Utility and Generalizability of Multistate, Population-Based Cancer Registry Data for Rural Cancer Surveillance Research in the United States

The National Cancer Institute characterizes cancer health disparities as adverse differences in cancer burden along the cancer control continuum from primary prevention to incidence to survivorship among specific population groups, including those defined by their geographic location (e.g., rural vs. urban; ref. 1). More than 46 million residents live in rural areas, but cancer disparities among rural populations are still relatively understudied (2). However, there is an increasing emphasis on the need to address cancer control in rural and other small populations through appropriate research methods, evidence-based public health and clinical interventions, and community engagement (3–6). Furthermore, recent descriptive analyses of population-based, cancer registry data have illuminated the cancer incidence disparities experienced by rural populations in the United States—highlighted in our work in this Rural Cancer Control CEBP Focus issue and other recent publications from Center for Disease Control and Prevention (CDC) and National Cancer Institute (NCI) researchers (6–8). Each of these analyses utilized different sources of multistate, population-based cancer registry data with varying extents of geographic coverage of the country. Our paper utilized the North American Association of Central Cancer Registries (NAACCR) public use dataset, which covers 93% of the United States (7). CDC researchers utilized data from the National Program of Cancer Registries (NPCR) and Surveillance Epidemiology and End Results (SEER) program (97% coverage), and NCI researchers utilized data from the SEER 18 program (28% coverage; refs. 6, 8). Interestingly, the NCI analysis of SEER 18 data indicated that the all cancers combined–incidence rate was higher in rural populations than in urban, while the CDC's analysis of NPCR/SEER data and our analysis of NAACCR data found that the all cancers combined–incidence rate was higher in urban populations.

Cancer surveillance data are useful for describing differences in incidence rates and trends among populations (e.g., rural and urban), but such data can also be used to assess cancer mortality, survival, risk factors, and outcomes and to help guide interventions and public policy (9). The strength of cancer registry systems is that they collect population-based data on individual cancer cases using consistent data standards that enhance the validity and generalizability of studies utilizing such data (10). Each central cancer registry is required to collect data on demographics, cancer identification, staging, treatment, follow-up, and death (11). However, each central cancer registry differs on what variables they may collect beyond required data, what processes researchers must utilize to obtain data, and what specific data elements may be available to outside researchers (12–13). While researchers may be interested in studying populations within a single state, NAACCR, NCI, and CDC all have avenues by which researchers can obtain multistate data for cancer surveillance research. Multistate data may be particularly meaningful for evaluating rural or other small populations nationally, assessing policy-relevant geographic contexts (e.g., the Delta Regional Authority or Appalachian Regional Commission), or exploring other important multistate categorizations (e.g., Medicaid expansion vs. non-Medicaid expansion states; refs. 14–16).

In light of the recent studies utilizing multistate cancer registry data to describe the rural cancer burden, this article aims to evaluate the utility and generalizability of multistate, population-based cancer registry datasets for rural cancer surveillance purposes.

To achieve our goal, our approach was two-fold. First, we described available multistate, population-based cancer datasets and their strengths and limitations for rural cancer surveillance research. Second, we explored two readily available population-based cancer registry datasets to assess their generalizability to rural populations.

Information on all incident cases of cancers in the United States are collected through either the CDC-funded NPCR or NCI-funded SEER registry programs (11). Together NPCR and SEER provide data on 100% of cancer cases diagnosed in the United States. There are mechanisms to request access to multistate data like the NPCR and SEER through the CDC and NCI, respectively. There are also mechanisms to access multistate, population-based cancer registry data through NAACCR, an umbrella organization for central cancer registries. NAACCR develops and promotes cancer registry data standards, provides training and certification, and publishes and promotes the use of cancer registry data for cancer surveillance research and cancer control (17). We evaluated the accessibility, geographic coverage and scale, extent of available variables, and strengths and weaknesses of five datasets available through NAACCR, CDC, or NCI, which are the only conduits to obtain multistate cancer registry data. These datasets included: the NAACCR public use dataset, the NAACCR Cancer in North America (CiNA) Deluxe File, SEER, NPCR public use dataset, and the United States Cancer Statistics (USCS) restricted access dataset (18–22).

