The development of a composite measure to identify priority areas of need for cancer screening interventions

Abstract

Background:

Cancer centers and health systems are tasked with deciding where to deploy community interventions to reduce the burden of cancer within their catchment areas. Few methods exist to prioritize communities in a systematic manner, considering features of individuals, populations, systems and policies. We developed a geographically informed index to prioritize census tracts based on community need, with an initial focus on identifying communities in need of breast cancer screening interventions.

Methods:

This study used publicly available data to select variables known to be associated with disparities in breast cancer screening rates. Variables were identified from five categories: economic stability, education access and quality, neighborhood and built environment, social and community context, health status, and healthcare access and quality. Data were analyzed at the census tract level across the Sidney Kimmel Comprehensive Cancer Center catchment (N=1216). Principal component analysis (PCA) was applied to 23 variables and five PCs were selected to construct a composite measure using a weighted sum. Resulting index values were used to stratify the dataset for further analysis and mapped for visualization.

Results:

The analysis produced the Community Need Priority Index – Breast Cancer Screening (CNPI-BCS), with values ranging from 0 to 1 (mean=0.259, sd=0.161). The top quintile (Q5, n=243) represented the highest need communities. Q5 tracts were primarily concentrated in Philadelphia, Camden, and Delaware counties. Philadelphia County had the highest average (mean=0.364, sd=1.78), and the most tracts in the top quintile (45%, n =175). Montgomery county had the lowest average (mean=0.169 sd=0.092).

Conclusion:

This novel methodological approach considered the complex nature of multiple, intersectional barriers to good health to identify priority areas of need within cancer center catchment areas.

BACKGROUND

Cancer centers have had a longstanding commitment to reduce the burden of cancer in the populations that they serve. In 2012, the National Cancer Institute (NCI) solidified this commitment by requiring that dedicated cancer centers identify and describe their catchment area, as well as document ongoing interventions that specifically address the cancer burden, risk factors, incidence, mortality, morbidity, and inequities in the catchment area¹. Most cancer center catchments include a wide geographic area², encompassing communities and populations with racial and ethnic diversity, a range of adverse social determinants of health, and environments that make accessing health care challenging. Identifying priority areas within catchment areas is important for maximizing resource allocation and impact.

Historically, assessing catchment areas and identifying communities in need has taken a broad range of forms and methods. Many cancer centers use a combination of publicly available data related to cancer incidence, mortality, health behaviors and social factors and their own collection of qualitative or quantitative data to dig deeper into communities and fully understand the complexities of their catchment area³. Tools like Cancer InFocus help to geographically visualize catchment area data to aid in identifying community characteristics⁴, but do not determine priority areas of need. Similarly, Robert Wood Johnson’s County Health Rankings uses 80 population health, well-being, and community conditions to compare counties to state and national averages but does not provide insight at the census tract level⁵. The Centers for Disease Control and Prevention created the Social Vulnerability Index (SVI) for prioritizing counties and census tracts, but it was developed to identify areas most at risk during hazardous events⁶. To our knowledge, the research presented here is one of the first of its kind to create a Community Need Priority Index (CNPI) that can be tailored to a particular cancer type or disease state. The current study describes the creation of a breast cancer screening (BCS)-focused CNPI, to identify and prioritize census tracts that may be in greatest need of breast cancer screening interventions.

Breast cancer screening is typically recommended annually or bi-annually to average-risk females between the ages of 40 and 74^7–9. Rates of routine screening have historically been lowest among racial and ethnic minority communities, those without a regular source of care, and those who are under- or uninsured. Geographic barriers to screening exist, with those living in rural areas or where there is a low supply of health care practices less likely to be screened for breast cancer^10,11. Research shows that neighborhood or area-level factors related to social determinants of health (SDOH), such as social vulnerability, may contribute to disparities in breast cancer screening¹². Taken together, disparities in breast cancer screening are multi-factorial and encompass a range of individual-, community-, and system-level factors.

