Re-evaluation of skin-explant model in graft-versus-host disease prediction

An in vitro skin explant model has been proposed for prediction of graft versus host disease and reported to be highly sensitive and specific for this purpose. In this study we have re-evaluated this model in both HLA full-matched BMT recipient-donor sibling pairs and also in HLA one haplotype-matched parent-children pairs. All assessments were made blindly by 3 independent observers. The predictive value of the test for the occurrence of clinical GvHD in 14 BMT patients was found to be less sensitive than reported previously (correlation coefficients were +0.019, +0.067 and -0.061 between clinical GvHD and in vitro primed allogeneic, primed mixed and unprimed allogeneic settings, indicating "poor" correlation). False positive and false negative results were high and there were also significant discrepancies between three blind observations in the grading of skin changes. Weighted kappa analysis revealed that there were "fair" correlations between the 3 observers (K = 0.25). These results indicated that the skin explant model is an unpredictable test system and there are great problems in standardization of the method.