Racial disparities in overall survival after the introduction of cyclin-dependent kinase 4/6 inhibitors for patients with hormone receptor-positive, HER2-negative metastatic breast cancer

CDK4/6 inhibitors (CDK4/6i) combined with endocrine therapy have improved HR + /HER2− metastatic breast cancer (MBC) outcomes. However, it is still unclear whether the response to CDK4/6i is similar for all races. Therefore, we aimed to assess overall survival (OS) trends stratified by race in patients with HR + /HER2− MBC after the approval of CDK4/6i, as part of the standard of care, in 2015. We performed a population-based study using the SEER database. Patients with HR + /HER2− MBC were divided into two time-based cohorts: 1) pre-CDK4/6i era (diagnosed in 2011–2013) and 2) post-CDK4/6i era (diagnosed in 2015–2017). We used propensity score matching and identified 2,684 patients in each cohort that matched in several characteristics. Kaplan–Meier methods were used to estimate 2-year OS. Association between cohort and OS was evaluated using marginal Cox proportional hazards models with robust sandwich variance estimator. We conducted competing risk analysis to estimate the risk of breast cancer death in both cohorts. The 2-year OS rate was 65% for the post-CDK4/6i era and 62% for the pre-CDK4/6i era (stratified log-rank p = 0.025). The 2-year OS for non-Hispanic White (NHW) patients improved in the post-CDK4/6i era compared to the pre-CDK4/6i era (67% vs. 63%, p = 0.033). However, OS did not improve for non-Hispanic Black (NHB) (54% vs. 54%, p = 0.876) or Hispanic (67% vs. 65%, p = 0.617) groups. The risk of breast cancer death decreased in the post-CDK4/6i era as compared to the pre-CDK4/6i era (2-year risk of breast cancer death: 33% vs. 30%, p = 0.015); however, this effect was observed only in NHW (sHR 0.84, p = 0.005) women, but not in NHB (sHR 0.94, p = 0.630) or Hispanic (sHR 0.91, p = 0.550) women. Our study confirms that outcomes for HR + /HER2− MBC have improved after CDK4/6i were introduced in 2015. However, this effect is primarily driven by the improved OS in NHW patients, without significant improvement in OS in NHB or Hispanics.


Background
Breast cancer is the most common malignancy and the leading cause of cancer mortality among women worldwide [1,2]. Breast cancer outcomes have improved in the last three decades, primarily mediated by earlier detection and advances in therapeutic options. However, about 30% of women with early breast cancer develop metastatic breast cancer (MBC), an incurable disease with a 5-year overall survival (OS) rate of only 30%. Despite the widespread use of routine mammography screening, the incidence of de novo MBC has remained 1 3 relatively steady for decades suggesting that mammography screening does not eradicate the emergence of biologically aggressive MBC [3,4]. Hormone receptor-positive (HR +)/ Human epidermal growth factor receptor 2 negative (HER2−) is the most common breast cancer subtype, accounting for 70% of all breast cancer [5]. Endocrine therapy has been the cornerstone of systemic treatment for HR + /HER2− MBC; however, novel targeted therapies have emerged as treatment options for these patients.
The addition of cyclin-dependent kinase 4/6 inhibitors (CDK4/6i) such as palbociclib, ribociclib, or abemaciclib to endocrine therapy (ET) has improved survival outcomes for patients with HR + /HER2− MBC. The landmark trial PALOMA-2 showed that adding the CDK4/6i palbociclib to letrozole as first-line treatment increased the median progression-free survival (PFS) by 10 months [22]. Subsequent studies demonstrated the clinical efficacy of CDK4/6i in patients with disease progression after endocrine therapy [23]. The Food and Drug Administration (FDA) approved CDK4/6i combined with endocrine therapy as first-line for patients with HR + /HER2− MBC in 2015, which became the standard of care for these patients [24]. Most patients enrolled in the CDK4/6i landmark trials were NHW. NHB patients were underrepresented in these studies representing less than 2% in CDK4/6i landmark trials; moreover, the proportion of Hispanic patients enrolled is unclear.
It is still unclear whether the response to CDK4/6i is similar for all races. Therefore, we aimed to assess OS trends, stratified by race, in patients with HR + /HER2− MBC before and after the approval of CDK4/6i as part of the standard of care in 2015.

