The 44^th Annual Meeting of the European Society for Blood and Marrow Transplantation: STATISTICAL SYMPOSIUM - POSTER SESSION

Copyright: Modified and published with permission from http://www.ebmt2018.org/

Sponsorship Statement: Publication of this supplement is sponsored by the European Society for Blood and Marrow Transplantation.

STATISTICAL SYMPOSIUM - POSTER SESSION

P901 Predictors of Outcome for Allogeneic Transplantation - Analysis of 1545 patients in the largest transplant center in Taiwan

Xiu-Wen Liao ¹, Jia-Hau Liu ^1,2, Ming Yao ², Chi-Cheng Li ^1,2, Bor-Sheng Ko ², Chien-Ting Lin ^1,2, Cheng-Hong Tsai ¹, Shang-Ju Wu ², Shang-Yi Huang ², Hsin-An Hou ², Meng-Yao Lu ³, Kai-Hsin Lin ³, Shiann-Tarng Jou ³, Jih-Luh Tang ^1,2, Yung-Li Yang ³, Dong-Tsamn Lin ³;

¹ Tai-Cheng Stem Cell Therapy Center, National Taiwan University, Taipei, Taiwan, Republic of China; ² Division of Hematology, Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan, Republic of China; ³ Division of Hepatology, Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan, Republic of China

Background: There are many patient, disease, donor, and transplant-related predictor model for the outcome of allogeneic hematopoietic stem cell transplant (allo-HSCT). However, these factors had never been confirmed in Chineses population in Taiwan. We have diverse HLA haplotype in its island history. In this retrospective analysis, we aimed to confirm these prognostic factors of different outcome of allo-HSCT.

Methods: Between 1984 Jan and 2016, 1545 consecutive patients receiving allo-HSCT at National Taiwan University Hospital were included. Their transplant data were collected following the EBMT Registry data collection forms and manuals. Clinical variables of recipients and donors were evaluated the Cox proportional hazardassumption. Transplant outcomes were measured in terms of 3-year overall survival rate (OS), 3-year non-relapse mortality rate (NRM), 3-year cumulative incidence of relapse (CIR), 100-day cumulative incidence of grade 2 to 4 acute GVHD (aGVHD), and 3-year cumulative incidence of chronic GVHD (cGVHD). Univariate and multivariate analysis were performed using Cox proportional hazard regression model to identify risk factors of different outcome. This study was approved by the hospital Research Ethics Committee.

Results: The median follow up time was 98 months. In multivariate analysis (Table 1), high risk disease status before allo-HSCT had significantly strong association with inferior OS (HR:1.74, 95% CI: 1.42–2.14, p < 0.001), NRM (HR:1.7, 95% CI:1.34–2.14, p < 0.001), and CIR (HR:1.4, 95% CI:1.06–1.87, p = 0.017). HLA disparity was significantly associated with worse OS and more aGVHD. Offspring donors and donors over 40 years old had double risk to develop grade 2 to 4 aGVHD (HR: 2.57, 95% CI: 1.53–4.33, p = 0.0004; HR: 1.84, 95% CI: 1.11–3.07, p = 0.0177). Interestingly, female donor to male recipient was associated with higher cGVHD (HR:1.33, 95% CI:1.08–1.62, p = 0.0066). Conditioning regimen with ATG had lower risks ofdeveloping cGVHD (HR:0.52, 95% CI:0.43–0.65, p < 0.0001). Peripheral blood stem cells and bone marrow combined with peripheral blood stem cells were significantly associated with higher cGVHD (HR:3.2, 95% CI:2.4–4.3, p < 0.001). CMV serology state was not significantly associated with any outcome in multivariate analysis.

Conclusions: Our data confirmed some factors being predictive to specific allo-HSCT outcomes, while the others are not. The differences maybe attributable to the ethnic and HLA diversity, conditioning regimen variability, or pharmacogenetic difference. Establishment of our own predictors for the allo-HSCT outcome may be important.

