國立高雄大學圖資館 |

Language: English

Back

Statistical Timeline Analysis for El...

The University of Wisconsin - Madison.

Statistical Timeline Analysis for Electronic Health Records.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Statistical Timeline Analysis for Electronic Health Records.
Author:	Weiss, Jeremy C.
Description:	132 p.
Notes:	Source: Dissertation Abstracts International, Volume: 75-11(E), Section: B.
Notes:	Adviser: C. David Page.
Contained By:	Dissertation Abstracts International75-11B(E).
Subject:	Computer science.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3629081
ISBN:	9781321057430

Statistical Timeline Analysis for Electronic Health Records.
Weiss, Jeremy C.

Statistical Timeline Analysis for Electronic Health Records. - 132 p.

Source: Dissertation Abstracts International, Volume: 75-11(E), Section: B.

Thesis (Ph.D.)--The University of Wisconsin - Madison, 2014.

This item must not be sold to any third party vendors.

Electronic Health Records (EHRs) now hold over 50 years of recorded patient information and, with increased adoption and high levels of population coverage, are becoming foci of public health analyses. The structure of EHR patient data limits existing clinical study paradigms, which fail to effectively capture the relational, temporal, and intermittent data characteristics. This dissertation develops statistical timeline analysis (STA), a set of algorithms that extend existing modeling approaches to address EHR data challenges.

ISBN: 9781321057430Subjects--Topical Terms:

199325
Computer science.

Statistical Timeline Analysis for Electronic Health Records.
LDR:03611nmm a2200361 4500 001 457679
005 20150805065219.5
008 150916s2014 ||||||||||||||||| ||eng d
020 $a 9781321057430
035 $a (MiAaPQ)AAI3629081
035 $a AAI3629081
040 $a MiAaPQ $c MiAaPQ
100 1 $a Weiss, Jeremy C. $3 708691
245 1 0 $a Statistical Timeline Analysis for Electronic Health Records.
300 $a 132 p.
500 $a Source: Dissertation Abstracts International, Volume: 75-11(E), Section: B.
500 $a Adviser: C. David Page.
502 $a Thesis (Ph.D.)--The University of Wisconsin - Madison, 2014.
506 $a This item must not be sold to any third party vendors.
520 $a Electronic Health Records (EHRs) now hold over 50 years of recorded patient information and, with increased adoption and high levels of population coverage, are becoming foci of public health analyses. The structure of EHR patient data limits existing clinical study paradigms, which fail to effectively capture the relational, temporal, and intermittent data characteristics. This dissertation develops statistical timeline analysis (STA), a set of algorithms that extend existing modeling approaches to address EHR data challenges.
520 $a Statistical timeline analysis models EHR data as patient-specific, relational timelines, where measurements and events occur in continuous-time instead of at fixed intervals. First, we adopt a relational forest algorithm and show improved performance at heart attack prediction compared to analogous non-relational algorithms. Then we turn to richer timeline models: continuous-time Bayesian networks (CTBNs), which model dependencies in rate among discrete variables over continuous time. We introduce partition-based CTBNs, a generalization that alleviates the exponential space constraints of CTBNs yet maintains the ability to model complex dependencies. We then develop a multiplicative forest learning algorithm with space linear in the number of forest splits that efficiently maximizes the partition-based CTBN likelihood.
520 $a To address CTBN inference challenges, we identify a general method for the improvement of sequential importance samples. Our method reduces sample weight variance by an order of magnitude, yielding a better approximation of the posterior distribution.
520 $a We also study point processes, which avoid CTBN inference challenges altogether. We show that the multiplicative forest learning algorithm applies and improves upon existing learning algorithms both in modeling dependences and as extracted features for forecasting heart attacks.
520 $a Finally we turn to attributable risk. The clinical study paradigm focuses on population-average changes in risk. However, the average outcomes of such studies are then applied to individuals when the application of the individual outcome is more appropriate. We show that individualized-risk modeling improves average individual outcomes and provides evidence of the EHR as an effective source for modeling individualized attributable risks.
520 $a Our contributions to statistical timeline analysis show algorithmic and performance improvements that address EHR data challenges. We expect further research combining these ideas to improve clinical understanding and patient care.
590 $a School code: 0262.
650 4 $a Computer science. $3 199325
650 4 $a Medicine. $3 193819
650 4 $a Statistics. $3 182057
690 $a 0984
690 $a 0564
690 $a 0463
710 2 $a The University of Wisconsin - Madison. $b Computer Sciences. $3 708692
773 0 $t Dissertation Abstracts International $g 75-11B(E).
790 $a 0262
791 $a Ph.D.
792 $a 2014
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3629081