The NAACCR public use dataset, which we analyzed in our published study in this CEBP Focus issue, covers 93% of the U.S. population, whereas SEER 18 covers 28% of the U.S. population across 18 different registries (7, 18, 20). There are three datasets available with the SEER program: SEER 9, SEER 13, and SEER 18. Each of these datasets includes an increasing number of registries and different years of data based upon the inclusion of each registry in the SEER program. To assess the generalizability of each dataset to rural populations, and to assess rural–urban differences in cancer incidence, we calculated age-adjusted cancer incidence rates and corresponding 95% confidence intervals for rural and urban populations in SEER 18 and the NAACCR public use dataset for cancers diagnosed between 2009 and 2013. Rate ratios with P-values less than 0.05 indicated statistically significant rural–urban differences in rates. Corresponding age-adjusted incidence rates and confidence intervals are reported from our study utilizing NAACCR's public use dataset. All analyses were done using SEER*Stat 8.3.4. We utilized these two datasets because, of the five datasets we reviewed, the NAACCR public use dataset and SEER 18 dataset are the only ones that are publicly available, include a rural–urban variable, and are able to be analyzed remotely under a data use agreement without additional approvals from the data-managing entity. To categorize rural and urban, we used the United States Department of Agriculture's Rural Urban Continuum Codes (RUCC), which classify counties by population size and proximity to a metropolitan area (2). We used a dichotomized RUCC designation (rural RUCC = 4–9; urban RUCC = 1–3), as this dichotomized designation was the only available rural–urban measure in the NAACCR dataset. We also performed this analysis stratified by race/ethnicity for non-Hispanic whites, non-Hispanic blacks, and Hispanics to evaluate the generalizability to the growing racial/ethnic diversity in rural populations (23). To further assess the generalizability of SEER 18 and the NAACCR public use datasets, we used 2009–2013 American Community Survey population overall estimates and stratified by race, ethnicity, poverty, and median household income to characterize rural populations covered by counties represented in SEER 18 and the NAACCR public use dataset and the United States population as a whole (24).

Table 1 summarizes the accessibility, geographic scale and coverage, variable availability, strengths, and weaknesses of several multistate, population-based cancer registry sources. With the exception of the USCS restricted access dataset, all datasets are remotely accessible through SEER*Stat software. However, only NAACCR CiNA Deluxe and SEER data are accessible through the case listing session for extraction of individual-level data to be analyzed in other software programs. Both the NAACCR CiNA Deluxe datasets and USCS restricted use datasets are only available for specific research projects after appropriate approvals from their respective entities. The geographic coverage varied from 28% of the U.S. population in SEER 18 to a potential of 100% coverage in the USCS restricted access dataset. While all data reported to central cancer registries are required to be geocoded to the census tract, data available through these mechanisms are only available most precisely at the county level. Such geographic specification is necessary for rural–urban characterization of cancer data (e.g., utilizing USDA rural–urban metrics), but each central cancer registry may restrict/limit the release of geographic information. The NAACCR public use dataset includes a dichotomous rural–urban variable, but identification of specific county is not available. The NPCR public use dataset does not include any substate geographic indicators, severely limiting the characterization of rural cancer burden. All datasets have data on tumor characteristics, but only the NAACCR CiNA Deluxe and SEER datasets have any data available on treatment or outcomes.

Table 2 displays the rural–urban age-adjusted cancer incidence rates utilizing 2009–2013 data from both the NAACCR public use dataset and SEER 18. The direction and magnitude of rural–urban all cancer incidence differences varied between the two data sources. The NAACCR dataset identified a slightly higher incidence rate in urban compared to rural populations for all cancers combined (448.7 vs. 446.4 per 100,000, respectively). SEER 18, on the other hand, identified a higher incidence rate in rural compared to urban populations for all cancers combined (461.8 vs. 449.1 per 100,000, respectively). Among 20 of the 23 individual cancers that we evaluated, both NAACCR and SEER 18 data showed either no rural–urban differences in rates or differences that were consistent in direction. For pancreas cancer, NAACCR identified a higher rate in urban populations, whereas there was no rural–urban difference in SEER 18. For melanoma, NAACCR identified a slightly higher rate in urban populations whereas SEER 18 analysis indicated a slightly higher rate in rural populations. For myeloma, NAACCR data indicated a higher rate in urban populations while SEER 18 data indicated no rural–urban difference in rates.

