Variability in Donor-Derived Cell-Free DNA Levels Predicts Mortality Risk in Heart Transplant Recipients

Abstract

<h3>Purpose</h3> Longitudinal surveillance of heart transplant (HTx) recipients using donor-derived cell-free DNA (dd-cfDNA) allows non-invasive detection of molecular injury patterns. We hypothesized that variability in dd-cfDNA levels over time identifies patients at risk of adverse outcomes. <h3>Methods</h3> 72 adult HTx patients monitored with gene expression profiling (GEP) and dd-cfDNA (AlloMap, AlloSure; CareDx Inc., Brisbane, CA) at our center were included. Variability (ASV) was defined as standard deviation of the 3 most recent (3-value) or all (all-value) sequential dd-cfDNA results and scaled x 10 for analysis (ASV*10). We used a Cox proportional hazards model to assess the association between ASV and mortality and a survival classification and regression tree (CART) algorithm to define a threshold for increased mortality risk. <h3>Results</h3> Patients had a median of 6 (IQR 4-9) paired GEP/dd-cfDNA results spaced 40.8 (IQR 31.5 - 66.1) days apart. 48.6% of patients had dd-cfDNA below the assay limit of detection (<0.12%); the remainder had a median dd-cfDNA of 0.17% (IQR: 0.12 - 0.45%) and a mean all-value ASV of 0.24 (SD: 0.40). ASV*10 of patients who died (2.62, SD 2.66) trended higher than of those who survived (0.54, SD 1.10) over a median follow-up of 480 days (IQR 244 - 859). In a univariate Cox model, the 3-value ASV*10 was associated with an increased risk of mortality (HR 1.66, 95% CI 1.14 - 2.41, p<0.009). Only a trend was evident after multivariate adjustment <b>[Figure 1a]</b>. Using a CART algorithm, a 3-value ASV*10 of 2.97 discriminated between patients at high and low risk of mortality after HTx <b>[Figure 1b]</b>. <h3>Conclusion</h3> In this proof-of-concept study evaluating the clinical utility of longitudinal dd-cfDNA results, we found that higher variability in dd-cfDNA over time was associated with an increased risk of mortality after HTx. Additional studies utilizing larger datasets are needed to validate and expand on our findings.