Levine, John E.

ABSTRACT Approximately 30% of patients with acute graft-versus-host disease (GVHD) develop steroid-refractory disease and have very poor outcomes. Ruxolitinib has become the standard of care for steroid-refractory acute GVHD, but it is unclear which patients derive benefit. The MAGIC Composite Score (MCS), an algorithm that combines clinical symptoms and biomarkers, has been validated to predict outcomes at the start of primary GVHD treatment. Here, we evaluated its performance at the initiation of second-line treatment in 278 patients. MCS stratified patients into three risk groups (MCS1-3), with the majority (88%) classified as intermediate or high risk. Increasing MCS score was associated with progressively higher 1-year non-relapse mortality (NRM) rates (16%, 41%, and 73%; p<0.001), lower 1-year survival (77%, 56%, and 24%; p<0.001), and lower complete response (CR) rates at day 28 (47%, 38%, and 20%, respectively; p<0.01). The area under the receiver operating characteristic curve (AUROC) for 1-year NRM was significantly higher with MCS compared to clinical symptoms alone (0.70 vs. 0.63; p=0.023). Among patients treated with ruxolitinib, higher MCS similarly predicted higher NRM and lower survival and CR rates. Patients classified as MCS2/3 had poor outcomes despite ruxolitinib, underscoring the need for novel therapies in this patient population. In conclusion the MCS is an accurate predictor of outcomes for patients who require second-line treatment and may be of use as an eligibility criterion for future clinical trials in this high-risk population. Key Points MAGIC Composite Scores that integrate clinical symptoms and biomarker scores predict outcomes of second-line therapy for acute GVHD more accurately than clinical symptoms alone. MCS identifies patients with high NRM despite treatment with ruxolitinib.

Abstract Background: A randomized multicenter trial (BMT CTN 1703) demonstrated that after reduced intensity conditioning HCT, PT-Cy-based GVHD prophylaxis led to significant reductions in acute and chronic GVHD, but was associated with significantly more Grade 2 (moderate) infections and similar overall survival vs Tac/MTX. A linked mechanistic study (BMT CTN 1801) found that PT-Cy led to pan-T cell depletion and reduced T cell receptor (TCR) repertoire diversity vs Tac/MTX. To discover the molecular mechanisms driving the association of PT-Cy with increased infections, we examined pathogen-specific T cell reconstitution in BMT CTN 1801 patients (n=165 PT-Cy, n=159 Tac/MTX). Methods: We performed deep TCR sequencing on 2,369 blood samples collected pre-HCT (recipient baseline) and on Days +7, 14, 28, 63, 100, 180, 270, 365, 730, and from the graft infusion product. We then applied a novel set of TCR classifiers (Adaptive Biotechnologies) capable of identifying pathogen-specific clones associated with CMV, EBV, SARS-CoV-2, HSV-1, HSV-2, parvovirus, RSV, influenza A, rhinovirus, adenovirus, HHV-6B, norovirus and Toxoplasma. Single-cell (sc) TCR/RNA-sequencing on a sample subset (n=26) was performed to determine the phenotype of pathogen-specific T cells. Results. By linking scRNA-seq with pathogen-specific TCR barcoding, we identified CD4 and CD8 memory T cells as their predominant T cell reservoir. We found that the total number of pathogen-specific singleton TCRs was significantly lower with PT-Cy, starting early post-HCT and persisting through 2 years. For example, at Day +14, PT-Cy patients had 7.8-fold fewer pathogen-specific singleton TCRs vs Tac/MTX (mean of 6.9 vs 54.1, p&lt;0.001). Consistent differences between PT-Cy and Tac/MTX were observed for TCRs specific for each pathogen individually. Patients who developed Grade 3 (severe) viral infections had significantly fewer pathogen-specific singleton TCRs at Day +14 compared to those who had no viral infections (mean of 4.0 vs 32.5, p=0.03). Leveraging intrinsic TCR barcoding of the HCT infusion (“graft”) and of the patient's baseline samples (“recipient”), we could assign the graft/recipient origin of the pathogen-specific cells with high fidelity. This analysis revealed pronounced differences in the origin of pathogen-specific T cells with PT-Cy vs Tac/MTX, especially prior to Day +100. The absolute number