Grade II Acute Graft-versus-Host Disease and Higher Nucleated Cell Graft Dose Improve Progression-Free Survival after HLA-Haploidentical Transplant with Post-Transplant Cyclophosphamide

Abstract

Compared with standard graft-versus-host disease (GVHD) prophylaxis platforms, post-transplantation cyclophosphamide (PTCy) after T cell–replete HLA-haploidentical (haplo) bone marrow transplantation (BMT) reduces the risk of grades III to IV acute (a) and chronic (c) GVHD, but maintains similar rates of grade II aGVHD. Given that mild GVHD has been associated with reduced treatment failure in HLA-matched BMT, we evaluated the risk factors for and effects of GVHD on survival in 340 adults with hematologic malignancies who engrafted after nonmyeloablative haplo-BMT with PTCy, mycophenolate mofetil, and tacrolimus. The cumulative incidence at 100 days of grade II and grades III to IV aGVHD were 30% (95% confidence interval [CI], 25% to 35%) and 2% (95% CI, 1% to 4%), respectively. The 1-year cumulative incidence of cGVHD was 10% (95% CI, 7% to 13%). In landmark analyses at 100 days, the 4-year probabilities of overall survival (OS) and progressionfree survival (PFS) were, 48% (95% CI, 41% to 56%) and 39% (95% CI, 32% to 47%) for patients without grades II to IV aGVHD, compared with 63% (95% CI, 53% to 73%) and 59% (95% CI, 50% to 71%) for patients with grade II aGVHD (P= .05 and P= .009). In multivariable modeling, when compared with patients who never experienced GVHD, the hazard ratio (HR) for OS and PFS in patients with grade II aGVHD was .78 (95% CI, .54 to 1.13; P= .19) and .69 (95% CI, .48 to .98; P= .04). Higher nucleated cell graft dose was also associated with improved OS (HR, .88; 95% CI, .78 to 1.00; P= .05) and PFS (HR, .89; 95% CI, .79 to 1.0; P= .05) and decreased risk of grades III to IV aGVHD (subdistribution HR, .66; 95% CI, .46 to .96; P= .03). PTCy reduces grades III to IV aGVHD and cGVHD, but retains similar incidence of grade II aGVHD, the development of which improves PFS. Higher nucleated cell graft dose goals may also improve survival after nonmyeloablative haplo-BMT with PTCy.

INTRODUCTION

Graft-versus-host disease (GVHD) is a major complication of allogeneic blood or marrow transplantation (BMT). Grades II to IV acute (a)GVHD occurs in 40% to 60% [1] and chronic (c)GVHD in 30% to 70% [2,3] of patients receiving HLA-matched sibling donor BMT using calcineurin inhibitor–based GVHD prophylaxis. aGVHD and cGVHD are leading causes of nonrelapse mortality (NRM) [4–9], but they also reduce quality of life as a result of end-organ damage and adverse effects related to immunosuppressive therapy [9–13]. However, the association of GVHD with these negative transplant outcomes is complicated by its protective effects against relapse [14–16]. Mild GVHD has been associated with improved overall survival (OS) in some studies of HLA-matched [16] and double umbilical cord blood transplant [17], protecting against relapse without significantly increasing NRM. These data have led to the concept of “good” GVHD, but attempts to dissociate “good” and “bad” GVHD have been largely unsuccessful.

Historically, HLA mismatching has also been associated with increased rates of both aGVHD [18–21] and cGVHD [13,21,22], and resulting NRM, making matched related donors the preferred donor source. However, modern transplant approaches [23], such as the use of post-transplantation cyclophosphamide (PTCy), have reduced the historically high rates of GVHD and NRM, making the safety and outcomes of HLA-haploidentical (haplo-) BMT now comparable with HLA-matched BMT [24–35]. In fact, earlier reports by our group and others found no significant association of the number of HLA mismatches between recipient and donor with either grades II to IV aGVHD or event-free survival after haplo-BMT with PTCy [36,37], but other potential risk factors for GVHD after haplo-BMT with PTCy have not been well defined.

Risk factors for GVHD in the HLA-matched setting using standard calcineurin inhibitor–based GVHD prophylaxis include older patient age [6,38,39], female donors for male recipients [6,38–40], cytomegalovirus (CMV) serology [41,42], donor age [43], intensity of conditioning [40,42], and peripheral blood stem cells (PBSCs) as the allograft source [42,44]. Recognized risk factors for cGVHD include unrelated donors [13,22], older patient age [2,13,45,46], PBSCs as the allograft source [47–49], and history of prior aGVHD [13,45,46]. Unfortunately, no factors have been identified to specifically predict for “good” (maximal grade II) GVHD. Given that cGVHD and aGVHD are associated with distinct risk factors and that PTCy has been shown to decrease grades III to IV aGVHD and cGVHD without similar decreases in the incidence of grade II aGVHD [23], it is possible that the risk factors for GVHD severity may also be distinct.

