Background
In a landmark article published more than 40 years ago, Dr. Philip Caves and colleagues described the percutaneous transvenous endomyocardial biopsy (EMB) and reported their experience with the 67 cardiac biopsies performed in 17 cardiac transplant recipients [
1]. Since then, serial EMB has become the standard of care for monitoring transplant rejection status in adult patients [
2]. Despite its wide acceptance, EMB has several limitations. First, patients who undergo EMB, particularly those undergoing biopsy surveillance for a prolonged period of time, are at increased risk of complications [
3]. Second, the pathological interpretation can be subjective [
4,
5]. Third, the procedure can only detect rejection after cellular infiltration and/or graft damage has occurred [
5], limiting a prognostic utility that may guide optimization of immunosuppression therapy. Finally, there is a noteworthy cost and patient inconvenience associated with EMB surveillance. Consequently, a significant interest in developing novel surveillance strategies has fueled many attempts at finding alternative non-invasive tests to monitor allograft status [
6].
A number of alternative tests have been developed but have either failed to deliver adequate performance or their implementation proved complex [
5]. In contrast, in a large randomized controlled trial, a non-invasive gene expression profiling (GEP) test (AlloMap®, CareDx, Brisbane, California) was found to be non-inferior to routine EMB for surveillance after heart transplantation in selected low-risk patients with respect to clinical outcomes [
7]. The test is performed on a peripheral blood sample; the results are reported as a single GEP score (range from 0 to 40) [
8]. Since the GEP test became available in 2005, more than 69,500 tests have been performed on more than 15,000 unique heart transplant recipients. This surveillance methodology has been generally used in adult patients older than 15 years for identification of heart transplant recipients at a low risk of acute cellular rejection at time of testing [
2].
In addition to the diagnostic utility of the single GEP score, Deng and colleagues noted an association between low variability of serial GEP scores and the clinical stability in patients [
9]. These initial findings have been independently confirmed by post-hoc multivariate analyses of 602 heart transplant patients enrolled in the Invasive Monitoring Attenuation by Gene Expression Profiling (IMAGE) study [
10]. This analysis suggested that the variability of GEP scores from an individual may predict the risk of future allograft dysfunction or death in that individual. The prognostic utility of serial GEP scores in predicting a clinical event was found to be independent of, and complementary to, the original use of a single GEP score in estimating the probability of acute cellular rejection at the time of testing. However, further validation of a serial characteristic of GEP scores was suggested [
10].
Therefore, in the current study, we expanded on the earlier findings by conducting an analysis of an independent patient population from the Cardiac Allograft Rejection Gene Expression Observational (CARGO) II study to determine: (1) the prognostic utility of within-patient variability of GEP scores in predicting future significant clinical events; (2) the negative predictive value (NPV) and the positive predictive value (PPV) of GEP score variability in predicting future significant clinical events.
Discussion
The most important finding of this study is that the GEP score variability may be useful in estimating the probability of future events of death, re-transplantation or graft failure in heart transplant recipients undergoing surveillance with GEP testing ≥315 days post-transplantation. A low variability of sequential GEP scores (≤0.6), which was found in 24.2 % of the study population, rendered NPVs of ≥97.0 % (Table
2), indicating clinical utility of GEP score variability in the identification of patients at a low risk for future clinical events of greatest concern. This finding may have important clinical implications in the longer-term management of heart transplant recipients as the low risk patients may be good candidates for optimization (i.e. reduction) of their immunosuppressive drugs. Consequently, we speculate that this may reduce the rate of unwanted adverse effects associated with chronic immunosuppression, particularly infections and nephrotoxicity. We believe that future GEP studies should focus on personalization of long-term immunosuppressive therapy and potential improvements in long-term outcomes of heart transplant recipients.
Using peripheral-blood specimens, the AlloMap gene expression test translates the complex gene expression patterns of the mononuclear blood cells into a single score (0–40) using a proprietary algorithm (Additional file
2). The risk of rejection is considered to be higher immediately after transplantation and lower by six to 12 months post-transplant [
13]. In the IMAGE study, the use of a single gene expression profiling score (below the 34 threshold) to assess a risk of acute cellular rejection in clinically stable patients six months post cardiac transplant resulted in a significant decrease in the number of EMB performed. This reduction was achieved without adversely affecting patient outcomes [
7]. More recently, another randomized clinical trial has shown similar results in patients who began the AlloMap surveillance two months post-transplant, which is considered a higher risk rejection time frame [
14]. Primarily based on the IMAGE study results, the International Society of Heart and Lung Transplantation (ISHLT) suggested that a single AlloMap score can be used to rule out the presence of acute cellular rejection of ≥ 2R in selected low-risk patients [
2]. In contrast, the GEP score variability analysis was performed to aid clinicians in assessing the future risk of a significant clinical event (death, re-transplantation or graft failure). In the field of transplant cardiology, there is a paucity of biomarkers that can reliably predict which patients are at risk of adverse clinical events [
15]. There is a growing need to develop evidence-based personalized strategies to optimize immunosuppressive therapy, improve long-term outcomes and reduce complications [
15]. Therefore, and in addition to the individual GEP score and clinical assessment, the GEP score variability may provide useful complementary information that may help personalize long-term care of heart transplant recipients.
