Utilizing SEER Cancer Registries for Population-Based Cancer Survivor Epidemiologic Studies: A Feasibility Study

This article is featured in Highlights of This Issue, p. 1697

The current U.S. cancer survivor population of about 17 million is expected to grow to over 20 million by 2026 (1) as a result of advances in the screening, diagnosis, and treatment of cancer. These cancer survivors face an uncertain future with the possibility of a myriad of physical, psychological, financial, and social consequences, some predictable and others unknown. Information on long-term effects of newer therapies is sparse, especially for vulnerable populations under-represented in clinical trials such as the elderly, minorities, young adults, and those with multiple comorbid conditions. Population-based research is needed to characterize the long-term effects of cancer, including the impact of evolving cancer treatments, and to identify strategies to mitigate the adverse effects of cancer and its treatment.

Observational studies provide critical information to address gaps in knowledge concerning the long-term survivor experience. Leveraging the existing resources of central cancer registries can improve the efficiency in the conduct of studies while ensuring adequate representation of diverse populations. While the primary role of cancer registries is to provide vital information for cancer surveillance and control (2), they provide an opportunity to perform population-based observational studies (3). Because cancer is a reportable disease, central cancer registries capture data about persons diagnosed with cancer, including patient demographics, primary tumor site, tumor morphology, stage at diagnosis, first course of treatment, and follow-up for vital status (4). Furthermore, the adoption of e-Path reporting in many cancer registries enables rapid case identification.

Depending on the research questions, participant involvement in a study can range from completing a survey to intensive in-person examinations, donation of biological specimens, and the sharing of personal health information. Since 1973, the NCI's Surveillance, Epidemiology, and End Results (SEER) Program has provided high quality, authoritative cancer incidence and survival data for specific states, regions, and population groups (5). This study was undertaken as a pilot project to assess the feasibility of using SEER registries to rapidly identify, recruit, and enroll individuals for population-based cancer survivor research studies and to assess the extent of their willingness to engage in a variety of potential research activities.

In 2016, six SEER cancer registries were selected among those responding to a request for proposals for a pilot study to determine the feasibility of obtaining patient reported outcomes from cancer survivors to enhance SEER registry data: the Louisiana Tumor Registry at Louisiana State University School of Public Health-New Orleans; the Iowa Cancer Registry; the Metropolitan Detroit Cancer Surveillance System; the New Jersey (NJ) State Cancer Registry; the Los Angeles (LA) Cancer Registry; and the Utah Cancer Registry. The Institutional Review Board at each site approved that site's study protocol and materials.

The target populations in this study were individuals diagnosed at any stage with early age–onset multiple myeloma or colorectal, breast, prostate, or ovarian cancer. Early age–onset was defined as under 50 years of age at time of diagnosis for breast or colorectal cancer, under 55 years of age for prostate cancer, and under 65 years of age for multiple myeloma or ovarian cancer. Two groups, defined by time since diagnosis, formed the sampling frame for each cancer type. The first group included those recently diagnosed (within 1 year of diagnosis) and the second group included longer-term survivors, diagnosed more than 3 years prior to the study start date for ovarian cancer or multiple myeloma cases or more than 5 years prior to the start date for breast, prostate, and colorectal cancer cases. The expectation was that each participating registry would recruit a minimum of 10 cases in each category of time since diagnosis for each cancer type. Overall target sample sizes ranged from 100 (Iowa) to 200 (NJ). Identification of sampling frames occurred in either late 2016 (Utah) or early 2017 (Louisiana, LA, Detroit, NJ, Iowa). The total number of individuals sampled from each SEER registry ranged from 320 (Iowa) to 1301 (NJ; Table 1A and B,Table 1B).

Participant diversity was encouraged; three registries had recruitment strategies to increase the representation of certain subgroups in their study sample. Utah oversampled Hispanics and residents of rural counties for the longer-term survivors, and Iowa oversampled non-Whites. Detroit limited recruitment to White and Black survivors and oversampled Black survivors.

All of the registries sent initial recruitment mailers for the study, some containing the paper questionnaire or a link to access the survey online if that option was available for that registry. The procedures for follow-up of cancer survivors who did not respond to the initial mailing varied by registry (Table 1A and B). For all registries, multiple attempts were made by mail and/or phone to request study participation if there was no response to the first mailing or the mailer was not returned as being undeliverable. E-mail was not used to initiate recruitment, as e-mail addresses are not routinely collected by SEER registries. An incentive for participation was not part of the protocol at any of the study sites.