Given the need to deploy interventions in our catchment area that serve those in greatest need, and the lack of a standardized methodology to identify priority areas, we developed the Community Needs Priority Index (CNPI). The index is a composite measure that can be tailored to reflect factors most related to the screening, health behavior, or outcome that specific interventions aim to impact. For our initial use of the CNPI, we aimed to identify communities where factors known to contribute to decreased screening were prevalent and where an intervention to increase screening rates was most needed. Hence, the development of the CNPI-Breast Cancer Screening (CNPI-BCS) is described herein. The purpose of this study is to highlight the methodology for creating such an index, when very few exist.

METHODS

Study Setting

The Sidney Kimmel Comprehensive Cancer Center (SKCCC) is an NCI-designated cancer center based in the Mid-Atlantic region of Greater Philadelphia. Its catchment area consists of seven counties across two states: Philadelphia, Bucks, Montgomery, and Delaware counties in Pennsylvania and Burlington, Camden, and Gloucester counties in New Jersey. It is a densely populated, urban catchment area, although there are large areas of suburban communities and pockets of rurality. The more than five million residents who live in the SKCCC catchment area are part of some of the most diverse populations in the nation¹³. In some counties, almost two-thirds of residents identify with a minority race or ethnicity. Residents of the catchment area often experience adverse social determinants of health; for example, almost one in four residents of Philadelphia County live in poverty, the highest rate in the nation for large U.S. cities¹⁴.

Data Sources

All analysis was completed using publicly available, community-level datasets, including the American Community Survey (2019–2023 5-year)¹⁵ and Decennial Census (2020) (US Census Bureau)¹⁶, CDC Places/Behavioral Risk Factor Surveillance System (BRFSS) (2022) (Centers for Disease Control and Prevention (CDC))¹⁷, and Mammography Facility Database (2024¹) (US Food and Drug Administration (FDA))¹⁸. The latest release data were used for each dataset. All census data were retrieved using the R package, tidycensus¹⁹.

Study Population and Inclusion/Exclusion Criteria

Data were analyzed at the level of census tracts, across the 7-county SKCCC catchment area: Starting from the total census tracts in the catchment/study area (N=1241), only tracts with a residential female population were included in the analysis (N=1216).

Data Cleaning and Processing

Variables selected from the American Community Survey (ACS) were combined to identify specific subpopulations that correspond with screening guidelines, based for average risk demographic profiles. The most recent guidelines from the US Preventive Service Task Force (USPSTF) states that there is a net benefit for women aged 40 to 74 years to receive biennial mammography screening⁷. To focus our analysis on this target population, we limited key variables to the sex and age groups as close as possible to the USPSTF screening criteria. For example, variables were specified, as Black female 35–74 years old, and No health insurance female 35–74 years old². Percentages were computed with different population denominators, selected for each variable. These included total residential population, female population, and various female subpopulations (16 and older, 18 and older, 25 and older). Fourteen variables were imputed, using a k-nearest neighbors algorithm (k=5)²⁰. Eight data points had one variable imputed, and four data points had 2, 3, 9, and 11 variables imputed. Sensitivity analysis was conducted to assess the inclusion of imputed values. Methods are described below. Population denominators, data sources, and imputed value counts are listed in Supplement 1. All analysis was done using R software version 4.4.1²¹.

Principal Component Analysis (PCA)

Principal component analysis provides dimensionality reduction, while addressing multicollinearity across variables. All variables are continuous and were normalized prior to analysis. Each principal component captures the maximum variance in the data along a specific direction in the feature space, with each component being orthogonal to the others. Each principal component is a linear combination of all original variables, with loading values representing the correlation between each variable and the component, indicating their relative contribution to the variance explained by that component. Principal components are evaluated by identifying those with eigenvalues greater than or equal to one (following the Kaiser rule²²), which indicates that the component explains more variance than any single variable alone, making it a meaningful dimension to include in analysis. Each component is also evaluated for variance explained, with the first component contributing the largest percentage to the cumulative proportion of explained variance. These proportions are then used as weights for each component in the calculation of the composite index. Each component is evaluated for correlation with mammography screening rates, using two-side Pearson’s r tests, to assess the direction of the relationship (positive or negative). This is used to standardize component scores when combining them in the index, so that high and low scores can be comparatively interpreted across all components.

