Health Disparities in Acute Myeloid Leukemia Patients Undergoing Treatment with Tyrosine Kinase Inhibitor (TKI) Therapy Targeting FLT3, IDH1, or IDH2

Introduction

Acute Myeloid Leukemia (AML) is the most common and fatal acute leukemia in adults, with an incidence of 3 to 5 cases per 100,000 in the US.¹ The use of Tyrosine Kinase Inhibitors (TKIs) represents a major advancement in the treatment of AML, with FDA approved therapies for AML with mutations in FLT3, IDH1 and IDH2; among others becoming common use.² While incidence of these targetable mutations appears to be evenly distributed across different races and ethnicities in the US,^3,4 evidence has shown that African American patients overall have generally worse health outcomes when compared to Non-Hispanic White (NHW) patients, likely attributed to socioeconomic factors and structural racism.⁵ Little has been published regarding the clinical effectiveness of these TKIs across different racial/ethnic groups. Additionally, there is growing evidence that real world (RW) clinical outcomes often underperform thresholds established by clinical trials, likely due to patient selection and other confounding factors,⁶ and this has been noted in other cancer settings.⁷ Finally, underrepresentation of ethnic minorities in clinical trials has remained an ongoing concern⁸ despite efforts,⁹ and RW data can offer insights on how diverse populations outcomes may differ¹⁰ as well as potentially confirming improvements in efforts to bring about equitable care for diverse populations.

Materials and Methods

This study used the US-based, electronic health record (EHR)-derived de-identified Flatiron Health Research Database, comprising patient-level structured and unstructured data curated via technology-enabled abstraction.¹¹ During the study period, the data originated from approximately 280 United States cancer clinics (~800 sites of care). Patients were included in the study if they had a confirmed diagnosis of AML (excluding promyelocytic and mixed phenotype leukemia), were ≥18 years at diagnosis, and had documentation of treatment with a TKI targeting FLT3, IDH1, or IDH2 as monotherapy or with an HMA from January 1, 2015, until December 31, 2023. Patient demographics, baseline characteristics, AML-related information, and treatments were captured via structured and unstructured documents in the EHR. The European Leukemia Net (ELN) 2017 risk classification was derived from available molecular and cytogenetic data. We addressed missing covariate values using the missing indicator method for most variables to maintain sample size, with the exception of race/ethnicity for which a complete case analysis approach was applied. ELN risk was obtained through utilization of a combination of baseline cytogenetic and molecular testing data. To ensure risk assessment reflected baseline disease characteristics, we included cytogenetic and molecular results from specimens collected within 30 days of the initial diagnosis. Samples collected prior to the diagnosis date or those with missing collection dates but associated with the diagnostic workup were also included. Cytogenetic criteria first looked for the presence of an “adverse” marker, which included the presence of t(6;9), 11q23.3, t(9;22), inv(3), or deletions of chromosomes 5, 7, or 17; then “intermediate” included presence of t(9;11); and finally, “favorable” included presence of t(8;21) or inv(16). Similarly, molecular criteria first looked for the presence of an “adverse” marker, which included mutations in GATA2, RUNX1, ASXL1, or TP53, or the absence of NPM1 mutation in the presence of FLT3-ITD; then “intermediate” included the co-occurrence of NPM1 mutation and FLT3-ITD, or the absence of both NPM1 and FLT3-ITD; and finally, “favorable” included presence of CEBPA mutation, or NPM1 mutation in the absence of FLT3-ITD. For patients with both cytogenetic and molecular data available, the final risk category was assigned based on a hierarchical “worst-case” approach. If a conflict existed between the two modalities, the “adverse” classification took precedence over “intermediate”, and “intermediate” took precedence over “favorable.” In instances where molecular or cytogenetic data were missing, unavailable, or did not meet the specific criteria for favorable or adverse risk, patients were conservatively classified as “intermediate”. For racial/ethnic identification, patients were classified either as non-Hispanic/Latinx white (NHW) and POC (patients of color including Hispanic/Latinx patients; patients who were both Non-Hispanic/Latinx or an unknown ethnicity as well as Black, Hispanic/Latino, Asian, or a race not previously listed).

The primary outcomes of this study included real-world event-free survival (rwEFS) and real-world overall survival (rwOS). For rwEFS, this analysis considered any of the following to be an event: induction failure (ie., induction treatment failure or primary refractory disease), relapse (ie., reappearance of blasts in the peripheral blood, >5% blasts in the bone marrow, or the development of extramedullary disease), or death. For rwOS, the Flatiron Health database captures death by using information from EHR data supplemented with external commercial sources and US Social Security Death Index data and has been benchmarked against the National Death Index.¹² For the purposes of de-identification, date of death was only available at the month-level and/or year-level granularity.