Conflict of interest: None

P902Changes, challenges and real-life data about stem cell transplant in multiple myeloma: The Granada population-based registry

Rafael Ríos-Tamayo ¹, María Esther Clavero Sánchez ¹, Dolores Sánchez Rodríguez ², Ana Beatriz Rivera Ginés ¹, Juan Sainz Pérez ³, Pedro Antonio González Sierra ¹, Elisa López Fernández ¹, Lucía Moratalla López ¹, Trinidad Roldán Benítez ¹, Teresa Rodríguez Ruiz ⁴, José Luís García de Veas Silva ⁴, Pilar Garrido Collado ¹, Estefanía Morente Constantín ¹, Jon Badiola González ¹, Manuel Jurado Chacón ¹

¹ University Hospital Virgen de las Nieves, Hematology, Granada, Spain; ² FIBAO, Monoclonal Gammopathies Unit, Granada, Spain; ³ GENYO, Center for Genomics and Oncological Research, Granada, Spain; ⁴ University Hospital Virgen de las Nieves, Clinical Laboratory, Granada, Spain

Background: The outcome of multiple myeloma (MM) is significantly improving in the past decades in terms of overall survival (OS)(Ríos-Tamayo et al, 2015) due to the widespread use of new drugs, a better supportive care, and an optimized use of stem cell transplant (SCT).

Autologous SCT (ASCT) remains a standard of care in candidate MM patients and the chronological age is not longer an insurmountable barrier. Tandem ASCT is recommended for high-risk (HR) patients. Allogeneic SCT (AlloSCT) is increasingly used particularly in the context of late relapses. Therefore, SCT units have to face a significant increase in its workload.

Methods: We evaluated all newly diagnosed MM patients (NDMM) in our population-based registry (1985–2017) to analyze the burden and outcome of SCT over the past three decades in periods of five years, in terms of progression-free survival (PFS) and OS. HR smoldering MM patients were included. PFS and OS were estimated according to the Kaplan-Meier method and differences compared using the log-rank test.

Results: 742 NDMM patients have been included in our registry by December 2017, 367 males (49,3%) and 377 females, median age 67 years (12–93). 179 patients had a first SCT from 1995 (first transplant), 102 males (57%) and 77 females, median age 56 years (12–70). 175 patients had ASCT (ASCT1) and 4 myeloablative AlloSCT. 25 (14 %) of these 175 ASCT1 had a second SCT: 6 ASCT (2 tandem auto-auto, 4 as salvage therapy) and 19 reduced-intensity AlloSCT (RICAllo) (13 tandem auto-RICAllo, 6 as salvage therapy). Two patients had a third transplant (both RICAllo, one as tandem and one at relapse).

Median time to first ASCT was 10.3 months (m) (3.4–114). Median CD34 was 3,14 x10⁶. With a median follow-up of 29.2 m for surviving patients, the median PFS was 13 m. Median OS for ASCT1 was 94.5 m (73.9–115.2). Median OS for patients with only ASCT1, second auto (ASCT2) and second RICAllo (AlloSCT2) was 86.7 m, not reached and 94.5 m, respectively (p = 0.33). After excluding the two patients with a third transplant, median OS is 100.1 for ASCT2 and 94.5 m for AlloSCT2. We did not found statistically significant differences in OS in patients who had ASCT2 by the type of transplant (tandem vs salvage).

The evaluation of response after transplant has a significant impact on OS (p < 0.001) with a median OS not reached for patients achieving CR. In relation to the impact of HR cytogenetic abnormalities by FISH, patients with 1q+ had a significantly poorer OS (p = 0.05).

According with calendar periods, the number of ASCT1 transplant was: 1995–99:11, 2000–04:31, 2005–09:26, 2010–14:47, 2015–17:64.

Conclusions: MM is a heterogeneous relapsing disease. Despite the progressive improvement in OS, patients not achieving CR or those with early relapse after ASCT1 have a suboptimal outcome. A second transplant may benefit a subset of these patients. Besides the use of new drugs and the current debate about the use of chemotherapy, SCT remains a cornerstone in the management of MM patients, with an increasing use in both real-life and clinical trials patients.

Conflict of interest: Nothing to declare

P903

Abstract previously published

P904EBMT RISK SCORE IMPACT ON MORTALITY RATE FOR ALLOGENEIC HEMATOPOIETIC STEM CELL TRANSPLANTATION

Valdas Peceliunas, Orinta Mickeviciute, Rita Cekauskiene, Igoris Trociukas, Laimonas Griskevicius; VUL Santaros Klinikos, Vilnius, Lithuania

Background: The European Group for Blood and Marrow Transplantation (EBMT) risk score provides a simple tool to predict an outcome of allogeneic hematopoietic stem cell transplantation (alloSCT). EBMT risk score includes five factors : stage of the disease, recipient age, time from diagnosis, donor type (related/unrelated) and donor -recipient gender combination. The range of EBMT index can vary from 0 till 7. The higher index the higher risk of mortality.