Table 3 displays the race-ethnicity specific age-adjusted cancer incidence rates for rural and urban populations utilizing 2009 to 2013 data from both the NAACCR public use dataset and SEER 18. Both the NAACCR and SEER 18 analyses indicated elevated overall rates in urban areas among non-Hispanic whites, non-Hispanics blacks, and Hispanics. For non-Hispanic whites, both NAACCR and SEER 18 indicated the rural–urban differences across all cancer types. In general, the age-adjusted incidence rates in the analysis of SEER 18 data were higher than the analysis of NAACCR data for both rural and urban non-Hispanic whites. For non-Hispanic blacks, the NAACCR analysis identified two small rural–urban differences (Hodgkin lymphoma and myeloma) that were not identified in the SEER 18 analysis. Rural–urban differences in kidney and urinary bladder cancers were identified amongst non-Hispanic blacks in the SEER 18 analysis.

Table 4 displays the sociodemographic representation of SEER 18 and the NAACCR public use datasets compared to the nation as a whole. Rural counties represent 10.6% and 14.6% of the SEER 18 and NAACCR populations, respectively, while nationally, 14.8% of the population is rural. Rural counties represented in SEER have higher proportions of black, Asian, and Hispanic populations compared with the NAACCR population and the national populations (e.g., SEER 18 Hispanic populations were 9.1% compared to 7.8% in NAACCR and 7.7% nationwide). SEER 18 rural counties had a slightly higher percentage of the population living in poverty (19.7%) compared with the counties represented in NAACCR (18.3%) and the U.S. rural population (18.1%). The regional distribution of rural populations in SEER 18 are notably different from the U.S. population overall, as the Northeast and Midwest are considerably underrepresented whereas the South and West are overrepresented.

This commentary provided an overview of cancer surveillance datasets that can be used for rural cancer surveillance research and described the generalizability of two datasets—the NAACCR public use dataset and SEER 18— with varying geographic coverage (93% and 28% representation of the U.S. population, respectively). Datasets also varied by opportunity to perform analyses beyond rates and trends and potential to explore cancer outcomes beyond incidence/tumor characteristics. Age-adjusted incidence rates were largely similar among both NAACCR and SEER 18 analyses for individual cancer types, but they differed in all cancer combined analyses. The NAACCR analysis indicated a slightly elevated rate in urban populations, while the SEER 18 analysis indicated a higher rate among rural populations. However, when the SEER 18 analysis was stratified by race/ethnicity higher rates in urban populations were identified across all groups. Furthermore, the geography of SEER 18 registries indicated underrepresentation of rural populations.

Data accessible through the CDC have greater geographic coverage (96–100% of the United States) than other sources. A recent study by Kuo and colleagues utilized the USCS restricted use dataset to evaluate its generalizability to the U.S. population and recommends that researchers take greater advantage of this resource because of its more complete geographic coverage compared to SEER 18 (25). Indeed, both the USCS restricted use dataset and the NPCR public use dataset have great geographic coverage, but only have variables relevant to tumor characterization (e.g., site, stage). The restricted use dataset, which has the potential to characterize rural areas, is only available for analysis at research data centers, which may be a barrier to some researchers. In addition, the NPCR public use dataset while remotely accessible, does not include any characterization of rural and urban populations, as there is no geographic indicator available more granular than state. Furthermore, this dataset only allows for analysis of incidence rates and trends.

The CDC has made addressing rural disparities an increasing priority, as highlighted by a series of surveillance reports on areas such as cancer disparities and health behaviors (26). Because of this increasing priority, we offer two recommendations. First, we suggest that data on treatment and outcomes be integrated into the USCS restricted use dataset to allow for quantification of rural–urban disparities/differences at points along the cancer control continuum (e.g., treatment, survival) to be explored using a generalizable, population-based data source. Second, we suggest that the public use dataset incorporate a rural–urban variable(s), such as a USDA rural–urban measure, much like that available in the NAACCR public use dataset, to allow researchers to utilize this generalizable data source to describe differences/disparities in cancer rates and trends. This would allay potential confidentiality and privacy concerns when working with counties with small numbers/populations while providing researchers the opportunity to evaluate rural–urban differences in cancer incidence.

