Percentile regression analysis of correlated antibody responses

Antibody responses following vaccination usually are analysed by comparing geometric mean concentrations across levels of relevant covariates and by comparing the proportions of vaccinees responding. In the regression setting, the analyses are done on log-transformed concentrations, estimating geometric mean responses conditional on a vector of covariates. More detailed analyses examining the relationship of covariates to different parts of the response distribution may be performed through the application of asymmetric least squares estimation of regression percentiles. We present a method for accounting for correlation in percentile regression analyses of longitudinal antibody response data. We illustrate the procedures with measles antibody response data from Haitian children who participated in a randomized trial of high titre vaccines. The strongest dose and strain effects were seen in the low end of the antibody concentration distributions.