Optimizing Patient Selection for Reduced Non-Relapse Mortality after Reduced-Intensity Conditioning: A CIBMTR Analysis

Graphical Abstract

Introduction

Allogeneic hematopoietic cell transplantation (HCT) is a potentially curative therapy for various hematologic malignancies. Traditional myeloablative conditioning (MAC) regimens exert strong anti-tumor effects but are often highly toxic for certain patient populations, such as older adults or those with comorbidities. The introduction of reduced-intensity conditioning (RIC) has expanded access to HCT by enabling its use in patients who are ineligible for MAC, thereby substantially broadening the transplant-eligible population [1–6].

Following the advent of RIC, numerous prospective and retrospective studies have compared RIC with traditional MAC regimens [1, 5–16]. These studies generally found that RIC is associated with lower non-relapse mortality (NRM) but a higher risk of relapse compared to MAC, resulting in overall similar survival outcomes [17–20]. However, these findings reflect only average treatment effects across study populations [21]. Given that the relative benefits and risks of conditioning intensity likely vary based on individual patient characteristics [9, 15, 22–24] and disease-related factors [10, 25], what clinicians ultimately seek is to determine the optimal conditioning intensity for each patient. To address this, we previously developed the Risk Assessment for the Intensity of Conditioning Regimen in Elderly Patients (RICE) score, which identified patients likely to experience reduced NRM with RIC, based on a large nationwide Japanese registry [26]. While the RICE score successfully stratified patients who may benefit from RIC in terms of NRM reduction, differences in baseline characteristics—such as race and donor source—between Japanese and Western cohorts raise the need for external validation. The present study thus aimed to assess the robustness of the RICE score using a publicly available dataset from the Center for International Blood and Marrow Transplant Research (CIBMTR).

Methods

Patient selection

This study utilized a publicly available dataset from the CIBMTR registry used in a previously published study [10], including patients aged 40 to 65 years with acute myeloid leukemia (AML) or myelodysplastic syndrome (MDS) who underwent first allogeneic HCT between 2009 and 2015. To reduce heterogeneity associated with HCT for active disease, all MDS cases were included regardless of disease status, while AML cases were limited to those in first or second complete remission. The analysis was further restricted to patients who received conventional GVHD prophylaxis with a calcineurin inhibitor plus either methotrexate or mycophenolate mofetil, and one of four commonly used conditioning regimens: (1) cyclophosphamide plus total body irradiation (CyTBI); (2) busulfan plus cyclophosphamide (BuCy); (3) fludarabine plus busulfan (FluBu); or (4) fludarabine plus melphalan (FluMel). Posttransplant cyclophosphamide as a GVHD prophylaxis were excluded because of the limited sample size (n = 89).

Data source

The CIBMTR maintains a comprehensive database of HCT by collecting consecutive patient-level data from global transplant centers. Clinical and follow-up data, including longitudinal information obtained through annual updates, are submitted to a centralized statistical coordinating center jointly operated by the CIBMTR headquarters at the Medical College of Wisconsin and the National Marrow Donor Program. All observational research conducted by the CIBMTR adheres to applicable federal regulations to ensure the protection of human subjects. For this analysis, the Institutional Review Board at the Medical College of Wisconsin, along with the institution’s Privacy Officer, approved a waiver of informed consent. The study was conducted in full compliance with the Health Insurance Portability and Accountability Act (HIPAA) requirements.

Definition

The intensity of the conditioning regimen was defined by the CIBMTR consensus criteria [27]. Revised disease risk index (DRI) and HCT Comorbidity Index (HCT-CI) scores were calculated as previously described [28, 29]. The RICE score in this study is based on three factors: advanced recipient age (>60 years), high comorbidities (HCT-CI ≥2), and use of umbilical cord blood (UCB). The RICE score is calculated as the sum of these factors: scores of 0 or 1 indicate a low RICE score, while 2 or 3 indicate a high RICE score [26]. The threshold for advanced age in the original paper was defined as ≥60 years; however, we used >60 years because the CIBMTR dataset provided only a categorical variable for age over 60.

Statistical analysis

The primary endpoint was NRM, consistent with our original publication[26], and was analyzed by comparing MAC and RIC, stratified according to the RICE score. Secondary endpoints included disease-free survival (DFS), overall survival (OS), and relapse. NRM was defined as death occurring in the absence of disease relapse. Cumulative incidences of NRM and relapse were estimated using Gray’s method. The Fine–Gray hazard model was employed to assess the effect of conditioning intensity on both NRM and relapse, accounting for competing risks—relapse for NRM and non-relapse death for relapse. The Kaplan–Meier method and Cox proportional hazards regression were used to evaluate the associations between conditioning intensity and both DFS and OS. Interaction was assessed on the additive scale using the relative excess risk due to interaction (RERI), along with its 95% confidence interval (CI) and P value [30].

