Variable selection methods for predicting clinical outcomes following allogeneic hematopoietic cell transplantation

Allogeneic hematopoietic cell transplantation (allo-HCT) is a potentially curative procedure for a large number of diseases. However, the greatest barriers to the success of allo-HCT are relapse and graft-versus-host-disease (GVHD). Many studies have examined the reconstitution of the immune system after allo-HCT and searched for factors associated with clinical outcome. Serum biomarkers have also been studied to predict the incidence and prognosis of GVHD. However, the use of multiparametric immunophenotyping has been less extensively explored: studies usually focus on preselected and predefined cell phenotypes and so do not fully exploit the richness of flow cytometry data. Here we aimed to identify cell phenotypes present 30 days after allo-HCT that are associated with clinical outcomes in 37 patients participating in a trial relating to the prevention of GVHD, derived from 82 flow cytometry markers and 13 clinical variables. To do this we applied variable selection methods in a competing risks modeling framework, and identified specific subsets of T, B, and NK cells associated with relapse. Our study demonstrates the value of variable selection methods for mining rich, high dimensional clinical data and identifying potentially unexplored cell subpopulations of interest.

Automated summary

Allogeneic hematopoietic cell transplantation is a potential curative treatment for many diseases, but the main barriers to success are relapse and graft-versus-host disease. Studies have shown that variable selection methods can help in exploring rich clinical data and identifying cell subsets associated with clinical outcomes.