We performed a retrospective cohort study of adults diagnosed with AML from 1/2014-8/2022 using the Flatiron HealthTM (FH) electronic health record (EHR)-derived database, a longitudinal database comprising de-identified, patient-level, structured and unstructured data curated via technology-enabled abstraction from ~280 academic and community cancer clinics (800 sites of care) across the United States [13, 14]. It is not claims-based but instead uses data abstracted from the EHR. While the FH database has similar demographic and geographic distributions to other national databases including the Surveillance, Epidemiology and End Results Program and National Program of Cancer Registries, ~75% of patients in the FH database are treated in community practices, and it has greater representation of the American South.

Our primary objective was to characterize sociodemographic associations with novel therapy use, with a novel therapy defined as a drug that received FDA approval during the study period. These were glasdegib, venetoclax, ivosidenib, midostaurin, gemtuzumab ozogamicin, gilteritinib, enasidenib, and CPX-351. Exploratory objectives were to determine sociodemographic associations with venetoclax use (the most commonly prescribed novel therapy) and site-level characteristics associations with early novel therapy adoption. The study received approval from the Dana-Farber/Harvard Cancer Center Office for Human Research Studies. Restrictions apply to the availability of these data, which were used under license from FH. The data that support the findings of this study have been originated by FH, Inc. They are de-identified and subject to obligations to prevent re-identification and protect patient confidentiality. Requests for data sharing by license or by permission for the specific purpose of replicating results in this manuscript can be submitted to dataaccess@flatiron.com.

Sociodemographic characteristics analyzed in the primary analyses included patient age, sex, and race-ethnicity, time from FDA approval to treatment event (defined below), practice type (academic or community), practice ID, and physician ID. Exposure time, defined as days between diagnosis and death or last follow up, was used as an offset variable. US population-weighted quantiles of an area-level measure of SES based on the Yost Index was available for patients treated at community sites of care [15]. SES rank-grouped patients based on a factor analysis of area-level characteristics from the American Community Survey (2015–2019) at the census block group level, including median household income, rent, poverty, employment, and education. SES was analyzed as low (bottom two quantiles) versus medium/high (top-three quantiles). Exploratory analyses of community sites were performed including SES as a covariate. The outcome measure was a treatment event, defined as the start of a new therapeutic regimen excluding hydroxyurea or research-based regimens. Treatment events were determined by oncologist-defined, rule-based lines of therapy specific for AML from technology-enabled abstraction and EHR-documented treatment administration data, then dichotomously categorized as including or not including a novel therapy. Treatment events were included starting with first-line treatment following diagnosis of AML and all subsequent line-of-therapy initiations until the end of the study period, end of EHR activity, or all-cause death determined from a combination of data from structured and unstructured EHR, commercial sources and the Social Security Death Index [16].

In primary analyses, race-ethnicity was defined dichotomously as POC or non-Hispanic (NH)-White, with POC an aggregate of individuals listed as NH-Asian, -Black, and -Other races and Hispanic Black, White, and Other races. An aggregated measure was used as small numbers for multiple race-ethnic groups made disaggregated models unstable. These categorizes and aggregations were based on the mutually exclusive combinations of race and ethnicity used by the FDA and the American Medical Association’s (AMA) definition of POC [17, 18]. The data source had two variables that indicated Hispanic status: a Hispanic indicator variable and a Hispanic race variable category. Hispanic status was identified through the Hispanic indicator in 99.5% of cases, and in 3.2% of cases, Hispanic was listed as a race category. In the primary analysis the POC term included individuals with Hispanic status identified through either variable. A sensitivity analyses were performed for each outcome using the race variable only, comparing POC as Hispanic, Asian, Black, or Other to White.

To determine site-level characteristics associations with early novel therapy adoption, we characterized sites as “early adopting” if they were below the median time to novel therapy use across the cohort (empirically identified as 91 days after FDA approval). Transformation to site-level variables was as follows. Race-ethnicity was transformed to the proportion of patients at a site who were POC; age and SES were transformed to the median values for patients at the site. Other site-level characteristics included the number of physicians in the dataset treating patients with AML and the number of treatment events at the site.

Baseline characteristics were reported descriptively; Chi-square or Wilcoxon-Rank-Sum tests examined differences between those who did or did not receive a novel therapy. Differences in novel therapy treatment events by race-ethnicity were assessed through mixed-effects Poisson regression with exposure time as an offset variable and with patient race-ethnicity, age, sex, and treatment site type as fixed effects and practice ID as a random effect [19]. An exploratory analysis of community sites was performed using a similar model with SES included as a fixed effect. Differences in use for the most frequently used novel therapy, venetoclax, were assessed at community sites using multinomial regression with the same variables and controlling for the competing risk of treatment with another novel therapy. Multivariable logistic regression was used to assess associations between treatment site characteristics and being an early adopting site. Due to multicollinearity between number of treating physicians, total treatment volume, and number of patients treated, only number of treating physicians was included in the model. Similarly, SES and site region was strongly collinear and, thus, only SES was included.

All p-values were two-sided, and the significance level was set to 0.05. No adjustment for multiplicity was applied. All analyses were performed using R version 4.2.1 (the CRAN project, www.cran.r-project.org) and RStudio (Posit Software, Boston, MA). Packages lme4 (for the mixed-effects Poisson regression) and nnet (for the multinomial regression) were used.