In a large, single-center, nonmyeloablative (NMA) haplo-BMT with PTCy cohort, we evaluated the effects of GVHD on survival and found that grade II aGVHD improved outcomes. Accordingly, we then sought to identify unique risk factors for the degree of aGVHD severity.

METHODS

Patients and Transplantation Procedures

Institutional Review Board approval was granted for this retrospective review of 372 consecutive adult patients with hematologic malignancies who underwent an NMA haplo-BMT with PTCy at Johns Hopkins between 2002 and 2012. Patients were treated on clinical trials or off-study and their outcomes included in previous publications [36,50–52]. Donors were first-degree or half-sibling related with 1 to 5 antigen-or allelic-level mismatches at HLA-A, -B, -Cw, -DRB1, and -DQB1 in either the graft-versus-host or hostversus-graft direction.

All patients received NMA conditioning consisting of fludarabine (30 mg/m² i.v. days –6 to –2), Cy (14.5 mg/kg i.v. days −6 and −5), and total body irradiation (200 cGy day –1) [51]. T cell–replete bone marrow grafts collected with a targeted total nucleated cell count of 4.0 ×· 10⁸/kg of recipient ideal body weight were administered on day 0 to all patients followed by PTCy (50 mg/kg ideal body weight i.v.) on days +3 and +4 with mesna. Recipients received mycophenolate mofetil and tacrolimus starting on day +5. Mycophenolate mofetil was stopped without taper on day +35. In the absence of GVHD, tacrolimus was discontinued without taper on either day +90 (n = 45) or +180 (n = 328), depending on protocol. Filgrastim was given from day +5 until neutrophil recovery.

Patients were excluded from study if they failed to engraft, were receiving second allogeneic transplants, or any conditioning or post-transplant regimens other than those defined above. However, patients receiving post-transplant antitumor maintenance were included. Sixteen of 24 acute lymphoblastic leukemia patients in this study had Philadelphia chromosome– positive acute lymphoblastic leukemia, 13 of whom received post-transplant prophylaxis with a tyrosine kinase inhibitor. Fifty-eight of 166 lymphoma patients received post-transplant rituximab, which was typically given weekly for 4 consecutive weeks at the time of count recovery.

Outcome Definitions

NRM was defined as death without progression, relapse, or unplanned treatment of disease persistence. NRM was a competing risk when estimating cumulative incidence (CuI) of relapse and vice versa. Events for progression-free survival (PFS) were relapse/progression, unplanned treatment of disease persistence, or death from any cause. An event for OS was defined as death from any cause. Patients still alive were censored at the time of last follow-up. Relapse, donor lymphocyte infusion, and death were competing events when estimating the CuI of GVHD. Grades III to IV aGVHD was also a competing risk when estimating the CuI of maximal grade II aGVHD. The modified Keystone criteria [53] and the 2005 National Institutes of Health consensus criteria [54] were used to diagnose and grade aGVHD and cGVHD, respectively.

The first or second author performed initial GVHD scoring by retrospective chart review as well as discussion with the patient’s primary transplant physician followed by a second independent assessment by the Johns Hopkins GVHD specialist (J.B.-M). In the event of a disagreement between the first and second authors and J.B.-M, the case was discussed by all 3 and a consensus reached. Given that grade I aGVHD is frequently transient, not routinely biopsied or treated, and can be difficult to distinguish from a drug or viral rash, patients with suspected maximal grade I aGVHD were included in the group of patients categorized as not having GVHD. However, if a patient experienced grade I aGVHD that progressed to either grade II or grades III to IV aGVHD, the patient was categorized in the higher grade GVHD group at the time of GVHD progression. Both a senior Johns Hopkins BMT physician (E.J.F) and the clinical director of the HLA lab (M.P.B.) performed HLA mismatch assignments.

Statistical Analysis

The primary objective of this study was to evaluate the risk factors for GVHD development and the relationship of grades of GVHD with OS, PFS, relapse, and NRM after NMA haplo-BMT using PTCy. Thirty-two patients with graft failure were excluded from the analysis to distinguish the impact of GVHD on survival outcomes from its association with engraftment. The CuI of GVHD was estimated via Gray’s method [55]. Fine and Gray’s regression model was used to assess baseline risk factors for GVHD via subdistribution hazard ratios (HRs) [56]. Multivariable model components for risk of GVHD were selected based on backward elimination using a P ≤ .10 for retention from Fine and Gray’s regression models with patient age at BMT as a continuous variable adjustment and stratification of BMT year, where BMT year was cut at the median (2002 to 2008 versus 2009 to 2012). Covariates considered in the assessments of risk for GVHD and the impact of GVHD on outcomes included pretransplant disease status (minimal residual disease or active disease versus complete remission), hematopoietic cell transplantation–specific comorbidity index (HCT-CI, treated as a continuous variable), total nucleated cell graft dose (treated as a continuous variable), CD3⁺ allograft cell dose (treated as a continuous variable), female-into-male allografting, recipient CMV serostatus (seropositive versus seronegative), donor age (treated as a continuous variable), and number of antigen or allele group level HLA mismatches in HLA-A, −B, −Cw, DRB1, and DQB1 (4 to 5 versus ≤3). The effects of developing grade II aGVHD or grades III to IV aGVHD on cGVHD were evaluated via a cause-specific hazard model with aGVHD as a time-dependent covariate, considering donor lymphocyte infusion, relapse, and NRM as competing risk events.

