Performance of treatment response assessment at day 7 by baseline acute graft-vs-host disease severity

Abstract

Background:

Response to first-line treatment in acute graft-versus-host disease (GVHD) is typically assessed at day 28 (D28) in clinical trials, but this convention was established without accounting for onset severity and thus was optimized for mild-moderate GVHD that comprises the majority of cases. Furthermore, the initiation of second-line therapy, which is considered primary treatment failure, is not based on standardized criteria and thus remains subjective and inconsistent for patients regardless of clinical trial participation.

Objective:

In this study we hypothesized that an early assessment of treatment response at D7 would accurately predict long-term outcomes for patients with severe GVHD and support the initiation of second line therapy in non-responders.

Study Design:

We analyzed six-month outcomes by D7 and D28 response for 1561 patients receiving systemic therapy for acute GVHD in two large trial cohorts - one observational (n=1008) and one interventional (n=553) after stratification for onset severity using Minnesota risk criteria.

Results:

Patients with Minnesota standard risk GVHD comprised approximately 80% of each cohort. D7 responses predicted much smaller differences in 6-month NRM (observational: responders: 9% vs non-responders: 23%; interventional: 12% vs 24%) than D28 responses (observational: 7% vs 35%; interventional: 9% vs 36%) and second line therapy was deferred in ~85% of patients who had not responded by D7. More than half of this “wait and see” group proved to be slow responders with low 6-month NRM of <10% and as a result the D28 response more accurately predicted 6-month NRM than D7 response (AUC: observational; 0.73 vs 0.63, p<0.001; interventional: 0.70 vs 0.60, p=0.002). Subset analyses confirmed the superiority of D28 over D7 in children with Minnesota standard risk GVHD and in patients with little or no lower gastrointestinal (GI) GVHD (stage 0 or 1) but not patients with stage 2 GI GVHD. In contrast, among Minnesota high risk patients, D7 (observational: 26% vs 54%; interventional: 20% vs 56%) and D28 (observational: 20% vs 57%; interventional: 22% vs 62%) responses both predicted large differences in 6-month NRM with similar AUCs (observational; 0.65 vs 0.69, p=0.171; interventional: 0.68 vs 0.71, p=0.581). Subset analyses demonstrated similar AUCs for both D7 and D28 in children with Minnesota high risk GVHD and in patients with severe GI GVHD (stage 2–4). Notably, second line therapy was deferred for 70% of high risk patients without a response at D7. The “wait and see” approach was common even after the approval of ruxolitinib for steroid-resistant GVHD, and their 6-month NRM was ~50%.

Conclusion:

These findings support the use of D7 response as an actionable assessment timepoint in high risk acute GVHD and highlight the need for severity-adapted response criteria in both clinical practice and trial design.

Introduction

Acute graft-vs-host disease (GVHD), a major cause of morbidity and non-relapse mortality (NRM) following allogeneic hematopoietic cell transplantation (HCT), affects approximately 45% of patients despite advances in prevention that have reduced the incidence of severe cases^1–5. High dose systemic corticosteroids continue to be the standard first line treatment⁶, despite multiple clinical trials of alternative therapies^7–12. Treatment response is commonly assessed by overall response (OR)¹³, which varies significantly according to initial disease severity - ranging from approximately 40% in patients with Minnesota high risk GVHD to 70% in those with Minnesota standard risk disease^14,15.

In clinical trials OR is typically assessed at day (D) 28 because earlier assessments (e.g., D7) were deemed too early to reflect true treatment effect¹⁶. However, this convention was established without accounting for GVHD onset severity and was thus effectively optimized for standard risk disease that comprises the majority of cases^13,16,17. Furthermore, the initiation of second-line therapy, considered as primary treatment failure, is not based on standardized criteria, and thus decisions remain subjective and inconsistent. These limitations highlight the need for evidence-based criteria to start second-line treatment in order to improve both patient outcomes and trial interpretation, especially in light of the recent availability of approved second-line options such as ruxolitinib and mesenchymal stromal cells^18,19. Minnesota risk categories are now commonly used as eligibility criteria for clinical trials for first line treatment of acute GVHD, but these categories had not yet been established when the D28 endpoint was promulgated. In this study, we assessed the performance of D7 OR in patients categorized by Minnesota risk at the start of treatment. We analyzed two large, multicenter cohorts of patients: first, those who received standard of care therapy and second those enrolled in clinical trials of first line treatment for GVHD. We also compared D7 OR to D28 OR to evaluate its potential as an early decision point for additional treatment according to GVHD severity.

