Utility of an AI Model in Predicting Risk of Kidney Allograft Rejection

The discrimination performance of AlloView, an artificial intelligence (AI) model to predict kidney transplant rejection (KTR) risk, has been validated in a large cohort. However, there are few data describing a reference range or threshold to guide clinicians in using AlloView data when deciding whether to perform biopsy.

A group of researchers, including S.V. Shah, MD, compared AlloView results by histological diagnosis. They presented results at the American Transplant Congress 2024. In their primary analysis, the researchers compared median AlloView results stratified by histological diagnosis: acute cellular rejection (ACR), acute antibody-mediated rejection (AMR), borderline rejection, and no rejection or acute tubular injury/necrosis (no rejection). They used nonparametric tests to analyze categorical and numerical variables. Their analysis included KTR occurring between January 1, 2017, and December 31, 2020, with initial donor-derived cell-free DNA (dd-cfDNA) result within 90 days of transplant, at least one biopsy, and a dd-cfDNA result ≤30 days before biopsy.

In sum, 41 AlloView results from 41 KTRs met the criteria for analysis. There were three biopsies with ACR, eight AMR events, seven borderline rejections, and 23 no rejection or acute tubular necrosis episodes. There were no significant differences in most baseline characteristics among different groups.

The median AlloView score was significantly higher in the ACR (62.9%; interquartile range [IQR], 51.1%-65.7%) and AMR (62.6%; IQR, 50.6%-76.4%) groups versus the no rejection group (15.5%; IQR, 9.3%-24.6%; P=.0273 and P=.0006, respectively). Borderline rejection had a median AlloView score of 28.9% (IQR, 20.4%-54.7%), statistically similar to the no rejection group (P=.0418).

The study authors concluded, “The significant differences seen highlight the utility of AlloView in discriminating patients’ individual risk of rejection. With further validation, these references may support interpretation of the model.”

Source: Shah SV, Voora S, Hanson, PJ, et al. Utilization of artificial intelligence (AI) in predicting the risk of kidney allograft rejection. Abstract #D148. Presented at the American Transplant Congress 2024; June 1-5, 2024; Philadelphia, Pennsylvania.