Abstract 463

We have recently described an algorithm combining biomarkers and clinical characteristics, such as severity, that, when assessed at the onset of GVHD symptoms, predicts outcomes such as response to treatment and non-relapse mortality [Levine, Blood, 2012]. This approach maximizes the information (both laboratory and clinical) utilized to develop the predictive algorithm. We have now evaluated whether a combination of pre-HCT clinical characteristics and GVHD biomarkers measured early post-HCT can predict the future occurrence of GVHD with sufficient accuracy and lead time to facilitate the preemptive treatment of a large majority of patients at high risk prior to the onset of GVHD symptoms.

We studied 393 patients who received a first allogeneic HCT from related donors between 2001 and 2011 at the University of Michigan whose clinical data and plasma samples were collected prospectively on an IRB-approved protocol. The cumulative incidence of grade 2–4 GVHD by day 100 was 31% (n=121; biopsy proven in 91%); at onset of symptoms (median = day 40), GVHD affected skin alone in 40 pts (33%), GI tract alone in 48 (40%), liver alone in three (2%), and multiple organs in 30 (25%). We divided patients randomly into a training set (n=264) to develop an algorithm to predict acute GVHD using clinical and biomarker parameters by logistic regression, and a validation set (n=129) to test the algorithm. There were no statistically significant differences in pre-HCT clinical parameters between the two sets. We measured a panel of 4 informative GVHD biomarkers (IL-2Rα, TNFR1, elafin, and REG3α) on plasma samples taken at day 7 after HCT by ELISA.

We postulated that day 7 biomarker concentrations would be most accurate in predicting early acute GVHD, and we therefore used grade 2–4 GVHD onset in the first two months post-HCT as the predictive endpoint. We developed three predictive models for GVHD in the training set: one model used 5 validated clinical characteristics alone (patient age, graft source [BM], HLA-match, GVHD prophylaxis, myeloablative conditioning [MA], use of TBI [Jagasia, Blood 2012]); the second used 4 biomarker concentrations alone; and the third used the combination of all 9 clinical and biomarker parameters. We postulated that an algorithm with a specificity of 50% or higher would be useful, so we compared all three models at a specificity of 50% for their respective sensitivities. The 5 pre-HCT characteristics alone provided 51% sensitivity; 4 biomarkers alone provided 66% sensitivity; and the combined 9 clinical and biomarker parameters provided 77% sensitivity (Table; combined vs clinical model, p<0.001; vs biomarker model, p=0.07). We then tested the sensitivity of the combined model in the validation set and found it to be 75%. The similar sensitivities in the two sets allowed us to combine them for further analyses.

We used the algorithm to classify patients as “high risk” (n=211) or “low risk” (n=182) for the occurrence of acute GVHD. Patients with a high risk panel were much more likely to develop grade 2–4 GVHD (Figure 1A; cumulative 38% vs 20%, p<0.001) which developed almost one month earlier than in the low risk group (median day of onset = 39 vs 65). The greater incidence of acute GVHD in high risk patients resulted in significantly greater non-relapse mortality (NRM) by day 180 post-HCT (12% vs 3%; p=0.001; Figure 1B). The relapse rate was identical in both groups (24%) and thus overall survival was significantly better in the low risk group (84% vs 73%, p=0.004). At one year post-HCT, the differences between high and low risk patients remained significant for NRM (17% vs 6%, p<0.001) and overall survival (61% vs 72%, p=0.01).

In conclusion, the best algorithm to predict acute GVHD in recipients of related donor HCT combined a panel of 4 biomarkers at day 7 after HCT and 5 pre-HCT clinical characteristics. The algorithm successfully stratified patients into high and low risk groups for acute GVHD, six month NRM and overall survival. We hypothesize the use of such an algorithm one week after HCT may permit preemptive treatment of patients who are at greatest risk early in their transplant course. We are currently analyzing whether biomarkers at a later time point will identify patients at greatest risk for late-onset GVHD.


Combined predictive algorithm

Score = -3.57 + 0.54xAge–16.83xBM + 1.35xMismatch - 0.08xGVHD prophy + 0.35xMA + 0.47xTBI + 0.37xlogIL2Rα – 0.06xlogTNFR1 – 0.12xlogElafin – 0.03xlogREG3α 
Score = -3.57 + 0.54xAge–16.83xBM + 1.35xMismatch - 0.08xGVHD prophy + 0.35xMA + 0.47xTBI + 0.37xlogIL2Rα – 0.06xlogTNFR1 – 0.12xlogElafin – 0.03xlogREG3α 


No relevant conflicts of interest to declare.

Author notes


Asterisk with author names denotes non-ASH members.

This icon denotes a clinically relevant abstract