The questionnaire was developed by NCI staff in collaboration with key personnel at each SEER registry. The questionnaire consisted of 28 items and included questions on demographics (sex, current employment, education), current health (comorbid conditions), and willingness to participate in various aspects of research studies, with slight variations for state-specific information (e.g., health insurance options vary by state).

The primary outcome variables were those pertaining to the respondent's willingness to participate in research studies and various aspects of studies; for example, whether the respondent would complete a single survey and/or multiple surveys, share their medical records, attend a clinic visit, or donate certain types of biospecimens. Other outcome variables included modality preference for completion of surveys (response choices: phone, paper, computer, smart phone or tablet, and other) and main reasons that they would be interested in participating in a research study [response choices: giving back to the medical community and helping those with cancer; learning more about cancer and relevant resources (including clinical trials); compensation; and other]. For the modality preference for survey completion and reasons for participating in a research study, some registries allowed multiple responses while the other registries asked the respondent to select only one choice.

Different methods were used to administer the questionnaire across registries (Table 1A and B); these methods included paper questionnaires sent through the mail (all registries except Detroit), a Web-based platform (Iowa, LA, Utah, NJ, Detroit), and telephone (Iowa, LA, Utah, Detroit, NJ). Utah conducted a randomized trial within this study to assess the response rate when offering a Web-based versus paper survey; potential respondents, thus, were offered only the survey type associated with the experimental arm to which they were assigned (6). At NJ, LA, and Utah, the questionnaire was available in Spanish.

Data on diagnosis date, age and stage at diagnosis, sex, and cancer type were abstracted from the SEER registry file and linked to each participant's questionnaire data by each registry. Cancer stage data were analyzed using the American Joint Commission on Cancer (AJCC) 6th edition staging manual categories (7), collapsed as 0 (in situ), I, II, III, and IV. Stage data in this format were not available for the analysis for the NJ and Detroit registries.

Response rates were calculated excluding individuals sent a mailing who were later determined to be ineligible. Personnel from each site abstracted and analyzed data for a limited set of SEER variables to compare enrollees with nonrespondents using χ² tests for categorical variables and Student t tests or Wilcoxon signed-rank tests for continuous variables. Data for each selected variable may not be comparable across the registries and results are show by registry (with the exception of LA, for which results were not available).

The associations between the willingness to participate variables and participant characteristics were examined using χ² tests. Analyses were carried out initially within study site strata, and Breslow–Day tests were conducted to examine effect modification of each association by study site. All associations were similar across the study sites; thus, combined results are presented. Multivariable logistic regression analyses were conducted to examine the associations between each of the participant characteristics and the willingness to participate variables adjusted for the other participant characteristics variables; the results were similar to the bivariate analyses and, thus, only the bivariate analyses results are shown.

All analyses were conducted using SAS v9.4. A P value of less than 0.05 was considered statistically significant.

The registries used a variety of methods to recruit participants, follow-up with nonresponders, and survey survivors, with varied response rates from 24.9% (NJ) to 46.9% (Utah; Table 1A and B). Five of the registries compared SEER data from those enrolled to nonrespondents (Supplementary Table S1); there were few statistically significant differences that were observed consistently among the registries. In Louisiana, Iowa, and Utah, the enrolled participants were more likely to be white than the nonrespondents; in contrast, in Detroit, the enrolled cancer survivors were more likely to be Black than their non-respondent counterparts. For other variables (e.g., age at diagnosis, cancer type, sex), there were no statistically significant differences, or inconsistent directions of association, between enrollees and nonresponders.

Five registries exceeded their overall sample size target; Louisiana achieved 98% of their overall sample size goal. Among the 992 total participants, the majority completed paper surveys at five of the registries; in Utah, the randomized trial conducted within the study showed that offering a paper survey only yielded a nonstatistically significant higher response rate than offering the Web survey only (6). At Detroit, the majority of participants (66.9%) completed the survey via a phone interview with a registry staff member.

