Prognostic value of response after upfront therapy for acute GVHD

Abstract

One challenge in designing clinical trials for treatment of acute GVHD (aGVHD) is the lack of an established standardized end point to measure the success of therapies. To facilitate assessment of end points in clinical trials for treatment of aGVHD in the current allo-SCT era, a national workshop was recently organized. In this study, which was presented at the workshop, we evaluated the prognostic value of response to upfront therapy in a cohort of 83 patients who had been enrolled on two clinical trials testing novel therapies for aGVHD at our institution. Our results indicate that patients whose aGVHD has a CR or PR by day 28 after initiation of systemic therapy have a significantly lower 6-month cumulative incidence of non-relapse mortality (NRM) (16%) than patients whose aGVHD did not respond to therapy by day 28 (48%, P=0.005). Multivariate analysis based on the Cox proportional hazards regression analysis showed that the impact of response on NRM is independent of patient and aGVHD characteristics. Our data confirm the validity of using day-28 response as a primary end point in clinical trials for upfront therapy for aGVHD.

Introduction

One challenge in designing clinical trials of acute GVHD (aGVHD) treatment is the lack of established outcomes to measure the success of therapies.¹ Survival end points, including OS and non-relapse mortality (NRM), have traditionally been used as end points in aGVHD therapeutic trials; however, these have proved to be of limited value. Experience has shown that successful control of aGVHD manifestations does not necessarily lead to improved survival, and statistical considerations required to show survival advantage usually necessitate large sample sizes, which are practically difficult to achieve.² Because of these limitations, and in the light of recent United States Food and Drug Administration policies indicating the acceptance of non-survival end points for drug approval, experts in the field have recommended the use of measures directly related to the control of GVHD manifestations, rather than survival, as primary end point of aGVHD therapeutic trials.¹ To test the validity of the use of response to therapy as primary end point for clinical trials of upfront aGVHD therapy, we assessed the independent prognostic value of response following the initiation of systemic therapy for aGVHD. Our study population was derived from participants in two clinical trials of upfront aGVHD treatment conducted at The University of Texas MD Anderson Cancer Center.^3,4 Pooling data from both clinical trials, we systematically evaluated the independent prognostic value of response to therapy on day 7, day 14 and day 28 as a predictor of the rate of NRM and OS. We chose this patient population because it reflects the characteristics of patients likely to be eligible for enrollment in trials of upfront aGVHD therapy. We assessed NRM at 6 months after therapy initiation, when the impact of aGVHD on outcome is most likely to manifest.

Patients and methods

Study design

We included all 83 patients who had been enrolled in two prospective clinical trials of first-line aGVHD therapy at MD Anderson. Details of trial design, patient and transplant characteristics, and aGVHD assessment and therapy were reported previously.^3,4 The first trial, conducted between August 2000 and July 2003 with 57 patients was a single-center, open-label, randomized phase III trial assessing the efficacy of methylprednisolone plus infliximab vs methylprednisolone alone.⁴ The second trial was part of a national multicenter randomized four-arm phase II trial conducted through the Blood and Marrow Transplant Clinical Trials Network (BMT CTN), and designed to evaluate the safety and efficacy of four agents in combination with a corticosteroid for aGVHD treatment.³ Between August 2005 and March 2008, 180 patients were enrolled in the multicenter trial, 26 of them at MD Anderson between November 2005 and February 2008. All patients in both MD Anderson trials signed informed consent before receiving therapy, in accordance with the Declaration of Helsinki. The protocols, amendments and informed consent documents were approved by the appropriate institutional review boards.

aGVHD diagnosis and assessment of treatment response

For both trials, aGVHD organ stage was scored by the attending physician, and an overall grade was determined by consensus criteria through a computerized algorithm integrated in our departmental database.⁵ Response to aGVHD therapy was evaluated by the attending physician in real time and was based on the maximum aGVHD stage in each organ. Diagnosis and scoring of aGVHD for each patient were confirmed retrospectively by the principal investigators (DRC and AMA). For the infliximab trial, formal aGVHD assessment was required twice weekly for the first 30 days after study enrollment. Each organ involved was assigned a stage at follow-up, and response was identified as the maximal response 1, 2 and 4 weeks after treatment initiation. For the BMT CTN trial, organ involvement was assessed weekly through week 8 and at 3, 4, 6 and 9 months after study enrollment. Scoring of organ stage and overall aGVHD grade were based on the consensus criteria.

