Characteristics of the overall cohort

Table 1 summarizes the characteristics of the 27,459 patients in the cohort. Median age at metastatic diagnosis was 64 (IQR 54–73). Grouped by stage, 57.3% of patients were diagnosed with early-stage breast cancer that later metastasized, 31.0% presented with de novo breast cancer, and 11.8% had an unknown stage. Among breast cancer subtypes, hormone receptor-positive (HR+)/human epidermal growth factor receptor 2-negative (HER2−) was the most prevalent, representing 66.1% of cases. This was followed by HR+/HER2+ at 12.3%, HR−/HER2− at 11.6%, and HR−/HER2+ at 4.0%. A small portion, 6.1% of patients, had an unknown subtype. By practice type, 79.9% of patients were associated with community oncology practices, while 20.1% were associated with academic practices. Supplemental Table S1 presents a comparison of the characteristics by race and ethnicity. Compared to White patients, Latinx and Black patients were younger (median age: Latinx = 58, Black = 60, White = 65) more likely to reside in the least privileged areas (ICE Q1: Latinx = 36.2%, Black = 58.4%, White = 7.6%) and differed in insurance coverage (% with Medicaid coverage: Latinx = 5.8%, Black = 4.4%, White = 1.5%). Information on the distribution of first-line treatment among the cohort can be found in Supplemental Table S2. The most common first-line treatment was Aromatase inhibitors (32.1%), followed by chemotherapy (19.6%), Aromatase inhibitors + cyclin-dependent kinase 4/6 inhibitors (17.7%) with the remainder of patients receiving other therapies (30.6%).

Table 1 Demographic and clinical characteristics of the mBC cohort at diagnosis: overall and by neighborhood privilegeCharacteristics of the cohort by neighborhood privilege

Table 1 summarizes the characteristics of the cohort by neighborhood privilege. The smallest share of the cohort came from the least privileged neighborhoods (17.2%), and the largest share came from the most privileged neighborhoods (24.0%). Compared to patients from the most privileged areas, those from the least privileged areas exhibited several differences. They were typically younger (median age: 62 vs 64), more likely to be Black (36.9% vs 2.6%) or Latinx individuals (13.2% vs 2.6%) were less often diagnosed with early-stage disease (54.2% vs 58.9%) and more frequently had the aggressive HR−/HER2− subtype (15.6% vs 9.9%).

Median time and adjusted HR of treatment initiation

Figure 2 depicts Kaplan–Meier curves of the risk of initiating first-line treatment within 90 days of metastatic diagnosis by neighborhood privilege. For ease of interpretation, we limited the figure to two curves—comparing the least and most privileged areas—and plotted the inverse risk of initiating first-line treatment (to denote that more patients received treatment as time progressed). At nearly all points in time, patients in the least privileged areas (represented by the orange line) had a lower risk of initiating treatment than patients in the most privileged areas (represented by the blue line).

Fig. 2

Kaplan–Meier estimates of treatment initiation comparing patients from the least and most privileged neighborhoods. Treatment initiation denotes first-line treatment initiation within 90 days of metastatic diagnosis. Plot generated in R using the survminer package (version 0.4.9). For ease of interpretation, this figure was limited to two curves—comparing the least and most privileged areas—and plotted the inverse risk of initiating first-line treatment (to denote that more patients received treatment as time progressed). Treatment initiation estimates excluded 3463 patients with a recorded therapy starting prior to metastatic disease that continued beyond 14 days after the index date of metastatic disease

Table 2 presents median TTI by neighborhood privilege. Patients in the least privileged areas had a longer median TTI than patients in the most privileged areas (38 days, 95% CI 36, 40 vs 31 days, 95% CI 29, 32). In addition, those in the least privileged areas had an adjusted HR indicative of less timely treatment initiation (HR 0.905, 95% CI 0.863, 0.950) relative to those in the most privileged areas (reference group).

Table 2 Kaplan–Meier estimates and hazard ratios of treatment initiation and overall survival by neighborhood privilegeMedian time and adjusted HR of OS

Figure 3 depicts Kaplan–Meier curves of the risk of death within 5 years of metastatic diagnosis by neighborhood privilege. At nearly all points in time, patients in the least privileged areas (represented by the orange line) had a greater risk of death than patients in the most privileged areas (represented by the blue line).