Deng and colleagues computed GEP score variability and performed post-hoc analyses on IMAGE patients who had at least two single GEP scores before an event or study end [
10]. In that study, multivariate analyses revealed that only GEP score variability was significantly associated with future clinical events; gender, race, age and cytomegalovirus serological status were not. In the cohorts presented in this article, we prospectively analyzed clinical outcome data of the CARGO II patients who had four GEP scores preceding a first clinical event (event group) or had 4 sequential GEP scores without a subsequent event (control group). In addition to the fact that a mean of 3.5 or 4.4 tests are performed within a two year surveillance interval [
10,
12], our decision to use four serial GEP scores was based on calculating 85 % probability of being within twofold of true variance (using 3 scores yields 76 % probability). Although we included two more GEP scores than Deng and colleagues in our calculation, the formula for computing GEP score variability remained the same (Additional file
3) [
10]. Importantly, enrollment in the current study began in 2005. The latest follow-up clinical data was provided by the end of 2011, allowing us to predict a significant clinical event up to 6 years following heart transplantation. Finally, we also computed NPV, PPV, sensitivity and specificity for the range of GEP score variability between 0.1 and 2.1.
Our results must be interpreted while considering the limitations of this study. First, our composite primary clinical outcome is a mixture of clinical conditions including subtypes of events that were rare (e.g. only four events of re-transplantation). In the future, we would aim to collect larger numbers and/or more specific subtypes of endpoints with sufficient power to be separately analyzed (e.g. deaths from infections or cancers due to long term sequelae of over-immunosuppression). Additionally, the expression levels of the informative genes used to compute GEP scores in some cases may reflect a feature of the status of the recipient’s immune system that is not directly associated with the allograft [
10]. Previous studies have indicated that prednisone doses above 20 mg/day may reduce the GEP score by inhibiting the expression of IL1R2, FLT3, and ITGAM genes [
10]. Most recently, it has been reported that acute cytomegalovirus infections may be associated with increased GEP scores in the absence of acute cellular rejection [
16]. Our retrospective analysis of this case-cohort study may introduce potential, unintentional bias in the selection of patients. To minimize selection bias, we included all patients who met our predetermined definition of clinical events in the event group. All other non-event patients from the CARGO II study who had four GEP scores were assigned to the control group. Moreover, one may argue that the limited sample size (36 in the event group and 55 in the control group) and the imbalance between the number of patients in each study group could affect our findings. The use of unequal assignments ≤ 3:1 between study groups does not significantly reduce the power of the study [
17]. Also, the baseline characteristics of study patients were well-matched at entry.
In our study, the performance of the GEP score variability in predicting outcomes for patients who were monitored with GEP testing ≥315 days post-transplantation (AUC ROC of 0.72, 95 % CI 0.61 to 0.81) is similar to the performance of the established single GEP score in ruling out acute cellular rejection for patient samples ≥2-6 month post-transplant (AUC ROC 0.71, 95 % CI 0.56 to 0.84). The limited PPV of the GEP score variability (Table
2) will undoubtedly be compared with interpretation of EMB findings for acute cellular rejection of cardiac transplants. In another report, observed agreement between local and central pathologists for EMB scoring was 28 % for grade ≥ 2R (2004 ISHLT grading system) [
4]. Despite the ISHLT simplification of the 1990 grading classification, suboptimal EMB readings are still common [
4]. Furthermore, the GEP score variability may predict future clinical events rather than be used for ruling out acute cellular rejection at the time of a test. In this study, the estimated PPVs do not exceed 36 % for variability values of up to 1.7. This limitation of the PPV is in part attributable not only to the underlying sensitivity of this measure, but to the low prevalence of events in the study population. Since the NPV exceeds 97 % for variability values of ≤0.6, this test result may be best suited for use to “rule-out” rather than for use to “rule in” the likelihood of a future clinical event (Table
2). Specific NPV and PPV values (Table
2) are provided to aid clinicians in estimating the likelihood of death, re-transplantation and graft dysfunction occurring in patients beyond 315 days post-transplant. However, the clinician may choose how to use the particular GEP score variability threshold based on a practical experience and make an associated clinical interpretation based on overall clinical presentation of the individual patient. Based on the results of this study, we suggest use of GEP score variability to predict a future composite clinical event (death from any cause, re-transplantation or graft failure) only for GEP tests collected at ≥315 days post-transplant. For example, the patients with GEP score variability ≥1.6 (approximately 8 % of the CARGO II population) may be at a higher risk of experiencing an event (PPV ~30 %; Table
2). These patients may be candidates for more vigilant surveillance and appropriately timely interventions. On the other hand, patients with a low GEP score variability, particularly scores ≤ 0.6 (approximately 25 % of our population) may be at a low risk of experiencing a clinical event (NPV ≥97 %) and may be potential candidates, depending on the full clinical circumstances in each case, for reduced immunosuppression.
Competing interests
Jörg Stypmann, Uwe Schulz, Paul Mohacsi, Andreas Zuckermann and Mario C. Deng received study support from CareDx. Mario C. Deng serves on the advisory board of CareDx. James P Yee is an employee owning equity in CareDx. Emir Deljkich, David Hiller and Lane Eubank are employees of CareDx. No other potential conflict of interest relevant to this article was reported.
Authors’ contributions
MGCL, JS, US, AZ, PM, CB, HR, JP, MZ, RF, DH, MD, PL, JV acquired data and supervised the CARGO II in-life conduct in respective centers. MGCL, JS, JV and JY designed the study and provided initial interpretation of the GEP variability results. DH and LE provided the core statistical analyses. ED drafted the manuscript. JV, MGCL and JS revised the manuscript. All authors read and approved the final manuscript.