izpis_h1_title_alt

Strategije za uravnoteženo izbiro retrospektivnih podatkov za simulacijo prospektivnih raziskav
Smodiš, Aleš (Author), Bratko, Ivan (Mentor) More about this mentor... This link opens in a new window

URLURL - Presentation file, Visit http://eprints.fri.uni-lj.si/3229/ This link opens in a new window

Abstract
Porast raziskav v medicini poraja vedno več ugotovitev, ki imajo lahko za posledico nova ali izboljšana obstoječa zdravljenja. Obenem pa potrebe po novih raziskavah pripeljejo do težav pri zagotavljanju zadostnega števila bolnikov za prospektivne raziskave vseh obetajočih zdravljenj. Po drugi strani lahko z restrospektivno raziskavo na obstoječih podatkih bolnikov do določene mere simuliramo prospektivno raziskavo. Glavna težava pri tem pristopu je, da imajo obstoječi podatki običajno neuravnotežene porazdelitve karakteristik po množicah bolnikov, na katerih izvajamo retrospektivno raziskavo, kar oteži vrednotenje učinkov zdravljenja. Predstavljen je algoritem za uravnoteževanje množic bolnikov z danimi karakteristikami, ki z uparjanjem in z izločanjem izbranih bolnikov ustvari uravnotežene podmnožice bolnikov. Algoritem uporablja Pearsonov test hi kvadrat za merjenje kvalitete medsebojne uravnoteženosti množic in vsoto uteženih razlik vrednosti karakteristik za določanje parov elementov med dvema množicama. Uvedeni sta dve novi strategiji uparjanja elementov: s požrešno metodo preko matrike podobnosti parov, ter z algoritmom minimin na drevesu stanj do predpisane globine za izbiro naslednjih dveh elementov za uparjanje. Uvedena je mera kvalitete uparjenosti med dvema množicama. Rezultati kažejo, da požrešna metoda daje boljše rezultate od izvirnega algoritma, medtem ko se algoritem minimin izkaže za časovno zahtevnega zaradi kombinatorične zahtevnosti in pri globinah, ki so glede tega še praktične za izvajanje algoritma, daje kvečjemu primerljive rezultate izvirnemu algoritmu, vendar slabše od požrešne metode. Metode so bile eksperimentalno primerjane na realnih podatkih iz medicinskih raziskav zdravljenja raka.

Language:Unknown
Keywords:retrospektivne raziskave, simulacija prospektivnih raziskav, uparjanje, uravnoteževanje množic, hevristično preiskovanje, hevristična ocena kvalitete uravnoteženosti, Pearsonov test hi kvadrat
Work type:Not categorized (r6)
Organization:FRI - Faculty of computer and information science
Year:2016
COBISS.SI-ID:1536769731 Link is opened in a new window
Views:476
Downloads:105
Metadata:XML RDF-CHPDL DC-XML DC-RDF
 
Average score:(0 votes)
Your score:Voting is allowed only to logged in users.
:
Share:AddThis
AddThis uses cookies that require your consent. Edit consent...

Secondary language

Language:Unknown
Title:Strategies for Balanced Selection from Retrospective Data for Simulation of Prospective Studies
Abstract:
The increase of medical research generates more and more findings which can result in new or enhanced existing treatments. This increase of medical research leads to problems at ensuring a sufficent number of patients for prospective studies of all of the promising treatments. On the other hand a prospective study can be simulated to a certain degree with a retrospective study using existing data. The main problem with this approach is, that the existing data usually have unbalanced distributions of characteristics over the sets of patients, which makes it difficult to evaluate effects of treatment. An algorithm is described for balancing sets of patients with given characteristics, which creates balanced subsets of patients using pairing and elimination of selected patients. The algorithm uses Pearson's chi-squared test for measuring the balance quality between two sets, and the sum of weighed differences between the characteristics for defining element pairs between sets. Two new element pairing strategies are introduced: a greedy method using an element similarity matrix, and the minimin algorithm using a state tree with limited depth for choosing the next elements to pair. A measure for the quality of a match between two sets is introduced. Results show that the greedy method gives better results from the original algorithm, whereas the minimin algorithm turns out to be time demanding because of the combinatorial complexity. At depths at which the algorithm is still practical to use, it gives results at best comparable to the original algorithm, but worse than the greedy method. The methods were experimentally compared on real data from medical studies in cancer treatment.

Keywords:retrospective studies, simulated prospective studies, pairing, data set balancing, heuristic search, heuristic evaluation of balance quality, Pearson's chi-squared test

Similar documents

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

Comments

Leave comment

You have to log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back