Background:

Cervical cancer disparities exist in the United States with the highest incidence in Hispanic women and the highest mortality in Black women. Effective control of cervical cancer in the population requires targeted interventions tailored to community composition in terms of race, ethnicity, and social determinants of health (SDOH).

Methods:

Using cancer registry and SDOH data, geospatial hot spot analyses were carried out to identify statistically significant neighborhood clusters with high numbers of cervical cancer cases within the catchment area of an NCI-Designated Cancer Center. The locations, racial and ethnic composition, and SDOH resources of these hot spots were used by the center's community outreach and engagement office to deploy mobile screening units (MSU) for intervention in communities with women facing heightened risk for cervical cancer.

Results:

Neighborhood hot spots with high numbers of cervical cancer cases in south Florida largely overlap with locations of poverty. Cervical cancer hot spots are associated with a high percentage of Hispanic cases and low SDOH status, including low income, housing tenure, and education attainment.

Conclusions:

A geospatially referenced cancer surveillance platform integrating cancer registry, SDOH, and cervical screening data can effectively identify targets for cervical cancer intervention in neighborhoods experiencing disparities.

Impact:

Guided with a data-driven surveillance system, MSUs proactively bringing prevention education and cervical screening to communities with more unscreened, at-risk women are an effective means for addressing disparities associated with cervical cancer control.

With an estimated 11,500 new cases diagnosed and 4,000 fatalities each year (1), cervical cancer is the third most common gynecologic malignancy in the United States, behind uterine and ovarian cancers. It is estimated that 90% of cervical cancer cases can be attributed to infections with human papilloma virus (HPV; ref. 2), a sexually transmitted disease agent for which vaccines have been available in the United States since 2006 (3). Although cervical cancer can be effectively prevented in the population with HPV vaccination in adolescents between ages 9 and 26 (4) and screening with HPV tests and Pap smear tests in adults between 21 and 65 (5), cervical cancer disparities by population group have been observed in the United States, with the highest incidence seen in Hispanic women and the highest mortality in Black women (1). Fuzzell and colleagues (6) noted that not all population subgroups (e.g., racial and ethnic minorities, rural residents, gender minorities, groups with different religious beliefs, non-English speakers, and groups with various medical conditions) participate in screening equally. In general, socioeconomically disadvantaged and minority groups show lower rates of being screened. Addressing barriers to achieving equal screening rates requires interventions targeting the individual needs of the subgroups.

Community outreach and engagement (COE) is a public health approach that can address many individual barriers to HPV vaccination and cervical screening (6). COE with mobile screening units (MSU), which are typically large vehicles outfitted with screening equipment and health care workers, are a viable approach for offering screening and HPV vaccination to targeted population groups for whom barriers to services at clinics or hospitals exist, such as those living in poverty and/or rural communities (7). In addition, for community outreach and MSU deployments to be effective, it is necessary that targets for intervention be identified with evidence from data containing the geographical location as well as the composition (e.g., races, ethnicities, income levels, and insurance status) of the population at risk (i.e., identified with historic trends of cervical cancer incidence and mortality). A surveillance system built with geospatially-referenced cancer registries and social determinants of health (SDOH) data (e.g., race, ethnicity, household income, and education) can serve as the platform needed to identify population targets for cervical cancer prevention at different geographic scales (e.g., neighborhoods, cities, counties, states, and multi-state regions) by integrating data on cancer registries, screening rates, and SDOH (8).

A geospatial hot spot refers to a location (i.e., census tract) where a value for a particular feature (e.g., number of women diagnosed with cervical cancer) tends to be higher than nearby locations (9). For cervical cancer research in the United States, geospatial hot spot analysis has been applied to identify geographic clusters of late-stage cervical cancer incidence at the county level for the entire United States (10) and at Census Block Group level for the state of California (11). Another study evaluated spatial clustering of female breast, cervical, and colorectal cancer incidences in the city of Baltimore, Maryland (12). Although these studies confirmed the existence of cervical cancer hot spots as well as associations between incidence and disadvantaged socioeconomic status at different geographic scales, the findings do not have the necessary resolution in terms of geographic scale and neighborhood characteristics for targeted interventions with respect to race, ethnicity, SDOH, and local cervical screening uptake (13, 14).

The purpose of this article is to present innovative COE efforts for cervical cancer control within the catchment area of the Sylvester Comprehensive Cancer Center (SCCC), the NCI-Designated Cancer Center for the highly populated metropolitan area of south Florida. The Sylvester Catchment Area (SCA) consists of four counties: Miami-Dade, Broward, Palm Beach, and Monroe, with an estimated population of over 6 million in 2021. These COE efforts are unique in that a surveillance system, SCAN360, is used to identify targets for deploying MSUs [known as the Game Changer (GC) vehicles] for prevention education and cancer screening in these communities. Using cancer registry and SDOH data in SCAN360, geospatial hot spot analyses were carried out to identify statistically significant neighborhood clusters with high numbers of cervical cancer cases. SDOH variables including race, ethnicity, income, education level, housing tenure, transportation means, and cervical screening uptake within these clusters were analyzed to identify population segments in need of community outreach intervention. Results of the analyses and choropleth maps showing locations of the hot spots were used by the COE team to plan for deployment of the GC vehicles.

Scan360

Sponsored by the SCCC, SCAN360 is a surveillance system that provides a web-based user interface built with interactive tables, figures, and maps for a “360-degree view” of cancer incidence, SDOH, and risk factors in neighborhoods throughout Florida (15). Cancer incidence and mortality data in SCAN360 come from the Florida Cancer Data System (FCDS), the state's official cancer registry. SDOH data are derived from dozens of sources including those provided by the U.S. Census Bureau, such as the 2010 decennial census and American Community Survey (ACS) by U.S. Census Bureau (RRID:SCR_011587). The PLACES data from the Centers for Disease Control and Prevention (CDC) provide geocoded estimates of behavioral risk factors such as smoking, alcohol use, access to health insurance, and cancer screening rates at both the census place and tract levels.

