Intelligent Database Systems Lab 國國國國國國國國 National Yunlin University of Science and Technology 1 Improved Propensity Matching for Heart Failure Using Neural Gas and Self-Organizing Maps Leif E. Peterson, Sameer Ather, Vijay Divakaran, Anita Deswal, Biykem Bozkurt, Douglas L. Mann IJCNN, 2009 Presented by Hung-Yi Cai 2010/12/01
18
Embed
Improved Propensity Matching for Heart Failure Using Neural Gas and Self-Organizing Maps
Improved Propensity Matching for Heart Failure Using Neural Gas and Self-Organizing Maps. Leif E. Peterson, Sameer Ather , Vijay Divakaran , Anita Deswal , Biykem Bozkurt , Douglas L. Mann IJCNN, 2009 Presented by Hung-Yi Cai 2010/12/01. Outlines. Motivation Objectives Methodology - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Intelligent Database Systems Lab
國立雲林科技大學National Yunlin University of Science and Technology
1
Improved Propensity Matching for Heart Failure Using Neural Gas and Self-Organizing Maps
Leif E. Peterson, Sameer Ather, Vijay Divakaran, Anita Deswal, Biykem Bozkurt, Douglas L. MannIJCNN, 2009
Motivation· Heart failure (HF) has a poor prognosis and is a
major cause of morbidity and mortality.· Unfortunately, the body of information on risk
information for HF among elderly patients is based to a large extent on older heart transplant patients with more severe conditions and more comorbidities.
· Therefore, data obtained from older cases are less amenable for non-biased studies of HF.
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
4
Objectives· The purpose of this paper is to study the effect of
four methods of propensity matching on the relative hazard of mortality among NYHA class III-IV heart failure patients vs. patients in NYHA class I-II.
I. M.Logistic Regression· In statistics, logistic regression is used for
prediction of the probability of occurrence of an event by fitting data to a logit function logistic curve.
6
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Neural Gas
7
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Self-Organizing Map
8
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Time-to-event survival analysis for
propensity-matched subjects using Kaplan-Meier analysis and Cox proportional hazards regression.
9
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Number of matched data based on logit-based
propensity matching (N=3,332).
10
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· NG was based on M = 50 nodes. SOM was
based on a 7 x 7 square map, and thus M = 49. CKM was based on k = 50.
11
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Linear fit results for regressing Cox PH Martingaler
Residuals on Logits and best-winning nodes for NG and SOM.
12
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Kaplan-Meier plot of all-cause mortality as a function of
having NYHA I-II vs. NYHA III-IV for N = 3,332 subjects propensity matched with logistic regression.
13
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Kaplan-Meier plot of all-cause mortality as a function of having NYHA I-
II vs. NYHA III-IV for N = 3,996 (normalized features) and N = 4,262 (standardized features) subjects propensity matched with neural gas.
14
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Kaplan-Meier plot of all-cause mortality as a function of having NYHA I-
II vs. NYHA III-IV for N = 4,004 (normalized features) subjects and N = 4 ,282 (standardized features) propensity matched with a 7 x 7 (M = 49) self-organizing map.
15
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.Experiments· Kaplan-Meier plot of all-cause mortality as a function of having NYHA I-
II vs. NYHA III-IV for N = 4,248 subjects propensity matched with crisp K-means cluster analysis using k = 50 clusters.
16
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
17
Conclusions· Overall, NG resulted in an increased HR for
mortality and explained considerably more variation in Martingale residuals when compared with logit-based propensity matching.
Intelligent Database Systems Lab
N.Y.U.S.T.
I. M.
18
Comments· Advantages
─ The NG algorithm presents result better than other match method.
· Applications─ The classification of medical treatment