A Population Health Assessment in a Community Cancer Center Catchment Area: Triple-Negative Breast Cancer, Alcohol Use, and Obesity in New Castle County, Delaware

Beginning in 2012, the NCI mandated that NCI-designated cancer centers define and describe their catchment area with regard to cancer burden, risk factors, and health inequities (1). The results of these assessments are to be used to prioritize research activities and guide the development of more effective clinical and policy interventions designed to equitably reduce the population-level burden of cancer (2). Although the NCI's catchment area mandate only directly applies to designated cancer centers, community cancer centers provide an estimated 85% of cancer care in the United States (3, 4) and therefore offer the potential to significantly expand the impact of catchment-based approaches (5). Where community cancer center resources are limited regarding the collection of new primary data, cancer registry data can be leveraged by linking it to existing secondary risk factor data sources. Then spatial analytic methods can be employed to guide the delivery of more geographically targeted interventions (6, 7).

To demonstrate the utility of this approach for a community cancer center, this assessment focused on characterizing the spatial distribution of triple-negative breast cancer (TNBC) within the Helen F. Graham Cancer Center & Research Institute (HFGCCRI) catchment area (New Castle County, DE). TNBC is an aggressive subtype of invasive breast cancer negative for the expression of estrogen (ER), progesterone (PR), and HER2 receptors that disproportionately affects younger women and is associated with poorer survival relative to other subtypes (8, 9). Black women are twice as likely to be diagnosed with TNBC than White women (8, 10), which contributes to national racial disparities in breast cancer mortality (11). We chose to focus on TNBC because, according to at least one report, Delaware is the state with the highest TNBC incidence rate in the United States (12), with more cases concentrated in New Castle County relative to the other two counties in the state (13). Consistent with these observations, Delaware breast cancer mortality rates are 55% higher for younger Black versus White women (14).

Recent studies provide evidence that the spatial or neighborhood context in which women reside is related to TNBC incidence, stage at diagnosis, and survival, lending support to the benefits of a catchment-based approach. For example, Qin and colleagues found that for Black patients with breast cancer, residing in a low-socioeconomic status (SES) neighborhood was associated with a higher risk of TNBC relative to other breast cancer subtypes, even after adjusting for patient-level SES (15). Hossain and colleagues reported that compared with White women, Black women with TNBC were more likely to live in disadvantaged neighborhoods and neighborhood disadvantage was, in turn, associated with a later stage of diagnosis and poorer stage-specific survival (16).

The relationship between neighborhood disadvantage and breast cancer characteristics may be at least partially mediated through disparate exposures, including the leading risk factors of obesity and alcohol use (9, 17). Luo and colleagues found that 48% of the racial disparity in TNBC incidence was mediated by metabolic dysfunction (e.g., larger waist circumference; ref. 18). Other research conducted by Shariff-Marco and colleagues suggests that a greater number of fast-food restaurants in low-SES neighborhoods helped to explain the higher odds of obesity among Black relative to White breast cancer survivors (19). Alcohol use, which contributes to an estimated 12% of breast cancer cases nationally and nearly one in five breast cancer cases in Delaware (20), may disproportionately affect TNBC risk for Black women. While a prior pooled analysis has not identified a relationship between alcohol use and TNBC in cohorts composed disproportionately of White women (21), Williams and colleagues found that consuming 7+ alcohol drinks per week was associated with an increased risk of invasive breast cancer among Black women across all tumor subtypes, including TNBC (22), but not among White women (23). Although historically, Black women have used alcohol at lower rates than White women, more rapid increases in disordered alcohol use have been observed for Black women in recent decades such that disordered drinking rates are now comparable between Black and White women overall (24). Among women of age 50 to 64, when TNBC rates are elevated relative to other breast cancer subtypes, disordered drinking rates are even higher among low-income Black women than their White counterparts (25). Similar to the potential relationship between fast-food restaurants and obesity, a greater exposure to alcohol retailers in predominantly Black neighborhoods may contribute to disordered alcohol use (26). However, the relationships between neighborhood characteristics and alcohol use patterns have yet to be examined in the context of TNBC.

The objectives of this study were to (i) characterize the spatial distribution of TNBC cases and (ii) assess the spatial relationships between TNBC and measures of obesity and disordered alcohol use within the HFGCCRI catchment area. The results of this assessment can guide future research and inform geographically targeted TNBC prevention interventions designed to reduce the adverse consequences of obesity and unhealthy alcohol use.