The majority of the study participants at each registry were female, were in good or very good health, and reported 0 or 1 comorbidities (Table 2). For the remaining characteristics, there were variations across the registries. For example, there were greater percentages of Black participants in the Detroit (36.6%) and Louisiana (22.8%) samples, and a greater percentage of Asians in the LA (14.4%) and NJ (13.0%) samples, compared with the other registry sites. Similarly, there were greater percentages of Hispanic participants in the LA (32.5%) and NJ (14.2%) samples compared with the other four registry sites. The majority of participants across the registries had at least some college education [range: 68.1% (LA) to 82.4% (Detroit)], with low percentages at each registry not having a high school degree [range: 1.4% (Detroit) to 15.0% (LA)]. Approximately half of respondents at all registries were employed full-time [range: 41.5% (Detroit) to 57.4% (Utah)]. There were some differences in the distribution of cancer types, cancer stage, and the time since diagnosis categories between the registries due to different initial sample size targets and varying success in recruiting within these strata.

Overall, a high percentage of respondents were willing to participate in various aspects of research studies (Tables 3 and 4A and B). Approximately 90% of participants answered that they would be willing to fill out a survey (Table 3). At four registries, the majority of respondents selected paper as the preferred survey type (Supplementary Table S2); at Utah, participants preferred computer-based, and in Detroit, the majority preferred delivery of the survey by phone. Of note, at all registries, a minority listed smartphone or tablet as the preference for survey delivery. Among those who preferred an electronically delivered survey, the majority at all registries selected a home computer as the preferred device with the exception of those in Louisiana, whose participants' most frequently stated preference was a mobile device.

Approximately 87% of the respondents were willing to undergo a clinical exam at their regular doctor's office and 56.2% stated their willingness to take part in a clinic-based study at a doctor's office other than their own (Table 3). Eighty-two percent reported that they would consent to have their medical records accessed for research.

Over 91% of respondents were willing to donate a biospecimen of some type (i.e., either blood, saliva, urine, stool or tissue; Table 4A and B,Table 4B). Overall, 77.8% of participants stated that they were willing to donate a blood sample for research and 83.5% were willing to donate their tumor tissue (Table 4A and B). In contrast, only about half of the participants responded that they would be willing to donate a stool sample.

Regarding the reasons to participate in research, over 78% of the total respondents at each registry stated that they would participate to give back to the medical community [range: 78.1% (LA) to 93.7% (Detroit); Supplementary Table S2]. The second most common reason for participating in research was to learn more about cancer and relevant resources such as clinical trials [range: 36.9% (LA) to 54.3% (Louisiana)].

Tables 3 and 4A and B show the associations, across the registries, for the demographic and health characteristics and selected willingness to participate variables. Those who had previously participated in a research study, were more educated, or were employed full-time, were significantly more likely to indicate a willingness to participate in a future study that required a biospecimen donation, requested medical record access, or included a clinic visit. Race and ethnicity were also significantly associated with the willingness to participate variables; White cancer survivors were the most likely, and Asian cancer survivors the least likely, to report willingness to participate in future studies that involved completion of a survey, accessing medical records, a clinic visit, or donation of biospecimens. Cancer survivors of Hispanic ethnicity were less likely than non-Hispanic cancer survivors to indicate willingness to participate in a future study involving survey completion, a clinic visit at their regular doctor's office, or donation of DNA; in contrast, Hispanic cancer survivors were more likely than their non-Hispanic counterparts to indicate a willingness to donate a stool sample.

By cancer site, breast cancer survivors were significantly less likely to report being willing to donate a stool sample for a future research study, and multiple myeloma survivors less likely to report willingness to donate a tissue sample, compared with participants with the other cancer types. Time since diagnosis and cancer stage were not significantly associated with any of the willingness to participate variables.

SEER and state cancer registries represent the most complete enumeration of cancer survivors in the U.S. population (3). While cancer registries have been successfully used to recruit research participants for studies in the past (examples: refs. 8–11), there have been research and societal changes that may affect use of registries for population-based research (12–16). For example, there is an increasing demand by researchers for biospecimen collection and access to all health records, as well as increasing awareness of privacy concerns and changes in technology communication patterns such as switch to cell phones and use of caller ID (3). Furthermore, the growth in rapid reporting mechanisms to cancer registries may open the door for registries to efficiently recruit recently diagnosed cancer survivors. For these reasons, this pilot study was conducted to assess the feasibility of leveraging SEER resources to recruit and engage both long-term and newly diagnosed cancer survivors in population-based research.

