Secondary Logo

Share this article on:

Low Variability of Intraindividual Longitudinal Leukocyte Gene Expression Profiling Cardiac Allograft Rejection Scores

Deng, Mario C.1; Alexander, Gregory2; Wolters, Hans2; Shahzad, Khurram1; Cadeiras, Martin1; Hicks, Albert3; Rowe, Theresa3; Klingler, Tod M.2; Eisen, Howard J.3

doi: 10.1097/TP.0b013e3181e7e536
Letters to the Editor

1 Department of Medicine Columbia University College of Physicians and Surgeons New York, NY

2 Research Informatics XDx, Brisbane, CA

3 Department of Medicine, Division of Cardiology, Drexel University College of Medicine, Philadelphia, PA

This work was in part supported by XDx Inc. (M.C.D., H.J.E.), National Institutes of Health (M.C.D., H.J.E.), and World Heart Inc. (M.C.D., H.J.E.).

Alexander, Wolters, and Klingler work for XDx Inc. Shahzad, Cadeiras, Hicks, and Rowe do not report any financial conflicts of interest.

Address correspondence to: Mario C. Deng, M.D., F.A.C.C., F.E.S.C., Director of Cardiac Transplant Research, Columbia University College of P&S, PH12—Stem Room 134, 622 W 168th Street, New York, NY 10032; and Howard J. Eisen, M.D., F.A.C.C., Thomas V. Fischer Professor of Medicine, Division of Cardiology, Drexel University College of Medicine, 230 N Broad Street, 7th Fl, North Tower, Philadelphia, PA 19102.


Received 12 October 2009.

Accepted 13 May 2010.

Using a cross-sectional experimental design, a gene expression profiling (GEP) blood test was recently developed (1) and Food and Drug Administration cleared (2) for the identification of heart transplant patients with low likelihood of rejection. GEP test-based protocols were implemented by several US centers since 2005 (3). A randomized clinical trial has shown that a clinical protocol based on this GEP-based noninvasive test is noninferior to routine biopsies in helping patients to avoid serious complications of heart transplantation (4). The test has high negative predictive value (but low or moderate positive predictive value) to rule out rejection when a single measurement taken at a single time point is compared with a set of reference thresholds. As such, it does not take into account whether the current GEP measurement is close to the range of prior values observed from the same patient. During the initial clinical experience, it became apparent that GEP scores for the same patient often fall within narrow ranges over long time periods (5). Based on this observation, the specific aim of this retrospective study was to quantify the relative contributions of between-person and within-person variances to the total variance in GEP scores to assess the potential value of incorporating prior GEP scores, in addition to the current score, into the clinical decision-making process.

Back to Top | Article Outline


Clinical Sampling

Between January 2006 and December 2008, consecutive clinically stable heart transplant recipients in two large academic US centers managed by noninvasive rejection monitoring without protocol endomyocardial biopsy using GEP according to clinical protocol (every 2–4 weeks in the period from 3 to 5 months posttransplantation, every 1–2 months in the period from 6 to 12 months, and every 3–6 months thereafter) with at least three GEP results were analyzed. Patients and samples were identified by retrospective chart review. Clinical evidence of rejection was defined as new onset of heart failure or graft dysfunction according to Invasive Monitoring Attenuation Through Gene Expression (IMAGE) criteria (4).

Back to Top | Article Outline

Gene Expression Profiling

The GEP test uses quantitative real-time polymerase chain reaction technology to measure the expression of 11 rejection markers and 9 control genes (1). These genes were originally selected from microarray analysis of peripheral blood mononuclear cells and subsequently verified using real-time polymerase chain reaction. Using a multigene algorithm, a score ranging from 0 to 40 is reported, with higher scores associated with greater probability of acute cellular rejection. The score has been shown to discriminate between moderate or severe (International Society of Heart and Lung Transplantation [ISHLT] grade ≥2R) acute cellular rejection and the absence of acute cellular rejection (quiescence) in a large observational study (1). The area under the curve was 0.67 with 95% confidence interval from 0.56 to 0.78. The test has been used clinically since January 2005 (3).