In addition to the conditioning intensity (MAC vs. RIC), the following covariates were included in the multivariate analyses: recipient’s age at HCT (≤60 vs. >60), sex mismatch (female to male vs. others), disease (AML vs. MDS), DRI (Low or Intermediate vs. High or Very High risk), HCT-CI (< 2 vs. ≥ 2), donor source (HLA matched related vs. HLA matched unrelated vs. HLA mismatched related/unrelated vs. UCB), the use of anti-thymoglobulin (No vs. Yes), and year of HCT.

Two-sided P-values ≤ 0.05 were considered to reflect statistical significance. A statistical analyses were performed with R (version 4.3.2) and EZR (version 1.63) [31].

Results

Patient characteristics and long-term outcomes in the entire cohort

In total, 2,595 patients were included in the study, of whom 1,727 (66.6%) received MAC and 868 (33.4%) received RIC (Table 1). Proportion of patients aged >60 years was 21.3% in MAC cohort and 38.8% in RIC cohort. Statistically significant differences were observed between the two groups in terms of recipient age, primary disease, DRI, HCT-CI, donor type, donor source, use of ATG, and year of HCT. The FluBu based regimen was the most frequently used conditioning in both groups. The median follow-up period among survivors was 5.0 years in the MAC group and 4.0 years in the RIC group.

Table 1.

Patient characteristics

	Entire	MAC	RIC
	N (%) = 2595	n (%) = 1727	n (%) = 868	P values
Age, category
<=60	2043 (78.7)	1512 (87.6)	531 (61.2)	<0.001
>60	552 (21.3)	215 (12.4)	337 (38.8)
Sex match between recipient and donor
Female recipient or male to male	2124 (81.8)	1415 (81.9)	709 (81.7)	0.872
Female to male	471 (18.2)	312 (18.1)	159 (18.3)
Disease
Acute myeloid leukemia	1569 (60.5)	1105 (64.0)	464 (53.5)	<0.001
Myelodysplastic syndromes	1026 (39.5)	622 (36.0)	404 (46.5)
Disease risk index
Low / Intermediate	1751 (67.5)	1201 (69.5)	550 (63.4)	0.002
High / Very High	844 (32.5)	526 (30.5)	318 (36.6)
HCT-CI
<2	934 (36.0)	668 (38.7)	266 (30.6)	<0.001
≥2	1661 (64.0)	1059 (61.3)	602 (69.4)
Donor type
HLA matched related	948 (36.5)	669 (38.7)	279 (32.1)	<0.001
HLA matched unrelated	1261 (48.6)	834 (48.3)	427 (49.2)
HLA mismatched related/unrelated	324 (12.5)	198 (11.5)	126 (14.5)
Umbilical cord blood	62 (2.4)	26 (1.5)	36 (4.1)
Donor source
Bone marrow	284 (10.9)	225 (13.0)	59 (6.8)	<0.001
Peripheral blood	2249 (86.7)	1476 (85.5)	773 (89.1)
Umbilical cord blood	62 (2.4)	26 (1.5)	36 (4.1)
GVHD prophylaxis
CSA + MMF/MTX +/− others	204 (7.9)	129 (7.5)	75 (8.6)	0.315
TAC + MMF/MTX +/− others	2391 (92.1)	1598 (92.5)	793 (91.4)
The use of ATG
No	1871 (72.1)	1325 (76.7)	546 (62.9)	<0.001
Yes	724 (27.9)	402 (23.3)	322 (37.1)
Conditioning regimens
Cy/TBI	307 (11.8)	306 (17.7)	1 (0.1)	NA
Bu/Cy	623 (24.0)	623 (36.1)	0 (0.0)
Flu/Bu	1310 (50.5)	784 (45.4)	526 (60.6)
Flu/Mel	355 (13.7)	14 (0.8)	341 (39.3)
Median Year of HCT, (range)	2012 (2009–2015)	2012 (2009–2015)	2013 (2009–2015)	<0.001
RICE score
Low (0–1)	2166 (83.5)	1558 (90.2)	608 (70.0)	<0.001
High (1–2)	429 (16.5)	169 (9.8)	260 (30.0)

MAC, myeloablative conditioning; RIC, reduced-intensity conditioning; HCT-CI, hematopoietic cell transplantation-specific comorbidity index; CSA, cyclosporine; TAC, tacrolimus; MTX, methotrexate; MMF, mycophenolate mofetil; ATG, anti-thymoglobulin; Cy, Cyclophosphamide; TBI, total body irradiation; Bu, Busulfan; Flu, Fludarabine; Mel, Melphalan.

