Feasibility of visualizing cancer incidence data at sub-county level: Findings from 21 National Program of Cancer Registries (2024)

Associated Data

Data Availability Statement

Abstract

Monitoring cancer incidence data by geography is useful for planning public health activities. However, due to anticipated confidentiality and statistical reliability issues, data on cancer incidence and mortality are more often displayed at a national, state, or county level, rather than at more local levels. To address this gap in displaying cancer data at the local level, the CDC’s National Environmental Public Health Tracking Program and 21 National Program of Cancer Registries worked together on a pilot project to examine the feasibility of displaying sub-county-level incidence of selected cancer types diagnosed during 2007–2016. The results from this project are important steps for building sub-county cancer displays into data visualizations and using the data in a way that provides meaningful insights. The availability of sub-county cancer data may allow researchers to better examine cancer data at a local level which may help guide public health decisions regarding community-based interventions and screening services.

Keywords: Sub county, Environmental health, Cancer surveillance, Tracking, Cancer registries

1. Introduction

Data on cancer incidence are commonly displayed at the national, state, or county level, but not at more local levels due to anticipated confidentiality and statistical reliability issues. (Centers for Disease Control and Prevention 2020) In recent years, there has been a push to develop timely, locally-relevant health information systems. (DeSalvo et al., 2016) Despite new developments in technology, public health data remain spatially unresolved, lagging, or all together unavailable. (Werner and Strosnider, 2020) Innovation is uniquely challenging in public health, as problems tend to be complex, dynamic, and context specific. (Berney et al., 2015) Some states have used environmental public health tracking to enhance their public portal to benefit programs by mapping data and developing smartphone-friendly versions of the their state’s Web portal. (Jordan et al., 2015; Abookire et al., 2020)

While the benefit of using finer geographic detail to analyze cancer registry data has been discussed previously, more precise geocodes (e.g., to the census tract level) in cancer data have only been routinely used more recently. (Rushton et al., 2006; Howe, 1986) Often geographic data for cancer cases are aggregated to the county or state level before release in public-facing databases. (Yu et al., 2017) Finer resolution data can allow for health issues affecting specific populations to be examined and tailored interventions to be implemented. (Centers for Disease Control and Prevention 2013) The use of sub-county data can also increase public awareness of place-based factors and understanding how these relate to health. (Werner et al., 2018)

To address this gap in displaying cancer data at the local level, two signature programs funded by Centers for Disease Control and Prevention (CDC) worked together by testing aggregation schemes for common and rare cancer types. The National Environmental Public Health Tracking Program (Tracking Program) was established in 2002 to track exposures and health outcomes including selected cancers associated with environmental hazards and to bridge existing data gaps. (McGeehin et al., 2004) State and local Tracking Programs develop or enhance standards as part of the Nationally Consistent Data and Measures (NCDMs) for data calculation, dissemination, and display. A pilot project in 2014 suggested that NCDMs, as well as standardized geographies, be established for generating, analyzing, and sharing sub-county data to allow for comparison of different health outcomes over space and time. (Werner et al., 2018) Collaboration between the Tracking Program and jurisdictions who work with the Tracking Program is key to advancing efforts and making sub-county data available on a larger scale. (McGeehin et al., 2004)

The National Program of Cancer Registries (NPCR) collects data on cancer occurrence, the type of initial treatment, and outcomes for 46 states, DC and 3 territorial cancer registries representing 97% of the United States population. NPCR measures progress in preventing and treating cancer in the United States. (National Program of Cancer Registries 2020) CDC displays cancer data at the national, state, county, and Congressional District levels. (Centers for Disease Control and Prevention 2020)

In this project, the Tracking Program and 21 NPCR registries examined the feasibility of displaying cancer incidence at sub-county levels. Using data for selected cancer types diagnosed during 2007–2016, stability and confidentiality issues were examined to identify which spatiotemporal aggregation minimized suppression and instability while allowing for display of finer spatial resolution data at each sub-county spatial resolution. The participants tested different spatial and temporal aggregations for displaying lung cancer, breast cancer (females only), prostate cancer, colorectal cancer, liver cancer, melanoma of the skin, and non-Hodgkin lymphoma. The aim of this project was to suggest spatiotemporal aggregations for sub-county cancer data display and dissemination.

2. Methods

3. Results

Average median, minimum, and maximum case counts by cancer type for each aggregation level were compiled for all regions (Table 2). The median annual case count was not large enough to display any cancer type at the census tract level without any temporal aggregation. States decided on the below suggested spatiotemporal aggregations by discussing observed output during regional meetings. The best outcome for all states were selected and agreed upon during the final meetings. When deciding on the final display recommendations, states considered percent of geographies that were suppressed, percent of geographies with unstable rates, as well as confidentiality (suppressing case counts less than 16).

Suggested spatiotemporal aggregations by cancer type are displayed in Table 3. The average annual number of cases and incidence rate for lung cancer (overall and stratified by sex) at the 5000 person aggregation for a 5-year period and at the 20,000 person aggregation for a 5-year period were found feasible to display. Because of regional differences in lung cancer incidence, the recommendations for spatiotemporal aggregations varied by workgroup.

States determined that it was feasible to display the average annual number of cases and incidence rate for female breast cancer at the census tract level for a 10-year period, at the 5000 person aggregation for a 5-year period, and at the 20,000 person aggregation for a 3-year period (Table 3). The average annual number of cases and incidence rate for non-Hodgkin lymphoma was found feasible to display at the 20,000 person aggregation for a 5-year period. The average annual number of cases and incidence rate for colorectal cancer was found feasible to display at the 5000 person aggregation for a 5-year period and at the 20,000 person aggregation for a 3-year period. The average annual number of cases and incidence rate for melanoma of the skin was found feasible to display at the 5000 person aggregation for a 5-year period and at the 20,000 person aggregation for a 3-year period. The average annual number of cases and incidence rate for prostate cancer was feasible to display at the census tract level for a 10-year period, at the 5000 person aggregation for a 5-year period, and at the 20,000 person aggregation for a 3-year period. It was not feasible to display liver and intrahepatic bile duct cancer at census tract or any other aggregation level, even using a 10-year period because of the small number of cases.

