Primary graft failure (pGF) is associated with considerable morbidity and mortality. Salvage hematopoietic SCT (HSCT) can rescue pGF patients; however, the optimal preconditioning regimen and stem cell source are yet to be determined, particularly in children. In this study, we retrospectively analyzed 102 pediatric patients who received salvage allogeneic HSCT for pGF. Salvage HSCT from matched or one-Ag-mismatched related donors (rMM01) provided superior OS compared with that from two- or three-Ags-mismatched related donors (rMM23) or cord blood transplantation (CBT). CBT showed a trend toward a slightly lower engraftment rate and late engraftment achievement compared with rMM23; however, the OS rate was similar between the two groups (47.6±7.7% for rMM23 and 45.7±8.6% for CBT, at 1 year after salvage HSCT). Multivariate analysis showed that preconditioning regimens with fludarabine or irradiation were associated with a higher engraftment rate and those with alkylating agents were associated with better OS. In conclusion, our results showed that rMM01 was the most suitable donor for salvage HSCT for pediatric pGF, and that CBT was an equally important option compared with rMM23 for patients without rMM01.
Primary graft failure (pGF) is associated with considerable morbidity and mortality in hematopoietic SCT (HSCT) because of severe infection due to sustained neutropenia and organ toxicity caused by preconditioning for the first HSCT. The incidence of graft failure is assumed to range from 2 to 20%, and the frequency depends on various factors such as stem cell source1, 2 and preconditioning regimen,3, 4, 5 underlying disease,5 HLA disparity2, 6, 7 and the presence of anti-HLA Abs.8
Salvage HSCT can rescue pGF patients,9, 10, 11, 12, 13, 14 and recent studies on adult pGF patients have showed the feasibility of using reduced-intensity conditioning containing fludarabine (FLU) and low-dose irradiation,13, 15, 16, 17 including short-term preconditioning regimen10, 18 and an advantage of PBSC as the stem cell source for salvage HSCT.19, 20 However, the management of pGF patients is usually difficult because of their poor clinical status. Owing to the rarity of pGF and emergent condition of the patients, prospective controlled trials are difficult and the optimal preconditioning regimen and stem cell source have not been established.
Although some reports have been published on salvage HSCT for pGF in children,20, 21, 22 the number of patients included was limited and the prognostic factors of salvage HSCT for pediatric pGF patients remained unclear. Different factors may affect the outcome of salvage HSCT in children compared with that in adults because of their immature immune status, small body size or difference in underlying diseases. A recent study showed a good outcome of CBT in adults,13 however, the role of CBT as salvage HSCT for pediatric pGF is yet to be determined because previous reports regarding pediatric pGF were mainly focused on salvage HSCT from mismatched related donor.20, 21, 22
In this study, to investigate the prognostic factors of salvage allogeneic HSCT and to obtain fundamental information for establishing a standard approach for pediatric pGF, we performed retrospective analysis of 102 patients who received a salvage HSCT for pGF after allogeneic HSCT.
Materials and methods
Patients and transplantations
This study was approved by the institutional ethics committee of Saitama Children’s Medical Center. A total of 102 patients were analyzed based on data reported to the Japan Society for Hematopoietic Cell Transplantation (JSHCT) registry23 (Table 1). The patients were selected according to the following criteria: (1) age 18 years or younger at salvage HSCT; (2) engraftment not achieved by a prior allogeneic HSCT and salvage allogeneic HSCT was performed; (3) salvage HSCT performed within 60 days after the first HSCT; (4) both the first and salvage HSCTs performed between 1982 and 2010.
Engraftment was defined as the first day of 3 consecutive days with an ANC of ⩾500/μL. Platelet engraftment was defined as the first day of achievement of a platelet count of ⩾2 × 104 μL without requiring blood transfusion. Myeloablative conditioning was defined as the TBI of 8 Gy or more, or administration of BU at more than 8 mg/kg. All other regimens were analyzed as reduced-intensity conditioning.24
The OS probability was calculated using Kaplan–Meier estimates. The incidence of engraftment and non-relapse mortality (NRM) were expressed as cumulative incidence curves, and were used to adjust for death before engraftment and relapse for competing risks, respectively. Univariate analyses of OS were performed using the log-rank test, and multivariate analysis was performed using the Cox proportional hazard regression model. The factors that were found to be significant at P-value <0.1 were entered into the multivariate analysis. All statistical analyses were performed with R software 2.13.0 (The R Foundation for Statistical Computing, Vienna, Austria). A two-sided P-value <0.05 was considered statistically significant.