Index Construction

The CNPI-BCS was constructed using a weighted sum method²³, where for each census tract, $i$, the index is equal to the sum of the first $k$ principal component scores, $PC$, multiplied by each respective weight, $w$, and the directional association (1 or −1), $D$:

$$Inde{x}_i=\sum _{k=1}{D}_k\ast P{C}_{ki}\ast W_k$$

High index values indicate high need – these are areas, or communities with higher barriers to care and therefore expected to have lower screening adherence rates. Once the index was computed for each census tract, five quintile groups (equal parts each containing 20% of the data) were computed for index score values to prepare stratification for subgroup analysis. CNPI-BCS values were re-scaled from 0 to 1 for ease of interpretation. Summary statistics were calculated for the original variables to characterize each subgroup.

Sensitivity Analysis

The following series of tests were used to compare the data and analysis with and without imputation: 1) distribution of the 14 variables using a Kolmogorov-Smirnov tests; 2) results for the principal component analysis were compared with and without imputation, based on eigenvalues and explained variance of each component, Tucker’s congruence coefficient of scores for each component, and largest change of loading value’s for each variable; and 3) resulting index values at the census tract-level, based on percent change and change between quintile classification. Results are presented in Supplement 2.

RESULTS

Variable Selection

Twenty-three variables were selected for which there are known associations with breast cancer screening rates^24–27. These variables were identified from the following social determinants of health categories (based on Healthy People 2030)²⁸: economic stability, education access and quality, neighborhood and built environment, social and community context, and health status, healthcare access and quality. All variables used in the downstream analyses are listed in Table 1.

Table 1. Variables Included in the Community Need Priority Index – Breast Cancer Screening (CNPI-BCS).

Mean and median were computed for each of the twenty-three variables included in the analysis.

	n = 1241
	Mean (SD)	Median [IQR]
Social and community context
Female 40–74 (%)	42.1 (9.1)	43 [11]
Hispanic: Fem 35–74 (%)	3.8 (5.9)	2 [3.5]
Black: Fem 35–74 (%)	10.5 (13.6)	4.2 [12.8]
No Vehicle: Workers (%)	8 (11.6)	2.8 [10.3]
No internet or no computer: Households (%)	9.6 (7.4)	7.8 [8.8]
Foreign born: Fem 18+ (%)	13.3 (10.5)	10.4 [11.1]
Limited English-speaking household: Adults (%)	3.6 (5.7)	1.5 [3.9]
Spanish speaking: Households (%)	8 (11.9)	4.4 [6.5]
No spouse: Fem 40–74 (%)	73.2 (13.4)	73.1 [20.2]
Education access and quality
No Bach/Grad Degree: Female 25+ (%)	58.4 (21.2)	60.4 [32.2]
Economic stability
Median Household Income ($ USD)	95055.7 (43582.8)	90485.5 [57954.5]
Income at/below 200% of poverty level ($ USD)	25.8 (18.6)	20.3 [25]
Unemployment: Fem 35–74 (%)	4.9 (6.1)	3.1 [5.9]
Underemployment: Female 16+ (%)	59.9 (10.1)	60.1 [12.3]
Food Stamps - SNAP: Households (%)	15.6 (16)	9.4 [17.8]
Neighborhood and built environment
Rural Population (%)	3.9 (15.5)	0 [0]
Health status, healthcare access and quality
No Health Insurance: Fem 35–74 (%)	2.1 (2.5)	1.4 [2.6]
Public Health Insurance: Fem 35+ (%)	22.9 (13.2)	19.4 [14.9]
Disability: Fem 35–74 (%)	16.7 (12.1)	13.8 [14.5]
Annual Checkup in past year: 18+ (%)	78.2 (2.9)	77.9 [3.6]
Obesity (BMI ≥30.0 kg/m2): 18+ (%)	32.6 (6)	31.4 [8.3]
No physical activity/exercise in past month: 18+ (%)	23.3 (7.7)	21.5 [9.5]
Mammography facilites (kernel density estimate)	4.7 (4.9)	3.1 [6.2]

