Asafu-Adjei, Josephine
(2012)
Covariate Adjusted Discrimination with Applications to Neuroscience.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
In post-mortem tissue studies that compare regional brain biomarkers across different mental disorder diagnostic groups, subjects are often matched on several demographic characteristics and measured on additional covariates. The goal of our research is to integrate the results from these types of studies using two commonly used statistical discrimination techniques, namely, linear discriminant analysis (LDA) and classification trees based on the algorithm developed by Breiman, Friedman, Olshen, and Stone (BFOS), to identify the most discriminatory subset of biomarkers. Subject matching and covariate effects don't appear in the literature implementing these discriminatory methods in the analysis of post-mortem tissue studies (e.g., Knable et al. 2001; Knable et al. 2002).
Although there are methods that have been developed for LDA to account for covariate effects on the response or feature variables of interest, none of these methods addresses the fact that individuals may also be matched across several groups. One aspect of our research extends this work to handle group matching.
To develop the theoretical foundations required to account for covariate effects in classification trees, we describe how to implement the BFOS algorithm, which is non-parametric and traditionally implemented in a data based setting, when the feature variables come from a known distribution. We then extend this algorithm to the case where the feature variables come from a known distribution, conditional on a covariate value. From this development, we carefully formulate a semi-parametric model for the conditional distribution of the feature variables that allows the use of the BFOS algorithm to construct a covariate adjusted tree based on one unique set of feature variables, in both a theoretical setting and in the context of training data. Finally, the tree construction procedure we develop using this conditional model is extended to handle group matching.
Our adjustment methodology is successfully applied to a series of post-mortem tissue studies conducted by Sweet et al. (2003, 2004, 2007, 2008) comparing several neurobiological characteristics of schizophrenia subjects and normal controls, and to a post-mortem tissue study conducted by Konopaske et al. (2008) comparing brain biomarker measurements of monkeys across three treatment groups.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
| Creators | Email | Pitt Username | ORCID  |
|---|
| Asafu-Adjei, Josephine | jka7@pitt.edu | JKA7 | |
|
| ETD Committee: |
|
| Date: |
31 January 2012 |
| Date Type: |
Publication |
| Defense Date: |
29 August 2011 |
| Approval Date: |
31 January 2012 |
| Submission Date: |
1 December 2011 |
| Access Restriction: |
2 year -- Restrict access to University of Pittsburgh for a period of 2 years. |
| Number of Pages: |
151 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Dietrich School of Arts and Sciences > Statistics |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
linear discriminant analysis, classification trees, recursive partitioning algorithm, matched design, post-mortem tissue studies, schizophrenia |
| Date Deposited: |
31 Jan 2012 15:19 |
| Last Modified: |
15 Nov 2016 13:55 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/10627 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}