Racial inequities in second-line treatment and overall survival among patients with metastatic breast cancer

Background Black women in the USA have a higher incidence and mortality of metastatic breast cancer (mBC) than White women, while Hispanic women have lower rates. Previous studies have focused on first-line (1L) treatment, but little is known about racial differences in treatment beyond 1L and their impact on outcomes. Methods This analysis utilized data from an electronic health record derived de-identified database and included patients with HR+HER2- mBC initiating 2L treatment (including CDK4/6-inhibitor [CDKi]-based, endocrine monotherapy, everolimus combination therapy, and chemotherapy and other systemic therapies) between 2/3/2015 and 7/31/2021. Real-world overall survival (rwOS) was defined as time from 2L initiation to death. Multinomial logistic regression assessed the likelihood of 2L treatment between race/ethnicity groups. Median rwOS was estimated using the Kaplan–Meier method and adjusted hazard ratios were estimated using multivariable Cox proportional hazards models. Results Among all patients who received 2L, non-Hispanic Black (NHB) and Hispanic/Latino patients were less likely to receive 2L CDKi compared to non-Hispanic White (NHW) patients (36%, 39% vs 42%, respectively). Median rwOS was 20.4, 37.6, and 25.3 months, in NHB, Hispanic/Latino and NHW patients, respectively. The rwOS remained poorer among NHB patients after adjustment (HR = 1.16; p = 0.009). In stratified analysis, adjusted rwOS was similar between NHB and NHW patients among those who received 1L CDKi. Conclusions These findings suggest that among patients with HR+HER2- mBC, NHB patients had worse survival beyond front-line setting, mainly among the subset of women who did not receive CDKi at 1L. This inequities in rwOS between race/ethnicity groups was not observed among patients who received 1L CDKi. Supplementary Information The online version contains supplementary material available at 10.1007/s10549-022-06701-5.


Introduction
In the USA, African American or Black (AA) women not only have a higher incidence of metastatic breast cancer (mBC) but also increased breast cancer mortality compared to White women [1,2]. Although improved breast cancer screening and treatments have resulted in an overall decrease in the breast cancer mortality, the divergence in mortality trends between Black and White women remains large in the USA [3]. In contrast, both incidence and survival rates of mBC were lower among Hispanic/Latina women compared to White women [1]. The drivers of racial inequities in mortality are multifactorial and may include more aggressive tumor characteristics, unequal access to care, and poorer socioeconomic status, among AA women with BC, although previous cohort study using the Surveillance, Epidemiology, and End Results (SEER) database found that the disproportionate burden of death in AA women persisted in subgroups with higher socioeconomic status and less aggressive tumor types [2].
The systemic treatment of women with hormone receptor (HR) positive, HER2 negative (HR+HER2−) mBC has historically involved primary endocrine therapy (ET) with selective use of chemotherapy [4]. In the past, ET-based regimens were the preferred treatment option in the 1L, 2L, and subsequent-line settings [5]. However, resistance to ET usually leads to disease progression and ultimately deaths from breast cancer [6]. Recent evidence has established the addition of cyclin-dependent kinase 4/6 inhibitor (CDKi) to ET as a preferred approach for 1L and subsequent treatments [7,8]. Previous studies have shown patterns of care and possible health inequities in treatment delay [9], the use of CDKi as first-line (1L) treatments [2,10,11], and survival [12]. However, it is unknown whether the inequities persist after treatment initiation and beyond first-line treatment, and whether it may further result in inequities in patient outcomes. Meta-analysis of randomized clinical trials suggests that compared to ET alone, combination of a CDKi and ET improved overall survival (OS) in both 1L and 2L treatment subgroups [13]. However, trial subgroup analysis stratified by Black, Hispanic/Latino vs White race groups is not available in these trials [14]. Therefore, real-world evidence among patients with mBC treated in routine practice may provide important insights on this topic.