Dimensionality Reduction

Principal component analysis (PCA) was performed to identify potential underlying patterns in the twenty-three variables above. To explore the most relevant components, those with an eigenvalue below 1 were removed from downstream analyses as they were determined not to significantly contribute to the overall variation in the data, resulting in five principal components (PCs) (Fig 1A). Importantly, these five components described over 70% of the variance within the data, ensuring that the data was still accurately represented despite reducing dimensionality (Fig 1B). As shown in Table 2, data elements were stratified across five main components which generally separated into variables related to socioeconomic status (e.g. income, food insecurity), household characteristics (household language, family composition), and access to care (transportation, insurance status). To ensure each variable was contributing to one or more of the five PCs, we examined the loading values for every variable across each component and determined each was contributing to at least one component, with an absolute value of at least 0.25 (Supplement 3).

Fig 1.

(A) Scree plot of potential CNPI-BCS variables. Eigenvalues were reported for each PC, which decreased from above nine to less than zero. The first five components were selected, based on the Kaiser rule—eigenvalues greater than one offered more explanatory value than individual variables alone. (B) Cumulative variance of PCs. Cumulative variance was reported, which increased from 40% to above 71.6%, when the first five components were combined. CNPI-BCS, Community Need Priority Index-Breast Cancer Screening; PC, principal component.

Table 2. Loading values for data elements in each PC.

The first 5 principal components, and the top 5 most relevant variables based on absolute value of loading values were computed. Explained variance was calculated for each component, as well as the cumulative variance.

PC	Variable	Loading value
PC1 Explained Variance: 40.0%	Income at/below 200% of poverty level ($ USD)	0.308
	No physical activity/exercise in past month: 18+ (%)	0.303
	Food Stamps - SNAP: Households (%)	0.301
	Public Health Insurance: Fem 35+ (%)	0.287
	Median Household Income ($ USD)	−0.278
PC 2 Explained Variance: 11.5%	Annual Checkup in past year: 18+ (%)	0.409
	Limited English-speaking household: Adults (%)	−0.349
	Foreign born: Fem 18+ (%)	−0.346
	Mammography facilites (kernel density estimate)	−0.325
	Black: Fem 35–74 (%)	0.300
PC 3 Explained Variance: 9.7%	Female 40–74 (%)	−0.392
	No Vehicle: Workers (%)	0.369
	No spouse: Fem 45–74 (%)	0.340
	Hispanic: Fem 35–74 (%)	−0.318
	Mammography facilites (kernel density estimate)	0.304
PC 4 Explained Variance: 5.6%	Foreign born: Fem 18+ (%)	−0.548
	Annual Checkup in past year: 18+ (%)	−0.467
	Limited English-speaking household: Adults (%)	−0.295
	Spanish speaking: Households (%)	0.235
	Female 40–74 (%)	−0.235
PC 5 Explained Variance: 4.9%	Underemployment: Female 16+ (%)	−0.595
	No Health Insurance: Fem 35–74 (%)	0.344
	Rural Population (%)	−0.311
	Unemployment: Fem 35–74 (%)	−0.284
	Mammography facilites (kernel density estimate)	−0.271
Cumulative Variance: 71.6%