We conducted a descriptive analysis of patients’ baseline clinical and treatment-related characteristics overall, and by race/ethnicity groups. We conducted several time-to-event analyses using first TKI initiation as the index date. For rwEFS analyses, patients were censored at the last date recorded during which a patient had the potential to be observed to have an event (ie., the last time that a clinic note was recorded, a lab specimen was collected, or a peripheral blast was measured). For rwOS analyses, patients were censored at last confirmed patient activity (ie., the latest occurrence of patient vitals, medication administration, laboratory testing, or abstracted treatment data). A two-sided Log rank test was used to test for statistically significant differences at an alpha level of 0.05. Resultant p < 0.05 were considered “statistically significant”.

We used the Institute of Medicine (IOM) definition of health disparity to guide variable inclusion. The IOM framework defines disparities as differences not justified by clinical characteristics or patient preferences. The framework posits that socioeconomic factors lie in the pathway between race/ethnicity and outcomes and therefore should not be adjusted.¹³ We initially performed an unadjusted Cox analysis to assess for any difference in outcomes between NHW and POC groups and then adjusted for clinically appropriate factors that can justifiably be expected to cause differences in health outcomes (ie., age, sex, and ELN risk stratification).

Results

Baseline Demographic and Clinical Characteristics are displayed in Table 1, among all patients (N = 482), median age at diagnosis was 69 (IQR: 60, 76) while median age at TKI initiation was 70 (IQR: 61, 77). About half (52%) of patients were male; 70% were NHW, 5.6% Hispanic/Latino, 5.6% Black and non-Hispanic/Latinx or an unknown ethnicity, 3.3% Asian and non-Hispanic/Latinx or an unknown ethnicity, 8.1% either another race not listed and non-Hispanic/Latinx or an unknown ethnicity, and 8.1% race not documented; 48.1% had Medicare as their primary insurance; 52% received therapy in a community setting (versus an academic medical center); 52% of patients had an ECOG of 0–1; 18.4% received TKI+HMA; and 10% of patients received a stem cell transplant (SCT) after initiating TKI targeted therapy. Per ELN 2017 risk stratification, 6% of patients were favorable, 52% intermediate, and 42% adverse. Among patients receiving TKI in 2^nd line or later (N = 384), POC were more likely to be younger at diagnosis (median age 64 vs 68 years), receive treatment in a Community Hospital (57% vs 44%), have Medicaid as primary insurance (8.8% vs 2.4%) and to be in the lowest socioeconomic index quintile (24.2% vs 7.8%) when compared to NHW patients. Other characteristics were similar among groups, including ELN risk stratification and percentage of patients receiving TKI for each targetable mutation (FLT3, IDH1 and IDH2).

Table 1 Baseline Demographic Characteristics

No significant differences were noted in rwOS and rwEFS outcomes by race/ethnicity regardless of whether patients received a TKI as 1^st or 2^nd Line of Therapy. When receiving 1L TKI, rwEFS was 2.2 [95% CI: 0.8–5.0] vs. 2.6 [95% CI: 1.1–NR] months; rwOS was 13.8 [95% CI: 10.5–23.4] vs. 9.2 [95% CI: 8.5–not reported (NR)] months for NHW and POC patients, respectively. When receiving 2L+ TKI, rwEFS was 2.5 [95% CI 2.1–4.3] vs. 3.2 [95% CI: 1.8–6.5] months (Figure 1A); rwOS was 9.7 [95% CI: 8.3–11.5] vs. 9.9 [95% CI: 7.4–15.1] months for NHW and POC patients respectively (Figure 1B). POC (vs NHW) patients receiving TKI in 2L+ showed no significant race/ethnicity differences in the hazard of rwOS in unadjusted models (HR: 1.26, p = 0.160) or models adjusted for sex, age at TKI initiation, and ELN 2017 risk stratification (aHR: 1.27; p = 0.132).

Figure 1 Unadjusted Kaplan–Meier curves describing rwEFS and rwOS among AML patients (excluding promyelocytic and mixed phenotype leukemia) who were ≥18 years at diagnosis and treated with a TKI targeting FLT3, IDH1, or IDH2 as monotherapy or with an HMA from January 1, 2015, until December 31, 2023 from Flatiron Health Electronic Health Record (EHR)-derived data in 2L+ and who did not receive Stem Cell Transplant (SCT) prior to TKI Initiation. Index date is date of TKI initiation. (A) rwEFS (B) rwOS.