In both trials, CR was defined as resolution of all signs and symptoms of aGVHD in all organs without intervening salvage therapy, PR as an improvement of at least one stage in one or more organs without progression in any other organ, mixed response as improvement in at least one organ with progression or newly developed aGVHD in another organ(s), and progressive disease as worsening by at least one stage without improvement in any involved organ. No response was defined as the absence of response after at least 7 days of therapy for skin aGVHD or 3 days for gastrointestinal or liver aGVHD in the infliximab trial, and as deterioration or absence of improvement within 7, 14 or 28 days of therapy initiation in the BMT CTN trial.

Statistical methods

The primary end points of our analysis were 6-month NRM and 2-year OS. The cumulative incidence of NRM was estimated considering death with active disease as a competing risk. OS was estimated by the Kaplan–Meier method. Predictors of NRM and OS were assessed in univariate and multivariate analyses using the Cox proportional hazards model. Results were comparable when outcomes were assessed from therapy initiation or as a landmark analysis starting on the day response to systemic therapy was assessed (day 7, 14 or 28). Only results from the landmark analyses are presented here. Akaike’s information criterion⁶ was used to determine the goodness-of-fit of two non-nested multivariate models including, among other variables, day-14 or -28 response. Proportionality of hazards was assessed graphically and statistically. Associations of response with patient, transplant and aGVHD characteristics were assessed by the χ²-test for categorical variables or the Wilcoxon rank-sum test for continuous variables. Agreement between responses at days 14 and 28 was measured using the κ statistic. P≤0.05 was considered statistically significant. Factors significant at the 0.1 level in univariate analysis were included in multivariate analyses. Statistical analyses were performed using STATA 9.0 (StataCorp, College Station, TX, USA).

Results

The analysis included 83 patients, 69% from the infliximab trial and 31% from the BMT CTN trial.³ Median age was 49 years (range, 20–70 years), and 52% of the patients underwent transplantation for treatment of acute leukemia. In 31% of patients, the underlying disease was in remission at transplantation. The conditioning regimen was myeloablative in 52% of patients, and peripheral blood was the stem cell source in 54%; 49% of patients received a graft from a matched sibling, 40% from a matched unrelated and 11% from a mismatched related donor. aGVHD prophylaxis was tacrolimus and MTX in 87% of patients. At initiation of systemic aGVHD therapy, 5% of patients had grade I, 63% grade II and 32% grade III–IV aGVHD. Forty-nine patients (59%) had visceral organ involvement at therapy initiation: 22 with gastrointestinal, 20 with gastrointestinal and skin, 4 with liver and 3 with liver and skin involvement. The remaining 34 patients had skin-only aGVHD. aGVHD was diagnosed at a median of 25 days (range, 9–83 days) after transplantation and systemic therapy initiated at a median of 2 days (range, 0–81 days) after aGVHD onset (Table 1).

Table 1.

Patient (N=83), transplant and aGVHD characteristics

Age in years, median (range)	49	(20–70)
	N	(%)
Parent clinical trial
Randomized phase III infliximab	57	69
MP (2 mg/kg)	28	49
Infliximab + MP	29	51
Randomized phase II BMT CTN	26	31
Etanercept + MP (2 mg/kg)	5	19
Mycophenolate + MP	5	19
Denileukin + MP	6	23
Pentostatin + MP	10	38
Donor/recipient sex
Female/male	19	23
Diagnosis
AML/MDS	35	42
ALL	8	10
Aplastic anemia	1	1
CML/MPD	5	6
CLL	9	11
Non-Hodgkin’s lymphoma	13	16
Hodgkin’s lymphoma	4	5
Multiple myeloma	5	6
Solid organ	3	4
Conditioning regimen
Myeloablative	43	52
Other	40	48
Disease status at transplantation
Remission	26	31
Active disease	57	69
Donor type
Matched sibling	41	49
Matched unrelated	33	40
Mismatched	9	11
Stem cell source
Peripheral blood	45	54
BM	38	46
GVHD prophylaxis
Tacrolimus/MTX	72	87
Tacrolimus/MTX/pentostatin	9	11
Other	2	2
aGVHD grade
I/II	4/52	5/63
III/IV	27	32
Organ involvement
Skin only	34	41
Visceral organ	49	59
Days to aGVHD onset, median (range)	25	(9–83)
Days from onset to initiation of systemic therapy, median (range)	2	(0–81)