Materials and Methods

Study Design

Two patient cohorts receiving systemic treatment for acute GVHD were analyzed after stratification by Minnesota risk classification. The first (observational trial) cohort included patients enrolled in the Mount Sinai Acute GVHD International Consortium (MAGIC) database and biorepository who underwent allogeneic HCT between January 2015 and December 2023, received systemic corticosteroid treatment for acute GVHD, and had complete GVHD staging data at the time of treatment initiation as well as at days 7 and 28 post-treatment. To enhance comparability to patients in the interventional trial cohort, we excluded patients if the initial corticosteroid dose was less than 0.5 mg/kg/d methylprednisolone equivalent (MPE) for Minnesota standard risk GVHD or if the initial corticosteroid dose was less than 1 mg/kg/d MPE for Minnesota high risk GVHD. Patients in the interventional cohort who were treated with alternatives to steroid treatment (itacitinib or sirolimus) were included because both of these treatments produced D28 response rates similar to high dose corticosteroid therapy. Data for the observational trial cohort were collected and reviewed according to the PRoBE design (prospective-specimen collection, retrospective-blinded-evaluation) as previously described²⁰. The clinical severity of acute GVHD was staged using published guidelines²¹.

The second (interventional trial) cohort consisted of patients enrolled in five multicenter primary GVHD treatment trials that enrolled patients between 2005 and 2021. Three of the trials were conducted by the Blood and Marrow Transplant Clinical Trials Network (BMT CTN): BMT CTN 0302 (NCT00224874), a phase 2 randomized trial that evaluated high dose corticosteroids in combination with one of four agents, etanercept, pentostatin, mycophenolate mofetil (MMF), and denileukin difitox, for grades I-IV GVHD¹²; BMT CTN 0802 (NCT01002742), a phase 3 randomized controlled trial that tested high dose corticosteroids plus either MMF or placebo for grades I-IV GVHD¹⁰; and BMT CTN 1501 (NCT02806947), a phase 2 randomized trial that compared sirolimus monotherapy to high dose corticosteroids for low risk GVHD defined by clinical and biomarker criteria¹¹. Two other interventional trials were conducted by MAGIC. The first trial (NCT03846479) evaluated the selective JAK1 inhibitor, itacitinib, as an alternative to systemic corticosteroids for low risk GVHD defined by clinical and biomarker criteria²². The second trial (NCT02133924) tested the combination of high dose corticosteroids and natalizumab for high risk GVHD defined by biomarker criteria⁹. We excluded all patients treated for grade I GVHD from both cohorts because such patients often respond to topical therapy and are commonly excluded from current primary treatment trials for GVHD^23,24. Consort diagrams for both cohorts are provided in Figures S1 and S2.

All studies were conducted in accordance with the Declaration of Helsinki and were approved by the Institutional Review Boards of all participating centers. All patients provided written informed consent before study enrollment.

GVHD Response

Complete response (CR) was defined as complete resolution of GVHD symptoms in all target organs and partial response (PR) was defined as improvement by at least one stage in at least one affected organ without worsening in any other organ. Non-response (NR) was defined as stable disease in all involved organs, progression in any organ, initiation of additional therapy, or death.²²

CR and PR at D28 are typically combined into a single OR category due to similar long-term outcomes.¹⁷ We validated this approach for D7 response across subsets of differing GVHD severity (Minnesota standard and high risk) using the observational trial cohort (Figure S3). Accordingly, patients achieving CR and PR were categorized as is OR, consistent with standard clinical trial practice, and those with NR as non-responders for both time points. The term “slow responders” was used to describe patients who were non-responders at D7 but achieved CR or PR by D28 without additional therapy.

