Mixture Models of Phenotypic Markers in Population Disease Screening for Hemochromatosis and Iron Overload

• Activity Number: 103
• Type: Invited
• Date/Time: Monday, August 4, 2003 : 10:30 AM to 12:20 PM
• Sponsor: WNAR
• Author(s): Christine E. McLaren*+ and Victor R. Gordeuk and Kuo-Tung Li and David M. Reboussin and Phyliss Sholinsky and Beverly G. Mellen and Mark Speechly
• Companies: University of California, Irvine and Howard University and University of California, Irvine and Wake Forest University School of Medicine and Epidemiology and Clinical and Wake Forest University School of Medicine and University of Western
• Keywords: EM algorithm ; maximum likelihood ; finite mixture distributions ; resampling

Abstract

Hemochromatosis (HH), a common autosomal recessive disorder, increases susceptibility to develop iron overload. We describe methods developed to analyze the distribution of transferrin saturation (TS), a phenotypic marker of HH, and previously applied to data from National Health and Nutrition Examination Surveys. Mixtures of normal distributions were considered. The EM algorithm was applied for parameter estimation. Model maximized log-likelihoods were compared, and significance was assessed by resampling. Three subpopulations of TS were identified in data from African Americans and Caucasians that satisfied Hardy-Weinberg conditions for major locus effects. Subpopulations with increased mean TS had increased mean age-adjusted serum ferritin concentration, consistent with elevated iron stores. Identification of mutations in the HFE gene, responsible for most cases of HH in whites, enables estimation of the association between observed genotypes and the most probable individual assignment into TS subpopulations. We illustrate application of modeling techniques to data from the Hemochromatosis and Iron Overload Screening Study, a primary care-based sample of 100,000 adults.