Methods: We retrospectively analyzed patients who undergone alloSCT at a single institution transplanted during 2015–2017 years. Data about pre transplant EBMT risk score and survival data were collected.The statistical data analysis was conducted using SPSS program. The differences were considered statistically significant at p ≤ 0,05.

Results: Records of 153 allogeneic bone marrow transplantation (BMT) recipients transplanted during 2015–2017.10 in Vilnius University Hospital Santaros Klinikos were evaluated. EBMT risk score was calculated for 153 recipients and was further analyzed. Median age of BMT recipients was 54 (21–76 years). Main indication for allogeneic HCT was acute myelogenous leukemia (75 (49%) patients). Median risk score was 3 (range 1–6) . Most recipients (N 60, 39,2 %) obtained EBMT index 3. There were two main risk factors that caused higher index - age and donor type (most of the cases is unrelated donor). After median follow up of 8 months 98 patients (64,1%) were alive and 55 patients (35,9%) died. Proportion of surviving patients with different pre-transplant EBMT risk scores displayed in table 1. There was a tendency for longer overall survival (OS) for patients with pre transplant EBMT risk score of ≤3 in comparison to patients with risk score above 3 (median OS 16 months was not reached p = 0.059) (figure 1). EBMT score and mortality risk did not differ statistically significantly between male and female patients. EBMT index score did not affect overall survival significantly in age groups below and above 40 years.

Conclusions: Outcomes of 153 patients after allogeneic SCT were analyzed in respect of pre-transplant EBMT risk score. The most common EBMT risk score was 3. The main established risk factors were age and donor type. There was a tendency for shorter OS for patients with EBMT risk score >3.

Conflict of interest: Nothing to disclose

EBMT index	No of recipients	Status alive N (%) / dead alive (%)
1	5 (3,3%)	5 (100%) / 0 (0%)
2	21 (13,7%)	17 (81%) / 4 (19%)
3	60 (39,2%)	39 (65%) / 21 (35%)
4	35 (22,9%)	21 (60%) / 14 (40%)
5	28 (18,3%)	14 (50%) / 14 (50%)
6	4 (2,6%)	2 (50%) / 2 (50%)

[ [P904 Table] EBMT index in HCT recipients.]

P905Global variation in unrelated hematopoietic stem cell donation during 2010 to 2015

Monique Jöris, Lydia Foeken; WMDA, Leiden, Netherlands

Background: A key project of the World Marrow Donor Association (WMDA) is the yearly WMDA annual report that has become an established instrument to describe the current status of haematopoietic stem cell (HSC) donation worldwide. They are used to observe trends and changes in number and type of HSC donation as well as the global exchange of these products.

Methods: Over 100 organisations around the world are requested to report data on number of unrelated volunteer donors and cord blood products listed in their database, as well as various characteristics of these donors and cords and the number and type of products provided to which countries. Quality control measures include validity cross checks within the annual questionnaire system as well as crosschecks between years manually by the WMDA staff.

Results: Numbers of HSC-cord, -aphaeresis and -marrow donations from 2010 to 2015 are shown per continent in the figure below. In Africa donation of HSC-apheresis was stable throughout the period 2010–2015 with around 10 donations per year. In Asia a growth was reported initially for all three stem cell types. However, from 2013 the number of donations stabilised except HSC-cord donations are still increasing. In Europe we see a decline in HSC-cord donation while HSC-apheresis and -marrow are increasing. Organisations in North America initially saw a growth for all three stem cell types, reaching a peak in 2013. After that decreases are reported, especially for HSC-cord. In Oceania a decrease in the shipment of HSC-cord was seen. While in South America donations of HSC-apheresis and -marrow grew rapidly, the use of HSC-cord declined.

