Effect of classifying disease states in genetic association studies for paratuberculosis

Genetic association studies are a means to elucidate underlying genetic regulation of host-pathogen interaction, immune response, and the fate of infection. Diseases such as paratuberculosis in cattle lack definitive diagnostic criteria, thereby complicating the definition of infection status as an outcome for genetic association studies. A study was performed to evaluate the potential bias in estimates of effect and differences in statistical power associated with parallel test interpretation, latent probability of infection adjusted for imperfect test sensitivity and specificity, and multinomial outcomes in cohorts of cattle simulated using Monte Carlo sampling methods. Test results were simulated for microbial culture of feces for Mycobacterium avium subsp. paratuberculosis (MAP) and serum ELISA for anti-MAP antibody using estimates of test sensitivity and specificity. A range of disease allele frequencies and levels of association were considered. Case-control study populations were drawn from the simulated cohorts and the association between the disease allele and infection status was evaluated using logistic regression for binary outcomes and polytomous regression for multinomial outcomes. For the majority of the classification and analytical methods evaluated, estimates of effect were biased toward the null. Frequentist approaches to analysis of the latent probability of infection and multinomial classifications based upon results of culture of feces for MAP demonstrated the smallest degree of bias. Power to detect associations was generally low for all models, but improved with larger effects and higher allele frequencies. Imperfect specificity of serum ELISA was a major factor in the degree of bias observed and statistical power. The results of this study indicate that the method of classifying infection status must be considered carefully in genetic association studies for paratuberculosis and other diseases with similar challenges in defining infection status, and study designs should be modified to accommodate relative advantages and disadvantages of available methods.