Istraživanje čimbenika rizika za nastanak alergijskih dišnih bolesti s pomoću Bayesovih mreža temeljenih na podacima iz više izvora

doctoral thesis

• Title (english): Exploring the risk factors for the development of allergic respiratory diseases by Bayesian networks built on multiple-source data
• Author: Jelena Kovačić
• Mentor: Veda Marija Varnai (mentor)
• Mentor: Anamarija Jazbec (komentor)
• Committee member: nije dostupno (član povjerenstva)
• Granter: University of Zagreb; School of Medicine; Zagreb
• Defense date and country: 2017-04-24, Croatia
• Scientific / art field, discipline and subdiscipline: BIOMEDICINE AND HEALTHCARE; Public Health and Health Care
• Universal decimal classification (UDC): 61 - Medical sciences

Abstract

Pretpostavlja se da bi zajednička analiza više izvora podataka mogla pomoći u
razjašnjenju etiologije nezaraznih, kroničnih bolesti. Međutim, statističke metode
nisu dovoljno razvijene za situacije kad se različita istraživanja samo djelomično
preklapaju prema mjerenim čimbenicima.
U ovom je istraživanju predložena nova metoda za meta-analizu takvih podataka,
G-MA, i primijenjena u izradi Bayesovih mreža više izvora podataka. U računalnim
eksperimentima G-MA je imala manju srednju kvadratnu pogrešku i uže intervale
pouzdanosti u procjeni jačine veza u odnosu na analizu pojedinačnih izvora.
U istraživanju čimbenika povezanih s prisutnošću alergijskih dišnih bolesti Bayesova
mreža s G-MA metodom (Bayes-G-MA) primijenjena je na četiri epidemiološka
izvora hrvatske populacije. U usporedbi s analizama pojedinačnih izvora,
Bayes-G-MA je prepoznala manje izravnih povezanosti s bolesti (alergijske bolesti
u djetinjstvu, alergijske bolesti u roditelja, indeks tjelesne mase), dok su ostale
bile neizravne (spol, tjelesna aktivnost), ili nisu prepoznate (pušenje). Bayes-G-MA
predložila je dvije nove veze (povezanost depresije i konzumacije alkohola s bolesti)
između varijabli koje nisu niti u jednom izvoru ispitivane zajedno. Uočene razlike
mogu biti odraz stvarnih međuodnosa, nedostatne statističke snage ili karakteristike
metode da ne prepoznaje sve prisutne veze.
Zaključno, najveći doprinos metode očekuje se u procjeni jačine veza, uz pomoć
ekspertnog znanja pri procjeni uzročno-posljedičnih veza.

Abstract (english)

It is suggested that multi-source data analysis could help to elucidate the aetiology
of non-communicable, chronic diseases. However, statistical methods for
multi-source data analysis with partially overlapping data sources are not developed
enough.
The study proposes new method for meta-analysis of such data, G–MA, applying
it to Bayesian network of multi-source data. In simulation experiments, G-MA
showed lower mean squared error and narrower confidence intervals in parameter
estimation, when compared to single data source analyses.
Bayesian network with G-MA method (Bayes–G–MA) was applied to analyse the
factors related to the presence of allergic respiratory diseases (ARD) in four Croatian
epidemiological sources. Compared to single source analyses, Bayes–G–MA
recognised lower number of direct relationships with ARD (allergic diseases in childhood,
allergic diseases in parents, body mass index), while others became indirect
(gender, physical activity) or unrecognised (smoking). Bayes–G–MA suggested two
new relationships (with depression and alcohol consumption) between variables not
tested together in any single source. These results could reflect true relationships,
but also lack of statistical power or intrinsic property of the method to under-report
present relationships.
In conclusion, the main contribution of proposed method is in the assessment of parameters, with expert help needed in evaluation of causal relationships.

• Keywords: medicina
• Keywords (english): medicine
• Language: croatian
• URN:NBN: urn:nbn:hr:105:628040
• Study programme: Title: Biomedicine and Health Sciences; Study programme type: university; Study level: postgraduate; Academic / professional title: doktor/doktorica znanosti, područje biomedicine i zdravstva (doktor/doktorica znanosti, područje biomedicine i zdravstva)
• Type of resource: Text
• Extent: 145 str.
• File origin: Born digital
• Access conditions: Open access
• Created on: 2023-05-05 08:03:40