The effects of GVHD on the outcomes were analyzed univariately via a landmark approach, which looks at the time to an event after a prespecified time point. By starting the failure-time clock later, one can treat events before the landmark as baseline factors in the analysis. Landmark time points of 100 days and 1 year were chosen based both on prior literature [2,14–16] and the ability to capture most patients who experienced grade II aGVHD and cGVHD, respectively, while still selecting time points early enough to ascertain the effects of GVHD on outcomes. In landmark analyses OS and PFS were estimated by the Kaplan-Meier method and GVHD effects were tested using log-rank test. The corresponding 95% confidence intervals (CIs) for OS and PFS were estimated via Greenwood’s formula. CuI of relapse and NRM were estimated and tested via Gray’s method [56] in the landmark analysis. Subgroup analyses and multivariable modeling were conducted by constructing GVHD status as a time-dependent covariate to assesses the effects of GVHD on OS and PFS via Cox regression models and relapse and NRM via cause-specific hazard models. In subgroup analyses a forest plot was used to depict the consistency of GVHD effects on PFS in specific patient subgroups. Final multivariable models for GVHD effects followed the same procedures used for selecting risk factors for GVHD with the same considered variables listed above, with the addition of HCT-CI and Disease Risk Index (DRI; high/very high, intermediate, or low) [57]. The effects of grades III to IV aGVHD on outcomes were only examined in time-dependent GVHD models because few patients developed grades III to IV aGVHD.

All analyses were carried out using the statistical software R version 3.3.1 (R Foundation for Statistical Computing, Vienna, Austria). All reported P-values are 2-sided. Curves shown in the figures were truncated at 7.5 years after BMT because few patients were at risk beyond this point. P < .05 was interpreted as statistically significant.

RESULTS

Patient Characteristics

Overall patient and transplant characteristics are shown in Table 1. We observed a total of 181 death events with a median follow-up of 4.98 years (range, 125 days to 11.4 years) based on the reverse Kaplan-Meier method [58]. Median patient age was 56 years (range, 18 to 75). Twenty-one percent of patients had active disease and 39% had minimal residual disease at the time of transplantation [59–61]. Fourteen percent of patients were low risk, 67% intermediate risk, and 19% high or very high risk by the DRI [57]. Thirty-eight percent of patients had HCT-CI [62] scores ≥ 3 (high risk). Forty-five percent of recipients were CMV seropositive.

Table 1.

Patient, Donor, and Transplantation Characteristics (N = 340)

Variable	Value
Median patient age, yr (range)	56 (18–75)
Male sex	229 (67)
Diagnosis
AML	80 (24)
CR	51 (64)
MRD (defined as flow cytometry evidence of persistent abnormal myeloblasts < 5% or persistent karyotype abnormalities) or active disease	29 (36)
ALL	24 (7)
CR	14 (58)
MRD (defined as flow cytometry evidence of persistent abnormal lymphoblasts < 5%, persistent BCR-ABL by PCR) or active disease	10 (42)
MDS/MPN	24 (7)
CR	0 (0)
MRD (defined as decline in blast count from <5% to >5% but with persistent dysplasia or myeloproliferation) or active disease	24 (100)
Aggressive NHL (including mantle cell)	116 (34)
CR	52 (45)
MRD (defined as partial response) or active disease	64 (55)
Indolent lymphoma/CLL	50 (15)
CR	9 (18)
MRD (defined as partial response) or active disease	41 (82)
Hodgkin lymphoma	35 (10)
CR	7 (20)
MRD (defined as partial response) or active disease	28 (80)
Multiple myeloma	11 (3)
CR	1(1)
MRD (defined as very good partial response, partial response, minor response) or active disease	10 (99)
Year of BMT
2002–2008	138 (41)
2009–2012	202 (59)
Pretransplant status
CR without MRD	134 (39)
CR with MRD	135 (40)
Active disease	71 (21)
HCT-CI risk score
0 (low)	83 (24)
1–2 (intermediate)	125 (37)
3–4 (high)	102 (30)
≥5 (very high)	30 (9)
DRI
Low risk	47 (14)
Intermediate risk	230 (68)
High or very high risk	63 (18)
Recipient CMV serostatus
Negative	186 (55)
Positive	153 (45)
Unavailable	1 (0)
Median donor age, yr (range)	40 (10–79)
Female donor to male recipient	94 (28)
Antigen-level HLA mismatches
≤3	116 (34)
4–5	224 (66)
Median cell dose infused
TNC, ×108/kg	4.14 (.97–8.53)
CD34+ cells, ×106/kg	4.09 (1.4–11.4)
CD3+ cells, ×107/kg	3.83 (1.13–10.39)

Values are n (%), unless otherwise defined. AML indicates acute myelogenous leukemia; CR, complete response; MRD, minimal residual disease; ALL, acute lymphoblastic leukemia; MDS, myelodysplastic syndrome; MPN, myeloproliferative neoplasm (includes chronic myelogenous leukemia and chronic myelomonocytic leukemia); NHL, non-Hodgkin lymphoma; CLL, chronic lymphocytic leukemia; TNC, total nucleated cells; kg, kilogram of ideal body weight.