Statistical Analyses

Cumulative incidences of NRM were calculated treating relapse and second allogeneic HCT as competing risks; likewise, relapse incidence was calculated treating NRM and second allogeneic HCT as competing risks²⁵. Differences between cumulative incidence curves were compared with the Gray test²⁶ and p-values were adjusted for multiplicity of comparisons with the false discovery rate²⁷. Overall survival was estimated using the Kaplan-Meier method, and differences between survival curves were compared using the log-rank test²⁸. We computed the area under the receiver operating characteristics curve (AUC) for the response assessments using 6-month NRM as the endpoint and time-dependent AUC for censored event times with competing risks²⁹. We compared time-dependent AUCs using non-parametric bootstrap test with 3000 iterations. All analyses were performed with the R computing language, version 4.3.3³⁰.

Results

Patient Characteristics

A total of 1,561 patients who received systemic treatment for acute GVHD were included in this analysis, comprising an observational trial cohort (n=1,008) and an interventional trial cohort (n=553). Several baseline characteristics differed between cohorts (Table 1). The observational trial cohort included a larger proportion of children (16% vs 3%), more unrelated (69% vs 57%) and haploidentical donors (12% vs 3%), and more use of post-transplant cyclophosphamide for GVHD prophylaxis (18% vs 12%) than the interventional trial cohort. These differences likely reflect the predominance of adult transplant centers participating in the interventional trials and the inclusion of more recently transplanted patients in the observational trial cohort. The prevalence of severe GVHD as defined by Minnesota criteria was similar between cohorts (observational: 21%; interventional: 23%) although there was more lower gastrointestinal (LGI) GVHD in the observational trial cohort (53% vs. 39%). Similar differences between the observational and interventional trial cohorts were observed when patients were stratified by Minnesota risk (Tables S1–2).

Table 1.

Patient Characteristics

	Observational (N=1008)	Interventional (N=553)
Age at BMT, years
Median [Min - Max]	55.0 [0.0 – 79.0]	52.0 [6.0 – 76.0]
Age groups
< 18 years	159 (16%)	14 (3%)
≥ 18 years	849 (84%)	539 (97%)
Sex
Female	435 (43%)	218 (39%)
Male	573 (57%)	335 (61%)
Primary Disease
Acute Leukemia / MDS	651 (65%)	373 (67%)
Lymphoma	80 (8%)	68 (12%)
Other Malignant	222 (22%)	71 (13%)
Other non-Malignant	55 (5%)	41 (7%)
Cell source
PBSC	763 (76%)	404 (73%)
Bone Marrow	209 (21%)	114 (21%)
Cord Blood	36 (4%)	35 (6%)
Donor Type
Related	315 (31%)	238 (43%)
Unrelated	693 (69%)	315 (57%)
HLA match
Matched	736 (73%)	436 (79%)
Mismatched	154 (15%)	98 (18%)
Haploidentical	118 (12%)	19 (3%)
Conditioning intensity
Full Intensity	585 (58%)	326 (59%)
Reduced Intensity	423 (42%)	227 (41%)
GVHD prophylaxis
CNI based	794 (79%)	481 (87%)
PTCy based	179 (18%)	67 (12%)
T-cell depletion	31 (3%)	5 (1%)
Other	4 (0%)	0 (0%)
Minnesota risk at Treatment
High	207 (21%)	126 (23%)
Standard	801 (79%)	427 (77%)
GVHD grade at treatment
II	711 (71%)	403 (73%)
III	244 (24%)	130 (24%)
IV	53 (5%)	20 (4%)
Organ involvement at treatment
LGI ± Other	539 (53%)	214 (39%)
Liver ± Skin ± UGI	33 (3%)	25 (5%)
Skin ± UGI	334 (33%)	262 (47%)
Isolated UGI	102 (10%)	52 (9%)

MDS: myelodysplastic syndrome, CNI: calcineurin inhibitor, PTCy: post-transplant cyclophosphamide, LGI: lower gastrointestinal, UGI: upper gastrointestinal