Data collection
We used the Surveillance, Epidemiology, and End Results (SEER) research plus database (2021 submission) to identify eligible cases. The follow-up cut-off for this database is December 31st, 2019. We included patients diagnosed with de novo MBC from 2011 to 2017. Other inclusion criteria included age > 18 years, ER and/or PR positive, and HER2 negative. Exclusion criteria included patients diagnosed by autopsy or death certificate, prior malignancies, bilateral breast cancer, non-metastatic disease, unknown metastatic status, and unknown ER, PR or HER2 status. Since we aimed to assess outcomes after CDK4/6i approval, we created two time-based cohorts: 1) patients diagnosed in 2011-2013 (pre-CDK4/6i era), and 2) patients diagnosed in 2015-2017 (post-CDK4/6i era). The deidentified data were determined exempt from informed consent by the Albert Einstein Institutional Board review, which approved the study.

Variables
We obtained data on demographic variables such as age, sex, and race; this last variable was collected in combination with ethnicity leading to 4 racial groups: non-Hispanic White (NHW), non-Hispanic Black (NHB), Hispanic from all races, and other race/ethnic groups (non-Hispanic Asian/ Pacific islander and non-Hispanic American Indian/Alaska Native). Socioeconomic variables included rurality and marital status. Rurality was obtained from the rural-urban continuum code from the SEER dataset and further dichotomized as metropolitan area or non-metropolitan (rural). We also collected information on breast cancer clinicopathological variables such as hormone receptor status (ER and PR), HER2 status, tumor grade, and metastatic sites (bone, brain, liver, or lungs). Treatment variables collected included surgery, chemotherapy, and radiation therapy. Breast cancer outcomes were assessed using vital status, survival in months, and cause of death.

Statistical analysis
We used the Pearson's X-square test in the overall cohort to assess the association between categorical variables and the student t test for continuous variables. Baseline characteristics variables were summarized with descriptive statistics.
Our cohorts differed in time and baseline characteristics. To decrease selection bias, we performed propensity score methods using complete case analysis to create matched cohorts. Then, we performed propensity score matching for the covariates sex, age, race, PR, ER, tumor grade, chemotherapy, surgery, radiation therapy, rurality, marital status, and metastasis to bone, brain, liver, and lungs. A 1-to-1 matching, without replacement, was performed using the nearest neighbor method with a caliper width equal to 0.1 standard deviations [25]. Matching was carried out using the Matchit package in R (version 4.3.3). We examined balance in the baseline covariates in the matched data by using standardized mean differences and variance ratios [25]. Furthermore, a Love-plot was created to visualize covariate balance for all variables included in the propensity score.
We also compared baseline characteristics between both cohorts after matching. We used McNemar's or Friedman's test for categorical variables and paired student t tests for continuous variables.

Survival analysis
OS was defined as the time in months from MBC diagnosis to death from any cause. Since the follow-up time could be longer for the pre-CDK4/6i cohort, patients from both cohorts were censored at 24 months, and we proceeded to analyze the 2-year OS. We estimated OS in the pre-CDK4/6i and post-CDK4/6i cohorts using Kaplan-Meier methods in the matched data set and compared OS in both groups by using the log-rank test stratified on matched pairs. Association between cohort and OS was evaluated using marginal Cox proportional hazards models with a robust sandwich variance estimator to account for clustering within matched sets [26]. Hazard ratios (HRs) were estimated from these models.
We conducted a competing risk analysis to estimate the risk of breast cancer death in both cohorts. To compare the risk of breast cancer death between cohorts, we estimated cumulative incidence functions (CIFs) and marginal subdistribution hazard ratios (sHRs) from clustered Fine and Gray models that accounted for the within-pair clustering of outcomes [25].
We also assessed OS and breast cancer death risk among races by using stratified Cox proportional models and Fine and Gray models, respectively, which were adjusted for clinicopathological, treatment, and socioeconomic covariates such as age, hormone status, HER2 status, tumor grade, metastatic sites, chemotherapy, radiation therapy, surgery, rurality, and marital status.
Two-sided P values and 95% confidence intervals (CIs) are reported. An α equal to 0.05 was used for all hypothesis testing. Statistical analyses were performed in R (Version 1.4.1106).

