*Purpose: Discarding donor organs lengthens time patients spend on the kidney transplant (KT) wait list. To identify characteristics associated with short-term outcome in deceased donor (DD) KT recipients, we compared predictive models that included KDPI/recipient/peri-operative variables, molecular markers, and the combination of KDPI/recipient/peri-operative and molecular markers. Such a strategy may prove useful when evaluating composite scoring systems that are desperately needed.

*Methods: Affymetrix gene expression data from pre-implant (PI) biopsies and GFR one month post-transplant were available for 191 KT subjects. Short-term outcome was classified according to patients’ GFR at 1 month, as GFR<=40 (low) versus GFR>40 (high). To identify a model predicting GFR high versus low, we included KDPI and all recipient and peri-operative variables having a univariable p-value <0.10 in a logistic regression model and applied a backward elimination procedure. Thereafter we fit a model to predict high vs low GFR using the gene expression data and another model that included KDPI/recipient/peri-operative variables along with gene expression data. For each model, we estimated the area under the receiver operating characteristic curve (AUC) the net reclassification improvement for comparing these three models.

*Results: Among 191 deceased donor kidney transplant recipients, there were 55 (28.8%) with low GFR and 136 (71.2%) with high GFR. Donor type (P=0.002), donor age (P=0.001), recipient height (P=0.022), recipient body weight (P=0.001), KDPI (P <0.001) and KDRI (P=0.002) were significantly different between the two groups. Therefore, these variables were included in a multivariable logistic regression model predicting one month GFR. After performing backward elimination, only KDPI (P=0.0007) and recipient body weight (BW) (P=0.0022) remained. There were 16 probe sets differentially expressed when comparing the high vs low GFR groups using a Benjamini and Hochberg FDR<0.15. There were 15 probe sets in the gene only model and 15 probe sets in the KDPI+BW+gene model, with 11 in common. The areas under the ROC curves for the three fitted models were KDPI+BW (AUC=0.713, 95% CI: 0.633, 0.794), gene only (AUC=0.828, 95% CI: 0.76, 0.896), and KDPI+BW+gene (AUC=0.833, 95% CI: 0.761, 0.905). There was a significant difference between the KDPI+BW only model and the gene only model (P=0.0334) as well as the KDPI+BW+gene model (P=0.0302). However, there was not a significant difference between the gene only and KDPI+BW+gene models when comparing the AUCs (P=0.806).

*Conclusions: A panel of PI molecular markers may predict short-term outcomes more accurately than scoring systems currently in place.

« Back to 2019 American Transplant Congress