Impact of Initial GVHD Severity on the Utility of Early Response Assessment

We first analyzed patients with Minnesota standard risk GVHD (observational, n=801; interventional, n=427), who comprised approximately 80% of each cohort. We confirmed earlier studies and found that D7 responses predicted much smaller differences in 6-month NRM than D28 responses (Figure 1). Relapse rates in these patients were generally similar between the two groups (data not shown) and thus differences in overall survival were explained primarily by differences in NRM (Table S3). Of interest, second line therapy was deferred in ~85% of patients who had not responded by D7 and more than half of this “wait and see” group proved to be slow responders with low 6-month NRM of <10% (Table 2A). As a result of these slower responses the D28 response assessment was much better than the D7 assessment as measured by the AUC for 6-month NRM (observational: 0.73 vs 0.63, p<0.001; interventional: 0.70 vs 0.60, p=0.002). A subset analysis that included only children from both cohorts (n=135) revealed similar results (0.81 vs 0.65, p = 0.016).

Figure 1. Minnesota standard risk: Cumulative incidence of NRM by day 7 and 28 assessments.

Observational trial cohort (A-B): (A) D7 OR: responders: 9% (95% CI: 7% - 12%), non-responders: 23% (95% CI: 18% - 27%). (B) D28 OR: responders: 7% (95% CI: 5% - 9%), non-responder: 35% (95% CI: 29% - 42%). Interventional trial cohort (C-D): (C) D7 OR: responders: 12% (95% CI: 9% - 17%), non-responders: 24% (95% CI: 17% - 30%). (D) D28 OR: responders: 9% (95% CI: 6% - 12%), non-responders: 36% (95% CI: 27% - 45%). AUC comparisons are shown in Table 3.

Table 2.

Six-month NRM of non-responders at D7 who did not receive 2nd line treatment (“wait and see”)

A. Minnesota Standard Risk
	All “wait and see”		D28 responders “slow responders”		D28 non-responders
Trial cohort	n*	6-month NRM	n**	6-month NRM	n***	6-month NRM
Observational	259/314	19%	151/259	9%	108/259	32%
Interventional	141/162	19%	75/141	8%	66/141	32%
B. Minnesota High Risk
	All “wait and see”		D28 responders “slow responders”		D28 non-responders
Trial cohort	n*	6-month NRM	n**	6-month NRM	n***	6-month NRM
Observational	60/109	50%	19/60	32%	41/60	59%
Interventional	59/75	47%	31/59	29%	28/59	68%

^*number of D7 “wait and see”/ number of D7 non-responder

^**number of D28 responders/ number of D7 “wait and see”

^***number of D28 non responders/ number of D7 “wait and see”

The prognostic superiority of D28 compared to D7 assessments was clear in the subsets of patients who exhibited little or no lower GI (LGI) disease at GVHD onset but was lost among patients with significant (i.e., stage 2) LGI disease (Figure S4–6, Table 3). This observation led us to hypothesize that D7 and D28 responses would provide similar prognostic accuracy in patients with Minnesota high risk GVHD. Our analysis confirmed this hypothesis, showing large differences in 6-month NRM between groups predicted at both time points (Figure 2). Relapse rates were not significantly different by response status (data not shown) and thus response to treatment, regardless of whether it was assessed at D7 or D28, was the primary driver of differences in overall survival (Table S5).

Table 3.