This study aims to investigate breast cancer treatment inequities across the care continuum for women with HR+HER2− mBC by examining differences in 2L treatment options and real world overall survival (rwOS) across racial or ethnic groups. We evaluate the use of CDKi, ET and chemotherapy in 2L setting, and rwOS from 2L initiation between race/ethnicity groups.

Data source and population selection
This study used the nationwide Flatiron Health electronic health record (EHR)-derived de-identified database. The Flatiron Health database is a longitudinal database, comprising de-identified data originating from approximately 280 US cancer clinics (800 sites of care) [15,16]. Included patients were adult patients with confirmed diagnosis of mBC on or after 1/1/2011, who had at least two EHRdocumented visits on or after 1/1/2011 and had received at least two lines of therapy. Patients also must have had HR + /HER2− disease as identified by any positive HR and only negative HER2 status test results measured on or before 1L initiation. Patients were excluded if they were male, had unknown dates or results of biomarker tests only, had a > 90 day gap between metastatic diagnosis and first documented visit (in order to minimize risk of missing treatment data), or if they received clinical study drugs for 1L or 2L treatment. To ensure assessment of CDKi exposure within the era when such drugs were widely available, only patients who initiated 2L treatment for mBC between 2/3/2015 (first FDA approval of a CDKi) and 7/31/2021 (at least six months before the end of follow-up at Jan 31, 2022) were included in the analysis.
Institutional Review Board approval of the study protocol was obtained prior to study conduct, and included a waiver of informed consent.

Variables and endpoints
The primary exposure variable was documented race/ethnicity, categorized as Non-Hispanic White (NHW; reference group), Non-Hispanic Black (NHB), Hispanic or Latino, and Other/Unknown. Patient-level demographic and clinical characteristics were ascertained using structured and unstructured data, curated via technology-enabled abstraction. Key covariates include: age at metastatic diagnosis (years), stage at initial diagnosis (stage I-III, stage IV and unknown), practice type (community vs academic), 1L treatment group (CDKi, endocrine therapy, or chemotherapy and others) 5 , duration of 1L treatment (days), progression during 1L (yes/no), progression within the first 6 months of 1L start (yes/no), ECOG performance status at the start of 2L treatment, insurance status, and number and sites of metastasis on or before 2L start. ECOG performance status was categorized as 0, 1, ≥ 2, and unknown, based on the ECOG value measured within 30 days prior to and 7 days after 2L initiation. Insurance status was categorized as commercial health plan payer (reference group), Medicaid, Medicare, self pay/undocumented, and other payer, based on the last insurance record before 2L initiation. Site of metastasis was categorized as visceral, bone only and others. Number of metastases was categorized as 1, 2, and ≥ 3. In addition, distribution of area-level socioeconomic status (SES) index (quintile) were also summarized by race/ethnicity group. The SES index was constructed using a factor analysis from seven American Community Survey (ACS, 2015-2019) characteristics of census block group-level social determinants of health following the approach of Yost [17,18]. Missing values of the variables that were used for constructing SES index were imputed using multiple imputation by chained equations [19]. Additional categories of unknown were used for variables with missing values, e.g. ECOG PS, insurance, SES index.
The primary outcomes were 2L treatment groups and real-world overall survival (rwOS). Treatment groups at 2L were categorized as: (1) any CDKi (monotherapy or combination therapies); (2) endocrine therapy alone; (3) everolimus combination therapy; and, (4) chemotherapy and others (reference group). The rwOS was defined as time from start of 2L treatment to date of death, or censoring at last confirmed activity [20]. Vital status (alive/death) and date of death were determined using de-identified patientlevel structured and unstructured data from the EHR, curated via technology-enabled abstraction, obituary data, and the public Social Security Death Index [20].