The characteristics of the 102 patients are shown in Table 1. The median follow-up period in the surviving patients was 1502 days (range, 37–9101) after salvage HSCT. The estimated OS probability and s.e. at 1 year after salvage HSCT was 53.3±5.0%. The cumulative incidence of engraftment at 60 days was 55.7±5.0%, whereas NRM at 1 year was 31.7±4.7%. The median time to achieve engraftment was 15 days from salvage HSCT. Re-salvage HSCT was performed for 12 patients who failed to achieve engraftment in the first salvage HSCT. Of these, eight patients achieved engraftment and five were alive at the last follow-up.
The association between the outcome and clinical characteristics is shown in Table 2. Younger age (<6 years) was associated with poor engraftment probability (37.6±6.0% versus 65.6±8.4%, P=0.01); however, this difference did not remain significant in multivariate analysis (P=0.10) (Table 3). The OS at 1 year was similar between patients aged <6 (50.5±8.3%) and ⩾6 years (54.9±6.2%). Median interval between the first HSCT and salvage HSCT was 39.5 days. The interval of the two HSCT was not associated with the engraftment (P=0.12), NRM (P=0.65) and survival (P=0.72) of the salvage HSCT. Our patients suffered from 77 malignant diseases and 25 non-malignant diseases. OS was superior in patients with non-malignant disease than in those with malignant disease because of the relapse of malignant diseases (P=0.002). However, this difference was not statistically significant in multivariate analysis.
There were eight cases with >10% of donor chimerism at diagnosis of pGF, which might be poor graft function. Of the eight cases, one patient received salvage HSCT from the same donor and achieved engraftment. Four cases received salvage HSCT from rMM23 donors who were different from the first HSCT, and three cases achieved engraftment. The remaining three cases received salvage CBT and two cases achieved engraftment.
Stem cell sources
Detailed information of stem cell sources for the first and salvage SCTs is shown in Table 1. There was no salvage HSCT from unrelated donor except cord blood, as pGF patients required immediate rescue.
All 26 patients whose first HSCT was performed from an HLA-matched or one-Ag-mismatched related donor (rMM01) received salvage HSCT from an rMM01 donor. Of the 26 HSCTs, 5 salvage HSCTs were performed from rMM01 donors different from those for the first HSCT, whereas the remaining 21 salvage HSCTs were performed from the same rMM01 donor. All the five HSCT from a different rMM01 donor achieved engraftment, whereas the engraftment rate of HSCT from the same rMM01 donor was 42.1±11.8% (P=0.02). OS at 1 year for HSCT from a different rMM01 donor and the same rMM01 donor was 80.0±17.9% and 61.5±10.7%, and the estimated OS probability of the two groups of rMM01 did not reach statistical significance (P=0.70).
For the patients who received HSCTs from HLA 2- or 3-Ags-mismatched related donors (rMM23), 16 salvage HSCTs were performed from rMM23 donors and 2 were performed using CB. Of the 51 CBTs used as the first HSCT, stem cell sources for salvage HSCT were two of one-Ag-mismatched related donors, 24 of rMM23 and 25 of CB.
The engraftment rate at 60 days was 57.7±10.1% for rMM01, 57.1±7.7% for rMM23 and 42.2±9.4% for CBT (P=0.24). Univariate analysis showed that the stem cell source did not cause a statistically significant difference in engraftment probability (Table 2 and Figure 1a), although cumulative incidence curve of engraftment of rMM01 and rMM23 were similar, and CBT showed a trend toward a lower engraftment rate.
The median time for neutrophil engraftment after CBT was 22 days, which was significantly longer than those for rMM01 (15 days) and rMM23 (15 days) (P=0.02 and 0.0007, respectively). Platelet engraftment probability at day 100 was 41.1±12.5% for rMM01, 44.9±10.0% for rMM23 and 40.7±10.9% for CBT (P=0.77). The median time for platelet engraftment after salvage HSCT was 26 days for rMM01, 32 days for rMM23 and 63 days for CBT.