GVHD Incidence and Immunosuppression Discontinuation

Consistent with previous reports [26,27,35,50], the CuI at 100 days of grade II and grades III to IV aGVHD were 30% (95% CI, 25% to 35%) and 2% (95% CI, 1% to 4%), respectively. The 1-year CuI of grade II aGVHD and grades III to IV aGVHD were 33% (95% CI, 28% to 38%) and 6% (95% CI, 3% to 8%), respectively. The CuI of cGVHD at 1 year was 10% (95% CI, 7% to 13%) and at 2 years was 12% (95% CI, 8% to 15%). Mild, moderate, and severe cGVHD at 2 years occurred in 9% (95% CI, 6% to 12%), 1% (95% CI, 0% to 2%), and 2% (95% CI, 0% to 3%) of patients, respectively. Median time to development of grade II aGVHD was 43 days (interquartile range, 37 to 55), grades III to IV aGVHD was 108 days (interquartile range, 54 to 183), and cGVHD was 207 days (interquartile range, 152 to 269). Among patients who had grade II aGVHD, 19% later developed cGVHD. Compared with patients without aGVHD, patients who experienced grade II or grades III to IV acute GVHD had a significantly higher risk of cGVHD development with cause-specific HRs (CSHRs) of 2.83 (95% CI, 1.44 to 5.59; P = .003) and 4.03 (95% CI, 1.32 to 12.32; P = .01), respectively.

GVHD was also examined by the number of HLA antigen mismatches (4 to 5 versus ≤3) at HLA-A, −B, −Cw, −DRB1, and -DQB1 between the donor and recipient in the graft-versushost direction. The 1-year CuIs of grade II aGVHD were 28% (95% CI, 20% to 37%) and 36% (95% CI, 29% to 42%), grades III to IV aGVHD 4% (95% CI, 1% to 8%) and 6% (95% CI, 3% to 9%), and 2-year cGVHD 8% (95% CI, 3% to 13%) and 14% (95% CI, 9% to 18%) in patients with ≤3 versus 4 to 5 mismatches, respectively (Supplementary Figure 1). There was no significant difference in GVHD incidence between men who had received female allografts and all other recipient–donor sex combinations (Supplementary Figure 2).

The rate of successful discontinuation of all pharmacologic and phototherapeutic immunosuppression was assessed. Of the patients who were alive at 1 year after BMT, 78% had discontinued all immunosuppression and 22% remained on immunosuppression. Of all patients studied, 31% were alive, in remission, and had discontinued all immunosuppression by 1 year post-transplant.

Landmark Analysis of GVHD Effect on OS and PFS

To evaluate the effect of GVHD on OS and PFS, we first conducted landmark analyses with landmark times selected at 100 days and 1 year after BMT for aGVHD and cGVHD, respectively. The selected time points allowed us to capture the vast majority of GVHD events, where 90% of patients who developed maximal grade II aGVHD did so by 100 days and 85% of patients who developed cGVHD did so by 1 year post-transplant. Among patients who never experienced grades II to IV aGVHD, 23% died before the landmark time of 100 days. Among patients who experienced grade II aGVHD, 4% died before 100 days. Among patients who never experienced grades II to IV aGVHD or cGVHD, 36% died before the landmark time of 1 year. However, among patients who developed cGVHD, only 5% of patients died before 1 year. We did not examine the effects of grades III to IV aGVHD on PFS and OS by landmark analysis because of the limited number of events.

In landmark analyses patients who had experienced grade II aGVHD by 100 days were compared with patients who had not developed grades II to IV aGVHD by this time. In the cohort as a whole, 4-year OS and PFS were 48% (95% CI, .43 to .54) and 38% (95% CI, .33 to .44), respectively. By aGVHD status at 100 days, 4-year OS and PFS were 63% (95% CI, 53% to 73%) and 59% (95% CI, 50% to 71%) for patients with a grade II aGVHD event and 48% (95% CI, 41% to 56%) and 39% (95% CI 32% to 47%) for patients without grades II to IV aGVHD event (Table 2, Figure 1A; P = .05 and P = .009, respectively). For patients who were alive at 1 year after BMT, 4-year OS and PFS were 75% (95% CI, 60–93%) and 62% (95% CI, 45–86%) for patients with a cGVHD event and 65% (95% CI, 57% to 75%) and 63% (95% CI, 53% to 75%) for patients without any GVHD event by 1 year (Table 2, Figure 2A; P = .74 and P = .78, respectively).