For the purposes of this study, the HFGCCRI catchment area was defined as New Castle County, in which approximately 76% of all patients with breast cancer seen at HFGCCRI reside. Patient records came from the HFGCCRI cancer registry. The HFGCCRI is a part of the Christiana Care Health System, headquartered in New Castle County, DE, and provides care to an average of more than 600 analytic breast cancer cases each year, representing 85% of all cases in the surrounding county. This study was reviewed and approved by the Christiana Care Health System Institutional Review Board and was conducted in accordance of the U.S. Common Rule.

New Castle County has a population of 555,133 residents, of whom approximately 65% are White, 25% Black/African American, and the remainder are classified as “other” (27). Wilmington is the largest city in New Castle County and has a population of 70,904 residents, representing approximately 15% of the county's population. The racial makeup of Wilmington differs from the surrounding county: approximately 58% of city residents are Black/African American, 35% White, and the remainder classified as “other.”

Patient records with address data were not available from the Delaware Cancer Registry given privacy protections. A comparison of HFGCCRI records with New Castle County summary data made available by the Delaware Cancer Registry found that the HFGCCRI cases, in aggregate, were representative in terms of patient age, race, receptor status (ER, PR, and HER2), and stage (Supplementary Table S1).

This study drew on HFGCCRI cancer registry data for 3,449 adult female New Castle County residents who were diagnosed with invasive breast cancer between the years 2012 and 2020. This time frame was selected to maximize the number of breast cancer cases where subtype markers (i.e., ER, PR, and HER2 receptors) were routinely documented in the cancer registry. Patient residential address, demographic, and clinical data were abstracted from the registry. Patient addresses were manually cleaned and geocoded using ArcGIS 10.6, yielding a match rate of 95% (3,449/3,622). Unmatched patients did not significantly differ from matched patients by age, race, ethnicity, stage, subtype, or insurance payer.

Sociodemographic measures included age at diagnosis, race/ethnicity, and insurance payer status, which were all directly abstracted from the HFGCCRI cancer registry. Payer status (commercial, Medicaid, Medicare, or other/self-pay) was used as a proxy for socioeconomic status in lieu of other available data (28). Clinical measures included breast cancer stage and receptor status. Cases were classified into ‘TNBC' when ER, PR, and HER2 were all known negative; all other invasive cases were classified as “Not-TNBC,” including when the HER2 status was unknown (HER2?). Patient-level measures of obesity and alcohol use were not available from the HFGCCRI cancer registry.

Spatial measures of obesity and disordered alcohol use within the HFGCCRI catchment area were generated from Christiana Care Health System electronic health record (EHR) data for 20,310 unique adult New Castle County residents who were admitted to an inpatient unit between July 1, 2018 and June 30, 2019. These measures were used to estimate the potential impact of environmental or neighborhood factors on rates of obesity and unhealthy alcohol use, similar to the conceptualization of “obesogenic” environments (29), and as such included both women and men regardless of admitting diagnosis. The Christiana Care Health System operates two acute care hospitals in the county that together account for 1,227 inpatient beds. Similar to the large share of cancer care that the HFGCCRI provides in New Castle County, these two hospitals provide 88% of nonveteran adult acute care in New Castle County (45,278 hospital discharges/51,262 total discharges; ref. 30). Prior research has found that these EHR data can be used to generate valid point-level spatial estimates of health behaviors (e.g., smoking) in New Castle County (31). In addition, self-report data on alcohol use from national surveys (e.g., Behavioral Risk Factor Surveillance System) may have limited validity without multiple adjustments (32); the availability of clinical data can overcome these limitations. International Classification of Diseases (ICD) diagnosis codes abstracted from the EHR for obesity and alcohol use disorder (AUD) were used to categorize patients into “obese” or “not obese” and “AUD” or “no AUD” categories. Patient addresses were manually cleaned and geocoded using ArcGIS 10.6, yielding a match rate of 98% (20,310/20,706).

Spatial point-level measures of exposure to fast-food restaurants and alcohol retailers in New Castle County were generated from commercial data and publicly available records. Fast-food retailer data were obtained from SICCODE.com and included establishments with a North American Industry Classification System (NAICS) code of 722513 (33), consistent with established methods for characterizing fast-food and limited service food retailers (34). Alcohol retailer data were extracted from a public state business license database that was current as of April 17, 2019 (35). Guided by prior research that has documented a link between risky alcohol use and residential exposure to off-premise alcohol retailers (e.g., liquor stores), but not on-premise alcohol retailers (e.g., bars; ref. 36), we included only licensed establishments that sold alcoholic beverages for offsite consumption. All retail locations were geocoded using ArcGIS 10.6 with a match rate of 100% (fast-food restaurants N = 221, alcohol retailer N = 160).