Sylvester community outreach and engagement with the GC vehicles

Guided with information provided by SCAN360, Sylvester's Office of Outreach and Engagement deploys a multitude of interventions in the communities to increase awareness of the importance of early detection and screenings and by making different types of screening techniques available. Sylvester's COE team provides several options for cancer screening using self-sampling approaches for fecal immunochemical testing and HPV cervical screenings. Testing for prostate-specific antigen, hepatitis C, and human immunodeficiency virus can also be offered by deploying phlebotomists to communities. These COE events take place via MSUs called the GC vehicles, which are equipped with private rooms for blood work, screening activities, and in-person counseling for cancer risk avoidance (e.g., smoking cessation; ref. 16).

Geospatial hot spot analysis of cervical cancer cases

Geospatial hot spot analysis for this study was carried out with ArcGIS Pro 3.0, which implements the Gi* statistic formulated by Getis and Ord (9). For every location within a predefined spatial extent (e.g., within a 2-mile radius or for every 20 connected, neighboring locations), Gi* is calculated as a weighted sum of the feature values for all locations within the extent, divided by the sum of all features in the entire study area (17). For every hot spot analysis run, ArcGIS produces a Z-score (i.e., difference between Gi* and its expected value, divided by the SD of Gi*) to determine the significance of a Gi* (18). If a Z-score of a Gi* is above the significance thresholds (i.e., for a hot spot, Z >1.65 for 90% significance, Z > 1.96 for 95% significance; and Z > 2.5 for 99% significance; thresholds with the same absolute values but less than zero are used for cold spot determination), the location is considered a statistically significant hot (or cold) spot (18). In practice, a Getis-Ord Gi* analysis is preceded by an analysis for the global Moran's I statistic (19) to determine the spatial extent for Gi* evaluation (20, 21). The global Moran's I tests for statistical significance of spatial autocorrelation (i.e., clustering of locations with high values) in a study area with a hypothesized spatial extent (e.g., a circular boundary; ref. 22). A Moran's I analysis also produces a Z-score for inferential analyses. A large, above-threshold Z-score means that there is intense clustering of higher values in the catchment area within the hypothesized extent. In theory, the probability of finding statistically significant hot spots is highest when Gi* is evaluated with the spatial extent that produces the largest Z-score for Moran's I (20, 21).

For this study, the purpose of hot spot analysis is to identify neighborhood clusters with the highest number of cervical cancer cases for targeted intervention with the GC vehicles. For public health research, there has never been a consensus on the precise areal size for a neighborhood (23). We used census tracts as the areal units to represent the construct of neighborhood, because a census tract is the smallest census geography used by public agencies to share data (i.e., most public health data are not released at the smaller “block level” due to confidentiality considerations). To identify neighborhood hot spots of cervical cancer, we first tallied the FCDS invasive cervical cancer cases (i.e., no benign, borderline malignant, or in situ cases) diagnosed from 2010 to 2019 per census tract. Before determining the dependent variable for Gi* calculation, we explored the difference in spatial patterns between hot spots identified by number of cases and crude incidence rates (i.e., number of cases divided by number of at-risk populations for each census tract). Compared with hot spots by total case numbers, we found that hot spots by crude incidence rates shifted from census tracts with higher numbers of cases toward census tracts with fewer cases but smaller female populations. In the highly urbanized environment of south Florida, minorities and disadvantaged populations mostly live in high-density urban cores, for which high case numbers are weighted down by high populations in incidence calculations, while neighborhoods of higher income levels are less densely populated and showed higher incidence for smaller number of cases. It was determined that hot spots identified with crude incidence rates do not represent the intended targets for intervention by the GC vehicles, which were designed to address disparities in cancer control.

According to data released by the Florida Department of Health (24), from 2010 to 2019 annual age-adjusted incidence of cervical cancer within the SCA fluctuated from year to year between minimum 276 and maximum 342 per 100,000 women (i.e., the estimated numbers do not include incidences in Monroe County, of which the incidences were too small to be released for public viewing) with no significant trend of increasing or decreasing. Therefore, we decided to perform hot spot analyses by calculating Gi* statistics for the total number of cases in the 10-year period from 2010 to 2019, rather than for any single year, to reduce the possibility of identifying hot spots by random chance.

Estimating numbers of women unscreened

In identifying intervention targets for cervical cancer control, in addition to neighborhoods with a high number of cases, those with a high number of women ages 21 to 65 (i.e., at high risk for cervical cancer) who do not follow guideline-concordant screening also need to be located. Because cervical cancer screening rates in the SCA are uniformly high but the numbers of at-risk women vary from county to county (see Table 1 for descriptive statistics), screening rates alone cannot reflect variations of cervical cancer risk in different neighborhoods. We estimated the number of unscreened, at-risk women for a census tract by multiplying the percentage of unscreened women for the tract, identified from the 2022 release of CDC's PLACES data, with the number of women aged 20 to 64 (i.e., every 5-year age group beginning with 0 or 5 and ending with 4 or 9 in ACS data) for the same census tract obtained from the 5-year estimate of 2019 ACS data.

Table 1.

Descriptive statistics of cervical cancer cumulative incidences (2010–2019) and social determinants of health for 1,014 census tracts in south Florida's SCCC catchment area.