The incidence of GVHD at day 100 for rMM01 was lower (9.2±6.4%) than that for rMM23 (29.4±8.4%) or CBT (38.2±10.1%) (P=0.02) (Figure 1b). The low incidence of GVHD in HSCT from rMM23 was assumed to be caused by frequency of anti-thymocyte/lymphocyte globulin usage. anti-thymocyte/lymphocyte globulin was used in 18 of the 42 (42.9%) salvage HSCTs from rMM23, and in 7 of the 31 (22.6%) salvage CBTs.
NRM for each stem cell source was not statistically different; however, salvage HSCT from rMM01 showed a trend toward lower NRM (Figure 1c). The rMM01 group showed the highest OS (68.8±8.6%), whereas OS was similar in the rMM23 (47.6±7.7%) and CBT (45.7±8.6%) groups (Figure 1d). The cause of death was identified in 54 patients, in particular, infection in 22 (40.7%), organ dysfunction in 14 (25.9%), relapse of underlying disease in 13 (24.1%) and GVHD in 5 (9.3%). There was no difference in the distribution of the cause of death in the rMM23 and CBT groups.
Thirty-one salvage CBTs were performed in our cohort. The median TNC count was 5.3 × 107 cells/kg (range, 2.2–13.3), and the median CD34-positive cell count was 1.6 × 105 cells/kg (range, 0.13–5.3). There was no association between the cell counts and engraftment (Figures 2a and b). HLA Ag mismatch did not result in inferior engraftment (Figure 2c). Our analysis showed that cell count and HLA disparity were not related to the outcome of salvage CBT. Therefore we analyzed CBT as a single category.
The preconditioning regimens of salvage HSCT had a significant impact on engraftment and survival, whereas the regimens used for the first HSCT did not have a significant impact (Table 2). FLU, alkylating agents (CY or melphalan (L-PAM)), anti-thymocyte/lymphocyte globulin and irradiation were frequently used with various combinations, and all these components provided better engraftment probability (Table 2) without significant increase of NRM.
The best combination for a preconditioning regimen seemed to be FLU+CY/L-PAM+irradiation, which was used in 13 patients, and 12 patients achieved engraftment with a median of 20 days. The engraftment rate in the 16 patients without any preconditioning was estimated to be 12.5±8.7%. Nine of these sixteen patients died, including six patients who died before 28 days after salvage HSCT.
The irradiation dose varied, with a dose of 2 Gy in 12 patients, 3–5 Gy in 14 and >6 Gy in 14 patients. The engraftment incidence for these doses was 72.7±15.4, 92.9±8.9% and 61.5±14.7%, respectively. Although these differences were not significantly different (P=0.43) because of the limited number of patients, 3–5 Gy showed a tendency for good engraftment probability.
Multivariate analysis showed that rMM01 was significantly correlated with better engraftment and survival (Table 3). The hazards for engraftment and survival were not different between rMM23 and CBT, and were also found to be similar in multivariate analysis. Multivariate analysis showed also an advantage of FLU and irradiation in engraftment. The hazards for engraftment was 3.33 (1.62–6.83, P=0.001) and 2.68 (1.36–5.59, P=0.0005), respectively. On the other hand, alkylating agents had a major impact on survival probability, and the usage of alkylating agents was related with superiority in survival.
There are several reports on salvage HSCT for graft failure, with the majority including both primary and secondary graft failure.9, 11, 15, 25 However, evidence on management of pGF in children is limited; therefore, analysis of pediatric pGF is important in order to establish a standard treatment strategy against this rare event. This study included the largest number of children with pGF investigated to date and showed that an rMM01 was the most suitable donor for salvage HSCT, and that CBT was an equally important option compared with a haploidentical related donor for patients without a well-matched related donor.
Our results suggested that a matched related donor or one-antigen-mismatched donor was associated with better engraftment and survival rate. The engraftment probability was better for HSCT from a different donor, providing better engraftment probability than that from the same donor in this group. However, a previous study did not find statistically significant differences between engraftment with the same and different donors, mainly because of the limited number of patients.9, 16 We have little detailed information on donor chimerism at the diagnosis of graft failure and the presence of anti-HLA Abs, but our result suggests that there may be an undetectable problem in grafts of the first HSCT, and that graft dysfunction may have caused pGF in some HSCTs. Therefore, if available, a different matched or one-Ag-mismatched related donor should be selected as the stem cell source for salvage HSCT. If there was no different rMM01 donor, our results showed that salvage HSCT from the same rMM01 donor could provide better OS probability than that of rMM23 or CB.