of pathogen-specific TCRs originating from the graft was significantly higher with Tac/MTX vs PT-Cy. This difference was observed from day +7 through 6 months post-HCT, with up to 7.1-fold more graft-derived pathogen-specific TCRs in Tac/MTX vs PT-Cy (p&lt;0.001 at each timepoint before day +100). Within patients who received Tac/MTX, the balance of graft vs recipient pathogen-specific clones significantly favored graft clones (for example, a mean of 14.9 graft vs 8.6 recipient pathogen-specific TCRs at Day+14, p=0.02). By contrast, with PT-Cy, in addition to having substantially fewer pathogen-specific TCRs overall, the proportion of these TCRs was not significantly different between graft and recipient clones (mean of 2.1 graft vs 3.3 recipient pathogen-specific TCRs at Day+14, p=0.42). These data uncover that, in addition to the reduction in pathogen-specific T cells with PT-Cy, there was a relative skewing of the anti-microbial T cell reservoir towards recipient cells that persisted after conditioning. Conclusions: PT-Cy led to in vivo T cell depletion across all T cell subsets, which prominently included clones tracked from the allograft itself into the HCT recipient, with a relative sparing of recipient T cell clones that survived transplant conditioning. While the mechanism for relative sparing of recipient cells has not yet been determined, these conditioning-exposed cells may be less proliferative than newly infused graft T cells. The depletion of graft T cells included CD4 and CD8 memory cells specific for infections, which may have contributed to relative deficits in protective immunity against infection in patients receiving PT-Cy. Importantly, these data suggest that with PT-Cy, persisting recipient TCRs may play a key role in preserving anti-infectious immunity. Taken together, these results provide a mechanistic basis for key post-transplant clinical outcomes of the landmark BMT CTN 1703 study. They also underscore the importance of recipient immune status with PT-Cy, creating the opportunity for targeted infection risk assessment in these patients.

Abstract Background Acute graft-versus-host disease (GVHD) is a major cause of non-relapse mortality (NRM) after allogeneic hematopoietic cell transplantation (HCT). High-dose steroids are standard first-line therapy for grade II–IV acute GVHD, whereas systemic treatment (tx) is generally not recommended for grade I. However, in practice many patients (pts) with grade I GVHD receive systemic steroids. A prior GITMO randomized trial (n=173) of grade I GVHD to either systemic steroid tx or “watch and wait” found that preemptive steroid tx reduced progression to grade II but not grade III/IV GVHD, increased serious infections, and resulted in similar chronic GVHD, NRM and survival (Bacigalupo, Haematologica 2017). In this study we analyzed real world evidence for early systemic steroid use and the potential impact of serum biomarkers in grade I GVHD. Methods We analyzed clinical data and serum samples from pts with grade I acute GVHD at 24 Mount Sinai Acute GVHD International Consortium (MAGIC) centers between 2014 and 2022 who were treated according to physician discretion. Pts were divided into a preemptive steroid tx group or a watch-and-wait group (systemic steroids administered only if progression to grade II–IV GVHD). Preemptive steroid use was treated as a time-dependent covariate and a frailty proportional hazards model with random effects was used to adjust for center effects to account for possible heterogeneity in clinical practices across transplant centers. Serum samples collected at the diagnosis of grade I GVHD were analyzed for ST2 and REG3α to calculate validated Ann Arbor (AA) scores (1, 2, 3). Results Grade I GVHD was diagnosed in 1143 pts; 591 (52%) received preemptive tx and 552 (48%) were treated by a watch-and-wait strategy. In the preemptive tx group, steroids were initiated at a median of 0 days and within 7 days for &gt;90%. The median initial dose of steroids was 1 mg/kg of prednisone or its equivalent. In the watch and wait group, only 184 pts (33%) required systemic steroids for progression to grade II-IV GVHD at a median of 14 days from grade I diagnosis. Preemptive steroid tx reduced the risk of grade II–IV