Abbreviations: aGVHD=acute GVHD; BMT CTN=Bone Marrow Transplant Clinical Trials Network; MDS=myelodysplastic syndrome; MP=methylprednisolone; MPD=myeloproliferative disorder.

Response to first-line therapy

The proportion of patients who responded to first-line therapy (CR/PR) varied slightly over time, ranging between 60% and 65% on days 7 and 28. However, among responders, the proportion with CR increased from 53% (27/51) at day 7 to 92% (46/50) at day 28 (Figure 1). Changes in response according to day-7 response were tracked, and are listed in Table 2. Notably, response was durable for most of the 27 patients who achieved CR by day 7, as 89% and 78% still had CR on days 14 and 28, respectively. In addition, at least half of patients who had PR (12/24), no response (7/14) or mixed response (4/7) at day 7 converted to CR by day 28 without additional lines of therapy, and two patients with no response at day 7 achieved PR by day 28. The relatively high conversion to CR/PR between days 7 and 28 was reflected in a low statistical agreement (35%) and a low κ statistic (0.14) between responses assessed on these days. Agreement was also low between responses at days 7 and 14. Although these statistics are based on a relatively small number of patients, our data indicate that day 7 assessment may underestimate treatment efficacy (and hence may not be a good indicator of long-term outcome). As day-7 responses are relatively unstable, we considered only responses at days 14 and 28 in evaluating predictors of long-term outcome. Agreement and κ value (0.48) were high between responses on days 14 and 28.

Figure 1.

Distribution (%) of response to therapy for aGVHD over time. Patients who received salvage therapy were coded as ‘‘PD’’ irrespective of response to salvage. Response on day 28 was unknown in two patients who died at 3 and 15 months from study entry due to underlying malignancy and stroke, respectively. NE, non-evaluable: two patients died of aGVHD on days 12 and 15 from study entry and were not evaluable for day-28 assessment. MR, mixed response; NR, no response; PD, progressive disease.

Table 2.

Changes in response over time according to day 7 response

Day 7 response	Response over time	Day 14	Day 28
CR (N=27)	CR	89%	78%
	PR	0	4%
	PD	4%	7%
	MR	7%	0
	UNK	0	7%
PR (N=24)	CR	37%	50%
	PR	33%	4%
	PD	25%	42%
	NR	4%	4%
MR (N=7)	CR	29%	57%
	PR	43%	0
	PD	14%	43%
	MR	14%	0
NR (N=14)	CR	29%	50%
	PR	14%	14%
	PD	14%	36%
	NR	43%	0
PD (N=11)	CR	9%	18%
	PR	9%	0
	PD	82%	64%
	NE	0	18%

Abbreviations: MR=mixed response; NE=not evaluable; NR=no response; PD=progressive disease; UNK=unknown.

Predictors of response to first-line therapy

We found no association between day-14 or -28 response and aGVHD, patient or transplant characteristics, except for a marginally significant association between disease status at transplantation and response: a higher proportion of patients with active disease at transplantation achieved response than of those in remission (P=0.05) (Table 3). There was no significant difference in response between patients enrolled in the BMT CTN and infliximab clinical trials or between patients with grade I–II and with grade III–IV aGVHD at therapy initiation. Our data imply that response to therapy is not a surrogate of favorable aGVHD or patient characteristics in our patient population.

Table 3.