Table 2.

Landmark Analyses for GVHD Effects on OS, PFS, Relapse, and NRM

Outcomes*	Landmark 100 Days		Landmark 1 Year
	No aGVHD	Grade II aGVHD	No GVHD	cGVHD
OS, no. events/no. at risk	110/208	38/95	46/128	10/31
4-year probabilities	.48	.63	.65	.75
(95% CI)	(.41−.56)	(.53−.73)	(.57−.75)	(.6−.93)
	P = .054		P = .735
PFS, no. events/no. at risk	109/184	38/91	33/95	11/29
4-year probabilities	.39	.59	.63	.62
(95% CI)	(.32−.47)	(.50−.71)	(.53−.75)	(.45−.86)
	P = .009		P = .780
Relapse, no. events/no. at risk	91/184	23/91	30/95	4/29
4-year CuI	.50	.27	.33	.13
(95% CI)	(.42−.58)	(.17−.37)	(.22−.44)	(.00−.27)
	P < .001		P = .057
NRM, no. events/no. at risk	18/184	15/91	3/95	7/29
4-year CuI	.11	.13	.04	.25
(95% CI)	(.06−.16)	(.06−.21)	(.00to.09)	(.06−.43)
	P = .26		P < .001

^*OS and PFS were estimated from Kaplan-Meier method; relapse and NRM were estimated from cause-specific hazard models.

Figure 1.

(A) In patients who were alive at day 100, the probabilities of OS (P = .05) and PFS (P = .009) were significantly better in patients who had developed grade II aGVHD (OS, dark blue line; PFS, hash-marked light blue line) when compared with patients who had not developed grades II to IV aGVHD (OS, black line; PFS, hash-marked gray line). (B) In patients who were alive and without relapse at day 100, CuI of relapse was significantly less (P < .001) in patients who had developed grade II aGVHD (dark blue line) when compared with patients who had not developed grades II to IV aGVHD (black line). CuI of NRM was not significantly different in patients who had developed grade II aGVHD (hash-marked light blue line) when compared with patients who had not developed grades II to IV aGVHD (hash-marked gray line). no GVHD24 indicates patients alive and without grades II to IV aGVHD by 100 days; aGVHD2, patients alive and having already developed grade II aGVHD by 100 days; no, number at risk; no a24, patients alive and without aGVHD; a2 patients alive and having already developed grade II aGVHD

Figure 2.

(A) In patients who were alive at 1 year, the probability of OS and PFS was not significantly different in patients who had developed cGVHD (OS, dark yellow line; PFS, hash-marked light yellow line) when compared with patients who had not developed GVHD (OS, black line; PFS, hash-marked gray line). (B) In patients who were alive and without relapse at 1 year, CuI of relapse was significantly lower in patients who developed cGVHD (dark yellow line) when compared with patients who had not developed GVHD (black line). CuI of NRM was significantly different in patients who developed cGVHD (hash-marked light yellow line) when compared with patients who had not developed GVHD (hash-marked gray line). no GVHD indicates patients alive and without aGVHD or cGVHD by 1 year; cGVHD, patients alive and having developed cGVHD by 1 year.

Multivariable Analysis of GVHD Effect on OS and PFS

Next, we examined the effects of developing GVHD on survival outcomes in multivariable models that adjusted for patient and transplant factors and were stratified by year of BMT (2002 to 2008 versus 2009 to 2012). With no prior GVHD as the reference, the HR for OS was .78 (95% CI, .54 to 1.13; P = .19) for patients with grade II aGVHD, 1.99 (95% CI, 1.03 to 3.84; P = .04) for patients with grades III to IV aGVHD, and .89 (95% CI, .50 to 1.58; P = .69) for patients with cGVHD (Table 3). Older patient age as a continuous variable was associated with decreased OS (HR, 1.02; 95% CI, 1 to 1.03; P= .01). Higher nucleated cell graft dose as a continuous variable was associated with improvements in OS (HR, .88; 95% CI, .78 to 1.00; P= .05). Intermediate-risk or high-/very-high-risk disease by DRI were associated with inferior OS (HR, 2.19 [95% CI, 1.29 to 3.72; P = .003] and HR, 3.21 [95% CI, 1.88 to 5.70; P < .0001], respectively) when compared with low-risk disease (Table 3).

Table 3.