acute GVHD (HR, 0.71; 95% CI, 0.56–0.89; P=0.003), but not the risk of grade III/IV acute GVHD (HR, 1.03; 95% CI, 0.69–1.52; P=0.899) or chronic GVHD (HR, 1.03; 95% CI, 0.69–1.52; P=0.899), consistent with the GITMO trial. The 12-month incidence of NRM was higher for preemptive tx vs. watch and wait (13% vs 7%, P=0.002). In multivariate analysis preemptive steroid use independently predicted an increased risk of NRM (HR, 1.63; 95% CI, 1.06–2.50; P=0.026) as did several other known risk factors (older recipient age, umbilical cord blood donor, HCT-CI ≥3, and myeloablative conditioning). Greater NRM in the preemptive tx group was driven primarily by 3-fold more infectious deaths; deaths due to uncontrolled GVHD were similar for both tx groups. The risk for relapse was unexpectedly lower in the preemptive tx group (HR 0.64; 95% CI 0.47-0.88; P=0.006) and overall survival was therefore not different between groups, but risk factors for relapse such as minimal residual disease status were not available precluding further analysis of this finding. Serum was collected in 871 (76%) pts at onset of grade I GVHD for retrospective biomarker analysis. AA scores were AA1 (80%), AA2 (15%), and AA3 (5%). AA scores were highly predictive of progression to grade III/IV GVHD (AA1 vs AA2 vs AA3: 9% vs 14% vs 44%, all P&lt;0.001) and the risk of 12-month NRM (8% vs 16% vs 39%, all P&lt;0.001); this pattern was observed in both tx groups. The negative effect of preemptive steroid tx was most pronounced in pts with grade I/AA1 GVHD who experienced 5-fold more infectious related deaths compared to pts managed with a watch and wait approach. Conclusion Treatment of grade I GVHD with systemic steroids did not reduce the incidence of grade III/IV GVHD or deaths due to uncontrolled GVHD and instead increased the risk of NRM from infections compared to pts who were treated only after GVHD progressed. Importantly, 2/3 of the watch and wait pts never needed systemic steroid tx. These findings remained robust after adjusting for multiple confounders, including center effects, and were consistent with the GITMO trial. MAGIC biomarkers at the onset of grade I GVHD predicted outcomes and a watch and wait strategy may be best for the 80% of pts with AA1 GVHD. Better treatment options are needed for the 20% of pts with grade I and AA2/3 GVHD.

Abstract Introduction: Graft Failure (GF) is a rare but devastating outcome of HCT. Primary GF (PGF), defined as a failure of neutrophil recovery by Day+28, is straightforward to diagnose. Diagnosing secondary GF (SGF) is more challenging, with its broad time-range, multiple confounding diagnoses, and lack of predictive biomarkers. In BMT CTN1703/1801, we analyzed patients receiving RIC HCT for heme malignancies with either Tac/MTX (n=159) or PT-Cy (n=165) GVHD prophylaxis. SGF was defined as donor chimerism &lt;5% after initial donor engraftment. With Tac/MTX, there were 3 PGF and 1 SGF diagnoses. With PT-Cy, there were 4 PGF and 6 SGF. Median SGF diagnosis was Day +64 (range: Day+28-215). There were too few Tac/MTX patients to analyze SGF, but sufficient events with PT-Cy. To identify PT-Cy SGF predictors, we leveraged lymphocyte, T, B, and NK cell reconstitution analysis. In SGF we found an early, profound deficit in the reconstitution of all major lymphocyte populations, including total lymphocytes, T, B, and NK cells, as early as Day+28. This enabled the modeling of a SGF risk classifier based on the Absolute Lymphocyte Count (ALC). Methods: Clinical ALC measurements were performed on all SGF patients (n=6) and on non-GF controls with ALCs available (146 of 155 non-GF patients). Flow cytometry was performed on all PT-Cy SGF patients (n = 6) and from a subset of non-GF controls (n = 18: controls were chosen as patients without relapse or severe GVHD, to reduce confounders introduced by immunologic interventions, and for whom all samples, including from the graft infusion, were available). T, B, and NK counts, as well as T cell subsets, were compared (using Welch's T test) on Days 7, 14, 21, 28, 42, 63, 98, 180, 270, 365, 730, with SGF patients censored on the day of GF diagnosis. To interrogate the optimal ALC cutoff, the cumulative incidence of SGF was computed at Days +28 and +42, with