Evaluation of predictors of response to upfront therapy for aGVHD

	Total	Responders (CR/PR), day + 14, N=54	All other, day + 14, N=29		Responders (CR/PR), day + 28, N=52	All other, day + 28, N=31	P
	N=83	N (%)	N (%)	P	N (%)	N (%)
Age (years) median		49 (20–70)	49 (25–64)	0.6	49 (20–67)	50 (27–70)	0.7
Parent clinical trial
BMT CTN	26	17 (65)	9 (35)		18 (69)	8 (31)
Infliximab	57	37 (65)	20 (35)	0.9	34 (60)	23 (40)	0.3
Donor/recipient sex
Female/male	19	12 (63)	7 (27)		41 (65)	22 (35)
Others	63	42 (67)	21 (33)	0.4	10 (53)	9 (47)	0.2
Disease status
Remission	26	13 (50)	13 (50)		12 (46)	14 (54)
Active disease	57	41 (72)	16 (28)	0.05	40 (70)	17 (30)	0.03
Stem cell source
Peripheral blood	45	31 (69)	14 (31)		30 (67)	15 (33)
BM	38	23 (61)	15 (39)	0.5	22 (58)	16 (42)	0.3
Conditioning regimen
Myeloablative	43	25 (58)	18 (42)		27 (63)	16 (37)
Other	40	29 (73)	11 (27)	0.2	25 (62)	15 (38)	0.9
Donor type
Matched sibling	41	29 (71)	12 (29)		28 (68)	13 (32)
Matched unrelated	33	20 (61)	13 (39)		19 (58)	14 (42)
Mismatched	9	5 (56)	4 (44)	0.3	5 (56)	4 (44)	0.3
aGVHD grade
I-II	56	37 (66)	19 (34)		36 (64)	20 (36)
III-IV	27	17 (63)	10 (37)	0.8	16 (59)	11 (41)	0.7
Organ involvement
Skin only	33	21 (64)	12 (36)		22 (67)	11 (33)
Visceral organ	50	33 (66)	17 (34)	0.6	30 (60)	20 (40)	0.3

Abbreviations: aGVHD=acute GVHD; BMT TCN=Bone Marrow Transplant Clinical Trials Network.

Response to first-line therapy as predictor of long-term outcome

OS

With a median follow-up interval of 44 months (range, 12–80 months) in surviving patients, 61 deaths occurred. Only one death occurred within 14 days and three more deaths within 28 days of therapy initiation. Three of these four deaths were related to aGVHD. Two patients (with deaths on days 12 and 15) were considered non-evaluable for day-28 response.

In univariate analysis, only disease status at transplantation was a significant predictor of OS. Landmark analysis starting on days 14 and 28 since therapy initiation showed that patients whose underlying malignancy was in remission at transplantation had a significantly lower mortality than those with active disease (day 14, hazard ratio (HR)=0.5, 95% confidence interval=0.3–0.9; P=0.04). Notably, response to aGVHD therapy did not significantly affect OS when assessed based on day 14 (HR=0.9; P=0.6) or day 28 (HR=0.7; P=0.2). However, stratified analysis according to disease status showed a stronger trend of improved OS in responders for patients in remission (day 28, HR=0.3, 95% confidence interval 0.1–1.1; P=0.06), than for patients with active disease (day 28, HR=0.7, 95% confidence interval 0.3–1.3; P=0.3). Notably, aGVHD severity was not significantly associated with OS. Two-year actuarial OS was 31% among patients with grade I–II and 22% among patients with grade III–IV aGVHD (P=0.1). Similarly, organ involvement, patient age, donor/patient sex, conditioning regimen, donor type, cell source or clinical trial had no significant impact on OS. Despite the comparable mortality rates, causes of death were strikingly different in responders and non-responders. NRM accounted for 95 and 96% of deaths among patients who did not respond by days 14 and 28, respectively, whereas relapse of underlying malignancy accounted for 56 and 61% of deaths among patients who responded by days 14 and 28, respectively.

Non-relapse mortality

Of the 61 deaths, 38 (62%) were from causes unrelated to the underlying malignancy. In univariate analysis, response (CR/PR) to systemic therapy was the most significant predictor of 6-month NRM (Table 4). Six-month NRM was 46 and 48% in nonresponders, and 18 and 16% in responders on days 14 and 28, respectively (day 14, P=0.01; day 28, P=0.005). The superior outcome in responders was maintained at 2 years. Estimated based on day-28 response, the 2-year cumulative incidence of NRM was 73% in non-responders and 26% in responders (HR=3.5, P<0.001).