Descriptive and bivariate statistics, and posthoc tests with Bonferroni-adjusted P values, were used to characterize the breast cancer patient groups (TNBC and Not-TNBC) for the sociodemographic and clinical variables.

Methods from the field of spatial statistics were used to characterize the spatial distribution of TNBC and its risk factors within New Castle County. The spatial concentrations of TNBC and Not-TNBC cases were visualized using the method of spatial intensity, defined as the expected number of events per area unit. Spatial intensity was estimated using the kernel density approach, which calculates a distance-weighted count of events per unit area within a fixed radius (bandwidth) of any generic location (37). A fine mesh of generic locations is placed over the study region, intensity estimated at each location, and the results mapped to display continuous spatial variation in concentrations of events. Spatial intensity was estimated for each of the mapped point locations of TNBC cases and Not-TNBC cases separately and the ratio of their intensities (TNBC vs. Not-TNBC) was used to nonparametrically estimate and visualize spatial variation in risk of TNBC, specifically the odds of TNBC relative to Not-TNBC (38, 39). Tolerance regions, denoting areas with significantly higher or lower risk of TNBC versus Not-TNBC (labeled here as hot and cold spots, respectively), were identified using the random labeling approach common with spatial point patterns of two types (40, 41). Similarly, spatial intensity was estimated for the AUD and obesity risk factors from the hospitalized adult population drawn from the Christiana Care EHR and the ratio of their respective intensities (AUD to no AUD and obese to not obese) used to estimate spatial variation in risk of AUD and obesity.

TNBC hot and cold spots, defined by the tolerance regions, were then compared by sociodemographic and clinical characteristics. In addition, the EHR data were used to estimate the percentage of the admitted adults with AUD or obesity within each hot and cold spot. Finally, measures of alcohol and fast-food exposure were quantified as the number of retailers of each type per 1,000 hospitalized adults for each hot and cold spot.

The intensity ratios of AUD and obesity across the county were simultaneously visualized using the method of bivariate choropleth mapping (42, 43), which is usually applied to aggregate area-level data (e.g., census tracts) but here was applied to the point-level scale at which spatial odds of AUD and obesity were estimated. The spatial odds for each outcome were sorted into the following risk tertials based on their distribution of values: ≤25^th percentile (low), 25^th to 75^th percentile (moderate), and >75^th percentile (high). These recoded values were used to create a 3 by 3 classification system to classify geographic areas that were low, moderate, or high in each AUD and obesity risk factor. The resulting classifications were symbolized using color and saturation to depict nine categories ranging from low in both values to high in both. The final map was overlaid with the TNBC tolerance regions (i.e., hot and cold spots) to display spatial overlap with areas with significantly higher or lower risk of TNBC.

Patient-level datasets generated for this study are not publicly available because they represent protected health information but may be made available in a deidentified format from the corresponding author on request. Spatial catchment area databases generated for this study based on publicly available sources (i.e., the alcohol and fast-food retailer point locations) are available from the corresponding author on request.

Table 1 reports the sociodemographic characteristics for the 3,449 identified invasive breast cancer cases (462 TNBC and 2,987 Not-TNBC). Consistent with prior research, compared with the Not-TNBC group, patients with TNBC were significantly younger (median age 59 vs. 63), more likely to be Black (38.7% vs. 20.0%), and more likely to be diagnosed with late-stage breast cancer (i.e., stage 3b and above; 14.7% vs. 7.1%). The younger average age for the TNBC group was largely due to a higher rate of cases under age 50 relative to the Not-TNBC group (23.8% vs. 17.0%). No group differences were observed by insurance payer.

Figure 1 visualizes the variation in the spatial odds of TNBC relative to Not-TNBC cases. Two hot and two cold spots were identified, reflecting areas with significantly higher (Wilmington and Bear) and lower (North Wilmington and Northwest New Castle County) spatial odds of TNBC relative to Not-TNBC cases, respectively. Of the total 462 TNBC cases, 32% (147) resided in hot spots that accounted for 11% of the catchment land area; 14% (66) resided in cold spots that accounted for 14% of the catchment land area. The remaining 54% of cases (249) were distributed throughout the remaining 75% of the catchment area.