Community Needs Priority Index – Breast Cancer Screening (CNPI-BCS) Creation

To create an index which utilized each of the components identified through the PCA, we utilized a weighted sum approach to ensure that components, which explained more variance within the data, were given a larger weight in the final index (Figure 2A). Two-sided Pearson’s correlation tests were used to assess the relationship between each component and self-reported mammography screening rates. The following correlation statistics and p-values were calculated – PC1: cor=−0.49, p <0.0001; PC2: cor=0.18, p <0.0001; PC3: cor=0.43, p <0.0001; PC4: cor=−0.31, p <0.0001; PC5: cor=0.03, p= 0.393. The signs of component scores for PC2, PC3, and PC5 were reversed to match the directionality of each component in the creation of the index. This assured congruence across components, to ensure higher index values correspond to lower screening rates. CNPI-BCS scores for census tracts were scaled from 0 to 1.0, with a mean of 0.259 (sd = 0.161) and median of 0.222 (IQR = 0.194). Next, as current mammography screening rates were not included in the index to decrease potential bias in our methods, we compared mammography rates with the index values to confirm association between the two variables across the study area. Two-sided Pearson’s correlation tests were used to assess this relationship. As shown in Figure 2B, the CNPI-BCS is well correlated (cor = −0.58, P < .0001) with BCS rate, aligning higher CNPI-BCS score with higher screening need. To understand areas that would most benefit from increased screening, we separated the census tracts into quintiles, with the bottom quintile (Q1) representing areas of lowest need and the top quintile (Q5) the highest need. A biplot is used to visualize the index construction and resulting output (Supplement 4). The input variables are represented as vectors to demonstrate the relationship between each variable and each of the first two principal components. CNPI-BCS quintiles are also shown in the biplot, demonstrating clear separation between quintile, along the first dimension (PC 1).

Fig 2.

(A) CNPI-BCS calculation. Each PC was multiplied by its relative weight and summed to create the initial index. Zero to one scaling was applied to increase the interpretability of the CNPI-BCS. Two-sided Pearson’s correlation tests were used to assess the relationship between each component and self-reported mammography screening rates. The following correlation statistics and P values were calculated: PC1: cor = −0.49, P < .0001; PC2: cor = 0.18, P < .0001; PC3: cor = 0.43, P < .0001; PC4: cor = −0.31, P < .0001; and PC5: cor = 0.03, P = .393. The signs of component scores for PC2, PC3, and PC5 were changed to match the directionality of each component in the creation of the index. This ensured congruence across components, which allowed higher index values to correspond to lower screening rates. (B) Correlation between CNPI-BCS and adherence to BCS. Mammography screening rates within each census tract are plotted against CNPI-BCS values within the same tract and correlation between the two data points calculated. Two-sided Pearson’s correlation tests were used to assess this relationship. CNPI-BCS, Community Need Priority Index-Breast Cancer Screening; PC, principal component.

CNPI-BCS Summary

To identify specific areas of need within our catchment area, we mapped the Breast Cancer screening CNPI-BCS across the SKCCC catchment area. CNPI-BCS scores for census tracts in Q1 range from 0 to 0.129, and Q5 range from 0.374 to 1.0. Summarized by county, Philadelphia County had the highest mean value of 0.364 (sd = 0.178), illustrating the highest screening need, whereas the lowest mean was in Montgomery County, which had a mean CNPI-BCS value of 0.169 (sd = 0.092). 45% (n = 175) of tracts in Philadelphia County were in Q5 – the highest proportion of any county, while Gloucester and Bucks had the lowest proportions (each 1.4%, n = 5 and n = 53, respectively). Montgomery county had 37.5% (n = 81) in the Q1 – the highest proportion of any county, followed by Bucks (36.3%, n = 53). Q5 tracts were primarily concentrated in Philadelphia, Camden, and Delaware counties. Camden and Delaware counties had the most evenly distributed tracts across all five quintile groups, while Bucks and Gloucester counties had the lowest.

Quintile Characterization

To better understand the populations within the CNPI-BCS quintiles, we analyzed the contributing data elements for significant differences across the tracts. Census tracts in Q5 (highest need) had a significantly higher percentage of workers with no vehicle (19.7%) than in Q1 (2.4%) (Table 3). In comparison, high need areas also exhibited the highest density of mammography facilities, likely due to the more urban nature of these census tracts compared to the lower quintiles. Tracts within the highest need quintile also tended to include more racial and ethnic minorities, with adults living in a limited English-speaking household significantly overrepresented in the highest quintile (8.8% in Q5 compared to 1% in Q1). Additionally, while there were similar rates of annual checkups between low and high need quintiles (which suggest similar access to primary care facilities), there were increased numbers of uninsured/publicly insured individuals, with a marked increase in females over 35 shown to have public health insurance in Q5 (42.4%) versus Q1 (11.9%).

Table 3.