CountyStatisticNo. of casesaCum. Inc.bAge-Adj. C.I.cP. Hisp.dP. NH BlackeP. NH WhitefW. Ages 20–64gMedian H. IncomehP. Worker No VehiclesiP. Renter HousingjP. Bachelor DegreekP. No InsurancelP. W. Unscreenm
Broward (N = 308) Mean  3.18 1.31 1.23 23.74 26.37 44.48 1735.31 55866.36 3.28 36.97 29.57 23.27 17.34 
 SD  2.21 0.94 0.95 12.79 25.97 24.32 865.55 24745.12 3.93 20.00 14.51 6.13 2.44 
 Skewness  2.00 3.17 2.83 52.58 126.72 −6.46 1.82 1.40 235.78 47.22 47.65 22.45 46.50 
 Kurtosis  6.68 21.04 16.04 −15.62 47.67 −95.44 8.57 2.67 715.84 −42.01 −51.36 −41.41 −1.86 
 Percentiles 25th 2.00 0.69 0.58 14.27 6.87 25.18 1154.25 38593.00 0.74 21.49 17.48 18.63 15.40 
  50th 3.00 1.11 1.02 23.05 16.62 46.32 1646.50 49754.50 1.95 34.67 27.85 23.00 17.25 
  75th 4.00 1.68 1.62 31.06 38.05 62.27 2205.75 66909.00 4.44 51.07 41.00 27.40 18.90 
Miami-Dade (N = 447) Mean  3.38 1.33 1.30 64.31 18.09 15.07 1720.06 48432.10 5.28 46.20 26.87 32.91 20.20 
 SD  2.19 0.82 0.86 28.02 27.66 15.99 727.514 27630.28 7.52 24.00 17.25 8.95 3.37 
 Skewness  1.24 1.48 1.57 −50.60 148.25 148.44 1.55 2.27 273.89 16.14 100.18 3.08 13.86 
 Kurtosis  1.72 3.00 3.31 −109.19 65.27 126.90 5.72 8.67 939.91 −104.78 30.87 −46.62 −42.19 
 Percentiles 25th 2.00 0.74 0.71 40.27 0.64 3.98 1245.00 29795.00 0.97 24.98 13.60 27.40 17.80 
  50th 3.00 1.13 1.10 70.27 2.83 7.72 1632.00 42219.00 2.32 44.65 21.90 32.70 20.00 
  75th 5.00 1.72 1.69 91.19 23.02 21.41 2087.00 59741.00 6.34 65.83 35.80 39.20 22.70 
Monroe (N = 18) Mean  2.44 1.94 1.48 19.80 4.23 73.01 854.17 63654.83 5.11 34.75 30.54 19.82 18.26 
 SD  1.38 0.77 0.61 11.85 3.67 14.79 461.283 21656.21 7.17 16.20 7.30 5.35 1.69 
 Skewness  2.24 0.10 0.54 51.34 83.84 −52.26 0.49 1.55 287.21 85.60 -27.01 67.73 −13.93 
 Kurtosis  6.75 -1.05 0.02 -65.98 6.13 −60.27 −0.56 3.12 978.07 21.04 52.82 −46.49 −105.35 
 Percentiles 25th 2.00 1.32 1.09 8.71 0.85 62.75 467.00 49128.00 1.00 22.94 25.18 15.00 16.80 
  50th 2.00 1.85 1.42 20.45 4.16 72.85 809.00 59467.50 2.67 29.47 30.50 19.40 17.95 
  75th 3.00 2.56 1.73 28.05 6.44 86.98 1122.00 70761.25 7.13 49.63 36.20 23.23 20.10 
Palm beach (N = 241) Mean  2.57 1.24 1.16 17.94 18.85 59.20 1366.73 55340.75 3.85 33.62 30.94 22.39 19.01 
 SD  1.63 0.72 0.86 14.15 23.19 27.07 690.051 26000.55 5.21 20.94 15.74 9.16 3.23 
 Skewness  1.15 0.97 1.26 147.55 176.49 −54.44 0.80 1.10 229.71 51.60 38.90 92.44 73.51 
 Kurtosis  1.10 1.07 1.99 186.66 235.80 −85.34 1.66 0.99 545.92 −46.86 −76.63 33.61 13.18 
 Percentiles 25th 1.00 0.65 0.48 8.14 3.42 37.67 864.00 35544.50 0.62 15.44 18.00 15.65 16.50 
  50th 2.00 1.10 0.95 13.73 8.73 65.49 1293.00 50260.00 1.85 31.05 27.80 20.00 18.50 
  75th 3.00 1.66 1.64 22.63 23.65 81.16 1845.00 69505.50 4.61 48.28 43.40 27.50 21.20 
CountyStatisticNo. of casesaCum. Inc.bAge-Adj. C.I.cP. Hisp.dP. NH BlackeP. NH WhitefW. Ages 20–64gMedian H. IncomehP. Worker No VehiclesiP. Renter HousingjP. Bachelor DegreekP. No InsurancelP. W. Unscreenm
Broward (N = 308) Mean  3.18 1.31 1.23 23.74 26.37 44.48 1735.31 55866.36 3.28 36.97 29.57 23.27 17.34 
 SD  2.21 0.94 0.95 12.79 25.97 24.32 865.55 24745.12 3.93 20.00 14.51 6.13 2.44 
 Skewness  2.00 3.17 2.83 52.58 126.72 −6.46 1.82 1.40 235.78 47.22 47.65 22.45 46.50 
 Kurtosis  6.68 21.04 16.04 −15.62 47.67 −95.44 8.57 2.67 715.84 −42.01 −51.36 −41.41 −1.86 
 Percentiles 25th 2.00 0.69 0.58 14.27 6.87 25.18 1154.25 38593.00 0.74 21.49 17.48 18.63 15.40 
  50th 3.00 1.11 1.02 23.05 16.62 46.32 1646.50 49754.50 1.95 34.67 27.85 23.00 17.25 
  75th 4.00 1.68 1.62 31.06 38.05 62.27 2205.75 66909.00 4.44 51.07 41.00 27.40 18.90 
Miami-Dade (N = 447) Mean  3.38 1.33 1.30 64.31 18.09 15.07 1720.06 48432.10 5.28 46.20 26.87 32.91 20.20 
 SD  2.19 0.82 0.86 28.02 27.66 15.99 727.514 27630.28 7.52 24.00 17.25 8.95 3.37 
 Skewness  1.24 1.48 1.57 −50.60 148.25 148.44 1.55 2.27 273.89 16.14 100.18 3.08 13.86 
 Kurtosis  1.72 3.00 3.31 −109.19 65.27 126.90 5.72 8.67 939.91 −104.78 30.87 −46.62 −42.19 
 Percentiles 25th 2.00 0.74 0.71 40.27 0.64 3.98 1245.00 29795.00 0.97 24.98 13.60 27.40 17.80 
  50th 3.00 1.13 1.10 70.27 2.83 7.72 1632.00 42219.00 2.32 44.65 21.90 32.70 20.00 
  75th 5.00 1.72 1.69 91.19 23.02 21.41 2087.00 59741.00 6.34 65.83 35.80 39.20 22.70 
Monroe (N = 18) Mean  2.44 1.94 1.48 19.80 4.23 73.01 854.17 63654.83 5.11 34.75 30.54 19.82 18.26 
 SD  1.38 0.77 0.61 11.85 3.67 14.79 461.283 21656.21 7.17 16.20 7.30 5.35 1.69 
 Skewness  2.24 0.10 0.54 51.34 83.84 −52.26 0.49 1.55 287.21 85.60 -27.01 67.73 −13.93 
 Kurtosis  6.75 -1.05 0.02 -65.98 6.13 −60.27 −0.56 3.12 978.07 21.04 52.82 −46.49 −105.35 
 Percentiles 25th 2.00 1.32 1.09 8.71 0.85 62.75 467.00 49128.00 1.00 22.94 25.18 15.00 16.80 
  50th 2.00 1.85 1.42 20.45 4.16 72.85 809.00 59467.50 2.67 29.47 30.50 19.40 17.95 
  75th 3.00 2.56 1.73 28.05 6.44 86.98 1122.00 70761.25 7.13 49.63 36.20 23.23 20.10 
Palm beach (N = 241) Mean  2.57 1.24 1.16 17.94 18.85 59.20 1366.73 55340.75 3.85 33.62 30.94 22.39 19.01 
 SD  1.63 0.72 0.86 14.15 23.19 27.07 690.051 26000.55 5.21 20.94 15.74 9.16 3.23 
 Skewness  1.15 0.97 1.26 147.55 176.49 −54.44 0.80 1.10 229.71 51.60 38.90 92.44 73.51 
 Kurtosis  1.10 1.07 1.99 186.66 235.80 −85.34 1.66 0.99 545.92 −46.86 −76.63 33.61 13.18 
 Percentiles 25th 1.00 0.65 0.48 8.14 3.42 37.67 864.00 35544.50 0.62 15.44 18.00 15.65 16.50 
  50th 2.00 1.10 0.95 13.73 8.73 65.49 1293.00 50260.00 1.85 31.05 27.80 20.00 18.50 
  75th 3.00 1.66 1.64 22.63 23.65 81.16 1845.00 69505.50 4.61 48.28 43.40 27.50 21.20 