If a well-matched related donor is not available for pGF patients, it is necessary to select either a CB or a mismatched related donor. A previous report including a small number of pediatric pGF patients showed an advantage of PBSC from mismatched related donors,19 whereas our analysis demonstrated that CBT provided a survival rate similar to that of HSCT from mismatched related donors. The survival curve for CBT was superimposed on the curve for HSCT from a mismatched related donor, although engraftment achievement tended to be lower and engraftment was late. In our cohort, the incidence of GVHD of rMM23 was not higher than that of CBT. This surprising result is assumed to be associated with the frequent use of anti-thymocyte/lymphocyte globulin, possibly related to the low incidence of GVHD; however, it may increase the incidence of infectious complications, leading to a similar survival rate between rMM23 HSCT and CBT.
CB is an important source for prompt HSCT because it is a readily available stem cell source, which does not require any manipulation of the donor. However, a recent study on engraftment probability of salvage CBT for pGF showed that a small fraction of children had a worse outcome than salvage HSCT from a mismatched related donor.19 Another study on salvage CBT for pGF reported the association of cell dose of >2.5 × 107 cells/kg with a better outcome.13 In our cohort, the median TNC count of CB was 5.4 × 107 cells/kg, which was considerably higher than that reported previously. It is difficult to obtain CB containing a large number of cells for adults; however, the small body size of children allows a relatively high number of CB to be obtained. Our results showed that CB with a sufficient cell dose could overcome HLA disparity. We consider that CB may become the first option for salvage HSCT for pediatric pGF. Although we could not use it during this study period, HSCT using double CB might enhance the role of CBT in salvage HSCT for pGF.
The performance status (PS) of pGF patients is often poor because of infection and toxicity of the prior preconditioning regimen for the first HSCT, which limits the preconditioning agents for the salvage HSCT. There are reports showing that some cases can achieve engraftment without any preconditioning regimen;26, 27 however, it is generally accepted that most pGF cases require some preconditioning treatment.14 We consider that myeloablative conditioning is not required for the salvage HSCT in pGF patients because their marrow is already hypocellular; however, reduced-intensity conditioning may help engraftment. Our study showed that FLU and alkylating agents provide better engraftment probability without increasing NRM. Of note, patients who receive less preconditioning agents may have a worse condition such as poor PS or severe organ dysfunction, usually resulting in poor outcome. Although we did not have sufficient detailed PS data to analyze the association between PS and outcome, our data suggested that a non-myeloablative preconditioning regimen including FLU, alkylating agents and irradiation should be used at a level that can be tolerated by the patients.
The optimal dose of irradiation has still not been determined; however, our results indicate that 3–5 Gy is sufficient for engraftment because >5 Gy of irradiation did not increase engraftment probability. This result is comparable with the result of a previous study that the addition of 2–4 Gy of irradiation improved the engraftment rate of RIST for aplastic anemia.28, 29
Of note, this study was a retrospective and uncontrolled study; therefore, there are some limitations of our data. For example, these include the selection criteria of a stem cell source, dosage of preconditioning agents and inconsistent indications for salvage HSCT. Further prospective studies in a large cohort and further biological research about allogeneic HSCT are therefore required to further improve the treatment of pediatric pGF patients after allogeneic HSCT.
In conclusion, our study provides valuable information on the role of the second salvage HSCT for pediatric pGF. CB should be ranked equally with mismatched related donors as a stem cell source for pGF in children without a matched-related donor.
Hansen JA, Gooley TA, Martin PJ, Appelbaum F, Chauncey TR, Clift RA et al. Bone marrow transplants from unrelated donors for patients with chronic myeloid leukemia. N Engl J Med 1998; 338: 962–968.
Shaw BE, Russell NH, Devereux S, Das-Gupta E, Mackinnon S, Madrigal JA et al. The impact of donor factors on primary non-engraftment in recipients of reduced intensity conditioned transplants from unrelated donors. Haematologica 2005; 90: 1562–1569.
Baron F, Sandmaier BM . Chimerism and outcomes after allogeneic hematopoietic cell transplantation following nonmyeloablative conditioning. Leukemia 2006; 20: 1690–1700.