Table 2 reports sociodemographic and breast cancer characteristics for cases in each of the hot and cold spots. There were significant differences by age, with younger median age (62) in the two hot spots relative to the two cold spots (median age 63 and 65 in Northwest New Castle County and North Wilmington, respectively). The hot spots, particularly Wilmington, also had significantly greater proportions of Black breast cancer cases (Wilmington: 71.1%, Bear: 40.1% vs. Northwest New Castle County: 5.0%, North Wilmington: 7.1%). No differences were observed by ethnicity, stage of diagnosis, or insurance payer.

Table 2 also reports on the catchment area risk factor characteristics for the TNBC hot and cold spots. Compared with hospitalized adults who resided in cold spots, a higher percentage of hospitalized adults who resided in the hot spots were diagnosed with an AUD and obesity. Wilmington's percentage of hospitalized adults with AUD (28.9%) was significantly greater than that of Bear (15.0%), Northwest New Castle County (15.8%), and North Wilmington (13.9%). Wilmington also had significantly greater density of offsite alcohol retailers per 1,000 hospitalized adults compared with the other areas (13.4 vs. 5.6, 6.1, and 7.9). Wilmington and Bear had comparable percentages of hospitalized adults with obesity (43.7% and 45.9%, respectively), which were significantly higher than those observed in Northwest New Castle County and North Wilmington (37.7% and 33.9%, respectively). No significant differences were found for fast-food restaurant densities between the hot and cold spots.

Figure 2 visualizes the spatial variation in risk for AUD and obesity within the catchment area. Figure 2A displays how the odds of AUD versus no AUD varied spatially such that higher odds of AUD was observed in the Wilmington hot spot relative to the Bear hot spot and the two cold spots. The largest concentration of alcohol retailers was also located within the Wilmington hot spot. Other areas within the catchment area but outside of the TNBC hot and cold spots had elevated odds of AUD versus no AUD, but only the area immediately south of the Bear hot spot was statistically significant. Figure 2B shows how the spatial odds of obesity versus non-obesity varied spatially such that higher odds of obesity were observed in the Bear hot spot relative to the Wilmington hot spot and the two cold spots. The spatial pattern of fast-food restaurants did not suggest a clear concentration in either hot spot.

Figure 3 is a bivariate choropleth map that visualizes the areas of the catchment area that were low, moderate, and high for the joint spatial odds of AUD and obesity in reference to the TNBC hot and cold spots. The entire Wilmington TNBC hot spot land area (locations at which the joint spatial odds were estimated) was classified as high for AUD and high or moderate for obesity. The largest share of land area in the Bear TNBC hot spot was classified as high for obesity and moderate or high for AUD, next followed by land area classified as moderate for both AUD and obesity, with smaller percentages of land area classified as high for both AUD and obesity and high for AUD and moderate for obesity. In contrast, the largest shares of land area in the Northwest New Castle County and North Wilmington TNBC cold spots were classified as moderate for AUD and low for obesity. In the Northwest New Castle County TNBC cold spot, a relatively small portion of the land area was classified as high AUD. No land area in either cold spot was classified as high for both AUD and obesity. See Supplementary Table S2 for additional details.

The results of this population health assessment in a community cancer center catchment area revealed significant spatial variation in the distribution of TNBC cases. Defining the catchment area as New Castle County, Delaware, the county with the greatest concentration of TNBC cases (13) in a state that leads the nation in TNBC incidence (12), we observed areas that were ‘hot' and ‘cold' for TNBC cases relative to other invasive breast cancer subtypes. The hot spots accounted for approximately a tenth of the catchment land area but nearly a third of all TNBC cases. These findings are consistent with prior research (15, 16) and suggest that spatial variation in TNBC incidence may be related to neighborhood characteristics, including disparate exposure to risk factors.

In support of the disparate exposure hypothesis, we found evidence of spatial variation in measures of obesity and unhealthy alcohol use that covaried with TNBC hot and cold spots. In the Wilmington hot spot, for example, a significantly higher percentage of adults who had been hospitalized had met criteria for an AUD compared with other geographic areas. Wilmington also had a significantly greater density of alcohol retailers relative to other areas, a risk factor for unhealthy alcohol use (36). Likewise, both Wilmington and the Bear TNBC hot spot had a significantly higher percentage of hospitalized adults who had been diagnosed with obesity relative to hospitalized patients who resided in the TNBC cold spots. However, no differences were observed for fast-food restaurant densities. Although the density of fast-food restaurants has been previously linked to obesity and obesity-related health outcomes (44, 45), other research suggests that the relationship between healthy food access and obesity is moderated by socioeconomic and other residential factors (e.g., segregation; ref. 46). The current study was not designed to detect such interactions, which warrant further investigation.