death without GF as a competing risk. Results: We have previously demonstrated that, compared to Tac/MTX, PT-Cy patients exhibited an early decrease in reconstitution of all T cell populations (with normal NK and B cell reconstitution). Here we focused specifically on PT-Cy patients with or without SGF. Graft CD34 counts/kg were not different between SGF patients and non-GF patients (mean CD34/kg = 1.38 x10e6 (SGF) vs 0.67 x10e6 (non-GF, p = 0.61) However, even amidst the overarching early suppression in T cell reconstitution with PT-Cy, SGF patients could be easily distinguished from the larger PT-Cy cohort, based on more profound deficits in ALC, T, B and NK cells reconstitution, measured using 2 strategies: (1) SGF patients demonstrated significant early (Day +28) quantitative defects in the reconstitution of the ALC (mean +/- SEM 262+/-32 cells/µL (non-GF) vs 60 +/- 25 (SGF, p&lt;0.0001), CD4 T cells (62 +/- 12 cells/µL vs 19+/-15, p=0.048), CD8 T cells (15 +/-3 cells/µL vs 3.5 +/-1 p=0.0009), all CD8 T cell subpopulations, as well as NK cells (116+/-31 cells/µL vs 2.9+/-1.4 cells/µL, p= 0.002) and B cells (3.5 +/-1.4 cells/µL vs 0.16+/-0.07 p =0.03). (2) In SGF, there was a significant deficit in the rate of rise of all major lymphocyte populations between Days 28-42-60 vs non-GF, including CD4 T cells (p&lt;0.0001), CD8 T cells (p = 0.0002), B cells (p&lt;0.0001), and NK cells (p =0.048). These discoveries suggested that a classifier could be identified to risk-stratify patients for SGF. We explored an ALC cutpoint, amenable to standard clinical lab analysis. A statistically significant threshold was identified at both Days+28 and +42, with Day+42 being most predictive: A threshold of 120 cells/µL was identified as optimal, with landmark analysis documenting a SGF rate of 34.7% below the cutpoint, and SGF of 1% above it (HR = 45.9, 95% CI 5.7 - 366). Conclusions: Despite the small number of events, PT-Cy patients with SGF demonstrated a distinctive reconstitution trajectory that encompassed an early, substantial, and sustained deficit in all lymphocyte counts, as well as a lack of their longitudinal expansion. This enabled the discovery of a candidate ALC biomarker cutpoint at Day+42 that could distinguish patients who were more likely to develop SGF. If confirmed, these data could generate a predictive biomarker for SGF, which would enable the design of trials evaluating early interventions (e.g. CD34+ boosts, DLI, modification of immunosuppression) to improve outcomes for these patients.

Abstract Acute kidney injury (AKI) frequently follows hematopoietic cell transplantation (HCT). Soluble urokinase-type plasminogen activator receptor (suPAR) is a biomarker of AKI in the general population. We evaluated suPAR and its association with AKI requiring dialysis (AKI-D) in patients undergoing allogeneic HCT (alloHCT). Performance of suPAR was compared with serum creatinine (sCr) and neutrophil gelatinase-associated lipocalin (NGAL). Data were obtained from the Blood and Marrow Transplant Clinical Trials Network 1202 cohort, an observational study of 1709 alloHCT recipients. Adults aged ≥18 years with AKI-D after HCT were included. Adults who did not develop AKI were included as controls and matched 1:1. Periodic serum samples (7-90 days) were analyzed for NGAL, suPAR, and sCr. The 1:1 matched case-control groups (n = 62 each) were balanced in demographic variables, except for graft-versus-host disease prophylaxis. The median time from transplant to AKI-D was 2.6 months (range, 0.03-20.39). The day +7 suPAR level after HCT was higher in patients with AKI (median, 2.7 ng/mL) than in controls (2.1 ng/mL; P = .002). In the multivariate analysis, the day +7 suPAR level was associated with the development of AKI-D (P = .009). The area under the curve for the receiver operating characteristic curve for day +7 suPAR levels was 0.75. Neither NGAL nor sCr were associated with AKI-D. Elevated day +7 suPAR levels predicted AKI and lower overall survival (OS). The suPAR level at day +7 after HCT may be an early prognostic factor for the development of AKI-D and OS. Future prospective studies could evaluate this at different stages of AKI. This trial was registered at www.ClinicalTrial.gov as #NCT01879072.