Final Multivariable Models for OS, PFS, Relapse, and NRM

Covariates*	OS		PFS		Relapse		NRM
	HR (95% CI)	P	HR (95% CI)	P	CSHR (95% CI)	P	CSHR (95% CI)	P
GVHD status (t)†
No GVHD(t)	1		1		1		1
Grade II aGVHD(t)	.78 (.54–1.13)	.19	.69 (.48−.98)	.04	.60 (.40−.89)	.01	1.52 (.70–3.31)	.29
Grades III- IV aGVHD(t)	1.99 (1.03–3.84)	.04	1.66 (.86–3.18)	.13	.82 (.30–2.24)	.70	8.04 (3.07–21.04)	<.0001
cGVHD(t)	.89 (.50–1.58)	.69	1.14 (.67–1.95)	.62	.69 (.33–1.44)	.32	4.53 (1.84–11.12)	.001
Patient age‡	1.02 (1.00–1.03)	.01	1.01 (1.00–1.02)	.04	1.01 (1.00–1.02)	.23	1.03 (1.00–1.05)	.05
DRI
Low risk	1		1		1	1
Intermediate risk	2.19 (1.29–3.72)	.003	2.51 (1.51–4.17)	.0004	3.73 (1.86–7.46)	.0002	NR
High or very high risk	3.21 (1.81–5.70)	<.0001	3.76 (2.17–6.50)	<.0001	6.44 (3.12–13.28)	<.0001
Nucleated cell graft dose × 108/kg	.88 (.78–1.00)	.05	.89 (.80–1.00)	.05	.87 (.77 to .99)	.04	NR
HCT-CI‡	NR		NR		NR		1.25 (1.08–1.44)	.002
CD3+ cell graft dose × 107/kg	NR		NR		NR		.78 (.61 to .99)	.04

GVHD(t) indicates time-dependent covariate of GVHD status; NR, not retained.

^*Stratification by BMT year 2009–2012 vs. 2002–2008.

^†Time-dependent covariate of GVHD status, GVHD(t), was constructed.

^‡Age as a continuous variable.

After adjusting for potential confounding risk factors, when compared with patients who never experienced GVHD, patients who developed grade II aGVHD had significantly improved PFS (HR, .69; 95% CI, .48 to .98; P = .04). Although patients with grades III to IV aGVHD had inferior PFS (HR, 1.66; 95% CI, .86 to 3.18; P = .13), the results did not reach statistical significance due in part to a low number of events. There was no difference in PFS in patients who developed cGVHD (HR, 1.14; 95% CI, .67 to 1.95; P = .62). Higher nucleated cell graft dose as a continuous variable was associated with improved PFS (HR, .89; 95% CI, .80 to 1.00; P = .05). When compared with low-risk disease by DRI, intermediate-risk or high-/very-high-risk disease had inferior PFS (HR, 2.51 [95% CI, 1.51 to 4.17; P = .0004] and HR, 3.76 [95% CI, 2.17 to 6.50; P < .0001], respectively; Table 3). Other assessed risk factors, including CD3⁺ graft cell dose (treated as a continuous variable), female-into-male allografting, recipient CMV serostatus (seropositive versus seronegative), HCT-CI, donor age (treated as a continuous variable), and number of antigen-level HLA mismatches were not significantly associated with OS or PFS.

Forest Plot for Subgroup Analysis of the Effects of Grade II aGVHD on PFS

Given the significant improvements in PFS with grade II acute GVHD when compared with no GVHD in multivariable analysis, we generated a forest plot to illustrate the effects of grade II aGVHD on PFS in different subgroups, such as disease type (Figure 3). In terms of the overall effect, grade II aGVHD improved PFS and individual disease type subgroups had overlapping CIs, all showing a benefit from grade II aGVHD. Only 11 patients had a diagnosis of multiple myeloma and were thus excluded from subgroup analysis. Recipients who were CMV seropositive had an HR above 1 for PFS, suggesting worse outcomes for those who experienced grade II aGVHD. CMV-seronegative recipients, however, had a strong benefit in PFS from grade II aGVHD. Patients with active disease, minimal residual disease, and complete remission at the time of BMT all benefited from grade II aGVHD in terms of PFS. Importantly, older patients and those with HCT-CI scores ≥3 also appeared to benefit from grade II aGVHD.

Figure 3.

PFS was examined in subgroups by forest plots. In the group as a whole (shown as Summary), patients benefited from grade II aGVHD. All individual subgroups, except for CMV-positive recipients, benefited from grade II aGVHD. AML indicates acute myelogenous leukemia; MDS, includes myelodysplastic syndrome and myeloproliferative neoplasms (chronic myelogenous leukemia and chronic myelomonocytic leukemia); ALL, acute lymphoblastic leukemia; Agg Lymph, aggressive non-Hodgkin lymphoma including mantle cell; Indo Lymph, indolent non-Hodgkin lymphoma including chronic lymphocytic leukemia; Hod Lymph, Hodgkin lymphoma; Mult Myeloma, multiple myeloma; NR, not run; CR, complete response; MRD, minimal residual disease; DAge, donor age; Not Fem2Male, includes male into male, male into female, or female into female allografts; Fem2Male, includes female into male allografts only; NC, total nucleated cells per kilogram of ideal body weight.