Propensity score matching
From the overall cohort, 2684 patients in the pre-CDK4/6i era were matched with 2684 patients in the post-CDK4/6i era (Fig. 1). The distribution of baseline covariates was adequately balanced in the matched data set (Table 1). Absolute standardized mean differences were < 0.1 for all covariates, suggesting a negligible difference between both cohorts; indeed, the largest standardized mean difference was 0.04 in the matched data set. Baseline characteristics between both groups were compared after matching. We found no statistically significant difference between cohorts in age, sex, race, marital status, rurality, tumor grade, hormone receptor status, surgery, chemotherapy, radiation, or metastatic pattern. Matched cohorts were used for further survival analyses.

Risk of breast cancer death before and after CDK4/6i approval
We used competing risk analysis to assess the risk of breast cancer death. The cumulative incidence function (CIF) curve for breast cancer death is depicted in Fig. 3. The estimated probability of breast cancer death at 2 years was 33% and 30% for the pre-CDK4/6i and post-CDK4/6i eras, respectively (p = 0.015) (Fig. 3). The risk of breast cancer death was reduced in the post-CDK4/6i era with an estimated SD standard deviation, Std. Mean Diff. standard mean difference, y years, NH non-Hispanic, ER estrogen receptor, PR progesterone receptor, NA not applicable, Met metastasis, Pos positive, Neg negative, Unk unknown  (Table 6 and SI 3).
In addition, we conducted competing risk analyses to identify factors associated with the risk of breast cancer death (both cohorts combined) (  (Table 5).

Discussion
Our study shows that the OS has improved for patients with HR + /HER2− MBC in the post-CDK4/6i era as compared to those in the pre-CDK4/6i era (2-year OS rate 65% vs. 62%, p = 0.025); however, this improvement was seen only for NHW patients (HR 0.87, p = 0.016), with no improvement for NHB (HR 0.98, p = 0.864) or Hispanic (HR 0.92, p = 0.578) women. The risk of breast cancer death decreased in the post-CDK4/6i era as compared to the pre-CDK4/6i era Fig. 1 Consort Diagram. From the Surveillance, Epidemiology, and End Results (SEER) research plus database, we identified 21,797 patients. Subsequently, patients with both estrogen and progesterone receptor-negative breast cancer, patients with HER2-positive tumors, and patients with unknown stage and bilateral breast cancer were excluded. A propensity score matching method was used to obtain two matched cohorts with 2,684. Abbreviation: PSM: Propensity score matching Our data suggest racial disparities in OS after the introduction of CDK4/6i in patients with HR + /HER2− MBC. Racial disparities in breast cancer outcomes have been well recognized [27]. Prior studies have demonstrated that Black women are diagnosed at more advanced stages [28] and are more likely to discontinue chemotherapy [29]. Studies have reported that early discontinuation of therapy, clinicopathological characteristics, and insurance might not completely explain the survival differences between NHB and NHW women [9,30]. In a population study, Huang reported that patients with MBC in the lowest socioeconomic status (SES) quintile had a significantly increased risk of breast cancer death compared to those in the highest SES quintile (sHR 1.  [31]. These results suggest that socioeconomic factors affect the risk of breast cancer death in patients with MBC. In our database, we only had access to information about rurality and marital status; therefore, we could not assess the full effect of SES on OS trends. Rurality has been associated with MBC outcomes, possibly mediated by distance to healthcare centers and access to screening and treatment [3,4,32]. A meta-analysis including 21 studies showed that women living in rural areas were 1.2 times more likely to be diagnosed with MBC when compared to women living in urban areas [33]. Our study also found that rurality was associated with worse OS (HR 1.16, p = 0.032). However, despite adjustment for rurality, only NHW demonstrated improvement in OS and breast cancer death in the post-CDK4/6i era.
NHB and Hispanic women were underrepresented in the landmark CDK4/6i trials; consequently, there is scarce literature on the response to CDK4/6i among different racial groups. In a meta-analysis of 4 trials that evaluated CDK4/6i as first-line therapy, there was an interaction between treatment effect on PFS and ethnicity (HR 0.56 for all, HR 0.9 for Asian, HR 0.62 for non-Asian, p for interaction = 0.002) [34]. However, this study was focused on Asian vs. non-Asian groups and did not provide information about NHB or Hispanic patients. PALOMA-3 and MONARCH-2 trials evaluated CDK4/6i after progression on ET and found no racial differences in PFS; however, their analyses were focused on White and Asian patients [35,36]. A real-world study by Agrawal and colleagues found that the median PFS for Indian women treated with palbociclib and letrozole was 20.2 months in the first line and 12 months in the second line, suggesting similar effectiveness to other real-world evidences [37]. Our study focused on NHW, NHB, and Hispanic patients. Asian composed only a small group of patients; therefore, they were combined with the other race group/ethnic in our study. We found no improvement in OS or breast cancer death risk from the pre-CDK4/6i to the post CDK4/6i era for the other race/ethnic group.