The 4-year cumulative incidence of NRM was comparable between the MAC and RIC groups, at 24.8% and 23.7% (P = 0.365), respectively (Table S1). However, the relapse rate was significantly lower in the MAC group compared to the RIC group (33.0% vs. 41.8%, P < 0.001). Consequently, DFS at 4 years was superior in the MAC group compared to the RIC group (42.2% vs. 34.4%, P < 0.001). In multivariate analysis, RIC was associated with a lower risk of NRM compared to MAC (HR, 0.81; 95% CI, 0.67–0.96; P < 0.001), but a higher risk of relapse (HR, 1.44; 95% CI, 1.25–1.67; P < 0.001), resulting in inferior DFS (HR, 1.17; 95% CI, 1.04–1.30; P = 0.008) (Table 2). Interaction analysis between each covariate and conditioning intensity suggested that advanced age (RERI = 0.21, P = 0.173), high comorbidities (RERI = 0.20, P = 0.137), and UCB (RERI = 0.82, P = 0.292) were the top three covariates with the highest RERI values, which are components of the original RICE score (Table S2).

Table 2.

Multivariate analyses on transplant outcomes stratified by the RICE score

	Entire		Low RICE score (0–1)		High RICE score (2–3)
	N=2595		n=2166		n=429
	HR(95% CI)	P values	HR(95% CI)	P values	HR(95% CI)	P values
NRM
MAC	1	Ref	1	Ref	1	Ref
RIC	0.81(0.67–0.96)	<0.001	0.86(0.70–1.06)	0.150	0.69(0.48–0.99)	0.048
Relapse
MAC	1	Ref	1	Ref	1	Ref
RIC	1.44(1.25–1.67)	<0.001	1.46(1.23–1.72)	<0.001	1.48(1.06–2.08)	0.023
DFS
MAC	1	Ref	1	Ref	1	Ref
RIC	1.17(1.04–1.30)	0.008	1.22(1.08–1.38)	0.002	1.01(0.79–1.29)	0.944
OS
MAC	1	Ref	1	Ref	1	Ref
RIC	1.08(0.96–1.22)	0.182	1.13(0.99–1.29)	0.065	0.96(0.74–1.24)	0.740

All models were adjusted for recipient’s age, sex mismatch, disease, DRI, HCT-CI, donor source, the use of ATG, and year of HCT.

Validation of the RICE score

A high RICE score was identified in 169 patients (9.8%) in the MAC group and 260 patients (30.0%) in the RIC group (Table 1). NRM at 4 years among patients with a low RICE score was 22.7% in the RIC group and 23.5% in the MAC group (P = 0.403) (Figure 1A). In contrast, among patients with a high RICE score, 4-year NRM was significantly lower in the RIC group compared to the MAC group (26.1% vs. 37.3%, P = 0.018) (Figure 1B). In the multivariate analyses, these findings were also confirmed in both patients with low RICE score (HR, 0.86; 95% CI, 0.70–1.06; P = 0.150) and high RICE score (HR, 0.69; 95% CI, 0.48–0.99; P = 0.048) (Table 2). In analyses limited to patients with a RICE score of 0 or 1, RIC did not reduce the risk of NRM compared to MAC (Figure S1), even after adjusting for confounders (Table S3), supporting the classification of a score of 1 as a low RICE score.

Figure 1. Cumulative incidence of NRM and relapse in patients with low and high RICE scores.

Cumulative incidence of NRM in patients with a low RICE score (A) and a high RICE score (B), and cumulative incidence of relapse in patients with a low RICE score (C) and a high RICE score (D).

We next evaluated secondary long-term outcomes. The RIC regimen was associated with a higher risk of relapse compared to the MAC regimen, regardless of RICE score (Figure 1C–D), which was confirmed in multivariate analyses (Table 2). Among patients with a low RICE score, MAC was superior to RIC in both DFS and OS, due to the lack of NRM reduction with RIC (Figure 2A–B and Table 2). In contrast, in the high RICE score subset, DFS and OS were comparable between MAC and RIC, as the increase in relapse and decrease in NRM appeared to offset each other (Figure 2C–D and Table 2).

