Abstract
Background: Prognostic models for AML patients (pts) predict outcomes based on clinical and cytogenetic features at diagnosis. However, as a pt lives beyond the initial date of diagnosis following treatment, his/her estimated survival period improves and these models lose accuracy. Conditional survival (CS) probability estimations use the time a pt has survived since diagnosis to improve on prognostic accuracy and provide personalized survival information. Here we report population-level, 5-yr relative cancer survival probabilities of AML pts conditional on having already survived ≥1 yr(s) after diagnosis.
Methods: A total of 13,933 adult AML cases diagnosed during 2000-2009, and followed through 2014 were assembled from all 18 registries of the Surveillance, Epidemiology and End Results (SEER). Pts who did not receive chemotherapy in the first 3 months after diagnosis were excluded. Survival probabilities were calculated using the SEER Survival System (SEER*Stat). We calculated relative survival (RS) probability estimates that compare survival in the pt cohort relative to the expected survival of the general population (matched for age, sex, race and calendar period over a standard follow-up time). Trends in 5-year RS estimates conditional on already having survived 0-5 years from diagnosis were modeled using regression modeling. Estimates were adjusted for age group [adolescents and young adults (AYA) (18-39 yrs); middle-aged (40-59 yrs) and elderly (> 60 yrs)], yr of diagnosis, years already survived (YAS), sex and race. We performed subset analyses for AML categories with adequate sample size for analysis in SEER - AML with myelodysplasia (MDS)-related changes or AML-MRC (n = 794), core-binding factor (CBF) AML (n = 462) and acute promyelocytic leukemia (APL) (n = 1499). Tests of significance were based on two-sided hypothesis at the .05 level. Analyses were performed using STATA.
Results: Figure 1A shows survival of all AML pts relative to the expected survival of the general population by age groups. At 10 yrs after diagnosis, the RS probabilities of AML (all categories) was 50% for AYA, 35% for middle-aged and 10% for elderly cohort. When conditioned on YAS, the estimated 5-year RS probabilities of AML pts consistently improved with successive yrs as shown in Figure 1B. Figures 2A-2C show RS probability estimates for specific AML categories by age subgroups - 10-yr RS was highest for APL [80% - AYA, 70% - middle-aged, 50% - elderly] as expected but surprisingly lower than expected in CBF-AML [65% in AYA group, 50% in middle-aged and 30% in elderly]. Figures 2D-2F illustrate the influence of AML categories and advancing age on CS estimates. The estimated 5-yr RS probabilities when conditioned on successive YAS increased for all age groups and for each AML category. The 5-yr relative CS was highest in APL pts among all AML categories and almost reached 100% for all three age groups, thereby approximating survival of the general population. Pts with CBF-AML had high 5-yr relative CS probabilities of 90% in all age groups, consistent with their favorable prognosis. The 5-yr conditional RS probability for AML-MRC exceeded 90% and 80% for AYA and middle-aged groups, respectively but reached only 60% in the older cohort. AML-MRC is considered a biologically heterogeneous disease and this finding suggests a biologically aggressive course in older patients. In multivariable analyses, the 5-yr RS estimates were significantly lower in > 60 yrs cohort (vs. 18-39 yrs, 36.09% drop (95% CI, 35.81-36.09)] but better in females (vs. males, 5.07% improvement (95% CI, 4.89-5.25), with YAS [conditioned on 1 to 5 YAS, 5-yr RS improved from 19.92% (95% CI, 15.03-24.81) to 53.77% (95% CI, 49.55-57.98)] and in recent treatment period [2006-2009 vs, 2002-2005, 9.07% improvement (95% CI, 8.93-9.21).
Conclusions: Following treatment completion, survival prognosis in AML can be further refined using a population-derived, 5-yr CS estimates based on YAS. While CBF-AML is often considered "favorable-risk" AML, long-term RS probabilities belied this, with unexpectedly lower survival probabilities (~30-65% at 10 yrs). This data suggests the need for a longer period of leukemia surveillance following treatment completion. RS and CS information can be applied clinically at the individual level and can be used in the design of clinical trials to determine appropriate length of follow up.
Advani: Takeda/ Millenium: Research Funding; Pfizer: Consultancy. Gerds: CTI BioPharma: Consultancy; Incyte: Consultancy. Majhail: Anthem, Inc.: Consultancy; Sanofi: Honoraria. Sekeres: Celgene: Membership on an entity's Board of Directors or advisory committees.
