Real Time Central Assessment of Kidney Transplant Indication Biopsies by Microarrays: The INTERCOMEX Study

​Am J Transplant. 2017 Nov;17(11):2851-2862.
PMID: 28449409 

Abstract
Halloran PF, Reeve J, Akalin E, Aubert O, Bohmig GA, Brennan D, Bromberg J, Einecke G, Eskandary F, Gosset C, Duong Van Huyen JP, Gupta G, Lefaucheur C, Malone A, Mannon RB, Seron D, Sellares J, Weir M, Loupy A.

​The authors conducted a prospective trial to assess the feasibility of real time central molecular assessment of kidney transplant biopsy samples from 10 North American or European centers. Biopsy samples taken 1 day to 34 years posttransplantation were stabilized in RNAlater, sent via courier overnight at ambient temperature to the central laboratory, and processed (29 h workflow) using microarrays to assess T cell- and antibody-mediated rejection (TCMR and ABMR, respectively). Of 538 biopsy samples submitted, 519 (96%) were sufficient for microarray analysis (average length, 3 mm). Automated reports were generated without knowledge of histology and HLA antibody, with diagnoses assigned based on Molecular Microscope Diagnostic System (MMDx) classifier algorithms and signed out by one observer. Agreement between MMDx and histology (balanced accuracy) was 77% for TCMR, 77% for ABMR, and 76% for no rejection. A classification tree derived to provide automated sign-outs predicted the observer sign-outs with >90% accuracy. In 451 biopsy samples where feedback was obtained, clinicians indicated that MMDx more frequently agreed with clinical judgment (87%) than did histology (80%) (p = 0.0042). In 81% of feedback forms, clinicians reported that MMDx increased confidence in management compared with conventional assessment alone. The authors conclude that real time central molecular assessment is feasible and offers a useful new dimension in biopsy interpretation. ClinicalTrials.gov NCT#01299168.

Figure 2: Classification tree to show how molecular measurements can be assembled for automated diagnostic sign‐out. The tree predicts class labels—the diagnoses assigned by one observer (P.F.H., i.e. T cell–mediated rejection [TCMR], antibody‐mediated rejection [ABMR], mixed, or no rejection [NR]) using time of biopsy after transplantation plus all of the rejection molecular measurements (Table S1) but with no histology information. Node headers (and node colors) correspond to the majority P.F.H. diagnoses (PFHDx) in that node. The six terminal nodes (i.e. leaves) of the tree are the predictions. The line of four numbers within each node/leaf corresponds to the proportion of P.F.H. diagnoses for ABMR, mixed, NR, and TCMR, respectively. For example, the tree predicts that P.F.H. called 24% of the 519 biopsy samples ABMR (leftmost leaf) and that assessment was correct (i.e. P.F.H. actually did diagnose ABMR) in 87% of those biopsy samples and called mixed in an additional 2%.