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:English
Title:Statistical methods for analysis of intensive longitudinal data in medicine
Abstract:
The work presents methods for modeling intensive longitudinal data. Wearable devices with sensors (e.g. smart watches) are making constant gathering of physiological data easier. Longitudinal data have their special characteristics, that stem from taking time into account in analyses - individuals are measured at multiple occasions. Intensive longitudinal data have very frequent measurements and allow modeling of processes at the level of the individual. This is opening new opportunities in the field of personalized medicine. In the first part of the text, an interactive tool for working with longitudinal data is presented. {\tt medplot} was developed to enable non-statisticians access to some advanced methods for analyzing longitudinal data. With the application and the explanations of interpreting the results we have offered biomedical users a tool for more advanced analysis of longitudinal data and have in this way contributed to the advancement of the field by easier adoption of the methods in practice. In the second part, the problem of periodic longitudinal data is tackled - this is data which values repeat with some period of time, e.g. yearly. An established method of modeling with restricted cubic splines (RCS) is upgraded, to take into account the periodicity of the data. By deriving periodic RCS and implementing them in the {\tt peRiodiCS} package, we have contributed to the field of modeling periodic data, as we have offered another tool, which has in some cases better characteristics as the existing ones, due to the smaller number of estimated parameters. In the third part, heart rate intensive longitudinal data is looked at with the purpose of predicting an upcoming event of atrial fibrillation. Bag of patterns representation of data is used with cross validation procedure to check the predictive properties at different parameter values. Data had the shortcomings of a relatively small sample. We believe the findings might represent a starting point or some guidance at additional research, when larger data sets of heart rate for individuals become available.

Keywords:intensive longitudinal data, periodicity, bag of patterns

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Back