Although data accessible from the CDC is more representative of rural populations, SEER underrepresents some rural populations in its geographic coverage. The distribution of rates by rural-urban designation in our study using NAACCR data corroborated those published by the CDC using similarly representative data, which indicated nominally higher overall rates in urban areas, whereas SEER indicated higher overall rates in rural areas (7, 8). Further, the sociodemographic characteristics of counties represented in SEER 18 differ from the U.S. rural population as a whole. This corroborates previous studies which indicated that geographic areas covered by SEER were more urban than non-SEER areas of the country, one of which also showed that SEER's rural representation has become less generalizable to the nation over time (25, 27). A strength of SEER is that there is purposeful inclusion of populations who have been historically underrepresented, particularly racial and ethnic minority populations. Indeed, over the years the coverage of SEER has expanded to include registries like those incorporating rural Georgia, an area with a high proportion of rural black residents. However, as a whole, rural populations are not well-represented in SEER. Like racial and ethnic minorities, rural populations have been historical underrepresented and understudied in cancer research. Two priorities in NCI's Division of Cancer Control and Population Sciences include both rural cancer control and enhancing SEER for research use (28). Further, one of the goals of the SEER program is to identify changes and differences in cancer incidence among geographically and otherwise sociodemographically defined population subgroups (29). To simultaneously address these priority areas and goals, the SEER program could improve its rural coverage by inclusion of additional registries to enrich the data for cancer disparities research purposes. This is especially important because SEER data are the most readily accessible population-based dataset to explore cancer treatment and survival. Additionally, other datasets are available that link SEER data to Medicare, Medicare-Health Outcomes Survey, and Medicare-Consumer Assessment of Healthcare Providers (30). Thus, it is a valuable resource for not only cancer surveillance research, but also for health services research that may elucidate rural-urban disparities in cost, utilization of tests and procedures, quality of life, and quality of care. For example, SEER-Medicare data have been used to explore temporal trends in cancer incidence and mortality, treatment, and costs among rural populations or in comparing rural-urban differences (31–33). Expanding SEER's geographic coverage will help researchers address rural health service research questions in addition to rural cancer surveillance research questions.

Aggregated analysis of SEER 18 data revealed elevated incidence rates in rural compared with urban areas. However, when rates were stratified by race/ethnicity, the urban rates were higher across all groups. Because of the solely descriptive nature of our commentary's analysis, this result may be attenuated with more rigorous analyses (34, 35). Further, it is indicative of an imbalance of the regional distribution of rural and urban populations by racial/ethnic groupings within SEER (i.e., a conditional disparity; ref. 36). This phenomenon was not identified with our NAACCR analysis, however, likely due to the fact that its distribution among geographic regions largely mirrors that of the nation. This phenomenon also underscores the importance of considering the interplay of race and rurality on health outcomes especially across multistate regions and the need to expand SEER 18′s geographic coverage (37).

NAACCR's CiNA Deluxe File provides high-quality data on multiple cancer outcomes, including incidence, site specific factors (e.g., breast cancer subtype), treatment, and survival. The potential availability of two rural–urban measures (RUCC and RUCA) and county of residence for cancer cases make it a valuable data source for studying cancer in rural populations, although release of county-specific data requires active consent from each registry. The CiNA Deluxe file has been used to explore cancer disparities utilizing data from as many as forty central cancer registries and to explore disparities within policy-relevant, multistate rural regions like Appalachia (38, 39). In addition, the extensive geographic coverage of the NAACCR public use dataset provides a generalizable data resource for rural cancer surveillance research. Although the dichotomous nature of the rural–urban variable within the public use dataset may limit more nuanced rural characterization, it provides data with broad geographic coverage. There is a tremendous opportunity for researchers to use these data for rural cancer surveillance research.

As rural cancer disparities are increasingly studied, it is important to ensure that population-level surveillance data are generalizable to the rural population and that these data can be utilized to assess multiple aspects of the cancer control continuum from prevention to diagnosis and to survival. We provide recommendations to entities who provide access to population-based data and to researchers who utilize these data to enhance the rural cancer surveillance research. We recommend that the CDC increase the available variables in their two, multistate datasets to enable researchers to perform research that is generalizable to the population and that NCI should expand SEER's geographic coverage to capitalize on the availability of treatment and outcome variables and linkage to Medicare and other datasets. Furthermore, we suggest that researchers take advantage of the USCS restricted use dataset and NAACCR's CiNA deluxe datasets for rural cancer surveillance research purposes.

No potential conflicts of interest were disclosed.

Conception and design: W.E. Zahnd, W.D. Jenkins, A.S. James

Development of methodology: W.E. Zahnd, G.A. Colditz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W.E. Zahnd

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W.E. Zahnd, W.D. Jenkins, A.S. James

Writing, review, and/or revision of the manuscript: W.E. Zahnd, W.D. Jenkins, A.S. James, S.R. Izadi, D.E. Steward, A.J. Fogleman, G.A. Colditz, L. Brard

W.D. Jenkins, A.J. Fogleman, and L. Brard (1P20CA192987–01A1) and A.S. James, S.R. Izadi, and G.A. Colditz (1P20CA192966–01A1) are funded by grants from the National Cancer Institute.

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