aTotal number of invasive cervical cancer cases, 2010–2019.

bCumulative incidences per 1,000 female residents, 2010–2019. Cumulative incidence = total number of 2010–2019 cases/2010 census population.

cAge-adjusted cumulative incidences per every 1,000 female residents, 2010–2019. Age-adjusted cumulative incidences = cumulative incidences age-adjusted to NCI SEER 2000 standard population.

dPercentage of Hispanic population.

ePercentage of non-Hispanic Black population.

fPercentage of non-Hispanic White population.

gNumber of women ages 20–64.

hMedian household income.

iPercentage of workers with no vehicle available.

jPercentage of renter occupied housing units.

kPercentage of adults ages 25 and older with a bachelor's degree or higher.

lPercentage of adults ages 18 to 64 with no health insurance.

mPercentage of women ages 21 to 65 without a Pap test in the last 3 years prior to 2020.

Data Sources: Florida Cancer Data System (a,b,c); 2010 US Census Decennial Survey (d,e,f); 2019 ACS 5-Year Estimates (g); 2015 ACS 5-Year Estimates (h,I,j,k); 2022 CDC PLACES (l,m).

According to the 2018 National Health Interview Survey, the prevalence of women ages 20 to 69 who had hysterectomies (i.e., including total and subtotal hysterectomies that preserve the cervix) is approximately 13% (25) for the entire United States, but data are not available for estimating the prevalence at census tract level. Thus, our estimation of unscreened women may be overestimated for census tracts with women who had hysterectomies. Because the purpose of this analysis is to identify census tracts with high numbers of unscreened women, overestimation by 13% is not expected to cause high-risk neighborhoods to be unrecognized.

Data availability

Except for cancer registry data from FCDS, all other data used for this study were derived from public sources (e.g., U.S. Census) and are available upon request from the corresponding author. Cancer-related data are available after obtaining authorization from FCDS.