Landmark Analysis of Effects of GVHD on Relapse and NRM

In the entire cohort the CuI of relapse was 48%. In patients who were alive and relapse-free at 100 days, the CuI of relapse was lower (P < .001) in patients who had developed maximal grade II aGVHD when compared with patients without grades II to IV aGVHD (4-year CuI, 27% [95% CI, 17% to 37%] versus 50% [95% CI, 42% to 58%], respectively) (Table 2, Figure 1B). In patients who were alive and relapse-free at 1 year, patients who had developed cGVHD had a lower (P = .06) relapse rate when compared with patients who had not developed GVHD (4-year CuI, 13% [95% CI, 27% to 33%] versus 33% [95% CI, 22% to 44%], respectively) (Table 2, Figure 2B).

In the entire cohort the 4-year CuI of NRM was 14% (95% CI, 10% to 18%). In patients who were alive and relapse-free at 100 days, the CuI of NRM was not statistically different (P = .26) between those who developed grade II aGVHD and those who did not develop grades II to IV aGVHD (4-year CuI, 13% [95% CI, 6% to 21%] versus 11% [95% CI, 6% to 16%], respectively) (Table 2, Figure 1B). In patients who were alive and relapse-free at 1 year, the CuI of NRM was significantly different (P< .0001) in patients who had developed cGVHD when compared with patients who had not developed GVHD (4-year CuI, 25% [95% CI, 6% to 43%] versus 4% [95% CI, 0% to 9%]) (Table 2, Figure 2B).

Multivariable Analysis of the Effects of GVHD on Relapse and NRM

Based on multivariable analyses, patients who developed grade II aGVHD had a significantly lower risk of relapse (CSHR, .60; 95% CI, .40 to .89; P = .01), but no significant difference in risk of NRM (CSHR, 1.52; 95% CI, .70 to 3.31; P = .29) when compared with patients without grades II to IV aGVHD (Table 3). In contrast, patients who developed grades III to IV aGVHD had no significant difference in relapse (CSHR, .84; 95% CI, .30 to 2.24; P = .70), but a significantly higher risk of NRM (CSHR, 8.04; 95% CI, 3.07 to 21.04; P ≤.0001). Patients who developed cGVHD had a nonsignificant lower risk of relapse (CSHR, .69; 95% CI, .33 to 1.44; P = .32), but a statistically significant higher risk of NRM (CSHR, 4.53; 95% CI, 1.84 to 11.12; P = .001) (Table 3).

Higher nucleated cell count as a continuous variable was associated with decreased relapse (CSHR, .87; 95% CI, .77 to .99; P = .05). When compared with low-risk disease by DRI, intermediate-risk or high-/very-high-risk disease had a higher risk of relapse (CSHR, 3.73 [95% CI, 1.86 to 7.46; P = .0002] and CSHR, 6.44 [95% CI, 3.12 to 13.28; P < .0001], respectively).

Older patient age and higher HCT-CI were associated with increased NRM (CSHR, 1.03 [95% CI, 1 to 1.05; P = .05] and CSHR, 1.25 [95% CI, 1.08 to 1.44; P = .05], respectively). Higher CD3⁺ cell graft dose treated as a continuous variable was associated with a significant decrease in NRM (HR, .78; 95% CI, .61 to .99; P = .04). Other assessed risk factors, including female-into-male allografting, recipient CMV serostatus (seropositive versus seronegative), donor age (treated as a continuous variable), and number of antigen-level HLA mismatches, were not significantly associated with relapse or NRM.

Multivariable Modeling of GVHD Risk Factors

In multivariable modeling, older patient age and higher CD3⁺ cell graft dose treated as continuous variables were associated with increased risks of grade II aGVHD development (subdistribution HR, 1.02 [95% CI, 1.0 to 1.04; P = .03] and subdistribution HR, 1.20 [95% CI, 1.03 to 1.39; P = .02], re-spectively) (Table 4). Greater number of HLA mismatches, female donor into male recipient allografting, disease remission status, nucleated cell graft dose, donor age, and HCT-CI were not significantly associated with grade II aGVHD.

Table 4.

Final Multivariable Models for GVHD

Covariate*	Grade II aGVHD		Grades III-IV aGVHD		cGVHD
	SDHR (95% CI)	P	SDHR (95% CI)	P	SDHR (95% CI)	P
Patient age†	1.02 (1–1.04)	.03	.99 (.95–1.03)	.56	1.00 (.98–1.03)	.68
Female-into-male allografting	NR		NR		NR
CD3+ cell graft dose per kg × 107/kg†	1.20 (1.03–1.39)	.02	NR		NR
Nucleated cell graft dose × 108/kg†	NR		.66 (.46-.96)	.03	NR

SDHR, subdistribution HR.

^*Stratified by year of bone marrow transplant (2002–2008 vs. 2009–2012).

^†As a continuous variable

In multivariable modeling of grades III to IV aGVHD, higher nucleated cell dose was associated with a decreased risk of grades III to IV aGVHD (subdistribution HR, .66; 95% CI, .46 to .96; P = .03). Donor age was not associated with grades III to IV aGVHD. In addition, we found no association of nucleated cell dose with donor age. All other assessed potential risk factors mentioned above were not significantly associated with grades III to IV aGVHD. Finally, in multivariable modeling, none of the baseline risk factors was associated with cGVHD.

