Concerns with the use of imputation to assign HLA allele-level typing in research predicting transplant outcomes
Tambur, A. R.; Gmeiner, M.; Manski, C. F.
Show abstract
Incomplete information on HLA allele typing is a persistent problem when analyzing the role of Human Leukocyte Antigen (HLA) in transplantation. To refine the predictions possible with partial knowledge of HLA typing, some researchers use HaploStats statistics on the frequencies of haplotypes within specified ethnic/national populations to impute complete HLA allele typing. We evaluated methods that use imputation to predict patient outcomes after organ transplantation, with focus on prediction of graft survival conditional on typing information of the donor and recipient. Logical arguments show that imputation yields no predictive power when predictions are conditioned on all observed HLA typing data. Computational experiments indicate that imputation does not have predictive power when applied to risk-assessment models that make predictions conditional on only part of the observable HLA data. We therefore caution against reliance on imputation to overcome incomplete measurement. We encourage high-resolution typing of HLA antigens to improve prediction of transplant outcomes and matching of donors with recipients. Similar considerations should likely apply in other clinical settings.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.