We are writing in response to the News & Views commentary by M. Abecassis and B. Kaplan (Biomarkers in transplantation — the devil is in the detail. Nat. Rev. Nephrol. 11, 204–205; 2015)1, which discusses our study that described the development of the non-invasive Kidney Solid Organ Response Test (kSORT) to identify kidney transplant recipients at high risk of acute rejection (The kSORT assay to detect renal transplant patients at high risk for acute rejection: results of the multicenter AART study. PLoS Med. 11, e1001759; 2014)2. We acknowledge the complexity of the statistical analysis used to evaluate the diverse cohort of independent samples included in this study in order to obtain an accurate molecular diagnostic, but are surprised by the misunderstanding of the authors regarding kSORT and the kSORT analysis suite (kSAS) algorithm.

As part of the AART study, 558 peripheral blood samples from 436 renal transplant recipients at eight centres were analysed using real-time quantitative PCR. A 17 gene set was defined and a novel correlation based algorithm — kSAS — was developed for the detection of acute rejection. The sensitivity and specificity of kSORT to detect acute rejection were 92.31% and 93.48%, respectively (area under the curve (AUC) 0.92, 95% CI 0.856–0.981)2.

Abecassis and Kaplan describe kSORT as representing “a form of model fitting that trains multiple models to determine the “best fit” by starting with 17 genes and reducing the number of genes until a best fit is found,” which is factually incorrect. kSORT measures the expression of 17 genes in blood and establishes the risk of acute rejection using a fixed algorithm (kSAS). kSAS is comprised of 13 models, each consisting of 12 of the 17 genes, and assesses the correlation (Pearson correlation coefficient) for each of these models between an unknown sample with predetermined acute rejection and no acute rejection reference profiles. In an aggregated risk analysis, a score is calculated for each sample based on all 13 models2 and the result reflects the patient's immune response at the time of assessment. kSAS scores ≥9 are defined as high-risk, scores ≤−9 are defined as low-risk and scores of >−9 to <9, are defined as indeterminate risk for acute rejection (Fig. 1). Thus, thresholds for risk of acute rejection are very well defined, and Abecassis and Kaplan are incorrect in their assertion that “no defined threshold values” exist. The details of the algorithm and the thresholds are provided in the original publication2.

Figure 1: The Kidney Solid Organ Response Test (kSORT) analysis suite (kSAS) algorithm.
figure 1

The kSAS algorithm correlates the gene expression for 13 gene models, each consisting of 12 genes, in unknown samples with reference profiles for acute rejection and no acute rejection, and provides a risk score for acute rejection (high, intermediate or low risk). Modified with permission from PLOS © Roedder, S. et al. PLoS Med. 11, e1001759 (2014), which is licensed under a Creative Commons Attribution 4.0 International License. To view a copy of this licence visit https://creativecommons.org/licenses/by/4.0/.

PowerPoint slide

Full size image

Although Abecassis and Kaplan criticize the heterogeneity of the AART study, this aspect is actually a strength of the study. AART was designed to represent the real-life post-transplantation setting to enable the development of a robust biomarker. Thus, AART deliberately included patients with different demographics, peripheral blood processing methodologies, and post-transplantation management strategies. These variables represent confounders for gene-expression-based biomarker development3,4,5,6,7 that should be accounted for when developing diagnostic assays. The AART study results demonstrate the ability of kSORT to detect clinical acute rejection in a large number of samples and subclinical acute rejection in a smaller number of samples at the time of biopsy assessment. The results further suggest an ability of kSORT to predict acute rejection up to 3 months before the development of clinical acute rejection2. The ability of kSORT to specifically detect subclinical acute rejection was further validated in the ESCAPE study, which demonstrated accurate classification of patients into high-risk or low-risk groups at the time of protocol biopsy in 89.2% of cases8. In this study, kSORT was able to predict the onset of subclinical acute rejection up to 6 months before a protocol biopsy with an AUC of 0.73 (95% CI 0.6–0.86). These data have also been validated in the interim results of the SAILOR randomized trial9. Furthermore, Immucor DX (Grand Rapids, Michigan, USA), an independent certified clinical laboratory, has validated the analytical accuracy, precision, sensitivity and specificity of kSORT, and has made this assay available as a Laboratory Developed Test for immune surveillance of renal transplant recipients in conjunction with the current standard of care.

Abecassis and Kaplan claim that the positive predictive value of kSORT would likely fall from 93.21% to 60% if applied to an at-risk kidney transplant population. We agree that the positive and negative predictive values can be important measures for clinical value and depend on prevalence. In transplantation, however, the true prevalence of acute rejection cannot be fairly determined because acute rejection evolves over time. The current gold standards to monitor for and detect acute rejection are limited and based on consensus, and subclinical acute rejection might be missed in patients who never undergo protocol biopsies. Given the high sensitivity and specificity of kSORT for detection of acute rejection, this assay might facilitate the development of precision diagnostics to monitor the immune response of kidney transplant recipients in a personalized manner.

In conclusion, the commentary by Abecassis and Kaplan1 includes misinterpretation of kSORT and the kSAS algorithm. kSORT has demonstrated high sensitivity and specificity to detect acute rejection at the time of assessment in a large number of patients and to predict the development of acute rejection in a smaller number of patients. Three ongoing clinical studies will provide further validation of this data and the clinical value of kSORT.