Figure 2. DFS and OS in patients with low and high RICE scores.

DFS in patients with a low RICE score (A) and a high RICE score (B), and OS in patients with a low RICE score (C) and a high RICE score (D).

We assessed the incremental value beyond established tools, including DRI and HCT-CI. Regardless of DRI risk, 4-year NRM was similar between RIC and MAC (Low/Intermediate: 22.7% vs. 22.7%, P = 0.405; High/Very High: 25.5% vs. 29.5%, P = 0.434). In multivariable analysis, DRI did not identify patients who benefited from lower NRM with RIC (Low/Intermediate: HR, 0.80; 95% CI, 0.63–1.01; P = 0.056; High/Very High: HR, 0.84; 95% CI, 0.63–1.12; P = 0.230). At 4 years, NRM among patients with low HCT-CI (< 2) was 22.2% with RIC and 20.9% with MAC (P = 0.974) (Figure 3A), whereas among those with high HCT-CI (≥ 2) it was 24.3% and 27.2%, respectively (P = 0.170) (Figure 3B). In multivariable analysis, RIC was not associated with lower NRM in patients with low HCT-CI (HR, 0.91; 95% CI, 0.65–1.28; P = 0.590), but was associated with reduced NRM in those with high HCT-CI (HR, 0.76; 95% CI, 0.61–0.94; P = 0.013).

Figure 3. Cumulative incidence of NRM stratified by HCT-CI.

Cumulative incidence of NRM in patients with HCT-CI <2 (A), HCT-CI ≥2 (B), HCT-CI ≥2 & low RICE score (C), and HCT-CI ≥2 & high RICE score (D).

We then assessed how RICE and HCT-CI differ in identifying patients who benefit from reduced NRM with RIC. Most patients with low comorbidity (HCT-CI < 2) had a low RICE score (n = 931), with only three exceptions. Conversely, among patients with high comorbidity (HCT-CI ≥ 2), 74.3% (1235/1661) had a low RICE score, driving the divergence between the two indices. In this discordant subset (HCT-CI ≥ 2, recipient age ≤60, non-UCB; i.e., RICE score=1), 4-year NRM was similar between RIC and MAC (23.4% vs 25.4%; P = 0.351) (Figure 3C). In multivariable analysis restricted to this discordant subset, RIC was not associated with a lower risk of NRM (HR, 0.82; 95% CI, 0.64–1.06; P = 0.140). By contrast, in the non-discordant high-comorbidity subset (HCT-CI ≥2 and high RICE score), RIC was associated with lower NRM than MAC (25.6% vs 37.3%; P = 0.014) (Figure 3D), which was confirmed in multivariable analysis (HR, 0.69; 95% CI, 0.48–0.99; P = 0.049). Accordingly, and also considering that NRM was comparable across conditioning intensities in patients with a RICE score of 1, a single assessment based solely on HCT-CI, recipient age, and UCB status is insufficient to identify patients likely to experience lower NRM with RIC.

Discussion

Over the past few decades, RIC has become increasingly utilized [8], extending beyond its original use in older or frail patients to include individuals who might otherwise tolerate MAC. Despite this broad adoption, selecting the most appropriate conditioning intensity for individual patients remains uncertain [17–20]. The RICE score, based on three factors: advanced age, high comorbidities, and the use of UCB, was developed from a Japanese registry analysis to support the selection of appropriate conditioning regimens. This study assessed this precision approach using the CIBMTR independent large registry cohort and successfully identified patients who benefit from reduced NRM with RIC regimens. In addition to demonstrating the reproducibility of the RICE score across different patient characteristics and clinical practices, this study also underscores the utility of a personalized approach based on large real-world evidence.

Reducing relapse is a key advantage of the MAC regimen, and the CIBMTR cohort consistently showed that MAC was associated with a lower risk of relapse regardless of the RICE score, aligning with findings from the RCT conducted in the United States (US) [15]. Consequently, among patients with a high RICE score, the benefits and risks of RIC and MAC appeared to offset each other, resulting in no significant difference in survival. In contrast, among patients with a low RICE score, the absence of NRM reduction with RIC rendered MAC superior to RIC in both DFS and OS. Notably, this relapse-reducing benefit of MAC was less evident outside the US. For example, the RCT conducted by the European Group of Blood and Marrow Transplantation (EBMT) did not show a significant difference in relapse rates between conditioning intensities [7]. Similarly, in the Japanese cohort, the impact of MAC on relapse was limited [26]. As a result, RIC was associated with improved survival among patients with a high RICE score due to reduced NRM without an increase in relapse, whereas no survival difference was observed among those with a low RICE score [26]. Although the clear reasons for regional variation in the relapse-suppressing effect of MAC remain unclear, it is possible that a higher proportion of patients in the US had residual disease, given the pronounced benefit of MAC in patients with minimal residual disease (MRD) [25]. Due to the lack of standardization and differences in the sensitivity of MRD detection techniques [32–35], our analysis primarily focused on NRM. While future studies should aim to develop a comprehensive risk model that incorporates relapse risk factors such as MRD status [36, 37] and molecular profiles of tumor cells [38–41], we believe that the externally validated RICE score provides important insights for clinical decision-making in selection of conditioning intensity.