Descriptive statistics of cervical cancer incidences and SDOH in SCA

Table 1 provides descriptive statistics for variables pertaining to cervical cancer incidence, race/ethnicity, and SDOH at census tracts in the four counties in SCA. Variables for the number of cases and cumulative incidence were based on FCDS invasive cervical cancer cases diagnosed from 2010 to 2019. Cumulative incidence for each census tract was calculated for every 1,000 female residents by dividing the total number of cases for the decade with 2010 population from the 2010 decennial census. Age adjustment was made by applying age group weights of the 2000 U.S. standard population defined by NCI's Surveillance, Epidemiology, and End Results Program (26). Table 1 shows that the distributions of cervical cancer and SDOH variables in all four counties are mostly skewed. Miami-Dade County, with a dominant 64% Hispanic population, has the highest number of cervical cancer cases and the lowest average median household income of the four counties. Second to Miami-Dade in number of cases is Broward County, which has a more balanced mix of races and ethnicities in the population and the highest mean number of women in the recommended screening ages for cervical cancer per census tract. The majority of the populations of Palm Beach and Monroe Counties is non-Hispanic White. These two counties have fewer women in the recommended screening ages for cervical cancer than Miami-Dade and Broward, as the percentages of the population ages 65 and over in Palm Beach and Monroe are higher.

Cervical cancer hot spot analysis

We used the K nearest neighbors (KNN) method to define the spatial extent for Moran's I and Getis-Ord Gi* analysis, as KNN is considered the most appropriate method when distribution of the feature values is skewed (27). We first calculated Moran's I statistic iteratively with different numbers of KNNs (i.e., census tracts) to identify the optimal K value for Gi* evaluation in hot spot analysis. The process of searching for the optimal K value is by increasing K value incrementally (i.e., beginning with K = 8, considered the minimum value by rule of thumb; ref. 28) for a Moran's I calculation and adjusting the increment according to the difference in the Z-scores between every pair of successive calculations until arriving at the optimal K value with the maximum Moran's I. At K = 37, the Moran's I calculation produced the largest Z-score, 11.5 (i.e., less than 1% likelihood that the clustering pattern resulted from random chance), suggesting that clustering of census tracts with high cervical cancer cases is most intense when spatial autocorrelation is evaluated for every 38 connected census tracts (i.e., 37 neighbors surrounding the census tract for which a component of Moran's I is being calculated, see 28 for mathematic formulation of I). Thus, when Gi* is evaluated with K = 37, the probability of finding statistically significant cervical cancer hot spots within the catchment area is highest. We then performed Getis-Ord analysis with 37 nearest neighbors and identified three major clusters of hot spots with high numbers of cervical cancer cases (Fig. 1). To evaluate socioeconomic disparities associated with the identified hot spots, we used data from the 2015 American Community Survey to identify the percentage of the population in poverty by census tracts in the catchment area. A census tract is identified as a poverty area if it has more than 20% of the population living in poverty, while one with more than 40% is defined as an extreme poverty area (28). By comparing locations of the identified cervical cancer hot spots (Fig. 1) with poverty areas in Miami-Dade and Broward Counties (Fig. 2), we can see that all three clusters of hot spots largely overlap neighborhoods of poverty, which are the main targets for intervention with the GC vehicles.

Figure 1.

Hot and cold spots identified by total number of cervical cancer cases (2010–2019) for 1,014 census tracts in south Florida. Figure shows the locations of hot and cold spots by census tracts, identified with number of cervical cancer cases diagnosed from 2010 to 2019, in south Florida, which consists of four counties. The total number of census tracts included in this analysis is 1,014. Hot spots are colored in three shades of red by significant levels calculated with the Getis-Ord Gi* analysis. Cold spots are similarly colored in shades of blue. Boundaries of four major cities in the area are placed on the map for geographic references.

Figure 1.

Hot and cold spots identified by total number of cervical cancer cases (2010–2019) for 1,014 census tracts in south Florida. Figure shows the locations of hot and cold spots by census tracts, identified with number of cervical cancer cases diagnosed from 2010 to 2019, in south Florida, which consists of four counties. The total number of census tracts included in this analysis is 1,014. Hot spots are colored in three shades of red by significant levels calculated with the Getis-Ord Gi* analysis. Cold spots are similarly colored in shades of blue. Boundaries of four major cities in the area are placed on the map for geographic references.

Close modal
Figure 2.

Percent population in poverty (2015) for 1,014 census tracts in south Florida. Figure shows the locations of poverty areas by census tracts, identified with 2015 American Community Survey data, in south Florida, which consists of four counties. The total number of census tracts shown on the map is 1,014. Levels of poverty by census tracts are differentiated by three shades of green. Census tracts in the darkest shade of green are in extreme poverty (i.e., more than 40% of the population meet federal poverty criteria), followed in order by poverty (i.e., more than 20% population meet the criteria), and non-poverty (i.e., less than 20% population meet the criteria). Boundaries of four major cities in the area are placed on the map for geographic references.

Figure 2.

Percent population in poverty (2015) for 1,014 census tracts in south Florida. Figure shows the locations of poverty areas by census tracts, identified with 2015 American Community Survey data, in south Florida, which consists of four counties. The total number of census tracts shown on the map is 1,014. Levels of poverty by census tracts are differentiated by three shades of green. Census tracts in the darkest shade of green are in extreme poverty (i.e., more than 40% of the population meet federal poverty criteria), followed in order by poverty (i.e., more than 20% population meet the criteria), and non-poverty (i.e., less than 20% population meet the criteria). Boundaries of four major cities in the area are placed on the map for geographic references.

Close modal

Table 2 summarizes SDOH variables associated with census tracts identified as hot, cold, and not significant in Getis-Ord Gi* analysis. A census tract whose Gi* value falls between the criteria for 90% significant hot spot and 90% significant cold spot is identified as not significant. Table 2 shows that the values of average numbers of cervical cancer cases per census tract decrease in the order: hot spots, not significant, and cold spots. Hispanics are the majority group associated with hot spots, while non-Hispanic whites represent the majority for cold spots. Except for the percentage of workers with no vehicle, SDOH resources are lowest for hot spots, followed in order by not significant and cold spots. It is noted that many cervical cancer cold spots in SCA are in areas with many retirement communities. As a result, the cold spots (i.e., with the smallest number of workers for percentage calculation) show the highest percentage of workers with no vehicles.

