Abstract

The use of three statistical models yielded different estimates of the odds ratio relative to the association between the polymorphism in the HLA- DQ region and insulin-dependent diabetes mellitus (IDDM). The models used were: (1) the allele-dosage model which assumes that the number of susceptibility alleles has a linear effect on the logarithm of the odds; (2) the reference cell coding method used with alleles of susceptibility as a risk factor; or (3) a model that uses a classification of ?/? heterodimers as a susceptibility factor. We suggest that models which imply a log-linear relationship between a susceptibility marker and disease such as the first model are not appropriate in the assessment of the HLA-IDDM association. In contrast, although both latter models are valid, the third model is more compatible with current hypotheses of the pathological process of the disease. Once an estimation of the odds ratio is chosen, we use such an estimation to calculate an approximation of the prevalence of IDDM according to the polymorphism in HLA-DQ region using the iterative procedure of Newton- Raphson. These approaches are illustrated with data from a case-control study previously conducted in the city of Santiago, Chile.