A comparison of three statistical models for IDDM associations with HLA

Research output: Contribution to journalArticle

Abstract

The association between HLA-DQ haplotypes and insulin-dependent diabetes mellitus (IDDM) was studied in 48 children from 44 families ascertained from the high incidence area around Umea, Sweden. Numerous hypotheses have been proposed to explain associations between HLA and IDDM, but comparisons of statistical models based on these hypotheses have not been attempted. The aim of the present study was to compare the goodness-of-fit and predictive abilities among different statistical models. A likelihood-based analysis rather than a conventional analysis based on contingency tables was therefore adopted. We first used parental haplotype information in a conditional likelihood analysis (1) and then compared this analysis with that of an unaffected control group which used information on geographically matched controls. Under the analysis conditional on parental haplotype, a statistical model motivated by the hypothesis that the entire DQ heterodimer is involved in IDDM pathogenesis fit the data significantly better and had greater predictive ability than either a model motivated by the explanation that an IDDM gene is linked to DQB1 or that the DQB1 chain itself is involved in IDDM pathogenesis, or a model arising from the hypothesis that single amino acids at codon 57 of DQB1 and codon 52 of DQA1, respectively, confer susceptibility. Under the case-control analysis, the identity of the best-fitting or most predictive statistical model was not as clear, although both approaches to analyzing risk suggested that the single-amino-acids model had significantly poorer fit compared to the remaining two models.

Details

Authors
External organisations
  • University of Washington
Research areas and keywords

Subject classification (UKÄ) – MANDATORY

  • Endocrinology and Diabetes

Keywords

  • Diabetes mellitus, insulin-dependent, Disease susceptibility, Genetic marker, Haplotype, HLA-DQ antigen, Likelihood function, Model, statistical
Original languageEnglish
Pages (from-to)1-14
Number of pages14
JournalTissue Antigens
Volume48
Issue number1
Publication statusPublished - 1996 Jan 1
Publication categoryResearch
Peer-reviewedYes
Externally publishedYes