A prognostic score for acute graft-versus-host disease based on biomarkers: a multicentre study

Abstract

Background: Graft-versus-host disease (GVHD) is the major cause of non-relapse mortality after allogeneic haemopoietic stem-cell transplantation (SCT). The severity of symptoms at the onset of GVHD does not accurately define risk, and thus most patients are treated alike with high dose systemic corticosteroids. We aimed to define clinically meaningful risk strata for patients with newly diagnosed acute GVHD using plasma biomarkers.

Methods: Between April 13, 2000, and May 7, 2013, we prospectively collected plasma from 492 SCT patients with newly diagnosed acute GVHD and randomly assigned (2:1) using a random number generator, conditional on the final two datasets having the same median day of onset, into training (n=328) and test (n=164) sets. We used the concentrations of three recently validated biomarkers (TNFR1, ST2, and Reg3α) to create an algorithm that computed the probability of non-relapse mortality 6 months after GVHD onset for individual patients in the training set alone. We rank ordered the probabilities and identified thresholds that created three distinct non-relapse mortality scores. We evaluated the algorithm in the test set, and again in an independent validation set of an additional 300 patients who underwent stem cell transplant and were enrolled on multicentre clinical trials of primary therapy for acute GVHD.

Findings: In all three datasets (training, test, and validation), the cumulative incidence of 6-month non-relapse mortality significantly increased as the Ann Arbor GVHD score increased. In the multicentre validation set, scores were 8% (95% CI 3-16) for score 1, 27% (20-34) for score 2, and 46% (33-58) for score 3 (p<0·0001). Conversely, the response to primary GVHD treatment within 28 days decreased as the GVHD score increased 86% for score 1, 67% for score 2, and 46% for score 3 in the multicentre validation set, p<0·0001).

Interpretation: Biomarker-based scores can be used to guide risk-adapted therapy at the onset of acute GVHD. High risk patients with a score of 3 are candidates for intensive primary therapy, while low risk patients with a score of 1 are candidates for rapid tapers of systemic steroid therapy.

Funding: The National Cancer Institute, the National Heart, Lung, and Blood Institute, the National Institute of Allergy and Infectious Diseases, the Doris Duke Charitable Fund, the American Cancer Society, and the Judith Devries Fund.

FIGURE 1. Outcomes by Ann Arbor score at GVHD onset

Cumulative incidence of non-relapse mortality is shown for the 328 patient training set (A, Ann Arbor 1 (grey) vs 2 (blue), p<0·0001, 2 vs 3 (red), p=0·0081), the 164 patient testset (B, Ann Arbor 1 vs 2, p=0·0069, 2 vs 3, p=0·024), and the 300 patient multicenter validation set (C, Ann Arbor 1 vs 2, p=0·0023, 2 vs 3, p=0·0021). Cumulative incidence of relapse was not significantly different in the training set (D), testset (E), or multicenter validation set (F). One-year survival is shown for the training set (G, Ann Arbor 1 vs 2, p=0·028, 2 vs 3, p=0·015), the testset (H, Ann Arbor 1 vs 2, p=0·067, 2 vs 3, p=0·026), and the multicenter validation set (I, Ann Arbor 1 vs 2, p=0·0062, 2 vs 3, p=0·024).

FIGURE 2. Response rate to primary GVHD therapy by Ann Arbor score

Shown are the proportion of patients with complete or partial response for (A) the training set, (B) the testset, and (C) the multicenter validation set. Complete response rates are shown in panels D–F. The numbers in parentheses are the proportion of patients assigned to each Ann Arbor score within each set.

FIGURE 3. Outcomes for the multicenter validation set for each Glucksberg grade by Ann Arbor score

(1, grey; 2, blue; 3, red). Cumulative incidence of non-relapse mortality is shown for (A) Glucksberg grade I, p=0·0051; (B) grade II, p=0·0012; and (C) grade III/IV, p=0·087. P-values relate to any pairwise comparison of curves. The corresponding proportion of patients with complete or partial response are shown in Panels D–F. The numbers in parentheses are the proportion of patients assigned to each Ann Arbor score within each subset.

FIGURE 4. Comparison of Ann Arbor scores and Glucksberg grades to predict NRM risk (multicenter validation set)

(A) Multivariate comparison of Ann Arbor scoring (red) to Glucksberg grading (blue) for non-relapse mortality. Hazard ratios (diamonds) and their 95% confidence intervals [CI] (lines) for mild (grade I or Ann Arbor 1) and severe (grade III/IV or Ann Arbor 3) are shown relative to the reference group, moderate (II or 2) GVHD of each staging system. (B) ROC curves for the biomarker algorithm (red) or Glucksberg grades (blue) for prediction of non-relapse mortality. Diamonds indicate the thresholds that define Ann Arbor 1 and 3 GVHD. The AUC for the biomarker algorithm is 0.71 (95% CI, 0.64 to 0.75) and for Glucksberg grades is 0.57 (95% CI, 0.55 to 0.68).