Back to Top | Article Outline

Statistical Analysis

To evaluate the extent to which GEP test scores from the same patient are correlated, we reviewed the Columbia and Hahnemann longitudinal data. Model-based estimates of within-person variance (sw 2), between-person variance (sb 2), and intraclass correlation (ICC=sb 2/[sb 2+sw 2]) were obtained using one-way random effects analysis of variance. Analysis of variance models with patient treated as a random effect were applied to all the data and additionally to the subset of scores having a corresponding biopsy with ISHLT grade=0R and to the subset with ISHLT grade=1R.

Back to Top | Article Outline


A total of 1135 GEP tests performed for 238 patients aged 15 to 81 years from 55 days to more than 20 years posttransplant were analyzed. An average of 4.8 GEP tests and 1.25 recorded biopsy grades per patient were retrieved for these patients. The median interval between consecutive tests from the same patient was 4.6 months (range 0.3–28.0 months). Table 1 shows estimates of variance components and calculated ICC=0.46 for the total data set (all), ICC=0.53 for the subset with ISHLT grade=0R, and ICC=0.60 for the subset with ISHLT grade=1R.



Back to Top | Article Outline

Discussion and Conclusions

Longitudinal AlloMap scores from the same patient in a clinically stable condition without clinical heart transplant rejection have significantly less variation over time than scores from different patients. The ICC estimate using all the data suggests that a considerable proportion (46%) of the total variation can be attributed to heterogeneity between individual patient mean GEP scores. Furthermore, although only a minority of the scores had associated biopsy grades, estimates of ICC suggest that the percentage of between-person variance remains substantial (53%–60%) in subsets of the data where the recorded ISHLT biopsy grade was consistent with having no or mild rejection. These retrospective data support the hypothesis that stable heart transplant recipients without clinical signs and symptoms of rejection display consistent leukocyte GEP patterns over time. This observation suggests that individualized patient-specific score ranges may be established based on their historical GEP measurements. Although proposing a method for incorporating prior GEP scores into the clinical decision-making process is beyond the scope of this discussion, one possibility might be to use an approach such as in Ref. 6 to augment the predictions based on the currently accepted population-based reference scores (1, 2) through joint modeling of longitudinal GEP response and rejection event histories.

Mario C. Deng1

Gregory Alexander2

Hans Wolters2

Khurram Shahzad1

Martin Cadeiras1

Albert Hicks3

Theresa Rowe3

Tod M. Klingler2

Howard J. Eisen3

1 Department of Medicine

Columbia University College of Physicians and Surgeons

New York, NY

2 Research Informatics


Brisbane, CA

3 Department of Medicine, Division of Cardiology

Drexel University College of Medicine

Philadelphia, PA

Back to Top | Article Outline


1. Deng MC, Eisen HJ, Mehra RM, et al; for the CARGO Investigators. Non-invasive detection of rejection in cardiac allograft recipients using gene expression profiling. Am J Transplant 2006; 6: 150.
3. Starling RC, Pham M, Valantine H, et al. Molecular testing in the management of cardiac transplant recipients: Initial clinical experience [invited editorial]. J Heart Lung Transplant 2006; 25: 1389.
4. Pham MX, Teuteberg JJ, Kfoury AG, et al; for the IMAGE Study Group. Gene expression profiling for rejection surveillance after cardiac transplantation. N Engl J Med 2010; 362: 1890.
5. Deng MC, Halpern B, Wolters H, et al. Patient-specific longitudinal pattern of AlloMap test scores [abstract]. J Heart Lung Transplant 2009; 28: S229.
6. Proust-Lima C, Taylor JMG. Development and validation of a dynamic prognostic tool for prostate cancer recurrence using repeated measures of posttreatment PSA: A joint modeling approach. Biostatistics 2009; 10: 535.
    © 2010 Lippincott Williams & Wilkins, Inc.