Racial Disparities in Triple Negative Breast Cancer

Toward a Causal Architecture Approach

Scott D. Siegel; Madeline M. Brooks; Shannon M. Lynch; Jennifer Sims-Mourtada; Zachary T. Schug; Frank C. Curriero

Disclosures

Breast Cancer Res. 2022;24(37) 

In This Article

Results

TNBC cases accounted for 14% of the invasive breast cancer cases in the study population (Table 1). Compared to those without TNBC, TNBC cases were significantly younger at diagnosis (mean age 60.2 vs. 63.0), twice as likely to be Black (39.5% vs. 20.9%), more likely to have Medicaid or no insurance (8.2% vs. 5.4%), less likely to have Medicare (35.8% vs. 42.2%), and twice as likely to present with a late-stage cancer (14.8% vs. 7.2%). Comparing ICE measures by census tract of residence, TNBC cases were significantly overrepresented among Q1 (the most disadvantaged quintile) tracts for ICE-Race (23.0% vs. 12.8%), ICE-Income (17.7% vs. 12.5%), and ICE-Race/Income (18.1% vs. 11.5%). TNBC cases were similarly underrepresented among Q5 (the most advantaged quintile) tracts across all ICE measures.

See Table 2 for descriptive statistics on TNBC cases stratified by race. Compared to White TNBC cases, Black TNBC cases were significantly younger (mean age 56.9 vs. 62.4) and more likely to have private insurance (63.7% vs. 50.0%) but less likely to have Medicare (24.0% vs. 43.4%). Differences in insurance status can likely be attributed to mean age differences between Black and White TNBC cases. No significant differences were observed for stage of diagnosis. Comparing ICE measures by census tract of residence, Black TNBC cases were significantly overrepresented among Q1 (the most disadvantaged quintile) tracts for ICE-Race (43.6% vs. 9.5%), ICE-Income (31.3% vs. 8.8%), and ICE-Race/Income (36.3% vs. 6.2%). Black TNBC cases were similarly underrepresented among Q5 (the most advantaged quintile) tracts across all ICE measures.

Multivariate and univariate regression analyses for both fixed and mixed effects models (with a random tract-level intercept) were tested. For both ICE-Income and ICE-Race/Income, the fixed effects only and mixed effects models produced similar coefficient results with no change in inference. However, the ICE-Race mixed effects model resulted in a singular fit and coefficients could not be estimated. This was likely due to insufficient between tract variation to support estimation of a tract-level random effect. With no covariates in the model, the variance of the random effect (measuring between-tract variance) was significant but small (var = 0.092, p = 0.014), which was reduced and became non-significant in most models once covariates were included (see Additional file 1: Table S1). Correspondingly, the intraclass correlation coefficient (ICC) for the census tract random effect was 0.027, indicating that only 2.7% of the total variance was attributed to between-tract variability. Given the similarity of results across fixed effects only and mixed effects models for ICE-Income and ICE-Race/Income and the singular fit for the ICE-Race random effects model, and the similarity of results between the fixed effects univariate and multivariate models for ICE-Race, results from the multilevel fixed effects only models are presented here for ease of interpretation (see Additional file 1: Table S2 for the full set of available random effects model results).

Table 3 shows the results of multilevel fixed-effects only models that separately test each ICE measure. In univariate models, increasing age of diagnosis was associated with lower odds of TNBC (OR: 0.93, 95% CI 0.89, 0.96), while Black race was associated with more than double the odds of TNBC relative to White race (OR: 2.48, 95% CI 2.01, 3.05). Models run with the insurance variable excluded patients age 65 and older and those with Medicare or unknown insurance to better model insurance as a proxy for health care access and SES (i.e., patients are eligible for Medicare beginning at age 65 regardless of SES). Insurance type was not significantly associated with TNBC, and therefore, the multivariate models were run for the full study population (i.e., including patients age 65 and older) and without the insurance status covariate. Across multivariate models for each ICE measure, increasing age of diagnosis and Black race were significantly associated with lower and greater odds of TNBC, respectively (p-values < 0.05). Quintiles Q1–Q4 of ICE-Race (corresponding to greater disadvantage relative to Q5) were associated with significantly higher odds of TNBC, even after adjustment for patient-level race and age of diagnosis, ranging from Q1 (AOR: 2.09) to Q4 (AOR: 1.76). Neither ICE-Income nor ICE-Race/Income quintiles were significantly associated with TNBC after covariate adjustment, though their adjusted odds ratios suggested positive associations with TNBC. No significant interactions were observed in models that included cross-level interaction terms between patient-level race and tract-level ICE (Additional file 1: Table S3).

