J. RICHARD LANDIS, Ph.D. Professor, Department of Biostatistics and Epidemiology University of Pennsylvania, School of Medicine
THURSDAY, MARCH 14, 2013 2:00 p.m.– 3:00 pm Clinical Research Building, Room 692
This talk will present a 3-level nested (clinic, subject, observer) category-specific variance components model to estimate agreement and disagreement patterns among multiple observers in unbalanced designs. These methods will be illustrated within a “forensic biostatistics” investigation of identity mis-alignment of genotype to phenotype data within a mini-GWAS. Alerted by PID mismatches of duplicate SNP data records, category-specific intraclass correlation coefficients (Landis and Koch 1977; Landis et al 2011) comparing self-reported and genetically-inferred race were implemented within subsets of PIDs known to have been processed separately, and then the presence of non-random clustering of race disagreement patterns was discovered on selected genotyping plates. Fortunately, a subsequent GWAS “fingerprinting” sub-study focusing on 24 “identity” SNPs common to both studies permitted realignment of these identities, so that discovery and validation research could proceed. By blocking on genotyping plate as the first level, these nested category-specific intraclass correlation coefficient results illustrate the need for implementing basic experimental design features prior to conducting a GWAS (Lambert and Black 2012), balancing the distribution of gender, race/ethnicity and the primary outcome across the genotyping plates.