Statistical analyses
Descriptive statistics were used to compare demographic and clinical characteristics at 2L between race/ethnicity groups. Rates of receiving 2L treatment groups were estimated by race/ethnicity, and further stratified by 1L treatment (CDKi, vs chemotherapy/ET). Multinomial logistic regression was used to assess the likelihood of 2L treatment groups between NHB, Hispanic/Latino, Other/Unknown, and NHW (reference) patients, adjusted for demographics and clinical factors. For survival analysis, median rwOS was estimated using the Kaplan-Meier method and log-rank test. Adjusted hazard ratios (aHR) between race groups were estimated using multivariable Cox proportional hazards models, additionally adjusted for 2L treatment. Stratified analysis was conducted by 2L treatment groups. Sensitivity analyses omitting variables with high levels of missingness (ECOG status and insurance) and excluding patients with missing values in these variables were performed for both primary analyses. In exploratory analysis, we further stratified the survival analysis by 1L treatment (CDKi vs chemotherapy/ ET) and 2L treatment groups.
All analyses were conducted using R version 3.6.1.

Discussion
In this study, we did not observe statistically significant differences in the use of 2L CDKi across race/ethnicity groups among patients with HR+HER2-mBC, after adjusting for demographic and clinical factors. However, we found that NHB patients had poorer rwOS after 2L initiation, and the observed inequities in rwOS were mainly among the subset of women who did not receive CDKi at 1L. Our findings suggest that racial inequities in rwOS measured from the start of 2L treatment may be attributable to outcomes among women who do not receive CDKi during 1L treatment.
Racial differences in 2L treatment choices could be related to socioeconomic factors and access to care, potential differences in tumor biology (e.g. less targetable mutations, such as the PIK3CA mutation, for subsequent endocrine CDKi cyclin-dependent kinase 4/6 inhibitor; ECOG Eastern Cooperative Oncology Group; IQR interquartile range  Fig. 1 Histogram of 2L treatment groups by race/ethnicity treatments, or possible provider concerns related to drug toxicities or efficacy. Pooled analysis from eight clinical trials among women with HR+HER2-advanced breast cancer reported that CDKi combination therapy is associated with higher risks of all-grade and grade 3 and 4 (G3-4) toxicities, including neutropenia, leukopenia, anemia and nonhematologic events [14]. However, whether NHB or Hispanic/Latino patients experience more toxicities than NHW patients when treated with CDKi is unknown as the clinical trials that demonstrated the efficacy of the CDKi did not enroll adequate numbers of NHB or Hispanic/Latino patients and rates of toxicities were not examined in the different racial and ethnic groups. Thus, without clinical trial data related to rates of toxicities in different racial and ethnic groups, it is possible that some providers may have increased concerns about toxicities and tolerance of CDKi in NHB or Hispanic/Latino patients based off their clinical experiences with these drugs, which could contribute to lower rates of 2L CDKi use in NHB or Hispanic/Latino compared to NHW patients. Additionally, it is possible that the increased monitoring for toxicities that comes with the use of CDKi could be a barrier to their use experience more among non-NHW patients, since they may face more adverse socioeconomic factors, such as insufficient or more stringent insurance coverage resulting in higher co-pays for visits and treatments, or challenges associated with the costs of o the medication itself, and transportation to follow-up appointments, clinic visits or lab tests. We found that NHB or Hispanic/Latino patients were more likely to have Medicaid or uninsured, suggesting that financial constraints may play a role in their ability to access treatment with CDKi and receive the supportive care needed to remain on treatment. While commercially insured patients may qualify for savings on the treatment, patients on Medicaid insurance can only obtain conditional support for palbociclib. We found that the racial difference in rwOS measured after 2L initiation was driven by the subset of patients who did not receive CDKi at 1L. Meta-analysis from randomized clinical trials suggests that combination of CDKi and ET improves overall survival among patients with metastatic breast cancer in both 1L and 2L subgroups, compared to ET alone [13]. In this study, we found that about 40% of the patients treated in routine clinical practice in our study received CDKi as part of both 1L and 2L (Supplemental Fig. 1), although there is limited evidence on the efficacy of sequential CDKi treatment after progression and is the CDKi cyclin-dependent kinase 4/6 inhibitor; cOR crude odds ratio; aOR adjusted odds ratio a Adjusted for age at metastatic diagnosis, stage at initial diagnosis, practice type, 1L treatment group, duration of 1L treatment, progression within 6 months of 1L start, ECOG status at 2L start, number and sites of metastasis and insurance group; Race reference group: Non-Hispanic White; 2L treatment reference group: 2L chemotherapy and others Among patients who received CDKi as part of 1L and 2L treatment, the estimated median rwOS was similar between race and ethnic groups in this context. On the other hand, among patients who received CDKi only at 2L but not at 1L, we found that NHB patients had worse rwOS after 2L CDKi initiation compared to White patients. This novel finding suggests that CDKi therapy may have a potentially larger impact as 1L among NHB patients, and administering CDKi earlier may help to reduce the racial inequities in survival outcomes among patients with HR+HER2− mBC. In addition, we observed better rwOS among Hispanic/Latino patients, which is consistent with previous population-based studies [21,22], and known as "Hispanic Paradox" [23]. This observation has not been fully understood, but researchers have hypothesized that it could be due to selective immigration of healthy Hispanic/Latinos [24], the return of foregin-born Hispanic/Latinos to their native countries after diagnosis [25], and/or environmental and behavioral factors [25][26][27].