The results of this study showed that, across six SEER registries, using various recruitment methods, it is feasible to rapidly recruit a geographically and racially diverse group of cancer survivors over a short time period to participate in a research study. In this pilot study, almost 1,000 cancer survivors (615 recently diagnosed and 377 longer-term survivors) were successfully contacted and responded to a survey during a 12- to 18-month study period. Similar to previously conducted survey-based studies that utilized central cancer registries for recruitment, response rates ranged from 24.9% to 46.9% (9, 10). Additional effort by the registry staff at each site to recruit for this study ranged from as little as an additional one-third full-time equivalent (FTE) to as much as two FTE depending on the goal number to enroll as well as the protocol (e.g., survey modality offered, number of follow-up contacts for nonresponders). Novel approaches for recruiting participants through SEER registries in this study included the use of multiple options for survey completion, such as a Web-based option at five of the six registries, as well as a Spanish version of the survey, which was available at three registries.

To assess the generalizability of the results, a comparison of enrollees to nonresponders was conducted for a limited set of SEER variables. In general, there were few statistically significant differences between enrollees and nonresponders in the registry-specific analyses, and for some of the significant associations, the directions of the associations varied by registry, possibly reflecting regional population differences, variations in recruitment methods, or chance effects. It should be noted that this study focused on the recruitment of early age–onset cancer cases, as this is a NCI area of interest; thus, no statement can be made about the generalizability of these results to older cancer survivors, who may differ in their willingness to participate in a research study. The benefit of utilizing registries is that they provide a well-defined source population, allowing investigators to assess how well a study sample reflects the population of interest, and, thus, the external validity of the results (3).

Successful recruitment using SEER or other central cancer registries depends on knowing how best to reach and engage the targeted “local” population. Each SEER registry used their own methods of contacting and recruiting participants that was informed, in part, by previous studies carried out by these registries. Most used multiple modalities for survey administration: five offered paper surveys, five allowed phone completion of the survey, and five had a Web-based survey option. However, even in an era where tasks are increasingly done electronically, it is interesting to note that five out of the six registries received most of the completed surveys via paper, which is consistent with what has been found in the survey methods literature (17–21); in addition, at four of the six registries, participants stated paper as their preferred mode of survey delivery. The stated preference results should be interpreted with caution, as there is some evidence that participants tend to prefer the survey modality that they just completed (22). The one study registry that received most of their surveys using a modality other than paper was Detroit, where most surveys were completed by phone, informed by their experience conducting the Detroit Research on Cancer Survivors (ROCS) study, which is recruiting newly diagnosed African American breast, prostate, colorectal, and lung cancer survivors through the Detroit-based cancer registry (8). At all registries, only a small percentage of enrollees listed smart phone or tablet as the preference for survey delivery, which may reflect either lack of access to or experience with the devices or past difficulty with completing surveys on small devices. For those registries offering both a Web-based option and a paper option of the survey, most participants completed the paper version which may reflect that the initial contact was via a mailed (paper) letter, since SEER registries do not routinely collect e-mail addresses for contact. In Utah, where potential participants were randomized to receive either the paper survey or Web-based survey only, there was a nonstatistically significant higher response rate for the paper versus Web-based survey (6).

Among the respondents, 90% indicated that they were willing to participate in at least some aspect of a research study, and the majority were willing to participate in aspects of research associated with a higher participant burden, such as a clinic visit or biospecimen donation. However, as seen prior studies (23–27), those with lower education and those who had not participated in research studies in the past were less willing than others, speaking to the need for additional outreach efforts to engage certain populations.

Across all races and ethnicities, the majority of respondents were willing to participate in research but Black and Asian respondents (as well as those who were of Hispanic ethnicity) were less likely than their White counterparts to report willingness to participate in certain components, including survey completion, a clinic visit, or collection of a biospecimen. Similar differences have been observed in the Breast Cancer Family Registry study where enrollment rates and biospecimen collection among the patients with breast cancer and her family member(s) were considerably lower among Asian Americans compared with non-Hispanic whites and other race and ethnic subgroups (28). Differences in the relationships between race and ethnicity with the willingness to participate variables highlight the importance of understanding how to engage underrepresented populations in research, which includes recognizing community members as partners in research, building trust between the community and investigators, being transparent regarding risk of research to participants and community, and establishing a line of communication between the researcher and the community during all phases of research (29).