When the spatial risk factors were considered jointly, we observed that the entire Wilmington TNBC hot spot was classified as high risk for AUD, with most of that area also classified as moderate risk for obesity, and a smaller portion was classified as high for both. Most of the Bear TNBC hot spot land area was classified as high risk for obesity, with smaller but notable percentages classified as high or moderate risk for AUD. The largest share of land area within the two TNBC cold spots, by contrast, were classified as moderate risk for AUD and low risk for obesity, with no land area classified as high for both risk factors. These findings would suggest that unhealthy alcohol use may be of greater concern for the Wilmington hot spot and obesity for the Bear hot spot, with smaller regions within each hot spot where high levels of risk for obesity and AUD overlapped. Overlapping risk was much less of a concern for the cold spots overall, with some evidence of moderate to high risk observed for AUD and less risk observed overall for obesity. Taken together, these results provide preliminary support for a spatial link between TNBC hot spots and potentially modifiable risk factors. If future research should confirm this link, community outreach and other prevention-based interventions can be targeted to neighborhoods that fall within the TNBC hot spots. For example, place-based obesity interventions tailored to Black and other minority communities (47) or alcohol retailer density reduction policies (48) offer the potential to reduce exposure disparities.

Several study limitations should be acknowledged. First, the results from this study come from a single catchment area and may not generalize to other geographic regions. Catchment area assessments, by definition, should be specific to the unique cancer burden faced by a population within a defined geographic area and should not be used to directly inform cancer control and prevention efforts in other catchment areas. Rather than produce generalizable results, our overarching goal was to describe a method for conducting catchment area assessments that draw on cancer registry and other secondary data sources. Such an approach may be more feasible for community cancer centers where limited resources often preclude the collection of new primary data. Second, given privacy protections, the data for this study came from the cancer center registry, reflecting only patients treated at that cancer center, and not the comprehensive state registry. A comparison of cancer center and state registry data showed that, in aggregate, the sample used in this study was representative of the total catchment area breast cancer population in terms of age, race, ethnicity, stage, subtype, and insurance payer. However, it remains unknown whether the sample was geographically representative of all cases in the catchment area, particularly within the identified hot and cold spots. Third, patient-level data for obesity and unhealthy alcohol use from breast cancer cases were not available. Instead, we postulated that the spatial data for hospitalized adults and retailers provided insight into the potential environmental effects on behavioral risk factors, similar to conceptualizations of ‘obesogenic' environments (29). Nevertheless, the use of weight and height to diagnose a patient with obesity in a hospital setting does not necessarily reflect the metabolic processes implicated in carcinogenesis (e.g., insulin resistance; ref. 49). In addition, the cross-sectional design of this study precludes drawing causal inferences or controlling for potential confounders (e.g., access to care). For these reasons, we join the call that Qin and colleagues (15) made for future studies to prospectively integrate neighborhood characteristics and patient-level data—including health behaviors (e.g., alcohol use), metabolic and other known risk factors, and biomarkers of exposure—to better understand how environmental contexts influence breast cancer risk.

In conclusion, the results of this community cancer center catchment area assessment highlight the benefits of employing spatial methods to analyze cancer registry and other secondary data sources. In a county with concerning rates of TNBC, an aggressive subtype of breast cancer that disproportionately affects young and Black women, we observed hot spots where TNBC cases were more concentrated. In addition, we observed higher levels of spatial risk for obesity and unhealthy alcohol use within these hot spots, established breast cancer risk factors that may have particular relevance for TNBC. These findings can be used to guide community outreach and engagement activities to provide more targeted and equitable cancer control and prevention interventions.

No disclosures were reported.

S.D. Siegel: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. M.M. Brooks: Conceptualization, data curation, formal analysis, visualization, writing–review and editing. J. Sims-Mourtada: Conceptualization, data curation, writing–review and editing. Z.T. Schug: Conceptualization. D.J. Leonard: Conceptualization, writing–review and editing. N. Petrelli: Resources, writing–review and editing. F.C. Curriero: Conceptualization, supervision, methodology, writing–original draft, writing–review and editing.

This project was supported by NIGMS (P20 GM103446) from the NIH and the State of Delaware (to S.D. Siegel). We thank Bob Hall-McBride and Pat Swanson for their technical and administrative assistance in supporting this project.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.