D7 vs D28 AUC for 6-month NRM: Subsets of baseline GVHD severity

Minnesota risk	Subset	Trial cohort	D7 OR AUC (95% CI)	D28 OR AUC (95% CI)	P-value
Minnesota standard risk	All	Observational	0.63 (0.59 – 0.68)	0.73 (0.68 – 0.78)	<0.001
		Interventional	0.60 (0.54 – 0.66)	0.70 (0.64 – 0.76)	0.002
	LGI 0–1	Observational	0.63 (0.58 – 0.68)	0.73 (0.68 – 0.78)	0.001
		Interventional	0.57 (0.51 – 0.64)	0.69 (0.62 – 0.75)	<0.001
	LGI 2	Observational	0.67 (0.55 – 0.80)	0.74 (0.62 – 0.86)	0.236
		Interventional	0.84 (0.74 – 0.95)	0.79 (0.62 – 0.96)	0.616
Minnesota high risk	All	Observational	0.65 (0.58 – 0.71)	0.69 (0.63 – 0.75)	0.171
		Interventional	0.68 (0.60 – 0.76)	0.71 (0.63 – 0.79)	0.581
All	LGI 0–1	Observational	0.65 (0.60 – 0.69)	0.74 (0.70 – 0.79)	<0.001
		Interventional	0.59 (0.53 – 0.65)	0.68 (0.62 – 0.74)	0.003
	LGI 2–4	Observational	0.66 (0.60 – 0.72)	0.71 (0.65 – 0.77)	0.151
		Interventional	0.72 (0.65 – 0.80)	0.76 (0.68 – 0.84)	0.517
	Grade II	Observational	0.63 (0.58 – 0.68)	0.72 (0.67 – 0.77)	0.001
		Interventional	0.58 (0.51 – 0.64)	0.69 (0.63 – 0.75)	<0.001
	Grade III-IV	Observational	0.67 (0.62 – 0.73)	0.73 (0.67 – 0.78)	0.045
		Interventional	0.71 (0.64 – 0.78)	0.72 (0.65 – 0.80)	0.815

AUC: area under ROC curve, CI: confidence intervals

Figure 2. Minnesota high risk: Cumulative incidence of NRM by day 7 and 28 assessments.

Observational trial cohort (A-B): (A) D7 OR: responders: 26% (95% CI: 17%–34%), non-responders: 54% (95% CI: 44%–62%). (B) D28 OR: responders: 20% (95% CI: 13%–29%), non-responders: 57% (95% CI: 47%–66%). Interventional trial cohort (C-D): (C) D7 OR: responders: 20% (95% CI: 10%–31%), non-responders: 56% (95% CI: 44%–66%). (D) D28 OR: responders: 22% (95% CI: 13%–32%), non-responders: 62% (95% CI: 49%–73%). AUC comparisons are shown in Table 3.

Surprisingly, second line therapy was also delayed for ~70% of these high risk patients without a response at D7, particularly in interventional trials, even when GVHD severity had increased compared to baseline (data not shown). After ruxolitinib was approved for steroid resistant GVHD in 2019 the proportion of non-responding patients in the observational cohort that received second line therapy by D7 increased from 24% to 51%; no change was observed in the interventional cohort, however, possibly due to the influence of clinical trial requirements.

Unlike the results in patients with Minnesota standard risk at onset, this “wait and see” approach in high risk patients was associated with high 6-month NRM, even among the slow responders, and was ~50% overall (Table 2B). As expected, increasing numbers of patients in both cohorts initiated second line treatment between D7 and D28, particularly after the approval of ruxolitinib (Figure S7). The agents used for second line treatment are provided in Table S5.

Among Minnesota high risk patients (observational, n=207; interventional, n=126), the prognostic ability of the D7 response was comparable to that of the D28 response as measured by the AUC for six-month NRM (observational: D28: 0.69 vs D7: 0.65, p = 0.171; interventional: 0.71 vs 0.68, p=0.581); this finding persisted through 12 months (Figure 3). A subset analysis that included only children from both cohorts (n=38) revealed similar results (0.75 vs 0.67, p = 0.304). The D7 response assessment was also comparable to D28 assessment for other groups with severe GVHD as judged by MAGIC criteria (grade III/IV) (Table 3, Figures S8) or stage 2–4 LGI involvement at initiation of therapy (Table 3, Figure 4). Taken together, these analyses support treatment response at D7 as an actionable timepoint in patients with severe GVHD.

Figure 3. Time-dependent AUC for NRM in Minnesota High Risk patients.

Observational trial cohort (A): 6-month AUC: D7: 0.65 vs D28: 0.69, p = 0.171; 12-month AUC: D7: 0.64 vs 0.68, p = 0.267 Interventional trial cohort (B): 6-month AUC: 0.68 vs 0.71, p = 0.581; 12-month AUC: 0.67 vs 0.69, p = 0.657.

Figure 4. All LGI stage 2–4 GVHD: Cumulative incidence of NRM by day 7 and 28 assessments.