Table 2.

SDOH variables associated with census tracts identified as hot spots, cold spots, and not significant.

Cervical cancer and SDOH variablesCold spotsNot significantaHot spots
Total census tracts 72 765 177 
Total cervical cancer cases 146 2,249 761 
Average cervical cancer cases per tract 2.03 2.94 4.30 
Percent Hispanic cases 21% 39% 48% 
Percent non-Hispanic White cases 58% 35% 14% 
Percent non-Hispanic Black cases 14% 23% 37% 
Average median household income $59,000 $55,000 $42,000 
Average percent bachelor's degree or higher 42% 30% 18% 
Average percent renter occupied 38% 40% 44% 
Average percent workers with no vehicle 10% 4% 5% 
Average percent with no access to health insurance 20% 27% 33% 
Average percent women unscreened 18% 19% 21% 
Average number of women unscreened per tract 198 307 360 
Cervical cancer and SDOH variablesCold spotsNot significantaHot spots
Total census tracts 72 765 177 
Total cervical cancer cases 146 2,249 761 
Average cervical cancer cases per tract 2.03 2.94 4.30 
Percent Hispanic cases 21% 39% 48% 
Percent non-Hispanic White cases 58% 35% 14% 
Percent non-Hispanic Black cases 14% 23% 37% 
Average median household income $59,000 $55,000 $42,000 
Average percent bachelor's degree or higher 42% 30% 18% 
Average percent renter occupied 38% 40% 44% 
Average percent workers with no vehicle 10% 4% 5% 
Average percent with no access to health insurance 20% 27% 33% 
Average percent women unscreened 18% 19% 21% 
Average number of women unscreened per tract 198 307 360 

aNot significant: A census tract whose Gi* value falls between the criteria for 90% significant hot spot and 90% significant cold spot.

Spatial distribution of women unscreened for cervical cancer

To demonstrate the importance of identifying the actual number of unscreened, at-risk women for targeted intervention at the census tract level, Fig. 3 shows a side-by-side comparison of choropleth maps for percent unscreened versus number of women unscreened by census tracts. In Fig. 3A, all 1,014 census tracts were divided into five quantiles, ranked by the percentage of women unscreened. In Fig. 3B, the same quantiles are ranked by number of women unscreened. For visual assessment of the associations between cervical screening and cervical cancer risk (i.e., measured by number of cases at each census tract), we overlaid census tract centroids, each of which is represented as a yellow circle at the center of a census tract, on the choropleth maps of percent women unscreened and number of women unscreened. The size of a yellow circle varies by the number of total cervical cancer cases (2010–2019) at the census tract. On each choropleth map, for a census tract with higher number of cases (i.e., a larger yellow circle), we expect to see the polygon underneath in darker shades (i.e., more women unscreened).

Figure 3.

Comparison of spatial distributions of census tracts differentiated by percent women unscreened (A) and number of women unscreened (B) in Miami-Dade and Broward counties, Florida. Figure shows a side-by-side comparison of choropleth maps for percent unscreened versus number of women unscreened by census tracts. A, All 1,014 census tracts were divided into five quantiles, ranked by percent women unscreened. B, The same quantiles are ranked by number of women unscreened. Census tract centroids, each of which is represented as a yellow circle at the center of a census tract, were overlaid on the choropleth maps of percent women unscreened and number of women unscreened. The size of a yellow circle varies by the number of total cervical cancer cases (2010–2019) at the census tract.

Figure 3.

Comparison of spatial distributions of census tracts differentiated by percent women unscreened (A) and number of women unscreened (B) in Miami-Dade and Broward counties, Florida. Figure shows a side-by-side comparison of choropleth maps for percent unscreened versus number of women unscreened by census tracts. A, All 1,014 census tracts were divided into five quantiles, ranked by percent women unscreened. B, The same quantiles are ranked by number of women unscreened. Census tract centroids, each of which is represented as a yellow circle at the center of a census tract, were overlaid on the choropleth maps of percent women unscreened and number of women unscreened. The size of a yellow circle varies by the number of total cervical cancer cases (2010–2019) at the census tract.

Close modal

By comparing the two maps in Fig. 3, we can see that census tracts differentiated by the number of women unscreened show more variation in shades, which appear to correlate with the sizes of corresponding yellow circles better than those in the percent unscreened map. For example, some census tracts in the southwest quadrant of Broward County have high numbers of cervical cancer cases, but the percentages of women unscreened here are mostly in the lowest quantile. However, the association between cervical cancer cases and lack of screening can be better seen when census tracts are shaded by number of women unscreened. An analysis of statistical correlation supports this visual assessment that number of cervical cancer cases correlate better with number of women unscreened than with percent unscreened. For all census tracts in the catchment area, Spearman correlation coefficient between number of cases and percent women unscreened is 0.304, which is lower than the coefficient for number of women unscreened at 0.466.

Targeted intervention for cervical cancer control with the GC vehicles

Before we began analyzing cervical cancer hot spots with FCDS data, community outreach activities by the GC vehicles were often conducted at disadvantaged neighborhoods by requests from community leaders. To assess the spatial extent covered by past GC activities as well as to plan for future outreach events serving the identified hot spots, we geocoded the locations of GC events from 2019 to 2021 (i.e., 600 events identified) and overlaid these locations on the hot spots map shown in Fig. 4, in which GC locations are represented as small yellow circles and cervical cancer hot spots with 95% confidence level are highlighted in red, overlaid on the choropleth map differentiated by number of women unscreened. Figure 4 shows that from 2019 to 2021 the GC vehicles had frequented locations covering the two hot spot clusters as well as some census tracts with high number of unscreened women in Miami-Dade County. However, with only 16 events conducted in Broward County, GC vehicles did not cover either the hot spots or census tracts with high number of unscreened women within the county. Census tracts with high number of cervical cancer cases and/or unscreened women in Palm Beach County were also not covered adequately.