Abstract Changes in the clinical symptoms of acute graft-versus-host disease (GVHD) are currently used to assess treatment responses. The Mount Sinai Acute GVHD International Consortium (MAGIC) consortium has recently revealed that the integration of serum biomarkers with clinical symptoms at the onset of treatment in a MAGIC composite score (MCS) more accurately predicts treatment response and 6-month nonrelapse mortality (NRM) than clinical symptoms alone. In this study, we evaluated whether the integration of serum biomarkers and clinical symptoms on day 28 (D28) would also better predict NRM than clinical response only (CRO). We analyzed data from 1135 patients receiving systemic treatment for acute GVHD and created a fourth MCS category for patients with complete resolution of symptoms and low-risk clinical biomarkers on D28. Using a classification and regression tree model with 6-month NRM as the end point, we identified status of MCS 0 or MCS 1 at D28 as responses, which we termed the MAGIC composite response (MCR). In the validation cohort (n = 309), MCR more accurately predicted 6-month NRM than CRO (area under the curve: 0.77 vs 0.69; P = .014) and demonstrated higher negative and positive predictive values. MCR correctly reclassified both clinical nonresponders and responders: 28 of 213 clinical responders (13%) became nonresponders with fivefold higher NRM (34.3% vs 6.8%, P &lt; .001) and a larger group (29/96, 30%) of clinical nonresponders became responders with sixfold lower NRM (7.6% vs 50.7%, P &lt; .001). These findings support the use of MCR as a superior surrogate end point for long-term GVHD control and survival in future clinical trials.

Abstract Overall response (OR) that combines complete (CR) and partial responses (PR) is the conventional end point for acute graft-versus-host disease (GVHD) trials. Because PR includes heterogeneous clinical presentations, reclassifying PR could produce a better end point. Patients in the primary treatment cohort from the Japanese Society for Transplantation and Cellular Therapy (JSTCT) were randomly divided into training and validation sets. In the training set, a classification and regression tree algorithm generated day 28 refined response (RR) criteria based on symptoms at treatment and day 28. We then evaluated RR for primary and second-line treatments, using the area under the receiver operating characteristic curve (AUC) and negative predictive value (NPV) for 6-month nonrelapse mortality as performance measures. RR considered patients with grade 0/1 at day 28 without additional treatment as responders. RR for primary treatment produced higher AUCs than OR with small improvement of NPVs in both validation sets: JSTCT (AUC, 0.73 vs 0.69 [P &lt; .001]; NPV, 92.0% vs 89.6% [P &lt; .001]) and the Mount Sinai Acute GVHD International Consortium (MAGIC; AUC, 0.71 vs 0.68 [P = .032]; NPV, 90.9% vs 89.8% [P = .009]). RR for second-line treatment produced similar AUCs but much higher NPVs than OR in both validation sets of JSTCT (AUC, 0.64 vs 0.63 [P = .775]; NPV, 74.5% vs 66.0% [P &lt; .001]) and MAGIC (AUC, 0.67 vs 0.64 [P = .105]; NPV, 86.8% vs 76.1% [P = .004]). Classifying persistent but mild skin symptoms as responses and residual lower gastrointestinal GVHD as nonresponses were major drivers in improving the prognostic performance of RR. Our externally validated day 28 RR would serve as a better end point than conventional criteria in future first- and second-line treatment trials.

