Prognostic Factors Identified Via a Cox Proportional Hazard Cure Model Explain Survival Differences After Allogeneic Hematopoietic Stem Cell Transplantation (HSCT) in Chronic Myeloid Leukemia (CML) Between Two Consecutive Patient Cohorts

Abstract

AbstractAbstract 3748 Introduction:By genetic randomization according to availability of a matched related donor, the two consecutive German CML studies III and IIIA (recruitment from 1995 to 2001 and from 1997 to 2004) evaluated the role of early HSCT compared to medical treatment. Transplantation in first chronic phase (1st CP) was performed in 113 of 135 patients (84%) randomized to HSCT in study III and in 144 of 166 (87%) patients in study IIIA. Although transplantation protocols were comparable and most centers participated in both studies, even after adjustment for the established EBMT risk score (Gratwohl et al., Lancet 1998 [1]), post-transplant survival probabilities in study IIIA were significantly higher (p = 0.0073). Patients and Methods:Our aim was the evaluation of these survival differences. The German Registry for Stem Cell Transplantation and the Swiss Blood Stem cell Transplants Group provided data for an independent assessment of survival after first HSCT with a related donor performed in 1st CP between 1995 and 2004. Matching inclusion and exclusion criteria with those of the 257 patients of the two German studies, data of 607 patients were retrieved from the two registries. Early transplant-related mortality as opposed to hardly any event in later years suggests a time-dependent hazard of death. Hence, for the identification of prognostic factors, the Cox proportional hazard cure model was used where the population is considered as a mixture of susceptible (prone to an event) and non-susceptible (cured) individuals (Sy and Tylor, Biometrics 2000). The established pre-transplant risk factors age, recipient sex, donor sex, time between diagnosis and HSCT, calendar year of HSCT, stem cell source, and HLA matching were investigated as potential prognostic factors for survival. Parameter estimation was performed by application of the SAS macro of Corbière and Joly (Comput Meth Prog Bio, 2007). Results:Five-year survival probabilities were 73% for recipients of a related donor HSCT in the registry and 65% and 79% in the German studies III and IIIA, respectively. In the independent dataset of the registry, survival after HSCT was more favorable if performed after the year 1999. Because the previously published cut-points “1 year” for time from diagnosis to HSCT ([1]) and “44 years” for age at HSCT (Maywald et al., Leukemia 2006) were independently confirmed to separate survival probabilities the best, dichotomization was considered as an alternative to the originally continuous scale. Applied to the German CML study data, in the best model the “probability of cure” was significantly influenced by age (≤44 vs. >44 years, p < 0.0001), time from diagnosis (≤1 vs. >1 year, p=0.0304) and calendar year of transplant (≤1999 vs. >1999, p =0.0176) whereas the survival probabilities among the failure patients were best explained by HLA matching (p = 0.0348) and, again, age (p = 0.0067). Under consideration of weights and interactions, the possible combinations of the identified factors could be summarized in 4 risk groups with significantly different survival probabilities (at 5 years: 98%, 74%, 57%, and 20%). With the lowest risk group as reference level, all other levels contributed to a significant discrimination of the “cure probability” as well as of the survival probabilities among failure patients (maximum: p = 0.0196). When added as a further factor, study origin (CML III vs. IIIA) had no significant influence, whether in the model with the original variables or in the model with the risk groups. Conclusions:In a direct modeling approach by a multiple Cox proportional hazard cure model, the established risk factors age at HSCT, HLA matching, time from diagnosis to HSCT, and calendar year of HSCT were confirmed as independent prognostic factors which had a significant influence on the cure proportion and/or post-transplant survival probabilities. Using these factors or their resulting risk stratification, study origin lost its influence on cure and survival probabilities. Including the difference in time, the more favorable risk distribution in study IIIA could explain the significantly better survival outcome in comparison to study III. In addition, random variation might have a share in the outcome discrepancy. Together with information from the independent patient cohort, the cure model provided a novel tool to assess survival differences in consecutive patient study arms treated under comparable conditions. Disclosures:Hochhaus:Novartis, Bristol-Myers Squibb: Research Funding. Scheid:Novartis: Honoraria. Hasford:Novartis: Research Funding. CML Study Group:Kompetenznetz Leukämie, European Leukemia Net, Roche, Essex, AMGEN: Research Funding.