Numbers of donations shipped nationally are also different for each continent. In Asia around 90% of the donations are shipped within Asia itself and this number has not changed from 2010 to 2015. In Africa (92% in 2010) and South America (99% in 2010) the percentage of nationally shipped HPC products is high too. However, in both regions it is reported that more products are shipped internationally in 2015. North America and Oceania ship about 70% nationally and no significant difference was reported between 2010 and 2015. Organisations in Europe ship the lowest percentage nationally (36%) and this number is steadily decreasing (31% in 2015).

Conclusions: The data shows the type and number of unrelated HSC donations varies widely across world regions.

Most remarkable observation is that In several regions a decrease in the use of CBUs was observed, mainly in Europe, North America and Oceania where the use of HSC-cord was long established. Coincidently, since 2010 there has been a marked increase of 96% in the number of transplants performed from haploidentical relatives. For some countries, HLA haploidentical HSCT from family members is an attractive approach because of the cost effectiveness. Unrelated donor registries and cord blood banks are expensive to set up and to maintain. The decrease is mainly seen in the adult patient population, while the use of cord blood for children patients remains stable (data not shown).

Conflict of interest: The authors have nothing to disclose.

[ [P905 Figure] Shipments by HSC type per continent 2010–2015]

P906Predicting Day-100 TRM in patients with Acute Leukemia who received Myloablative (MA) conditioning regimen and transplanted using matched related donor allo-HCT: A Machine leaning study

Tusneem Elhassan, Naeem Chaudhri, Syed Osman Ahmed, Walid Rasheed, Fahad Alsharif, Hazzaa Al Zahrani, Amr Hanbali, Riad Elfakih, Marwan Shaheen, Feras Alfraih, Saud Alhayli, Shahrukh Hashimi, Fahad Almohareb, Mahmoud Aljurf;

King Faisal Specialist Hospital & Research Centre, Oncology (Section of Adult Haematology/BMT), Riyadh, Saudi Arabia

Background: Predicting Day-100 TRM is an important measure of the success of allo-HCT. However, it is still faced with limitations regarding statistical methodology and type of risk factors proposed. Previous studies were conducted in a relatively heterogeneous group. This study aims to Identify risk factors associated with Day-100 in a homogenous group of patients using Machine Learning (ML) approach to confirm the previous finding.

Methods: This is a retrospective study based on a prospectively collected data of a uniform treatment protocols at KFSH&RC that included 764 patients with Acute Leukemia underwent allo-HCT between 1997 and 2013. Baseline characteristics such as age, donor-recipient-sex-match, disease stage, graft type, diagnosis and time from diagnosis to allo-HCT were used in model building. Synthetic Minority Over-Sampling Technique (SMOTE) was utilized to adjust for the imbalanced class distribution of Day-100 TRM. Feature selection (FS) algorithm was utilized to identify the most important predictive features. Applied ML algorithms included Naïve Bayes (NB), K-Nearest Neighbor (KNN), Multilayer perceptron (MLP), Bagging, Random Forest (RF), Random Committee (RC) and Logistic regression(LR). Learning curves were created to check model overfitting using different sample sizes. Model accuracy was evaluated using Area Under Receiver Characteristic curve (AUC). Predictive models were built using WEKA 3.9.1.

Results: FS revealed that Stem cell source, Donor-recipient-sex-match and Disease stage were the most important features in predicting Day-100 TRM for this group. However, Time from Diagnosis to allo-HCT and donor sex were of least importance, table (1). All learning curves plateaued when choosing (≥ 70%) of the data. There was no evidence of overfitting, Figure (1). All models had a good performance based on AUC measure. AUC was 0.91, 0.94, 0.92, 0.92, 0.92, 0.93 for J48, RF, KNN, MLP, bagging and RC compared to 0.87 for LR, p-value (<0.05). NB was not significantly different from LR, AUC = 0.85. In this study RF showed the best performance while NB and LR showed the least.

Conclusions: Machine Learning methods showed superiority in predicting Day-100 TRM compared to classical models such as LR. Stem cell source, Donor-recipient-sex-match and Disease stage were the most predictive features.

Conflict of interest: None declared

Features	Ranking
Stem cell source	62.1868
Donor-recipient-sex- match	59.3014
Disease Stage	58.694
Recipient sex	58.2384
Age	56.2642
Diagnosis	54.4419
Time from Diagnosis to all-HCT	52.0121
Donor sex	52.0121

[ [P906 Table] Features ranking using the FS algorithm]