Clift RA, Buckner CD, Thomas ED, Bensinger WI, Bowden R, Bryant E et al. Marrow transplantation for chronic myeloid leukemia: a randomized study comparing cyclophosphamide and total body irradiation with busulfan and cyclophosphamide. Blood 1994; 84: 2036–2043.
McCann SR, Bacigalupo A, Gluckman E, Hinterberger W, Hows J, Ljungman P et al. Graft rejection and second bone marrow transplants for acquired aplastic anaemia: a report from the Aplastic Anaemia Working Party of the European Bone Marrow Transplant Group. Bone Marrow Transplant 1994; 13: 233–237.
Anasetti C, Amos D, Beatty PG, Appelbaum FR, Bensinger W, Buckner CD et al. Effect of HLA compatibility on engraftment of bone marrow transplants in patients with leukemia or lymphoma. N Engl J Med 1989; 320: 197–204.
Horan J, Wang T, Haagenson M, Spellman SR, Dehn J, Eapen M et al. Evaluation of HLA matching in unrelated hematopoietic stem cell transplantation for nonmalignant disorders. Blood 2012; 120: 2918–2924.
Spellman S, Bray R, Rosen-Bronson S, Haagenson M, Klein J, Flesch S et al. The detection of donor-directed, HLA-specific alloantibodies in recipients of unrelated hematopoietic cell transplantation is predictive of graft failure. Blood 2010; 115: 2704–2708.
Guardiola P, Kuentz M, Garban F, Blaise D, Reiffers J, Attal M et al. Second early allogeneic stem cell transplantations for graft failure in acute leukaemia, chronic myeloid leukaemia and aplastic anaemia. French Society of Bone Marrow Transplantation. Br J Haematol 2000; 111: 292–302.
Kanda J, Horwitz ME, Long GD, Gasparetto C, Sullivan KM, Chute JP et al. Outcomes of a 1-day nonmyeloablative salvage regimen for patients with primary graft failure after allogeneic hematopoietic cell transplantation. Bone Marrow Transplant 2012; 47: 700–705.
Min CK, Kim DW, Lee JW, Min WS, Kim CC . Additional stem cell therapy for graft failure after allogeneic bone marrow transplantation. Acta Haematol 2000; 104: 185–192.
Schriber J, Agovi MA, Ho V, Ballen KK, Bacigalupo A, Lazarus HM et al. Second unrelated donor hematopoietic cell transplantation for primary graft failure. Biol Blood Marrow Transplant 2010; 16: 1099–1106.
Waki F, Masuoka K, Fukuda T, Kanda Y, Nakamae M, Yakushijin K et al. Feasibility of reduced-intensity cord blood transplantation as salvage therapy for graft failure: results of a nationwide survey of adult patients. Biol Blood Marrow Transplant 2011; 17: 841–851.
Wolff SN . Second hematopoietic stem cell transplantation for the treatment of graft failure, graft rejection or relapse after allogeneic transplantation. Bone Marrow Transplant 2002; 29: 545–552.
Jabbour E, Rondon G, Anderlini P, Giralt SA, Couriel DR, Champlin RE et al. Treatment of donor graft failure with nonmyeloablative conditioning of fludarabine, antithymocyte globulin and a second allogeneic hematopoietic transplantation. Bone Marrow Transplant 2007; 40: 431–435.
Gyurkocza B, Cao TM, Storb RF, Lange T, Leisenring W, Franke GN et al. Salvage allogeneic hematopoietic cell transplantation with fludarabine and low-dose total body irradiation after rejection of first allografts. Biol Blood Marrow Transplant 2009; 15: 1314–1322.
Yoshihara S, Ikegame K, Taniguchi K, Kaida K, Kim EH, Nakata J et al. Salvage haploidentical transplantation for graft failure using reduced-intensity conditioning. Bone Marrow Transplant 2012; 47: 369–373.
Sumi M, Shimizu I, Sato K, Ueki T, Akahane D, Ueno M et al. Graft failure in cord blood transplantation successfully treated with short-term reduced-intensity conditioning regimen and second allogeneic transplantation. Int J Hematol 2010; 92: 744–750.