Observational trial cohort (A-B): (A) D7 OR: responders: 21% (95% CI: 15%–28%), non-responders: 50% (95% CI: 41%–59%). (B) D28 OR: responders: 17% (95% CI: 11%–23%), non-responders: 54% (95% CI: 44%–63%). Interventional trial cohort (C-D): (C) D7 OR: responders: 14% (95% CI: 6%–24%), non-responders: 57% (95% CI: 45%–68%). (D) D28 OR: responders: 16% (95% CI: 8%–25%), non-responders: 65% (95% CI: 51%–76%). AUC comparisons are shown in Table 3.

Discussion

In this study, we show that a response assessment at D7 after treatment initiation does not accurately predict long term outcomes in patients treated for mild to moderate GVHD, but performs well for patients treated for severe disease, especially in the lower GI tract. Surprisingly, second line treatment was often deferred for patients with severe GVHD who did not respond by D7, even after the availability of an approved second line therapy, suggesting a missed opportunity to intervene earlier in patients with predictably poor outcomes. These findings were consistent in two large independent cohorts despite heterogeneity in demographics and transplantation practices.

For patients with mild to moderate symptoms of GVHD at onset, D7 response predicted long term outcomes poorly and the “wait and see” approach for patients with an early non-response was successful due to the large number of slow responders. This was especially true for patients with little or no lower GI disease allowing a longer time frame to achieve disease control. By contrast, patients with high risk GVHD, often characterized by severe lower gastrointestinal involvement, required early disease control to prevent irreversible tissue damage^14,31.

This study has several limitations. The interventional trial data represent an older era of transplantation practice, with lower use of post-transplant cyclophosphamide and haploidentical donors. Both cohorts include large numbers of patients treated before a second line treatment for steroid-resistant GVHD became available. Thus, the generalizability of our findings to more recent patient populations warrants further study.

These databases also did not capture other clinical factors such as concurrent infections that may have impacted the decision to defer second line treatment in early non-responders; for some of these patients the long-term risks of a “wait and see” approach may be offset by the risks of additional immunosuppression. Moreover, three of the five clinical trials incorporated biomarker-based risk stratification in patient selection which enriched the populations for high and low risk GVHD^9,11,22 and may have influenced the relative performance of response assessments compared with a purely clinical classification strategy. The consistency of the findings across both real-world and clinical trial cohorts, despite differences in patient composition and era, however, does provide some reassurance in this regard. It is also important to note that a D7 assessment may not be appropriate to assess future novel agents that require a longer duration of administration to manifest benefit.

This study may have implications for the design of clinical trials. D28 was designated as the best time to measure response as a clinical trial endpoint for acute GVHD based on its association with downstream outcomes such as NRM and survival by an FDA-NIH public workshop in 2009^16,32. At the time, earlier time points such as D7 or D14 were rejected because they did not correlate as well as D28 with long term survival and may not have provided enough time for an optimal response to experimental interventions. These recommendations were largely based on databases dominated by patients with what is now classified as standard risk GVHD and responses did not account for baseline disease severity^13,17,33,34. Our results suggest that an earlier assessment of response may be appropriate in clinical trials, particularly in patients with high risk disease.

The response of lower GI disease to treatment is important for overall control of GVHD and is a major determinant of long-term outcomes. In this regard the ability of the GI crypts to regenerate the mucosal barrier is critical and validated serum biomarkers such as REG3α and ST2 that are released during tissue injury offer a direct measure of crypt damage throughout the small intestine^35–37. We recently showed that the combination of clinical severity and biomarker levels at D14 of treatment is a more accurate predictor of long-term outcomes than conventional clinical response³⁸. Future studies will examine whether the combination of changes in clinical symptoms with biomarker levels can prove beneficial at even earlier timepoints such as D7. For example, low biomarkers at D7 might identify a subset of high risk non-responders for whom a “wait and see” approach might be appropriate.

In conclusion, our findings support the use of treatment response at D7 as an actionable timepoint for patients with high risk acute GVHD provided an effective treatment is available. Adopting a severity-adapted approach to response evaluation, using earlier assessments in patients with high risk disease who are more likely to need experimental therapies, may improve outcome prediction, aid in clinical trial design and inform more individualized treatment strategies in acute GVHD.