Statistical methods for matched case-control studies and cDNA microarray /
This dissertation presents a method using the lognormal distribution for quantitative analysis of cDNA microarray, methods to detect effect modification by a matched covariate in matched case-control studies, and methods to fit a semiparametric regression splines model in matched case-control studie...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
2002.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=764789731&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | This dissertation presents a method using the lognormal distribution for quantitative analysis of cDNA microarray, methods to detect effect modification by a matched covariate in matched case-control studies, and methods to fit a semiparametric regression splines model in matched case-control studies. First, the ratios of average gene expression levels for individual genes is used as a measure of differences in expression levels. The distribution of the ratio is derived by assuming a lognormal distribution for gene expression level. Our approach is compared with previous work by Chen et al. (1997), where the gene expression level is modeled with a normal distribution rather than a lognormal distribution. We show that lognormality is a reasonable assumption using Shapiro-Wilks test and Q-Q normal plot. Second, methods for assessing effect modification by a matching covariate in matched case-control studies are developed, one is a new graphical method based on a semiparametric model. The graphical method is developed using a varying coefficient model (Carroll, Ruppert and Welsh, 1998; Tibshirani and Hastie, 1987). The other is a method based on a generalized additive model which requires specialized software. The simulation results show that the graphical method, when based on linear and quadratic effect modification, can be more powerful than a fully nonparametric generalized additive model fit, which is also a new approach. Third, we apply regression splines to matched case-control studies. Our main method uses approximate cross-validation to estimate the smoothing parameter, the method is based upon a higher order expansion for the leave-one out estimate. We also develop Monte Carlo Expectation Maximization (MCEM) and Bayesian methods to fit the regression spline model. We compare the approximate cross-validation approach, MCEM and the Bayesian approach using simulation, showing that they appear approximately equally efficient, with the approximate cross-validation method being computationally the most convenient. |
|---|---|
| Item Description: | Vita. "Major Subject: Statistics". |
| Physical Description: | xiii, 73 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 61-63). |