Fuji S, Nakamura F, Hatanaka K, Taniguchi S, Sato M, Mori SI et al. Peripheral blood as a preferable source of stem cells for salvage transplantation in patients with graft failure after cord blood transplantation: A retrospective analysis of the registry data of the Japanese Society for Hematopoietic Cell Transplantation. Biol Blood Marrow Transplant 2012; 18: 1407–1414.
Lang P, Mueller I, Greil J, Bader P, Schumm M, Pfeiffer M et al. Retransplantation with stem cells from mismatched related donors after graft rejection in pediatric patients. Blood Cells Mol Dis 2008; 40: 33–39.
Dvorak CC, Gilman AL, Horn B, Cowan MJ . Primary graft failure after umbilical cord blood transplant rescued by parental haplocompatible stem cell transplantation. J Pediatr Hematol Oncol 2009; 31: 300–303.
Liu H, Wang X, Geng L, Tang B, Tong J, Yao W et al. Successful second transplantation with non-myeloablative conditioning using haploidentical donors for young patients after graft failure following double umbilical cord cell transplantation. Pediatr Transplant 2010; 14: 465–470.
Atsuta Y, Suzuki R, Yoshimi A, Gondo H, Tanaka J, Hiraoka A et al. Unification of hematopoietic stem cell transplantation registries in Japan and establishment of the TRUMP system. Int J Hematol 2007; 86: 269–274.
Bacigalupo A, Ballen K, Rizzo D, Giralt S, Lazarus H, Ho V et al. Defining the intensity of conditioning regimens: working definitions. Biol Blood Marrow Transplant 2009; 15: 1628–1633.
Gaziev D, Polchi P, Lucarelli G, Galimberti M, Sodani P, Angelucci E et al. Second marrow transplants for graft failure in patients with thalassemia. Bone Marrow Transplant 1999; 24: 1299–1306.
Larocca A, Piaggio G, Podesta M, Pitto A, Bruno B, Di Grazia C et al. Boost of CD34+-selected peripheral blood cells without further conditioning in patients with poor graft function following allogeneic stem cell transplantation. Haematologica 2006; 91: 935–940.
Remberger M, Ringden O, Ljungman P, Hagglund H, Winiarski J, Lonnqvist B et al. Booster marrow or blood cells for graft failure after allogeneic bone marrow transplantation. Bone Marrow Transplant 1998; 22: 73–78.
Deeg HJ, Amylon ID, Harris RE, Collins R, Beatty PG, Feig S et al. Marrow transplants from unrelated donors for patients with aplastic anemia: minimum effective dose of total body irradiation. Biol Blood Marrow Transplant 2001; 7: 208–215.
Deeg HJ, O'Donnell M, Tolar J, Agarwal R, Harris RE, Feig SA et al. Optimization of conditioning for marrow transplantation from unrelated donors for patients with aplastic anemia after failure of immunosuppressive therapy. Blood 2006; 108: 1485–1491.
We thank all the clinicians and leaders of hospitals and centers who provided precise data via a registry of the Japan Society for Hematopoietic Cell Transplantation (JSHCT). Script kindly given by Dr Yoshinobu Kanda was used for data manipulation.
MK, KM, RS and TF designed the research. HY, MI, HK, JI, K Koh, YH, HT, AS, AK, HS, K Kawa and K Kato collected the data. MK analyzed the data, and MK and TF wrote the manuscript. All the authors discussed the results and commented on the manuscript.
The authors declare no conflict of interest.
About this article
T cell replete-haploidentical second hematopoietic stem cell transplantation for primary graft failure in pediatric patients with hematologic malignancies
Pediatric Transplantation (2017)
Autologous Cord Blood Cells Infusion as Salvage Therapy for Engraftment Failure After Haploidentical Hematopoietic Stem Cell Transplantation in Acute Myeloid Leukemia
Pediatric Blood & Cancer (2016)
Successful Engraftment and Durable Graft-versus-host Disease Control with Haploidentical Peripheral Blood and a Short-term Conditioning Regimen for Primary Graft Failure
Internal Medicine (2015)
Successful early unmanipulated haploidentical transplantation with reduced-intensity conditioning for primary graft failure after cord blood transplantation in hematologic malignancy patients
Bone Marrow Transplantation (2015)
Donor-specific anti-human leukocyte antigen antibodies were associated with primary graft failure after unmanipulated haploidentical blood and marrow transplantation: a prospective study with randomly assigned training and validation sets
Journal of Hematology & Oncology (2015)