Table 4.

Univariate analysis assessing predictors of non-relapse mortality at 6 months from initiation of aGVHD therapy

	Outcomes as of day 14			Outcomes as of day 14
	HR	95% CI	P	HR	95% CI	P
Response to systemic therapy
CR/PR	0.3	0.1–0.8	0.01	0.3	0.1–0.7	0.005
Other	Ref.			Ref.
aGVHD grade at study entry
I–II	Ref.			Ref.
III–IV	1.6	0.7–3.8	0.2	1.6	0.7–3.8	0.3
Organ involvement at study entry
Skin only	0.8	0.4–1.9	0.7	0.96	0.4–2.3	0.9
GI only	Ref.			Ref.
Skin + GI	Ref.			Ref.
Liver±skin	Ref.			Ref.
Age, years
≤49	Ref.			Ref.
>49	2.6	1.1–6.4	0.03	2.9	1.1–7.6	0.03
Parent clinical trial
BMT CTN	1.1	0.5–2.7	0.8	1.1	0.4–2.6	0.9
Infliximab	Ref.			Ref.
Donor/patient sex
Female/female	Ref.			Ref.
Female/male	0.6	0.2–1.9	0.4	0.7	0.2–2.3	0.6
Male/male	0.5	0.2–1.4	0.2	0.5	0.1–1.6	0.2
Male/female	0.5	0.1–1.6	0.2	0.5	0.1–2.0	0.4
Disease status at transplantation
Active disease	Ref.			Ref.
Remission	0.7	0.3–1.8	0.5	0.9	0.3–2.0	0.6
Conditioning regimen
Myeloablative	0.7	0.3–1.6	0.4	0.5	0.2–1.3	0.2
Other	Ref.			Ref.
Previous chemo regimens
p2	Ref.			Ref.
42	0.7	0.3–1.7	0.4	0.7	0.3–1.6	0.4
Donor type
Matched sibling	Ref.			Ref.
Matched unrelated	0.96	0.4–2.3	0.9	0.75	0.3–1.9	0.5
Mismatched	0.7	0.2–3.2	0.7	0.7	0.2–3.2	0.7
Source of stem cells
BM	Ref.			Ref.
Peripheral blood	2.1	0.85–5.0	0.1	2.3	0.9–5.9	0.09

Abbreviations: aGVHD=acute GVHD; BMT CTN= Bone Marrow Transplant Clinical Trials Network; CI=confidence interval; GI= gastrointestinal; HR=hazards ratio.

In addition to response, older age (>49 years) was associated with higher 6-month NRM when estimated based on day 14 (HR=2.6, P=0.03) or day 28 (HR=2.9, P=0.03). There was also a marginally significant trend toward higher 6-month NRM among patients who received a peripheral blood graft (P=0.1). Notably, aGVHD severity and organ involvement did not significantly affect 6-month NRM. Similarly, there was no significant impact for donor type, donor/patient sex, conditioning regimen, disease status at transplantation, number of previous chemotherapy regimens or clinical trial.

In multivariate analysis evaluating the independent effects of response to therapy, patient age and cell source, only response to systemic therapy and patient age remained significant predictors of NRM, with HRs comparable to those obtained on univariate analysis. Although results were comparable when assessment was based on day-14 or -28 response, the goodness-of-fit test based on Akaike’s information criterion indicated that day-28 response was statistically more predictive than day-14 response.

Stratified analyses showed that the impact of day-28 response on 6-month NRM was independent of aGVHD severity. Lack of response was associated with higher incidence of NRM in patients with grade I–II (HR=3.4, P=0.03) and with grade III–IV (HR=4.3, P=0.04) aGVHD (Figure 2). Furthermore, the incidence of NRM in responders was similar for patients with grade I–II (29%) and with grade III–IV (25%) aGVHD (HR=0.98; P=0.9), and the incidence among non-responders was similar for patients with grade I–II (76%) and with grade III–IV (82%) aGVHD (HR=0.6; P=0.2).

Figure 2.

Cumulative incidence of NRM according to aGVHD grade and response to upfront therapy.