An example of maps produced by each pilot state are shown in Figures. Fig. 1 displays a Minnesota state map for lung cancer incidence rates stratified by sex at the 20,000 person aggregation for a 3-year period (2014–2016). Fig. 2 displays a North Carolina map for colorectal cancer incidence rates, using 5000 person aggregation over a 5-year period (2012–2016). Fig. 3 displays Missouri maps for breast cancer incidence rates, using census tract aggregation over a 10-year period (2007–2016).

4. Discussion

The CDC Tracking Program and NPCR worked together on this pilot project to gain insight into displaying cancer data at the sub-county level by testing aggregation schemes for common and rare cancer types. We found suggested spatiotemporal aggregation combinations varied by each cancer type examined. The results from this project are important steps for building sub-county cancer displays into data visualization portals and using the data in a way that provides meaningful analyses leading to public health actions. Easy-to-use public portals widen the potential audience for the results of a spatial analysis of health data to go beyond scientists to include the public, policymakers, the media, and a host of others. (Bell et al., 2006)

Stability, reliability, and confidentiality were identified as issues when determining the best spatial and temporal aggregations to use. In 2017, the national incidence of cancer types examined in this project varied, with female breast (125.1 per 100,000 female persons), prostate (106.5 per 100,000 male persons), lung (55.2 per 100,000 persons), and colorectal (36.8 per 100,000 persons) cancers being more common than melanoma of the skin (22.7 per 100,000 persons), non-Hodgkin lymphoma (18.5 per 100,000 persons), and liver cancer (8.3 per 100,000 persons). (U.S. Cancer Statistics Working Group 2020) For rarer cancer types, obtaining sufficient numerators was difficult, due to having fewer cases even when aggregating, when using small areas. ¹¹ In determining which suggested spatial temporal aggregation combinations worked best for each cancer type, it was important to balance data stability and confidentiality issues while keeping geographic areas at the highest possible resolution and time periods as recent as possible. (Werner et al., 2018)

In this project, census tracts were the smallest area units used and we found that this spatial resolution required greater temporal aggregation. It is suggested that analyses use smallest area units with available data, and aggregation to larger units should be avoided unless there is appropriate rationale. (Pfeiffer et al., 2008) To allow for data users to have flexibility (i.e., finer spatial resolution with more temporal aggregation or coarser spatial resolution with less temporal aggregation), various sub-county aggregations were tested and included in the final recommendations.

Previous studies have examined the geographic variation in cancer incidence, mortality, treatment, and survival by mapping cancer data using geocodes and Geographic Information Systems (GIS). (Rushton et al.,; Sahar et al., 2019) One study found combining neighborhood measures and liver cancer rates can narrow down census tracts to target for liver cancer prevention more than standard approaches that use just demographic data (race/ethnicity and age) from the U.S. Census. (Lynch et al., 2020) Complete geocoded data on census tracts from all registries is desirable for national-level research related to spatial analyses. (Lynch et al., 2020) For diagnosis years 2001–2005 combined, valid information on census tract was submitted for 59% of the records in the NPCR dataset. (Singh et al., 2009) Further analysis of residential history may be helpful to access how residential history may impact the ability to get an accurate geocoded addresses.

The Tracking Program and NPCR is working on a second phase of this pilot project to further refine the work presented in this pilot project. This will include further discussion on counties not aggregated in this pilot project due to small populations, creating a 50,000 person minimum aggregation level to test rarer outcomes, including age-adjusted rates in analyses, creating appropriate cancer communication and messaging for the Tracking Network, including additional cancer types in analyses, building capacity for GIS and refining Tracking Program geocoding standards, and mapping additional interventions such as exposures and health outcomes and stratification by sex.

Findings of this study are subject to at least three limitations. One, age-adjusted rates were not tested in this phase of the pilot project. Second, a limited number of states (21) participated in phase one of the pilot project. Although we believe states that participated in the pilot project provided a good representation and diversity of NPCR registries, they are only a sample of the population. Additionally, sub-county areas are standardized within the Tracking Network but may not work for all purposes. The standardized sub-county areas are useful for providing nationally consistent data and measures to allow us to compare over time, space, and datasets, but some states may not be able to access or display sub-county data due to legislative reasons or have state-specific needs and may want or need to incorporate their own sub-county geographies. For example, areas with a minimum of 50,000 persons have been developed for 14 states for the purpose of disseminating cancer surveillance data. (Tatalovich et al., 2022)

The methods and results of this pilot project demonstrate how spatial and temporal aggregation can be used for cancer data at the sub-county level, which can be used for surveillance purposes beyond the Tracking Program in the United States to achieve stable estimates whilst considering confidentiality issues. (McGeehin et al., 2004) The availability of sub-county cancer data through the Tracking Network may allow public health researchers, policymakers, and the public to better examine cancer data at a local level, which may help guide decisions regarding community-based interventions and screening services. Preliminary data can be explored on CDC’s National Environmental Public Health Network at (https://ephtracking.cdc.gov/DataExplorer/). Under the select data section, individuals can select the cancer type they would like to review and the corresponding measures (age-adjusted rates and number of cases at the national and state level.

Acknowledgements

Footnotes

Data availability

Please contact corresponding author for data request. Crosswalks and other sub-county materials can be found at https://github.com/CDCgov/EPHTracking-Subcounty.

References