A pattern-mixture model for haplotype analysis of longitudinal traits with non-ignorable dropout

Current statistical methods allow the characterization of DNA sequence variants associated with interpersonal differences in a complex biological response. However, this process is significantly hindered when some subjects have to drop out early due to physiological side effects or limited duration. Here, we derive a pattern-mixture model for detecting functional nucleotide combinations (or haplotypes) responsible for longitudinal responses by making full use of information from those dropout data. The model was formulated within the maximum likelihood context, with the model parameters, haplotype frequencies, and haplotype effects estimated by implementing the EM and Newton-Raphson algorithms. One advantage of the model is to generate and address a number of clinically meaningful hypotheses about the genetic control mechanisms of longitudinal responses and time-to-event processes. By analyzing a pharmacogenomic data set, the model identified significant haplotype effects on heart rate increases in response to increasing doses of dobutamine. The statistical properties of the model and its usefulness and utilization were investigated through computer simulation. The new model can be used to unravel the genetic architecture of interpersonal variation in complex longitudinal responses with incomplete data and ultimately to materialize the idea of clinical genomics.