DISCUSSION

Given the reduction of severe aGVHD and all cGVHD when using PTCy after haplo-BMT, we sought to determine the graftversus-tumor effects of and the risk factors for GVHD in this distinctive transplant setting. Prior studies have used GVHD events as time-dependent covariates, performing landmark analyses to assess the antileukemic effect of GVHD [14,16]. In 1990 when studying HLA-matched transplantation, Horowitz et al. [16] showed an improvement in failure-free survival with mild aGVHD or cGVHD, but worse failure-free survival with moderate to severe aGVHD or cGVHD. Further-more, in a more recent analysis of reduced-intensity conditioning HLA-matched PBSC transplant, mild aGVHD was associated with a tendency toward increased OS, whereas grades III to IV aGVHD was associated with decreased OS [63]. Using similar landmark analyses we found that grade II aGVHD was significantly associated with improvements in OS and PFS and in multivariable models significantly associated with improved PFS. In addition, we found that grades III to IV aGVHD was significantly associated with inferior OS, whereas having cGVHD was not significantly associated with these outcomes (Figures 1 and 2, Table 3).

The beneficial effects of grade II aGVHD on survival were due to similar NRM, but a reduction in relapse in our study, which was consistent with a recent publication of double umbilical cord blood transplant [17]. All disease types had a PFS benefit from grade II aGVHD, despite the limited number of patients in each subgroup. PFS improvements with grade II aGVHD were also seen regardless of age, HCT-CI scores, DRI, donor age, donor sex, nucleated cell count, or disease stage. CMV-seropositive patients were the only patients without improved PFS with development of grade II aGVHD in our analysis. Given the overlapping CI of CMV-seropositive and CMV-seronegative recipients, this finding may be due to chance and not a statistical difference. However, GVHD has been shown in prior analyses to be a risk factor for CMV reactivation [64]. Thus, it is possible that increased risk of CMV reactivation led to increased risk of NRM, which outweighed the benefit of grade II aGVHD on relapse reduction.

Next, we analyzed potential risk factors for grade II aGVHD, grades III to IV aGVHD, and cGVHD. Although some historical risk factors were not relevant here because of the platform investigated (eg, high-intensity total body irradiation [40,42] and grafting with PBSCs [42,44]), the identification of only a few risk factors in this study may suggest that PTCy has modulated alloreactivity and potentially obviated the impact of some traditional risk factors for both aGVHD and cGVHD. For example, distinct from many prior analyses [6,13,38,40,45,46,65,66] female-into-male allografting was not significantly associated with GVHD in our study, which may have been due to either an absence of an association or the low number of cGVHD events limiting statistical power to detect a significant correlation. We did find that older patient age and higher CD3⁺ cell graft dose were associated with increased risk of grade II aGVHD, but these factors were not associated with either grades III to IV aGVHD or cGVHD. On the other hand, increasing nucleated cell dose was associated with less grades III to IV aGVHD, but was not associated with grade II aGVHD or cGVHD. These data suggest that PTCy may allow alloreactive responses to be controlled before they become more severe, such that factors that previously were associated with both mild and severe GVHD are now only associated with mild GVHD. Notably, higher CD3⁺ graft dose, despite its association with grade II aGVHD, had a protective effect on NRM. This result is consistent with findings in haploidentical transplantation with antithymocyte globulin by Dong et al. [67] in which CD3⁺ cell dose greater than 177 ×· 10⁶/kg was associated with less NRM and greater graftversus-tumor effects, but without increases in severe GVHD. Furthermore, in our analysis higher nucleated cell dose protected against both grades III to IV aGVHD and relapse, resulting in improved OS and PFS. This suggests that higher graft dose goals may benefit patients in terms of GVHD and relapse, which warrants further study.

PTCy has improved the safety of haplo-BMT, allowing the identification of a suitable donor for most patients. Given the relatively low NRM with NMA haplo-BMT when using PTCy, substantially improving outcomes further will depend on reducing relapse. Our analysis is inherently limited by its retrospective nature and also by the few number of grades III to IV aGVHD and cGVHD events and the heterogeneity of diseases included. However, we found that grade II aGVHD was associated with improvements in PFS across hematologic malignancies. Therefore, platforms that decrease grades III to IV aGVHD without decreasing grade II aGVHD, such as is seen with PTCy-based immunomodulation, may enhance survival after transplant. As such, transplant studies should distinguish grade II aGVHD from grades III to IV aGVHD to avoid lumping the “good” in with the “bad.” Finally, higher graft dose goals, earlier withdrawal of immunosuppression, utilization of PBSCs in high-risk patients [68], and the addition of novel immunotherapeutic post-transplant maintenance strategies to precipitate grade II aGVHD, augment antitumor immunity, and prevent late HLA-loss relapse [69,70] should be explored in this well-tolerated transplant platform.