Summary characteristics by CNPI-Breast Cancer Screening quintile. Mean and median were calculated for each variable, stratified by CNPI-BCS quintile

	CNPI-BCS Q1: Low Need		CNPI-BCS Q2		CNPI-BCS Q3		CNPI-BCS Q4		CNPI-BCS Q5: High Need
	n = 244		n = 243		n = 243		n = 243		n = 243
	Mean (SD)	Median [IQR]	Mean (SD)	Median [IQR]	Mean (SD)	Median [IQR]	Mean (SD)	Median [IQR]	Mean (SD)	Median [IQR]
Social and community context
Female 40–74 (%)	46.6 (7.9)	47.5 [8.4]	43.1 (8.1)	43.7 [8.9]	42.4 (8.3)	42.6 [9]	40.3 (9.8)	41.6 [11.2]	38.1 (9)	38.1 [11.4]
Hispanic: Fem 35–74 (%)	1.5 (1.7)	1 [1.7]	2 (1.8)	1.5 [2.4]	2.7 (2.4)	2.2 [2.6]	3.7 (3.4)	3.1 [4.5]	9.3 (10.4)	5.2 [13]
Black: Fem 35–74 (%)	2.1 (3)	1.1 [2.8]	4.1 (6.3)	1.8 [4.4]	7.4 (10.2)	3.9 [7.2]	15.1 (14.7)	10.4 [19]	24 (15.3)	22.2 [27.5]
No Vehicle: Workers (%)	2.4 (6.3)	0.7 [2]	4.8 (9.7)	1.6 [4]	4.1 (6.7)	2.1 [3.4]	9.2 (11.8)	5.5 [10]	19.7 (12.6)	17.6 [17.2]
No internet or no computer: Households (%)	3.7 (2.5)	3.2 [3.2]	6 (3.4)	5.4 [4.5]	8.5 (4.3)	8 [5.6]	11 (5.8)	10.3 [8]	18.6 (8.5)	17.8 [12]
Foreign born: Fem 18+ (%)	11.3 (6.7)	9.9 [8.8]	10.7 (6.9)	8.9 [8.9]	11.5 (7.8)	9.5 [9.9]	15.4 (11.8)	12.7 [12.7]	17.6 (14.8)	12.9 [19.9]
Limited English-speaking household: Adults (%)	1 (1.5)	0.4 [1.3]	1.5 (1.8)	0.9 [2.4]	2.1 (2.3)	1.3 [2.7]	4.4 (4.9)	3.1 [4.8]	8.8 (9.3)	5.9 [11.7]
Spanish speaking: Households (%)	2.8 (2.7)	2.1 [2.7]	3.5 (2.5)	3.1 [3.5]	5.2 (3.9)	4.4 [5.2]	7.8 (5.4)	7.2 [7.8]	20.6 (20.8)	13.5 [25.3]
No spouse: Fem 40–74 (%)	58.4 (10.6)	57.3 [12]	68.2 (10.3)	66.8 [14.6]	72.7 (8.2)	71.8 [10.1]	79.5 (8.7)	79.3 [13.2]	87.3 (7.2)	88.1 [9.8]
Education access and quality
No Bach/Grad Degree: Female 25+ (%)	35 (12.9)	35.8 [15.6]	47.2 (15.1)	49.9 [20.2]	60.6 (13.7)	62.3 [16.2]	67.6 (15.1)	71.2 [14.5]	81.5 (12.1)	83.7 [13.8]
Economic stability
