Page 84 - 2019_05-HaematologicaMondo-web

N. Gagelmann et al.
Introduction
Myelodysplastic syndromes (MDS) are a heteroge- neous group of clonal hematopoietic disorders that are characterized by abnormal cellular maturation resulting in cytopenia and a variable risk of progression to acute leukemia.1 Allogeneic stem-cell transplantation is still the only curative treatment option.2-4 In order to recommend MDS patients for transplantation besides considering only disease-specific factors, such as those proposed by the International Prognostic Scoring System (IPSS) and its revision (IPSS-R),5,6 a valid and readily reproducible trans- plant-specific scoring system is needed.7,8
Recently, two groups have developed prognostic sys- tems incorporating patient-specific as well as transplant- specific factors. The Gruppo Italiano Trapianto di Midollo Osseo (GITMO) registry developed a transplantation risk index consisting of the following factors: age (<50 or ≥50 years), IPSS-R score (low, intermediate, high, or very high), monosomal karyotype (yes or no), refractoriness to chemotherapy (yes or no), and Hematopoietic Cell Transplantatiom Comorbidity Index score (HCT-CI).9 The resulting risk index could clearly distinguish four dif- ferent groups (low, intermediate, high, and very high) with corresponding overall survival rates at 5 years of 76%, 48%, 18%, and 5%. The other prognostic score, by the Center for Blood and Marrow Transplant Research (CIBMTR) registry, included the following criteria: age (18-29, 30-49, or ≥50 years), Karnofsky status (90-100 or <90%), cytogenetics (very good to intermediate, poor, very poor or monosomal karyotype), blood blasts before transplantation (>3 or ≤3%) and platelet count before transplantation (>50 or ≤50 x 109/L). This system also dis- criminated four risk groups (low, intermediate, high, and very high) and showed corresponding overall survival rates at 3 years of 71%, 49%, 41%, and 25%.10 Both sys- tems resulted in relevant re-classification of patients in comparison with the IPSS-R while providing only modest improvement in predictability. While neither score has been externally validated nor investigated regarding prog- nostic power in large and independent cohorts, we hypothesized that both scores not only vary in design and follow-up but would also vary in prognostic ability when applied to the same cohort.7,8
We, therefore, aimed to develop and validate a clinical and transplant-specific prognostic score using data from a large cohort of MDS patients reported to the European Society for Blood and Marrow Transplantation (EBMT) registry and to validate and compare currently existing systems with respect to the resulting EBMT transplant- specific risk score.
Methods
Data source
The EBMT is a voluntary organization that comprises more than 500 transplant centers, mainly from Europe. Membership requires submission of the minimal essential data form A for all consecutive patients to a central registry from which patients can be identified by diagnosis of underlying disease and type of transplantation. The information in the minimal essential data form A data is updated annually. Informed consent to transplan- tation was obtained and data were collected locally according to
regulations that were applicable at the time of transplantation. All transplantation centers were required to obtain written informed consent before data registration with the EBMT in accordance with the 1975 Helsinki Declaration.
Patients
Adult patients (≥18 years) with MDS who underwent trans- plantation from an HLA-identical sibling or matched unrelated donor between 2000 to 2014 were included. Patients were eligi- ble if there was full information on their: diagnosis, donor data, cytogenetic risk, platelet count and blood blasts at transplanta- tion. Cytogenetic risk was stratified based on previously estab- lished systems.6,11 The prognostic subgroups were the following: del(11q) and -Y (very good); del(5q), del(12p), del(20q), and nor- mal karyotype (good); del(7q), +8, i(17q), +19, and other inde- pendent clones (intermediate); complex karyotype (three abnor- malities), inv(3), del(3q), and translocations involving 3q (poor), and very complex karyotype with more than three abnormali- ties (very poor). Monosomal karyotype was defined as mono- somy of two or more chromosomes or one single autosomal monosomy in the presence of other structural abnormalities.12 The IPSS-R was calculated prior to transplantation. In total, 1059 patients met the criteria and were included in the EBMT cohort. To evaluate possible selection bias, outcomes were compared between the final cohort and remaining patients not included in the analysis due to missing data in the registry (n=5122). Within the EBMT cohort, 519 and 876 patients had full data on all fac- tors included in either the GITMO or CIBMTR score.
Score development
The development of the transplant-specific risk score consist- ed of two steps. First, a Cox proportional hazards model using backward and forward selection was used to identify significant covariates for overall survival.13 Then, the hazard ratios (HR) obtained were classified as large (HR >1.59), intermediate (1.25< HR <1.60) and small (HR <1.25). Subsequently, a scoring rule was defined in which large effects were assigned two points, intermediate effects were assigned one point and small effects were assigned zero points. Scores were grouped based on asso- ciated hazard ratios into low-, intermediate-, high-, and very high–risk groups, providing group-based risk predictions for MDS. A second score was developed, based on the β coefficients derived from the model defined above, to provide individual- ized/patient-specific risk predictions. Second, both developed scores were validated and then compared to existing systems by assessing each score’s prognostic performance.
Statistical analysis
Overall survival and relapse-free survival were estimated using the Kaplan-Meier method and compared with the log-rank test in univariable analysis. Non-relapse mortality and relapse were ana- lyzed in a competing risks framework by using the cumulative incidence estimator and the Gray test for univariable analysis.14 Cox proportional hazards regression of complete data was used to develop the two new scores. Maximum likelihood from the Cox model was used to establish cutoffs for continuous variables. Score performance was analyzed using the concordance index (C): the probability that a patient who experienced an event had a higher risk score than a patient who did not (C >0.5 suggesting predictive ability).15,16 Each system was validated using 5-fold cross-validation with 100 repetitions. P values <0.05 were consid- ered statistically significant. Analyses were performed using SPSS for Windows version 24 (SPSS, Chicago, IL, USA) and R package version 3.4.3 (The R Foundation, Vienna, Austria).
930
haematologica | 2019; 104(5)