There were few differences in the willingness to participate in various aspects of research by cancer type. Individuals diagnosed with multiple myeloma were less likely than individuals diagnosed with the other cancer types to be willing to donate a tissue sample for research; furthermore, the percentage of respondents overall who were willing to donate a stool sample was lower than those who were willing to donate the other biospecimen types, with breast cancer survivors being the least likely to report being willing to donate a stool sample. We did not assess the underlying reasons for the choices associated with these cancer types, but the results speak to the need to clearly communicate issues about research, such as the type of tissue required for donation, use of existing tissue samples, and the importance of collections that are perceived more negatively (e.g., stool collections) for the relevant research.

Previous research has shown that altruism is one of the primary reasons that individuals participate in research (30, 31)—this was echoed by the participants in this study as well. While altruism is a major driving force for study participation, most of the sites noted that use of an incentive, which was not provided here, has helped in other studies. In their analysis of data from the 17 studies conducted from 2007 to 2016 that utilized the Utah SEER registry for recruitment, Millar and colleagues (32) found that the odds of recruitment increased by 62% with an incentive. Interestingly, all of the Utah SEER studies from 2007 to 2017 used a postincentive (32), those promised at the end of study completion, which has been shown to be less effective than unconditional preincentives (33, 34). Other recommendations from the registries after the completion of this study included building an informational website for participants; offering multiple modality options for completing a survey, with consideration of the sequence on how the modalities are offered (35); providing information on how the study results will be used (i.e., ensure that they know the importance of the research); minimizing participant burden; and sharing study results and providing study updates to engage survivors in continued study participation.

Cancer survivors in this pilot study reported willingness to participate in all aspects of research studies, including an in-person visit, blood collection, and access to medical records. Caveats are that these results reflect those who were willing to take part in this study in the first place and, intention does not always lead to the intended behavior. However, the response rates observed here are commensurate with several other survey-based studies conducted using central cancer registries for recruitment (9, 10). It is unknown whether a research study requiring multiple surveys, biospecimen donation, clinical exams, or medical record abstraction would have similar response rates as are reported in this manuscript, although in the analysis of the 17 Utah registry-based studies, results showed that having a biospecimen donation component did not affect response rates (32).

Overall, this study demonstrated the feasibility of leveraging population-based cancer registries to recruit and engage a geographically and racially diverse group of cancer survivors across cancer types and lengths of time since diagnosis. SEER registries represent 35% of the U.S. population (5), and state cancer registries cover the remaining population, making these resources an invaluable network for recruiting cancer survivors into research studies and utilizing the data collected within these registries.

No potential conflicts of interest were disclosed.

Conception and design: L. Gallicchio, J.W. Elena, M.M. Millar, L.E. Paddock, X.-C. Wu, K.J. Helzlsouer

Development of methodology: L. Gallicchio, J.W. Elena, A.S. Hamilton, T.A. Hastert, C.F. Lynch, J. Milam, L.E. Paddock, C. Sweeney, E.J. Trapido, M.M. West, X.-C. Wu, K.J. Helzlsouer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Gallicchio, M. Carter, A.S. Hamilton, T.A. Hastert, L.L. Hunter, J. Li, C.F. Lynch, M.M. Millar, D. Modjeski, L.E. Paddock, A.R. Reed, L.B. Moses, A.M. Stroup, C. Sweeney, E.J. Trapido, M.M. West

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Gallicchio, J.W. Elena, S. Fagan, J. Li, A.R. Reed, E.J. Trapido, K.J. Helzlsouer

Writing, review, and/or revision of the manuscript: L. Gallicchio, J.W. Elena, S. Fagan, M. Carter, A.S. Hamilton, T.A. Hastert, L.L. Hunter, J. Li, C.F. Lynch, J. Milam, M.M. Millar, D. Modjeski, L.E. Paddock, A.R. Reed, A.M. Stroup, C. Sweeney, E.J. Trapido, M.M. West, X.-C. Wu, K.J. Helzlsouer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Gallicchio, J.W. Elena, M. Carter, A.S. Hamilton, T.A. Hastert, J. Li, M.M. Millar, D. Modjeski, A.R. Reed, A.M. Stroup, M.M. West

Study supervision: L. Gallicchio, J.W. Elena, M. Carter, A.S. Hamilton, T.A. Hastert, J. Li, D. Modjeski, L.E. Paddock, A.M. Stroup, M.M. West, K.J. Helzlsouer

Research reported in this publication was supported by the NCI of the NIH under contract numbers HHSN261201300020I, HHSN261201300011I, HHSN261201300017I, HHSN261201300016I, HHSN261201300021I, and HHSN261201300004I.

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.