Survival and Prognosis in Patients with First-Line Imatinib Treatment Under Particular Consideration of Death Due to Chronic Myeloid Leukemia

Introduction: The IN-study section of the European Treatment and Outcome Study (EUTOS) registry comprises data on imatinib-treated patients with chronic myeloid leukemia (CML) who were enrolled between 2002 and 2006 in prospective, controlled clinical trials. Of those, 2290 adult patients had Philadelphia chromosome-positive chronic-phase (CP) CML and were eligible for analysis and prognosis of long-term survival. Improved survival increased the percentage of deaths not related to CML. While adjusting for this, our analyses put death due to CML into focus.

Aims: Based on the observed survival in our patient sample and on survival in matched population data, relative survival (RS) probabilities attributable to the excess hazard of CML should be calculated for the 2290 patients. These results were to be opposed to cumulative incidences of mortality (CIM) when only death due to CML is considered as an event and all other causes of death as competing risks. The ability to discriminate CIM of dying from CML should be assessed for the established prognostic models Sokal, Euro, and EUTOS score and a possibly identified new model. Candidate factors were age, sex, spleen enlargement, hemoglobin, platelets, leukocytes, and percentages of blasts, eosinophils, and basophils in peripheral blood.

Methods:Survival time was calculated from the date of start of treatment to death or to the latest follow-up date. Survival was censored at the time of allogeneic stem cell transplantation in first CP. As “death due to CML”, only death after recorded disease progression was regarded. Progression was given by observation of accelerated phase or blast crisis, both defined in accordance with the recommendations of the ELN (Baccarani et al Blood 2013). RS probabilities were calculated by the method of Pohar-Perme (Comput Biol Med 2007) and CIM by the cumulative incidence function. Population data was downloaded from the Human Mortality Database (www.mortality.org). All prognostic factors were measured at baseline and the influence on CIM due to CML was estimated by the Fine and Gray (FG) model. Level of significance was 0.05.

Results:The 2290 patients came from study groups in Germany, France, Italy, Spain, the Netherlands, and the Nordic study group and had a median observation time of 6.4 years. Observed 8-year overall survival probability was 89% [95% confidence interval (CI): 87-90%] and 8-year RS probability 96% [95% CI: 93-97%]. Cause of death was due to CML in 92 of 208 cases (44%), unrelated to CML in 104 (50%), and unknown in 12 cases (6%). Eight-year CIM were 4% [CI: 4-5%] for causes of death due to CML and 7% [CI: 6-8%] for causes of death due to other reasons, including the unknown causes where no progression prior to death was observed. From low to high risk groups, in 2205 evaluable patients, the Sokal score resulted in 8-year CIM of 3% [95% CI: 2-4%], 4% [95% CI: 3-6%], and 7% [n=499, 95% CI: 5-10%] and the Euro score in 8-year CIM of 4% [95% CI: 3-5%], 3% [95% CI: 2-4%], and 12% [n=222, 95% CI: 8-17%]. The EUTOS score suggested two groups with 8-year CIM of 4% [95% CI: 3-5%] and 9% [n=232, 95% CI: 5-13%]. Higher age, more blasts, a bigger spleen size enlargement, and low platelet counts significantly increased the CIM of dying from CML. The four factors were combined in a new prognostic model. Here, 8-year CIM were 2% [n=1349, 95% CI: 1-3%], 6% [n=596, 95% CI: 4-8%], and 11% [n=260, 95% CI: 8-16%].

Conclusions: An 8-year RS probability of 96% corresponded to an estimated 4% probability of dying due to CML which actually was the same result as the one calculated for the CIM. However, while for the first method, access to matched population data is necessary but no knowledge on the cause of death, in the second case, investigators need to assess whether an individual died from CML or not. Using the “progression prerequisite”, the FG model was most likely only based on “real” cases of death due to CML. As causes of death without prior progression, like infection or treatment-related toxicities, might well be attributable to CML, the CIM of death due to CML were supposedly underestimated.

For assessment of comparability between patient samples, prognostic models built from baseline variables remain important. In comparison to other scores, only the new model identified three risk groups with pairwise significantly different CIM and led to the largest high-risk group with an 8-year CIM above 10%. Independent data for further comparisons are collected.