Median Household Income ($ USD)	154628.7 (34817.7)	148807.5 [43778.5]	114036.3 (20391.2)	112120 [22202]	90571.7 (17234.9)	89800 [22188]	70926.4 (17184.3)	68778 [24477.5]	44870 (15403.4)	42569 [19433.5]
Income at/below 200% of poverty level ($ USD)	8.2 (4.2)	7.7 [5.4]	13.4 (4.8)	13 [6.2]	20.3 (6.4)	19.7 [7.4]	32.3 (9.6)	31 [13.5]	54.9 (13.3)	54.5 [17.5]
Unemployment: Fem 35–74 (%)	2.8 (3.2)	2 [3.8]	3.4 (3.4)	2.6 [4.6]	4.5 (5.1)	3.3 [5.4]	5.1 (5.6)	3.6 [7.6]	8.5 (9.2)	5.9 [10.8]
Underemployment: Female 16+ (%)	57.6 (10.2)	58.3 [12.1]	56.7 (9.6)	57.4 [12.2]	58.5 (8.8)	59.1 [10.9]	60.1 (8.8)	60.2 [11.7]	66.7 (10)	66.6 [12.8]
Food Stamps - SNAP: Households (%)	3.2 (2.6)	2.6 [3.4]	5.4 (3.7)	4.6 [4.8]	9.8 (5.2)	9.2 [6.2]	18.5 (8.6)	17 [11.9]	41 (14.6)	40.3 [20.8]
Neighborhood and built environment
Rural Population (%)	7.4 (19.7)	0 [0]	7.2 (21.8)	0 [0]	2.9 (13.7)	0 [0]	1.7 (10.3)	0 [0]	0.1 (0.8)	0 [0]
Health status, healthcare access and quality
No Health Insurance: Fem 35–74 (%)	0.7 (0.9)	0.5 [1]	1.4 (1.6)	0.9 [1.7]	2 (1.8)	1.7 [1.9]	2.7 (2.8)	1.9 [2.8]	3.7 (3.3)	2.9 [4.3]
Public Health Insurance: Fem 35+ (%)	11.9 (4.4)	11.4 [5.2]	15.4 (4.8)	14.8 [6.5]	19.4 (5.3)	19 [5.6]	25.6 (6.9)	25.1 [8.6]	42.4 (13.2)	41.3 [20.1]
Disability: Fem 35–74 (%)	6.5 (3.7)	5.7 [4.5]	10.1 (4.8)	9.8 [6.5]	14.7 (5.9)	14.4 [8.1]	20.4 (8.5)	19.7 [11.2]	32 (13.6)	29.9 [16.4]
Annual Checkup in past year: 18+ (%)	78.7 (2.4)	78.8 [2.6]	77.7 (2.3)	77.7 [2.6]	77.9 (2.8)	77.4 [2.7]	78 (3.3)	77.5 [3.7]	78.8 (3.4)	78.6 [5.9]
Obesity (BMI ≥30.0 kg/m2): 18+ (%)	27.9 (2.5)	28 [2.9]	29.4 (3.4)	29.8 [4.4]	31.7 (4)	31.6 [5]	34.3 (5.1)	34.9 [7.1]	39.7 (5.6)	41 [7.5]
No physical activity/exercise in past month: 18+ (%)	16.2 (2.4)	16.5 [3.1]	18.2 (2.9)	18.6 [3.5]	21.8 (3.1)	21.9 [3.5]	25.3 (4.1)	25.5 [5.2]	34.9 (6.2)	34.1 [8.1]
Mammography facilities (kernel density estimate)	3.1 (3.8)	1.7 [4]	4.3 (4.9)	2.6 [5.5]	3.7 (4.7)	1.6 [5]	5.1 (5.4)	3.4 [6.3]	7 (4.5)	6.3 [6.7]

