In contrast to acute cellular rejection, non-invasive assessment of antibody-mediated rejection (AMR) remains challenging. The routine surveillance endomyocardial biopsies (EMB) approach to monitor AMR suffers from multiple limitations. The development of an individual AMR risk score may allow for individualization of the EMB protocol.

We analyzed a prospective and deeply phenotyped cohort of patients transplanted between 2012 and 2017 at our center (n=700). Patients were randomly distributed in a training (2/3) and in a test set (1/3). We applied a mixed effect logistic regression with a random intercept to identify predictive variables associated with AMR≥pAMR1 at the level of each EMB. An AMR risk score was derived by applying the β-coefficients attributed to each predictive variable.

A total of 6403 EMB were analyzed among which 412 (6.4%) were diagnosed as AMR≥pAMR 1. In the training set, five predictive variables were independently associated with AMR: time post-transplant (P<0.001), pre-transplant sensitizing event (P=0.003), circulating donor-specific antibody at the time of EMB (P=0.010), left ventricular dysfunction (P=0.018) and prior history of pAMR2 (P<0.001). In the test set, the calibration and the discrimination of the model were good (area under the ROC curve=0.801, Fig. 1A). We observed a stepwise increase in the risk of AMR with increasing AMR risk score, ranging from 1.6% (1.1–2.5%) for score≤1 point to 36.8% of EMB (22.8–53.6%) for scores between 4 and 5 (Fig. 1B).

We identified 5 independent predictive variables associated with AMR and derived an AMR risk score that may help clinicians in individualizing the EMB protocol.