No significant differences in NRM were observed among patients with a RICE score of 1, suggesting that RIC can reduce the risk of NRM when two or three risk factors are present. These findings are consistent with previous reports indicating that recipient age alone should not guide clinical decision-making [2, 42, 43]. Furthermore, we also demonstrated that the RICE score has an advantage in identifying patients who are likely to experience lower NRM with RIC, compared with commonly used scoring systems, including DRI and HCT-CI. Experienced physicians typically determine conditioning intensity by integrating multiple clinical factors. The strength of the RICE score lies in its data-driven approach, which offers an alternative to physician-dependent judgment. It provides a simple and practical tool that may further assist clinicians in selecting appropriate conditioning intensity.

RCT is the gold standard for evaluating the effect of an intervention, as it eliminates selection bias and minimizes both measured and unmeasured confounders. For ethical and practical reasons, sample size calculations in RCTs are typically designed to detect differences in the primary endpoint across the overall population. As such, RCTs often lack sufficient power to detect heterogeneity of treatment effects across baseline characteristics, so-called “effect modification”, which has become an area of increasing interest [21, 44]. While retrospective studies are inherently subject to selection bias and unmeasured confounding, their larger sample sizes may allow for exploratory assessment of effect modification. The reproducible performance of the RICE score in this independent cohort suggests that such data could address the limitations of RCTs and may help personalize treatment strategies.

This study has several limitations. First, due to the limited number of patients who received posttransplant cyclophosphamide as GVHD prophylaxis, these patients were excluded from the analysis, and the RICE score should not be applied to this group. Given the rapid adoption of posttransplant cyclophosphamide in both haploidentical and non-haploidentical HCT [45], future studies are required to determine the optimal conditioning intensity in these populations. Nevertheless, in light of the limited evidence from large randomized trials employing MAC with post-transplant cyclophosphamide, it remains important to determine which patients derive reduced NRM with RIC instead of MAC under conventional GVHD prophylaxis [46, 47]. Second, the RICE score assigns one point for UCB, but the CIBMTR cohort included a limited number of patients who received UCB (n = 62). Therefore, the validity of including UCB in the scoring system remains unclear in Western populations. Importantly, despite the low proportion of UCB in this cohort, the RICE score still identified patients likely to experience lower NRM with RIC and outperformed HCT-CI, suggesting its potential utility for clinical decision-making. Third, because this publicly available CIBMTR dataset included only patients aged 40 to 65 and did not provide recipient age as a continuous variable, we were unable to validate the RICE score using the same age range as the original Japanese cohort (50 to 69 years) [26]. For the same reason, although the original RICE score assigns one point for age ≥60, this study assigned one point for age >60. Fourth, the sample size in this study limited the power to detect interactions. Although the aim of this study was to validate the RICE score in the CIBMTR cohort, it is noteworthy that the highest signals for interaction were observed for advanced age, high comorbidity, and UCB, which are components of the original RICE score.

In summary, this external validation using the CIBMTR dataset demonstrated that the RICE score effectively identified patients who experienced reduced NRM with RIC, consistent with findings from the original Japanese cohort. This straightforward and externally validated scoring system may support the selection of appropriate conditioning regimens and contribute to improved transplant outcomes globally.

Highlights:

• The RICE score, incorporating recipient age (>60 years), high comorbidities (HCT-CI ≥2), and use of umbilical cord blood (UCB), effectively identified patients who benefited from reduced NRM with RIC.

• This study externally validated the RICE score in a CIBMTR cohort and suggests its clinical potential in guiding conditioning intensity selection based on individual patient characteristics.

• The interaction-based model was externally validated across independent cohorts, underscoring its potential clinical utility for future studies.

Data sharing:

The dataset is available on the CIBMTR website. https://cibmtr.org/CIBMTR/Resources/Publicly-Available-Datasets#