The association of a novel psychosocial and mobility assessment with outcomes of veterans with acute myeloid leukemia (AML)

Abstract Background: Real-world outcomes of AML patients are consistently inferior to those of pivotal clinical trials, partly due to psychosocial factors that hinder timely access to medical care and medication adherence. The independent risk conferred by psychosocial and mobility (PSM) challenges, beyond age, physical frailty, and AML-related variables has yet to be examined. Our objective was to determine the association electronic health records (EHR)-based indicators of PSM on outcomes of veterans with AML, treated within the national Veterans Affairs (VA) Healthcare System. Methods: Veterans diagnosed with AML between 1/1/2012-4/1/2022 treated with either cytarabine-based intensive chemotherapy (IC) or venetoclax combinations (VC) were included in this retrospective analysis. Clinical, demographic, and laboratory data were obtained from the VA Corporate Data Warehouse with manual annotation of AML-specific variables. With preliminary input from leukemia-focused oncologists, we came up with 13 structured measures of PSM including: area deprivation index (ADI) score 6-10 (versus 1-5), rural residence, body mass index (BMI) ≥30 kg/m2, unmarried status, unemployment history, immobility, home oxygen need, comorbid substance use disorder, psychiatric disorder, and/or tobacco dependence, unstable housing, limited socioeconomic support, and income below cohort median. Based on the presence (1 point [pt]) or absence (0 pts) of each measure at the time of AML diagnosis, a score was calculated for each veteran, with 13 pts indicating highest psychosocial vulnerability. Regression models assessed the association of 1) individual PSM domains and 2) increasing PSM score with objective response rate (ORR) and overall survival (OS), stratified by treatment and adjusting for demographics, electronic VA-frailty index, and AML-related variables. Results: Of 1,166 veterans, 97% were male, 74% were White, and median age was 69 (interquartile range [IQR]: 63-74). Most (61%) received IC. Veterans treated with VC were more likely to be ≥75 years (52% vs 7%), frail (73% vs 55%), present with adverse risk AML by EuropeanLeukemia Net Classifications (61% vs 46%), and platelet count<20,000/µL at diagnosis (14% vs 11%, p<0.001 for all). The median PSM score was 4 (IQR: 2-5) for both groups; 96 (8.2%) veterans had a score ≥7. By individual measures, the VC group had higher rates of immobility (39% vs 14% in IC, p<0.001) and unstable housing (13% vs 9% in IC, p=0.02). In contrast, ADI decile score ≥6 (higher neighborhood deprivation) was more prevalent in the IC group (57% vs 49% in VC, p=0.01) along with unmarried status (51% vs 43% in VC, p=0.002). During follow-up, 909 (78%) patients died; the median OS for VC was 7.1 months (95% CI=6.4-8.4) and 9.6 months (95% CI=8.5-10.3) for IC. On multivariable analyses for the VC group, unemployment history (OR=0.5, 95% CI=0.27-0.94), oxygen requirement (OR=0.1, 95% CI=0.01-0.39), and comorbid psychiatric disorders (OR=0.6, 95% CI=0.32-0.997) were associated with lower ORR; unemployment history (HR=1.5, 95% CI=1.08-1.96), oxygen requirement (HR=2.95, 95% CI=1.39-6.22), and housing instability (HR=1.6, 95% CI=1.03-2.5) were significantly associated with increased hazard of death. For VC treated veterans, every 1 pt increase in PSM was associated with an 18% decrease in ORR (95% CI=0.72-0.94) and a 9% increased hazard of death (95% CI=1.01-1.16). In the IC group, while tobacco dependence was associated with lower ORR (OR=0.6, 95% CI=0.41-0.99), other individual PSM measures and increasing PSM score were not associated with an increased hazard of death. Conclusions: We highlight initial efforts to develop an EHR-based PSM assessment for veterans with AML. In veterans treated with VC, 3 individual PSM measures negatively impacted ORR and survival, along with an increasing PSM score. In contrast, for veterans induced with IC, only tobacco dependence was associated with worse ORR. Even in the VA where there are fewer barriers to access than private health systems, there may be additive risk of inferior treatment response and higher mortality when measures of PSM accumulate across multiple domains. Oral/outpatient regimens as VC are likely more sensitive to the impact of PSM relative to inpatient regimens as IC where the burden of medical appointments and medication adherence are mitigated by structured observation and psychosocial support. We are actively refining the assessment to validate in an external cohort.