3
A single institution study of CDK4/6i by Knudsen et al. included Black patients in the non-European group accounting for ~ 8% of patients. The non-European group experienced shorter PFS than patients of European descent (p < 0.002). However, Black patients entered treatment with CDK4/6i disproportionately with more recurrent disease and treatment with fulvestrant, both of which are associated with shorter PFS [38]. Isaacs combined NHB and Hispanic patients enrolled in PALOMA-2 (n = 65) and PALOMA-3 (n = 48) trials, and conducted a survival analysis [39]. In the   [40]. It is noteworthy that this was a single-center study in an academic center in the Bronx, NY; and the study population reflects the racial distribution of The Bronx population, demonstrating similar access to CDK4/6i for both racial groups. These results suggest that similar access to CDK4/6i leads to similar outcomes for Black and non-Black patients treated with CDK4/6i. Furthermore, population studies in the US have demonstrated that Hispanics have lower breast cancer incidence and mortality compared to NHW women [41]. Nevertheless, our study suggests that the OS has not improved after introducing CDK4/6i for Hispanics. Hispanic is a very heterogeneous group of patients from different countries, which can also lead to different OS among Hispanic subgroups. Currently, there is scarce literature about disparities in breast cancer outcomes for Hispanic women treated with CDK4/6i.
Our study has certain limitations. First, this is an observational study that is still subject to unmeasured confounding despite a large sample size. Second, a 24-month follow-up is a relatively short period to assess survival differences; HR hazard ratio, ER estrogen receptor, PR progesterone receptor, NA not applicable, Met metastasis, + positive, -negative, CI confidence interval  however, we used the last update of the SEER database, which contained a decent number of events to allow our planned survival analysis. Future studies with longer follow-up are required to confirm our results. Third, the SEER database does not provide information about treatment with CDK4/6i; therefore, we cannot assess the specific effect of CDK4/6i on survival trends for the entire population or race-stratified groups. Fourth, the SEER database does not provide information about endocrine therapies, which can directly affect our survival in our study population. Fifth, SEER dichotomizes information about chemotherapy (yes or no) but does not provide details about specific chemotherapy regimens or the number of chemotherapy lines patients received, which could influence our survival analysis. Sixth, we were unable to include more socioeconomic variables that could lead to differences in treatments, or access to them, for both cohorts, such as insurance status, household income, social deprivation index, and many other factors. In conclusion, our study confirms that the OS in patients with HR + /HER2− MBC has improved after CDK4/6i was introduced in 2015. However, this effect is primarily driven by the improved OS in NHW patients, without significant improvement in OS in NHB, Hispanics, or other race patients. Studies with larger sample sizes and longer follow-up are required to confirm our results. Further studies need to assess whether a biological etiology could lead to differences in response to novel therapeutic agents (such as CDK4/6i) in some racial groups or if socioeconomic factors may affect access to these novel therapeutic agents in some racial groups.