Background: The majority of patients respond to steroid treatment for graft-vs-host disease (GVHD), but the typically high doses and prolonged courses can cause substantial morbidity. GVHD can be classified at onset into two groups with significant differences in treatment response and non-relapse mortality (NRM) by clinical criteria (Minnesota risk) but the more favorable standard risk group still experiences high rates of treatment failure and NRM [MacMillan, BBMT 2015]. We previously validated a serum biomarker-based (ST2+REG3α) risk stratification system, MAGIC algorithm probabilities (MAPs), that can stratify patients both at initiation and during treatment into risk groups for treatment response and NRM [Hartwell, JCI Insight 2017, Srinagesh, Blood Adv 2019]. GVHD that is Minnesota standard risk and has a low MAP (i.e., Ann Arbor 1) at start of treatment represents a low risk group with &gt;80% response rate to steroids and ~10% NRM [Etra, Blood 2023]. We hypothesized that serial monitoring of GVHD symptom severity and MAPs twice (days (d) 7 and 14) in patients with clinical and biomarker defined low risk GVHD could further stratify patients into clinically meaningful groups. Methods: We retrospectively determined MAPs at d7 and 14 in 450 patients with low risk GVHD who participated in a prospective, multi-center, observational trial and were treated with systemic steroids between 2015-2023. MAPs &lt;0.291 at days 7 and 14 were defined as low risk based on a previously validated threshold [Major-Monfried Blood 2018]. Clinical responses were scored as complete (CR), partial (PR), or non-response (NR) by standard criteria. Initiation of second line therapy was considered NR. Patients were categorized into three groups (A-C) based on clinical responses and MAPs at days d7 and 14. A: CR/PR by d14 and low MAP at both d7+14; B: stable or worse GVHD (NR) by d14 with low MAP at both d7 and 14; C: any response with a high MAP at d7 or 14. The endpoints were clinical response at day 28 (overall response rate, ORR=CR or PR), regardless of earlier response status, and 6-month NRM. Results: The median age of the study population was 54 years (range: 0-79); 19% of patients received PT-Cy for GVHD prophylaxis. Target organ involvement was skin ± upper GI (66%), lower GI (23%), isolated UGI (11%), and skin+liver (1%), and GVHD severity was grade I (35%), II (61%), and III (4%) at start of treatment. Patients in group A (n=310, 69%), clinical responders with low MAPs through day 14, had excellent d28 ORR (93%) and very low NRM (4%). GVHD in this group was considered ultra-low risk (ULR) and subset analyses revealed highly consistent d28 ORR and NRM, regardless of target organ severity, age (&lt;18, 18-59, &gt;60), HCT-CI score, conditioning intensity, and GVHD prophylaxis. Group B patients comprised clinical non-responders with low MAPs through day 14 (n=112, 25%); half of them responded by d28 and their NRM was low (8%), albeit double that of group A (p=0.079). Group C was small (n=28, 6%) and included both clinical responders (n=20, 4%) and clinical non-responders (n=8, 2%) who had at least one high MAP through day 14. Their d28 ORR was 54% but NRM increased four-fold (32%) and was significantly worse than patients who maintained low MAPs regardless of clinical response (p&lt;0.001). Conclusion: Serial monitoring by clinical response and biomarkers during the first two weeks of treatment identifies a small number of high risk patients whose GVHD is initially low risk (group C). This group experiences high NRM regardless of clinical response to steroid treatment and may benefit from treatment escalation. Patients with low risk GVHD who do not have an early clinical response to steroids but maintain low MAPs, (group B) have reasonably good long-term outcomes with standard treatment. Finally, serial monitoring identifies a very large subset of patients with ULR GVHD (group A) who respond exceptionally well to standard treatment and experience very low rates of 6-month NRM regardless of pre-transplant characteristics and GVHD prophylaxis. Such patients may benefit from treatment de-escalation strategies such as shorter courses of lower dose steroids, a strategy we are currently testing in a clinical trial (NCT05090384).

Background Endpoints for GVHD treatment trials use early response (i.e, day (D)28 of treatment) as a surrogate for long term survival. Overall response (OR) that combines complete (CR) and partial responses (PR) at day (D) 28 is the conventional endpoint for primary acute GVHD trials because of similar survival outcomes between CR and PR. D28 OR ignores symptom severity at baseline and D28, however, and the PR category encompasses a wide range of responses from minimal improvement to near complete resolution of symptoms. D28 OR is the only endpoint currently approved by the FDA for efficacy, but it has not been re-evaluated as an endpoint for primary treatment (1L) in fifteen years and little is known regarding the prognostic value of D28 OR as an endpoint for second line treatment (2L) of GVHD. We hypothesized that refinement of the current D28 OR criteria, particularly through recategorization of PR in the GI tract, the main driver of nonrelapse mortality (NRM), would more accurately reflect disease control and better predict survival outcomes than current D28 response. Methods The Japanese Society for Transplantation and Cellular Therapy (JSTCT) collaborated with the Mount Sinai Acute GVHD International Consortium (MAGIC) to develop and validate D28 refined response (RR) criteria for 1L and 2L treatment (tx) in patients with Grades II-IV acute GVHD. To maximize the correlation between treatment success (i.e, response) with long term survival, our primary measured metric was the negative predictive value (NPV). We first randomly divided the 1L tx cohort of JSTCT into the training (n=2341) and validation sets (n=1156). In the training set we used a classification and regression tree (CART) algorithm to generate RR criteria that predicted 6-month (mo) NRM based on symptoms both at tx onset and D28. We then compared the two D28 response endpoints (RR vs OR) as surrogates for 6-mo NRM in two validation sets of both 1L tx (JSTCT: n=1156; MAGIC: n=1141) and 2L tx (JSTCT: n=510; MAGIC: n=196). Results The CART algorithm defined a RR as GVHD Grades 0/I at D28 without additional tx; all other D28 grades were considered refined nonresponses. As expected, RR successfully stratified the risk of 6-mo NRM in both 1L tx validation sets (JSTCT: 8% vs 38%, p&lt;0.01; MAGIC: 9% vs 37%, p&lt;0.01). We defined a discordant nonresponse as a nonresponse by RR criteria but a response by OR criteria, which accounted for the majority (~75%) of differences between RR and OR. 6-mo NRM in discordant nonresponders was four fold higher than in responders (JSTCT: 35%; MAGIC: 30%). The large majority (&gt;80%) of discordant nonresponders had persistent GI GVHD at D28. Because the proportion of discordant nonresponders was small (&lt;7% of the total population) in the 1L setting, the improvements of an already high NPV were small (JSTCT: 93% vs 91%; MAGIC: 91% vs 90%). In the 2L tx validation sets, RR also stratified the risk of 6-mo NRM (JSTCT: 26% vs 56%, p&lt;0.01; MAGIC: 13% vs 55%, p&lt;0.01). At the time of 2L tx, the discordant nonresponders are more common and half of the discordant nonresponders experienced 6-mo NRM (JSTCT: 45%; MAGIC: 59%); as a result, RR achieved much higher NPVs than OR (JSTCT: 75% vs 66%; MAGIC: 87% vs 76%). Deaths in responders from acute GVHD (vs. other causes such as infection) were more frequent in OR than RR in both 2L tx validation sets (JSTCT: 41% vs 20%; MAGIC: 41% vs 13%) and 1L tx validation sets (JSTCT: 25% vs 18%; MAGIC: 32% vs 29%). D28 RR is superior to OR because it more accurately reflects control of serious GVHD in four independent validation cohorts, particularly in patients with GI disease. Conclusion We conclude that RR is a more accurate endpoint than OR in patients with grade II-IV GVHD primarily because it classifies persistent GI GVHD at D28 as a non-response instead of a response, which correlates with 6-mo NRM and achieves a higher NPV. This validated RR metric can serve as a better endpoint than OR in assessing treatment efficacy in both primary treatment trials that enrich for high risk GVHD, e.g. GI involvement, and second line treatment trials.

Abstract Acute graft-versus-host disease (GVHD) grading systems that use only clinical symptoms at treatment initiation such as the Minnesota risk identify standard and high-risk categories but lack a low-risk category suitable to minimize immunosuppressive strategies. We developed a new grading system that includes a low-risk stratum based on clinical symptoms alone and determined whether the incorporation of biomarkers would improve the model’s prognostic accuracy. We randomly divided 1863 patients in the Mount Sinai Acute GVHD International Consortium (MAGIC) who were treated for GVHD into training and validation cohorts. Patients in the training cohort were divided into 14 groups based on similarity of clinical symptoms and similar nonrelapse mortality (NRM); we used a classification and regression tree (CART) algorithm to create three Manhattan risk groups that produced a significantly higher area under the receiver operating characteristic curve (AUC) for 6-month NRM than the Minnesota risk classification (0.69 vs 0.64, P = .009) in the validation cohort. We integrated serum GVHD biomarker scores with Manhattan risk using patients with available serum samples and again used a CART algorithm to establish 3 MAGIC composite scores that significantly improved prediction of NRM compared to Manhattan risk (AUC, 0.76 vs 0.70, P = .010). Each increase in MAGIC composite score also corresponded to a significant decrease in day 28 treatment response (80% vs 63% vs 30%, P &lt; .001). We conclude that the MAGIC composite score more accurately predicts response to therapy and long-term outcomes than systems based on clinical symptoms alone and may help guide clinical decisions and trial design.