Discussion

Our results show that day-28 response is an independent predictor of NRM following therapy for aGVHD. Although day-14 response was also associated with NRM, day 28 was the strongest predictor statistically. Our data also show that response was not a surrogate of favorable patient or aGVHD characteristics and was equally predictive of NRM in patients with severe (grade III–IV) or moderate (grade I–II) aGVHD. Our findings are consistent with those reported recently in two studies from the University of Minnesota (UM)⁷ and from the BMT CTN.⁸ Collectively, the findings of our study along with those of these two studies validate day-28 response as the most appropriate end point for future clinical trials of initial aGVHD therapy. This end point provides an early gauge of therapeutic effectiveness that independently predicts long-term outcomes.

The validity of this end point was confirmed by the reproducibility of the results in three separate data sets despite differences in the study populations. Both the BMT CTN⁸ study and ours were based on patients enrolled in clinical trials of upfront aGVHD therapy, whereas the UM study⁷ was based on all consecutive patients who underwent allo-SCT between 1990 and 2007. In addition, patient and aGVHD characteristics differed between the UM study and ours, including a lower mean patient age, a higher proportion of patients with grade I aGVHD, the inclusion of umbilical cord transplants and the use of myeloablative TBI conditioning (in most patients) in the UM study. Despite these differences, our findings were strikingly similar. Using similar definitions of CR and PR, the proportion of patients with CR was 55% and the 2-year cumulative incidence of NRM in responders was estimated at 26%, in both studies. More than one-third of patients were classified as non-responders in both studies (35 and 38%, respectively), and non-responders had estimated 2-year cumulative incidences of NRM of 73 and 52%, in our study and the UM study, respectively. The BMT CTN study reported comparable results.⁸

We defined response to therapy as a CR or PR. However, because only 5% of our patient population had a PR at day 28, the prognostic value of response in our study is mainly based on achieving a CR. We could not compare the prognostic value of CR vs PR because of the sample size limitation. Data from the UM and BMT CTN studies show comparable outcomes in these two groups.

aGVHD severity is an established strong predictor of response to first-line therapy and of outcome, regardless of the scoring system used.^5,7–12 In this study, however, aGVHD characteristics, including grade and organ involvement at therapy initiation, were not significantly associated with either response to therapy or NRM. This could be attributed to the small proportion of patients with mild (grade I) aGVHD, which would have contributed to a greater disparity in outcomes of grade I–II vs III–IV aGVHD. Only 5% of patients had grade I aGVHD in this study, compared with almost one-third of patients in the UM study⁷ and 12% in the BMT CTN study.⁸

Organ involvement (skin-only vs visceral) was also not significantly associated with either response or NRM in this study. Similar findings were reported in the BMT CTN study⁸ in patients with grade II aGVHD. These data suggest that skin-only aGVHD that requires systemic therapy has a comparable mortality risk to visceral organ aGVHD. In contrast, skin-only involvement was associated with a higher rate of treatment-related mortality in the UM study.⁷ This could be attributed to the higher proportion of severe organ stage (stage III–IV) in patients with skin involvement (42%) than in patients with visceral organ involvement (5% at most). In this study, we showed that the impact of day-28 response to therapy may override the effect of aGVHD severity at initiation of first-line therapy; the incidence of NRM was comparably low in responders with grade I–II and with grade III–IV aGVHD, and comparably high in non-responders with grade I–II and with grade III–IV aGVHD. These findings need to be validated in larger data sets.

Our data showed that day-28 response was not significantly associated with OS despite its significant association with NRM. These findings are not surprising, and are consistent with the increasing recognition that effective control of aGVHD activity does not necessarily translate into prolonged OS, underscoring the complexity of factors that determine outcomes in patients with aGVHD, including the competing risk of relapse or progression of the underlying malignancy.^1,13 In this study, the lower NRM in day 28 responders was offset by relapse-related deaths, resulting in comparable OS rates in responders and nonresponders. These findings highlight the importance of considering disease status at transplantation in the design of clinical trials to ensure comparability of the underlying competing risk of disease relapse or progression across treatment arms.

In summary, our data validate the use of day-28 response as an end point for efficacy in clinical trials of upfront aGVHD therapies.