Figure 4.

Outreach locations (n = 600, 2019–2021) by the GC vehicles, cervical cancer hot spots (2010–2019), and number of women unscreened (2020) within the catchment area of Sylvester comprehensive cancer center. Figure shows geocoded event locations of GC vehicles from 2019 to 2021 (i.e., 600 events identified). Event locations are represented as small yellow circles and cervical cancer hot spots with 95% confidence level are highlighted in red, overlaid on the choropleth map of census tracts differentiated by number of women unscreened.

Figure 4.

Outreach locations (n = 600, 2019–2021) by the GC vehicles, cervical cancer hot spots (2010–2019), and number of women unscreened (2020) within the catchment area of Sylvester comprehensive cancer center. Figure shows geocoded event locations of GC vehicles from 2019 to 2021 (i.e., 600 events identified). Event locations are represented as small yellow circles and cervical cancer hot spots with 95% confidence level are highlighted in red, overlaid on the choropleth map of census tracts differentiated by number of women unscreened.

Close modal

Following this assessment, maps such as those shown in Fig. 3 and Fig. 4 as well as sociodemographic data associated with the hot spots were provided to the GC staff to plan (i.e., to determine the locations, target population segments, staffing, priorities, and frequencies) for future events addressing inadequate outreach (e.g., less than one event per 5 square miles per year) in hot spots as well as in not-significant census tracts with high number of cases (e.g., more than one case per year) and/or high number of unscreened women (e.g., over 500 women unscreened). A new electronic event log has also been developed to track details of each event, such as the location, duration, types of activities, and number of at-risk people screened and/or consulted. We will use the log data to quantify service metrics such as the average number of screenings and consultations per event. These metrics will help the GC staff better determine the optimal number and schedule for future outreach events to sufficiently cover the identified intervention targets. In the future, reassessment of GC coverage and planning for new events will occur periodically when new cervical cancer cases and screening data become available for updating the locations of hot spots and/or census tracts with high number of women unscreened.

By combining geospatial data from different sources, we identified the locations and specific population segments experiencing cervical cancer disparities for targeted intervention with MSUs. We were able to identify cervical cancer hot spots in neighborhoods experiencing high levels of socioeconomic disparities by analyzing hot spots with the number of cases rather than incidences, because cancer incidences can underestimate the need for intervention in poor neighborhoods with high numbers of cases and large populations. We also showed that neighborhoods with a higher number of cervical cancer cases are associated with lower SDOH resources, including lower income level, housing tenure, and education attainment. These findings are consistent with previous studies that revealed positive associations between cervical cancer hot spots and socioeconomic disparities at varying geographic scales (10, 11, 12). Our analysis of the numbers of women unscreened for cervical cancer revealed that neighborhoods with a satisfactory screening rate can still have a large number of at-risk women unscreened because of a dominant presence of women ages below 65. These findings exemplify a potential fallacy that can occur when the percentage of women unscreened is used as the sole source of information for determining community targets for intervention. As the size and age distribution of a population can vary greatly across different neighborhoods, ignoring communities that meet a set screening percentage threshold without considering the actual number unscreened can potentially leave many high-risk women unscreened.

With the application described in this article, we demonstrated the utility and potential of a surveillance system that combines cancer registry and SDOH data to address disparities associated with cervical cancer control by a regional cancer center. In addition to cervical cancer, timely screening is also critical for controlling other types of cancer such as colorectal, breast, and lung cancer. Our approach with cervical cancer can be applied to these cancers by incorporating additional data based on respective risk factors and prevention guidelines. However, we do note that both the utility and limitations of our approach hinge critically on the availability and quality of data. For example, we were not able to incorporate HPV vaccination in a hot spot analysis because data for HPV vaccination rates are only available at the county level. The percentage of women who had hysterectomies at regional and local scales is also not available. With the continuous development and field applications of eHealth and mHealth systems, it is expected that issues with data availability will lessen as more geocoded data become available for cancer control applications. With this study, we demonstrated that a geospatially referenced cancer surveillance platform integrating cancer registry, SDOH, and other available risk factor data can be an important asset for a cancer center to effectively allocate limited resources for cervical cancer control by identifying communities and population segments experiencing cancer disparities.

E.N. Kobetz reports a patent for SCAN360 pending. No disclosures were reported by the other authors.

The content of this work is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or agencies that published data used for this study.

M.S. Lee: Conceptualization, formal analysis, writing–original draft, writing–review and editing. N.S. Elliott: Conceptualization, writing–review and editing. V.D. Bethel: Conceptualization, writing–review and editing. R.R. Balise: Conceptualization, writing–review and editing. E.N. Kobetz: Conceptualization, writing–review and editing.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

