Can We Predict Graft Intolerance Syndrome After Kidney Transplant Failure? External Validation of a Previously Developed Model

Abstract

Previously we established a prediction model for graft intolerance syndrome requiring graft nephrectomy in patients with late kidney graft failure. The aim of this study is to determine generalizability of this model in an independent cohort. The validation cohort included patients with late kidney graft failure between 2008 and 2018. Primary outcome is the prognostic performance of our model, expressed as the area under the receiver operating characteristic curve (ROC-AUC), in the validation cohort. In 63 of 580 patients (10.9%) a graft nephrectomy was performed because of graft intolerance. The original model, which included donor age, graft survival and number of acute rejections, performed poorly in the validation cohort (ROC-AUC 0.61). After retraining of the model using recipient age at graft failure instead of donor age, the model had an average ROC-AUC of 0.70 in the original cohort and of 0.69 in the validation cohort. Our original model did not accurately predict the graft intolerance syndrome in a validation cohort. However, a retrained model including recipient age at graft failure instead of donor age performed moderately well in both the development and validation cohort enabling identification of patients with the highest and lowest risk of graft intolerance syndrome.

Keywords: external validation; graft intolerance syndrome; graft nephrectomy; kidney graft failure; prediction model.

FIGURE 1

Patient inclusion.

FIGURE 2

Cumulative incidence curves of study outcomes.

FIGURE 3

AUC of ROC curve by follow up time (A) retrained model in the development cohort (B) retrained model in the validation cohort. The above figures show the ROC-AUC estimates at various time points during follow-up. The discriminative performance is reasonably constant throughout follow-up with an average of 0.70 in the training data, and 0.69 in the validation data.