|Adjusting for Matching and Covariates in Linear Discriminant Analysis
In studies that compare several diagnostic or treatment groups, subjects may not only be measured on a certain set of feature variables, but also matched on a number of demographic characteristics and measured on additional covariates. Linear discriminant analysis is sometimes used to identify which feature variables best discriminate among groups, while accounting for the dependencies among the feature variables.
In this paper, Drs. Asafu-Adjei and Sampson (from Pitt's Department of Statistics) and TNP investigators, Drs. Sweet and Lewis, present a new approach to linear discriminant analysis for multivariate normal data that accounts for the subject matching used in a particular study design, as well as covariates not used in the matching. The motivation for this research came from analyses of human post-mortem brain tissue studies conducted in the Conte Center for the Neuroscience of Mental Disorders (CCNMD).
Applications were given for post-mortem tissue data with the aim of comparing neurobiological characteristics of subjects with schizophrenia to those of normal controls. In addition, the performance of the approach was investigated using a simulation study. In a typical study, schizophrenia subjects and normal controls are paired on age at death, gender, and post-mortem interval (PMI). Auxiliary data such as brain tissue storage time and brain pH are also collected for each subject. Pairs are processed at the same time in a balanced fashion to avoid possible confounding of biomarker measurements by varying reagent strengths, time, and processing personnel.
The proposed methodology produced more accurate classification results compared with both the traditional approaches, where the improvement in the approach became more pronounced as increasing numbers of pairs and increasing differences between the effects for groups were considered.
|Asafu-Adjei JK, Sampson AR, Sweet RA, Lewis DA. Adjusting for Matching and Covariates in Linear Discriminant Analysis. Biostatistics, in press.|