The multilevel fixed-effects only model of age and ICE-Race was stratified by patient-level race to further characterize the relationship between patient race, area-level segregation, and TNBC (Table 4). Increasing age of diagnosis was associated with decreased odds of TNBC for Black patients (AOR: 0.89, 95% CI 0.83, 0.95) but not White patients (AOR: 0.97, 95% CI 0.92, 1.01). Among Black patients, ICE-Race quintiles were no longer associated with TNBC (p-values > 0.05). The magnitude of the adjusted odds ratios suggested a positive association, which the relatively small Black patient sample (N = 776) may be underpowered to detect. ICE-Income and -Race/Income quintiles were also not associated with TNBC for Black patients (p-values > 0.05). Among White patients, ICE-Race quintiles Q1, Q2, and Q4 were associated with significantly greater odds of TNBC (p-values < 0.05). This result would suggest that White women living in predominantly Black census tracts were more likely to be diagnosed with TNBC, relative to other forms of invasive breast cancer, compared to White women living in predominantly White census tracts. For ICE-Income, only quintile Q2 was significantly associated with TNBC (AOR: 1.52, 95% CI 1.05, 2.20), with a non-significant trend in the expected direction observed for Q1 (AOR: 1.25, 95% CI 0.76, 2.01). This would suggest that White women living in low-income census tracts are at an elevated risk for TNBC, relative to other forms of invasive breast cancer among White women living in higher-income census tracts. No significant associations with TNBC were observed for ICE-Race/Income among White women.

Figure 1 shows the spatial covariation of TNBC prevalence and ICE measures across county census tracts. Tracts are symbolized as follows: (1) light gray—lower% TNBC and lower ICE-measured disadvantage; (2) magenta—higher% TNBC and lower ICE-measured disadvantage; (3) teal—lower% TNBC and higher ICE-measured disadvantage, and (4) blue—higher% TNBC and higher ICE-measured disadvantage. TNBC prevalence appears to correlate most strongly with ICE-Race, represented by a greater number of dark blue census tracts in Figure 1A relative to ICE-Income and ICE-Race/Income shown in Figure 1B, C, respectively. Across all maps, higher TNBC prevalence and higher ICE-measured disadvantage overlap in the greater Wilmington area, extending southwest and in the northeastern-most corner of the county. Figure 1B (ICE-Income) and Figure 1C (ICE-Race/Income) each depict two census tracts with higher TNBC prevalence but lower ICE-measured disadvantage in the southern part of the county. Each of the maps depict two census tracts with lower TNBC prevalence but higher ICE-measured disadvantage in the north-central part of the county.

The tract classifications based on quintiles of TNBC prevalence and ICE-Race appear to differentiate the affected populations and area-level systems of exposure (Table 5). As expected, the ten census tracts characterized as high in both TNBC and ICE-Race disadvantage ("high/high") had a 28.8% prevalence of TNBC among breast cancer patients, and 63.9% of their general population from these census tracts were Black. Compared to the 12 tracts that were characterized as both low in TNBC and ICE-Race disadvantage ("low/low"), residents living in the high/high tracts had greater rates of poverty (23.3% vs. 5.0%) and higher rates of completing less than a high school education (13.0% vs. 5.6%). Comparing high/high and low/low tracts, the former had more than double the count (19 vs. 8) and density (0.48 vs. 0.20) of alcohol retailers. Similar but weaker patterns were observed for fast-food retailer counts (14 vs. 12) and density (0.35 vs. 0.30). High/high tracts also had greater prevalence of AUD (25.4% vs. 14.9%) and obesity (43.4% vs. 33.9%). Additional file 1: Tables S4 and S5 present these descriptive statistics according to classifications based on ICE-Income and ICE-Race/Income, respectively, which were similar to those based on ICE-Race. Additional file 2: Figure S1 shows the distribution of alcohol and fast-food retailers and AUD and obesity prevalence by tract ICE quintiles.

Figure 1.

Spatial covariation of triple negative breast cancer (TNBC) prevalence and ICE measures, New Castle County, DE. AC depict quintiles of TNBC prevalence and ICE-measured disadvantage (by race, income, and race/income) at the census tract level in New Castle County, DE. The extremes of the classification system in light gray, magenta, teal, and blue represent the spatial covariation of both measures, ranging from lower in both to higher in both. Across all maps, higher TNBC prevalence and higher ICE-measured disadvantage overlap in the City of Wilmington, as shown by the blue tracts. ICE-Race (A) appears to correlate with TNBC more strongly than ICE-Income (B) or ICE-Race/Income (C), as map A has more tracts classified as low–low (light gray) or high–high (blue) in both measures. ICE-Race (A) correlates with higher TNBC prevalence in additional census tracts south of the City of Wilmington, which correspond to tracts that have relatively large Black populations but relatively less income deprivation measured by ICE (B, C)

processing....