Accounting for long-term survivors and population-expected survival improves survival outcomes estimation and prediction in older patients with core-binding factor (CBF) Acute Myeloid Leukemia (AML)

Abstract

Abstract Introduction While disease-related features such as recurrent genetic abnormalities are routinely used to inform patient (pt) prognosis, pt-associated features including age also impact pt outcomes. Older pts have worse outcomes and inferior performance of routine genetic response prognosticators, exemplified by the inferior survival of older CBF-AML pts, who, based on current risk stratification, should have favorable outcomes including cures with chemotherapy alone. This highlights the critical need to improve our analytical tools for survival outcome prediction. We hypothesize that in population (popn)-based studies, it is crucial to adjust for popn-expected mortality (PEM) as older adults are often subject to competing mortality risks from unrelated causes in addition to the disease in question. Moreover, when the data comprises long-term (LT) survivors (depicted by a long plateau in Kaplan-Meier curves), it is imperative to account for a potentially 'cured’ subgroup of pts. Thus, to accurately estimate survival, PEM-adjusted mixture cure models (MCMs) may be critical tools, as they postulate that the sample comprises a mixture of LT survivors who only experience mortality from unrelated causes, and susceptible pts who may additionally experience excess (disease-related) mortality..Methods We analyzed 264pts with CBF-AML [47% with inv(16)/t(16;16), 53% with t(8;21)] aged 60-85 years (yrs; 80% White, 43% female), diagnosed between 2014 and 2021 in a U.S. popn-based dataset from Flatiron Health. PEM was adjusted for by matching pts on age, sex, and year of diagnosis using U.S. life tables from the Human Mortality Database. Pts are considered cured when their survival time reaches or surpasses survival expected for persons in the general popn without AML, such that their excess mortality risk becomes negligible and they are only subject to background mortality. MCMs were fit univariately to estimate proportion cured (PC), net survival at 5 yrs (NS-5) and time-to-cure (TTC) across various subgroups and to identify baseline features that are associated with cure and/or NS. TTC was defined as the time at which AML-related mortality is negligible.Results In the classical model, NS-5 (95% CI) was 21% (16%, 26%). Adjusting only for PEM, NS-5 was 24% (19%, 30%). Adjusting only for cure but not PEM, NS-5 was 21% (17%, 27%) and PC was 16% (11%, 23%). However, when both cure and PEM were properly accounted for, NS-5 was 27% (21%, 34%), with a PC of 25% (18%, 33%). The higher estimates reflect our adjustment for PEM, which appropriately separates AML-specific mortality from PEM. Overall, the estimated TTC for the pt cohort was 5.4 (2.9, 8) yrs. Pts aged 60, 72 (median) and 85 yrs at diagnosis had a NS-5 of 51%, 23% and 9%, PC of 41.3%, 21.7% and 9.2%, and TTC of 9.2, 5.4 and 2.9 yrs, respectively. Because older pts have shorter expected lifespan, the apparent early TTC for older pts potentially reflects high PEM for much older pts rather than actual cure, as PC reduced with age. Consistent with prior studies, the classical survival model predicted significantly (sig.) worse survival for t(8;21) pts compared to inv(16)/t(16;16) (HR= 1.35, P=.04) with NS-5 (17% vs 26%). However, the PEM-adjusted MCM found no such sig. difference in older pts (HR=1.4, P=.16; OR for cure =.76, P=.48). Likewise, there was no sig. subtype difference for PC (22% vs 27%), NS-5 (23% vs 30%), or TTC (5 vs 6.7). Similar patterns were observed for sex. In univariable MCMs, bone marrow (BM) blasts (HR=1.26, P=.04), ECOG value (HR=1.6, P=.006), urea nitrogen (HR=1.03, P=.005), diastolic blood pressure (HR=.97, P<.001), chromosome 5 deletion [del(5q)] (HR=1.83, P=.02), NRAS (HR=0.4, P=.01), and IDH2 (HR=2.6, P=.03) mutations were all sig. associated with mortality risk among susceptible pts. BM blasts (OR=.15, P=.01), creatinine (OR=0.15, P=.01) and del(5q) (OR=0.18, P=.02) were sig. associated with cure.Conclusion Our study highlights the need to fit PEM-corrected MCMs when analyzing popn-based data, especially in older pts susceptible to non-negligible background mortality or in molecular pt subgroups that, based on their known biology, should include pts with chances of cure, to avoid underestimating the net survival experience. Among older pts, our PEM-adjusted MCMs predicted sig. differences in cure and NS for certain pretreatment predictors. These findings motivate further studies on potential disparities in outcomes among older CBF-AML pts.