Analysis of peripheral blood mononuclear cell data for finding a cellular signature of graft-versus-host disease

GvHD is the most important complication of BMT. It has high morbidity and mortality. In an attempt to provide early prediction of GvHD, we analyzed PBMC data, together with GvHD outcome information, from a group of patients who underwent BMT. Various statistical methods were applied to overcome the difficulty arising from the high-dimensional longitudinal observations but with relatively small sample size. Analysis of “windowlized” data at a fixed time point did not find one subtype of PBMC or a subset of PBMCs that was important in the prediction of aGvHD. longitudinal data analysis indicated that proportions of T_ cells over time might separate well the patients with aGvHD and without aGvHD. This is consistent with the main finding in Lee (2006). In addition, we considered a two-level hierarchical model to describe the proportions of PBMCs. It indicates that T_cells CD3-CD8bdimCD8- is marginally associated with and disease.