Discussion and Conclusion

In this RW analysis of a diverse cohort of patients with AML treated with TKIs for FLT3, IDH1, and IDH2, we found that rwEFS and rwOS were relatively short, however, they did not differ significantly among different ethnic groups nor did hazard ratios suggest disparities. Our cohort represents some of the largest RW data on TKI therapy in POC, particularly when compared to registrational trials for available targeted therapies, with a reported 14% POC patients (N=70/482), with 27 patients being Hispanic or Latino, 27 being black and 16 being Asian; with 77 patients having unknown or not reported race. In previously reported trials, for instance the randomized Phase 3 ADMIRAL study which compared Gilteritinib to salvage chemotherapy in Relapsed/Refractory (R/R) FLT3 mutant AML, there were 41.2% POC (N=102/247) patients, which was mostly made up from Asian patients (N=69), with only 14 patients being Black; interestingly both Asian and Black patients appeared to derived greater benefit from TKI therapy, which even reached statistical significance for Asian patients.¹⁴ For IDH1, the initial Phase I trial that granted Ivosidenib approval for IDH1 mutant R/R AML did not report any data on race/ethnicity,¹⁵ however, the Phase I/II trial which combined Ivosidenib and Azacitidine for newly diagnosed IDH1 mutant AML reported 24.7% POC (N=24/97), though no post-hoc analysis was performed to assess for difference in effectiveness across race/ethnicity.¹⁶ For IDH2, the initial Phase I trial that granted Enasidenib approval for IDH2 mutant R/R AML reported 22.6% POC (N=78/345), however 50 of these patients fell into the “not reported” category. Only 19 patients were Black and 27 were Hispanic or Latino.¹⁷ In the confirmatory randomized Phase III trial which compared Enasidenib to Conventional Chemotherapy Regimens in R/R IDH2 mutated AML, 27.2% patients were POC (N=43/158), however 29 of these patients fell into the “not reported” category, while only 2 patients identified as Black, 12 as Asian and 4 as Hispanic or Latino.¹⁸ Neither of these studies performed post hoc analysis across race/ethnicity and both include an important number of patients with “unreported” race/ethnicity which makes the true percentage of POC significantly smaller. While smaller sample sizes prohibited us from performing further race/ethnicity-stratified analyses, when analyzing POC as whole we did not see any significant differences in clinical outcomes including rwEFS and rwOS. This coincides with reported data showing that Hispanic¹⁹ and Black patients derive significant benefit from targeted³ and less-intensive therapies,²⁰ thus bridging the disparity gap, which persists in cohorts treated with intensive or standard chemotherapy agents,²¹ this finding is likely due to POC patients having a higher comorbidity burden,²² which can compromise their ability to receive and tolerate more intensive treatment regimens. Recent studies have shown potential differences in AML biology²³ both at the molecular level and in regard to pharmacogenomics and associated drug disposition,²⁴ but these potential biological differences did not manifest as clinical outcome differences in our cohort, however this could be a result of the overall short duration of rwEFS and rwOS for patients with AML in our cohort or due to the reported relatively equal distribution of targetable mutations across Hispanic⁴ and Black patients.³ Additionally, we found that some baseline sociodemographic characteristics were different between groups, with POC being generally younger at diagnosis, treated in a community setting, had Medicaid as primary insurance, and belonged to the lowest socioeconomic quintile. Regarding age, previous reports have shown that both African American²⁵ and Hispanic patients^20,26 are younger than NHW at diagnosis, which is reflected in our cohort, however, whether this is a reflection of underlying biological or environmental factors is unknown. POC are known to have decreased access to tertiary care institutions,²⁷ which would account for the higher rate of treatment in the community setting and to have higher rates of poverty and social disadvantage,²⁸ which would correlate with our findings; however, the lack of difference in outcomes in these groups may be driven by access to TKIs and other less-intensive chemotherapies in our cohort and the overall poor prognosis of patients with AML for all groups, this highlights the potential role TKIs and other less intensive therapies could have in decreasing disparities in this high-need population. These findings should alert clinicians to consider prioritizing the use of TKIs in diverse populations, while acknowledging that the overall survival outcomes for AML remain poor. Limitations in our study include the fact that POC were in one analytical group due to limited sample sizes, which could mask race- and ethnicity- specific signals and relatively low (n = 70) overall numbers of POC limits statistical power. Further studies should examine TKI treatment patterns and outcomes with more race/ethnicity granularity.