DISCUSSION

To our knowledge, the research presented here is one of the first of its kind to create a geographically based composite index that can be tailored to a particular health behavior, outcome, or service to determine where to deploy interventions to overcome barriers and disparities in a catchment area. In this case, we created the Community Needs Priority Index – Breast Cancer Screening to identify communities in our catchment area that are most in need of interventions to promote breast cancer screening. The CNPI-BCS includes a variety of publicly available data sources at the census tract level that contribute to our understanding of breast cancer screening disparities, including individual-level factors such as race, ethnicity, and insurance status as well as community-level factors like the density of mammography screening providers. From the CNPI-BCS, we identified 243 tracts in Q5 across the catchment area as priorities for future breast cancer screening interventions.

Just as the causes of health disparities are multifaceted, so are the types of interventions that can be deployed to mitigate the disparity. In the case of interventions to mitigate breast cancer screening disparities, our cancer center operates a mammography mobile screening unit (MSU), an innovative method to provide medical care outside of fixed clinical sites²⁹. MSUs increase community access by bringing health care services to convenient locations thus decreasing the distance and travel time needed to access and creating opportunities for awareness and health education. Our MSU covers the cost of breast cancer screening and follow-up care for those who either lack health insurance or the means to cover the costs, therefore also reducing financial barriers to screening. However, other types of evidence-based interventions could be deployed in the priority regions such as educational campaigns about the importance of breast cancer screening, direct outreach by mail or telephone to eligible patients, navigation for scheduling and appointment attendance, and training for providers and clinical staff to recommend screening³⁰. Selecting the appropriate intervention for the priority areas may require knowledge about the most common barrier within the priority region, as well as the availability and feasibility of deploying a specific type of intervention.

A novel aspect of the CNPI is that it can be tailored to identify priority census tracts for interventions related to any type of health issue, if data sources are available to represent known factors that contribute to the health issue or disparity. In addition to the CNPI-BCS, a CNPI could be created to identify census tracts in need of interventions related to colon or lung cancer screening, smoking cessation, or diabetes and lifestyle modification. This speaks to the generalizability and scalability of the CNPI, as others can benefit from the creation of issue-specific, tailored versions of the CNPI.

Despite the utility of the CNPI-BCS in identifying priority census tracts for future interventions, there are some limitations to the index. The CNPI-BCS does not contain data related to all factors known to contribute to screening disparities, as not all data is available at the census tract level. While we felt that it was important to be able to identify communities at the finest geographic scale available, we recognize that communities do not often align to census tracts and that these may be arbitrary boundaries in a catchment area. Further, the CNPI has not been validated, as there are no other measures that prioritize community need at the census tract level.

We intentionally did not include breast cancer screening rates in the CNPI-BCS, even though these rates are available, although with some methodological limitations – including that the data are sampled estimates, self-reported, and restricted to women over 50 years old. Our desire was to create an index that accounted for more than just low screening rates when identifying areas of need, such that SDOHs and environmental factors were also included. Therefore, we expect that the CNPI-BCS will identify areas of need better than screening rates alone. We expect that by launching an intervention in the highest CNPI-BCS census tracts – in our case, deploying the mammography MSU – the rate of females who indicate that they have never previously been screened will be higher than the overall rate in our catchment area, indicating that we are overcoming barriers to screening and thus providing validation of our measure.

In summary, the Community Needs Priority Index for Breast Cancer Screening (CNPI-BCS) represents an innovative and scalable method to identify regions in greatest need of intervention, focusing on the complex interplay of social, economic, and healthcare-related factors that impact screening rates. The CNPI-BCS method integrated diverse data sources at the census tract level, which allowed for precise targeting of interventions aimed at reducing disparities in breast cancer screening. Although the index has not yet been validated and is limited by the availability of certain data, we intend to further study its potential to guide resource allocation and inform the design of tailored, evidence-based interventions. As such, the CNPI-BCS aimed to serve as a model that may be adapted for use with other health issues, providing a replicable framework for addressing health disparities and improving access to critical healthcare services in underserved communities.

Fig 3.

Map of CNPI-BCS values across SKCCC Catchment. Census tract-level CNPI values across the SKCCC catchment area were mapped. Values were grouped by quintile, where the bottom quintile (dark blue) represents the lowest need for BCS interventions and the top quintile (orange) represents the highest need. CNPI-BCS, Community Need Priority Index-Breast Cancer Screening; SKCCC, Sidney Kimmel Comprehensive Cancer Center.

Context Summary.

Key objective:

Using an evidence-based approach for identifying community need, to determine which census tracts across a cancer center catchment area are highest priority for cancer screening interventions.

Knowledge generated:

A composite measure was produced, Community Need Priority Index – Breast Cancer Screening (CNPI-BCS), to evaluate each census tract in our catchment area. CNPI-BCS showed correlation with breast cancer screening rates, and 243 tracts were identified as high need.

Relevance*:

This study reinforces the value of using geographic information to augment standard approaches for the identification of communities in need of increased efforts to improve cancer screening rates.

* The relevance statement was written by Editor-in-Chief Jeremy L. Warner, MD, MS, FAMIA, FASCO