To our knowledge, this is the first study comparing treatment choices beyond front-line and corresponding survival outcomes across race/ethnicity groups among patients in the 2L setting for HR+HER2− mBC. Previous subgroups analysis by race among trial patients only reported Asian and non-Asian groups [14]. However, the difference in mortality is largest between NHB and NHW patients, and thus it is critically important that continued efforts are made to understand the drivers of this inequity and also implement solutions to mitigate it. Additionally, our study provides insight into treatment patterns in a patient cohort largely originated in community clinics, rather than large academic medical centers where the majority of clinical trial patients Table 3 Associations between rwOS and race by 2L treatment group a Adjusted for age at metastatic diagnosis, stage at initial diagnosis, practice type, insurance type, 1L treatment group, duration of 1L treatment, progression within 6 months of 1L start, ECOG performance status at 2L start,and number and sites of metastasis; Race reference group: Non-Hispanic White b Overall survival analysis for all patients who received 2L treatment was additionally adjusted for 2L treatment group c P-values for interaction term between race and 2L treatment groups were calculated in multivariate analysis were recruited from. Our study therefore supplements current knowledge of real-world treatment patterns and outcomes among patients with HR+HER2− mBC by leveraging high quality contemporary electronic health record data, and we were able to assess the longitudinal treatment patterns throughout the patients treatment journey. In addition, we were able to adjust for critical clinical characteristics, including detailed drug categories, oral treatment information and progression events, that were curated with abstraction from unstructured documents. We had sufficient long-term follow-up from 2L treatment initiation, and we were also able to assess survival outcomes using high quality real-world composite mortality data in our cohort. As a real-world retrospective cohort study, there are also some limitations. The treatment choices were not random. Although we adjusted for potential confounders including demographics, tumor characteristics and clinical factors at 1L, there is potential bias from confounding by unmeasured covariates. We are limited to information documented during the course of routine care. Our sensitivity analysis suggested that variables such as ECOG status and insurance might have been missing not at random, and the impact on missing values and observed racial differences warrant future investigations. Moreover, factors such as social network support and marital status that may also play a role in the patients' treatment decisions and disease outcomes are not available. Further studies are needed to understand the contribution of those factors to inequities. In addition, although we only included patients who had a documented visit within 90 days of metastatic diagnosis, there is still a possibility that patients may have got treatment outside our network. Lastly, we were not able to include other race/ ethnicity groups (e.g. Asian) in our analysis due to the small sample size in the patient population.
Our study is the first study to focus on the treatment choices and corresponding survival outcomes beyond the front-line among patients with HR+HER2− mBC. We found that the differences in treatment choices may be partially explained by tumor characteristics and socioeconomic status between race groups, and that administering CDKi earlier in the patient treatment journey may be associated with racial inequities in breast cancer survival outcomes. Our results