Fig. 3

Kaplan–Meier survival estimates comparing patients from the least and most privileged neighborhoods. Kaplan–Meier survival estimates within 5 years of metastatic diagnosis. Survival estimates excluded 181 patients whose recorded death occurred before the index date of metastatic diagnosis. Such exclusion likely results from recording the date of death as year-month for privacy reasons and the use of middle of the month in time-to-event calculations. Plot generated in R using the survminer package (version 0.4.9)

Table 2 presents median OS from metastatic diagnosis by neighborhood privilege. Patients in the least privileged areas had shorter median OS time (29.7 months, 95% CI 28.5, 31.5) than patients in the most privileged areas (39.2 months, 95% CI 37.9, 40.6). In addition, those in the least privileged areas had an adjusted HR indicative of an increased risk of death (HR 1.170, 95% CI 1.107, 1.237) relative to those in the most privileged areas (reference group).

Interactive and stratified results by patient race and ethnicity

Table 3 presents the results from adjusted hazard models incorporating an interaction term between neighborhood privilege and race and ethnicity. Within these models, the reference group denoted White patients from the most privileged neighborhoods. Relative to this group, nearly all other groups had HR indicative of a lower risk of initiating first-line treatment and a greater risk of death. In our analysis of TTI, HR were consistently below one for Black and Asian patients and were similar among those in more and less privileged areas. For Latinx and White patients, HR of TTI were in all but one instance below one, but were often closer to one among those in more privileged areas. In our analysis of OS, HR for Black and White patients were closer to one among those residing in more privileged neighborhoods. Among Asian and Latinx patients, trends across survival HR were less consistent.

Table 3 Adjusted hazard ratios with interactions between ICE quintile, race, and ethnicity

Supplementary Table S3 presents Kaplan–Meier estimates of treatment initiation stratified by neighborhood privilege and race and ethnicity and offers additional evidence of racial and ethnic inequities in TTI and rwOS. For example, median TTI was 29 days (95% CI 28, 31) among White patients from the most privileged areas compared to 47 days (95% CI 42, 56) among Latinx patients from the least privileged areas and 38 days (95% CI 35, 40) among Black patients from the least privileged areas. Likewise, for Asian, Black, and White patients, median rwOS was lowest among those from the least privileged areas. Supplementary Table S4 provides evidence of similar inequities in results from adjusted HR of TTI and rwOS stratified by race and ethnicity.

Associations with additional social determinants of health measures

Supplementary Table S5 presents adjusted HR of treatment initiation and OS using alternate constructions of ICE. Compared to our main results, which defined ICE in relation to high-income White and low-income Black households in an area, our results using ICE constructed in relation to low-income Latinx households and low-income households of Color offered similar evidence of inequities. For example, across these three constructions, HR among those from the least privileged areas ranged from 0.872 (95% CI 0.830, 0.915) to 0.905 (95% CI 0.863, 0.950) for TTI and from 1.143 (95% CI 1.081, 1.209) to 1.170 (95% CI 1.107, 1.237) for rwOS.

Table 4 lists correlations between ICE and three additional SDOH measures related to structural racism and SES. These additional measures exhibited strong, positive correlations (> 0.65) with one another, except for Percent Black and the Yost Index (correlation of 0.380). This weaker correlation may reflect the absence of an economic component in the Percent Black measure and the absence of a racial component in the Yost Index.

Table 4 Correlations among additional social determinants of health measures

Table 5 presents the results from an exploratory analysis that utilized these additional measures of structural racism and SES. The top panel presents adjusted HR of the risk of first-line treatment initiation. Among those from neighborhoods with varying concentrations of Black residents, with the exception of Q3, we found no difference in the risk of treatment initiation as all confidence intervals included one. Using the Yost Index, our results were similar to those from our main analysis of ICE, as patients from areas with the lowest SES (Yost Index Quintile 1) had a lower risk of treatment initiation (HR 0.902, 95% CI 0.861, 0.946) than those from the highest SES (Yost Index Quintile 5). Analyses leveraging the Structural Racism Indicator also revealed evidence of inequities in treatment initiation with a lower risk of treatment initiation among those from areas with the highest levels of structural racism (HR 0.954, 95% CI 0.909, 1.002), though the confidence interval included one.

Table 5 Adjusted hazard ratios of treatment initiation and overall survival using alternate measures of structural racism and socioeconomic status

The bottom panel of Table 5 presents adjusted HR of the risk of death. Similar to our main analysis of ICE, our analysis of additional, area-level measures revealed evidence of inequities in OS when comparing the top and bottom quintiles of each measure. Across models, HR among those from areas in the bottom quintile were 1.071 (95% CI 1.012, 1.134) using the Percent Black measure, 1.173 (95% CI 1.112, 1.237) using the Yost Index, and 1.180 (95% CI 1.117, 1.247) using the Structural Racism Indicator.