1.
Center for Disease Control and Prevention
.
Cervical cancer statistics
;
2022
.
Available from
: https://www.cdc.gov/cancer/cervical/statistics/index.htm.
2.
Center for Disease Control and Prevention
.
Cancers associated with human papillomavirus (HPV)
;
2022
.
Available from
: https://www.cdc.gov/cancer/hpv/basic_info/index.htm.
3.
United States Food and Drug Administration
.
Gardasil
;
2023
.
Available from
: https://www.fda.gov/vaccines-blood-biologics/vaccines/gardasil.
4.
Center for Disease Control and Prevention
.
HPV vaccination recommendations
;
2021
.
Available from
: https://www.cdc.gov/vaccines/vpd/hpv/hcp/recommendations.html.
5.
US Preventive Services Task Force
.
Cervical cancer: screening
;
2018
.
Available from
: https://www.uspreventiveservicestaskforce.org/uspstf/recommendation/cervical-cancer-screening.
6.
Fuzzell
LN
,
Perkins
RB
,
Christy
SM
,
Lake
PW
,
Vadaparampil
ST
.
Cervical cancer screening in the United States: challenges and potential solutions for underscreened groups
.
Prev Med
2021
;
144
:
106400
.
7.
Greenwald
ZR
,
El-Zein
M
,
Bouten
S
,
Ensha
H
,
Vazquez
FL
,
Franco
EL
.
Mobile screening units for the early detection of cancer: a systematic review
.
Cancer Epidemiol Biomarkers Prev
2017
;
26
:
1679
94
.
8.
Brotherton
JML
,
Wheeler
C
,
Clifford
GM
,
Elfström
M
,
Saville
M
,
Kaldor
J
, et al
.
Surveillance systems for monitoring cervical cancer elimination efforts: focus on HPV infection, cervical dysplasia, cervical screening and treatment
.
Prev Med
2021
;
144
:
106293
.
9.
Getis
A
,
Ord
JK
.
The analysis of spatial association by use of distance statistics
.
Geogr Anal
1992
;
24
:
189
206
.
10.
Rutherford
Y
,
Mobley
L
.
Examining spatial clusters of high & low proportions of late stage cervical cancer in the U.S.: a look at geographic disparities & associated risk factors
.
Ann Cancer Epidemiol
2020
;
4
:
5
.
11.
Saghari
S
,
Samuel
S
,
Ghamsary
M
,
Edirlei
S
,
Marie-Mitchell
A
,
Morgan
JW
.
Geographic distribution of cervical cancer in California: a population based study
.
JSM Women's Health
2016
;
1
:
1001
.
12.
Torres
AZ
,
Phelan-Emrick
D
,
Castillo-Salgado
C
.
Evaluating neighborhood correlates and geospatial distribution of breast, cervical, and colorectal cancer incidence
.
Front Oncol
2018
;
8
:
471
.
13.
Goding Sauer
A
,
Bandi
P
,
Saslow
D
,
Islami
F
,
Jemal
A
,
Fedewa
SA
.
Geographic and sociodemographic differences in cervical cancer screening modalities
.
Prev Med
2020
;
133
:
106014
.
14.
Bauer
C
,
Zhang
K
,
Xiao
Q
,
Lu
J
,
Hong
YR
,
Suk
R
.
County-level social vulnerability and breast, cervical, and colorectal cancer screening rates in the US, 2018
.
JAMA Netw Open
2022
;
5
:
e2233429
.
15.
Bailey
Z
,
Balise
R
,
Bouzoubaa
L
,
Kobetz
E
.
SCAN360: a resource for a 360-degree view of cancer prevention, risk, and survival
.
Prev Chronic Dis
2020
;
17
:
E149
.
16.
Sylvester Comprehensive Cancer Center. University of Miami Miller School of Medicine
;
SCAN360: game changer vehicles
.
2023
.
Available from
: https://www.scan360.com/OutreachVehicle.
17.
ESRI. ArcGIS Pro
.
How hot spot analysis (Getis-Ord Gi*) works
;
2023
.
Available from
: https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/h-how-hot-spot-analysis-getis-ord-gi-spatial-stati.htm.
18.
ESRI. ArcGIS Pro
.
What is a z-score? What is a p-value?
;
2023
.
Available from
: https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/what-is-a-z-score-what-is-a-p-value.htm.
19.
Moran
PAP
.
Notes on continuous stochastic phenomena
.
Biometrika
1950
;
37
:
17
23
.
20.
Stopka
TJ
,
Goulart
MA
,
Meyers
DJ
,
Hutcheson
M
,
Barton
K
,
Onofrey
S
, et al
.
Identifying and characterizing hepatitis C virus hotspots in Massachusetts: a spatial epidemiological approach
.
BMC Infect Dis
2017
;
17
:
294
.
21.
Stopka
TJ
,
Krawczyk
C
,
Gradziel
P
,
Geraghty
EM
.
Use of spatial epidemiology and hot spot analysis to target women eligible for prenatal women, infants, and children services
.
Am J Public Health
2014
;
104
:
S183
9
.
22.
ESRI. ArcGIS Pro
.
How spatial autocorrelation (Global Moran's I) works
;
2023
.
Available from
: https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-statistics/h-how-spatial-autocorrelation-moran-s-i-spatial-st.htm.
23.
Roux
A
.
Investigating neighborhood and area effects on health
.
Am J Public Health
2001
;
91
:
1783
9
.
24.
Florida Department of Health. FLHealthCharts
.
Cervical cancer incidence
;
2023
.
Available from
: https://www.flhealthcharts.gov/ChartsDashboards/rdPage.aspx?rdReport=NonVitalInd.TenYrsRpt&cid=449.
25.
Adam
EE
,
White
MC
,
Saraiya
M
.
US hysterectomy prevalence by age, race and ethnicity from BRFSS and NHIS: implications for analyses of cervical and uterine cancer rates
.
Cancer Causes Control
2022
;
33
:
161
6
.
26.
National Cancer Institute. The Surveillance, Epidemiology, and End Results Program
.
Standard populations (millions) for age-adjustment
;
2023
.
Available from
: https://seer.cancer.gov/stdpopulations/.
28.
United States Department of Agriculture. Economic Research Services
.
Poverty area measures
;
2023
.
Available from
: https://www.ers.usda.gov/data-products/poverty-area-measures/.