Comparison of Young Adult Female Cancer Survivors Recruited from a Population-Based Cancer Registry to Eligible Survivors

Advances in cancer therapy have improved survival rates (1). However, many treatments affect the long-term health and quality-of-life of survivors. Population-based cancer registries allow for identification of survivors within a defined sampling frame (2). Registry-based studies are particularly relevant for young adult survivors, who experience a wide range of cancer types (3–5) and diverse sociocultural conditions related to insurance, employment, mobility, and familial situation (5–7). Furthermore, treatment is not centralized (4–6), making identifying young adult survivors difficult in a clinical setting. Thus, population-based studies are important to represent the heterogeneity of these survivors, and cancer registries provide a unique resource.

However, few publications directly evaluate whether long-term survivors of young adult cancers recruited from population-based registries are representative of the eligible population (8–11) or examine why eligible survivors are missing from a study population (e.g., unable to contact, refusal) by survivor characteristics. The Furthering Understanding of Cancer, Health, and Survivorship In Adult (FUCHSIA) Women's Study is a population-based study of women's health in female survivors of young adult cancers 2–19 years after diagnosis. We evaluate the representativeness of participants compared with those eligible for the study with respect to demographic and cancer-related characteristics. Results from this study can identify characteristics of cancer survivors who may be underrepresented in registry-based studies to develop targeted protocols to enhance recruitment. Furthermore, our results provide sampling probabilities, which can be used to address selection bias.

Survivors were identified in the Georgia Cancer Registry (GCR), a statewide, population-based tumor registry. All reportable malignant cancers (12) and ductal carcinoma in situ were eligible. Additional criteria included being female, aged 20–35 years at first diagnosis, diagnosed during 1990–2009 in metropolitan Atlanta or 1999–2009 outside metropolitan Atlanta, at least 2 years post diagnosis, and 22–45 years old. Recruitment for thyroid cancer and melanoma was restricted to metropolitan Atlanta (1995–2008) because these cancers are common and unlikely to receive therapies of interest. The interview was only available in English. Because the study required identifying information, GCR protocol required permission from survivors before study contact.

GCR staff contacted cancer survivors using updated contact information from LexisNexis Accurint. Starting in September 2011, eligible women were sent a letter explaining the cancer registry, describing how the registry protects privacy, and introducing the FUCHSIA Women's Study. A form for updating contact information or refusing further contact and a prepaid envelope were provided. After 2 weeks, nonresponders were called at all available phone numbers at different times of the day on different days of the week, including the weekend. Messages were left on voice mail when possible for the first call to a valid number and every 2 to 3 calls after. In March 2012, women were classified into those contacted by the GCR, those unable to be contacted (no reply by mail and no phone contact), and those reported to be ineligible by a third party (e.g., deceased; Fig. 1). Those contacted were further classified as agreeing to study contact, refusing study contact, undecided (i.e., never refusing or agreeing), or ineligible.

Survivors willing to be contacted by the FUCHSIA Women's study were sent a letter about the study with information on informed consent and the Health Insurance Portability and Accountability Act (HIPAA). Shortly thereafter, women were called for an interview about their reproductive history, demographic characteristics, health, and lifestyle. Before the interview, they were also told that participation required consent to have cancer diagnosis and treatment information abstracted from their medical records. Interested women were screened and interviewed if eligible. Informed consent and the HIPAA authorization were administered by phone prior to the interview. Individuals completing the interview were sent a $25 gift card and a medical record release form. Recruitment lasted from May 2012 through February 2013. Women were called at different times of the day on different days of the week, including the weekend. Messages were left on voice mail when possible. A second letter, reminding survivors about the study, was sent to individuals who had not been reached by October 2012. After recruitment ended, each woman was assigned a recruitment outcome, including interviewed (partial or complete) or nonparticipation. Subcategories of nonparticipation included refused, ineligible, not locatable (women whose phone numbers changed and could not be updated), or unresolved. Unresolved included women who did not actively refuse or agree to study participation. This category included women who passively refused, who were unavailable during the study timeframe, who we could not talk to directly (e.g., spouse refused without consulting the eligible woman), and whose phone numbers were unconfirmed.

The GCR provided routinely collected demographic and cancer-related information on all eligible women. County of the mailing address was categorized using the 2006 National Center for Health Statistics Urban-Rural Classification Scheme.

We performed descriptive analyses by recruitment outcome. We fit a single logistic model to evaluate the associations between all demographic and cancer-related variables and nonparticipation (vs. interviewed, including partial and complete). We also fit a polytomous logistic regression model, including all covariates in the same model, wherever refused, unable to contact (unable to contact and not locatable), and unresolved (undecided and unresolved) were each compared with interviewed. Both models were restricted to eligible women living in Georgia without missing data (Supplementary Fig. S1). We evaluated the sensitivity of our results through the following additional analyses: (i) restricting to metropolitan Atlanta, (ii) excluding thyroid and melanoma survivors, and (iii) restricting to African American and white women because other race groups were small.

The study was conducted in accordance with recognized ethical guidelines and was approved by the Emory University (Atlanta, GA) and Georgia Department of Public Health Institutional Review Boards.

The GCR identified 5,423 potentially eligible women (Fig. 1). Of these, 61% were contacted, 38% could not be contacted, and less than 1% were ineligible (e.g., deceased, did not speak English) during GCR outreach. Among those contacted, 23% replied by mail, 71% by phone, and 6% by both methods. Among respondents, 78% agreed to study contact, 15% refused, 2% reported being ineligible, and 6% were undecided or passively refused. Among women providing a reason for refusal, 74% reported lack of interest. The median number of calls to survivors was 5 [interquartile range (IQR): 2–10], including calls to incorrect numbers (31% of all calls). GCR staff spent over 2,500 person-hours, calling survivors, mostly weekdays between 4:00 and 7:00 pm over 29 weeks.

Of women who agreed to be contacted by the study, 50% finished the interview, 2% completed part of the interview, 16% refused, 8% were ineligible, 7% had an incorrect phone number, and 18% were unresolved. Reasons for ineligibility included unable to complete the interview, aging out of eligibility, self-reporting an ineligible age at first cancer diagnosis, being deceased, being unaware of cancer diagnosis, and not answering a screening question. The median number of study calls to survivors was 16 (IQR: 5–30), including unanswered calls, with a median of two calls (IQR: 1–4) answered by anyone. Among participants, the median number of calls was 10 (IQR: 4–21) and of answered calls was three (IQR: 2–4).

Differences in being interviewed by age at attempted contact were small (Table 1). White women were slightly more likely to be contacted by the GCR than African American women, but once contacted, white and African American women were equally likely to agree to study contact. However, African American women were more difficult for the study to locate. This pattern generally persisted across demographic and cancer characteristics (Supplementary Fig. S2). GCR contact was most successful for women living in metropolitan Atlanta compared with other areas, but once contacted, women were equally willing to release their names to the study across locations. However, study contact was less successful for women outside metropolitan Atlanta. This pattern persisted across demographic and cancer characteristics (Supplementary Fig S3).

The proportion of women interviewed was similar across age at diagnosis but varied across years since diagnosis. Time since diagnosis did not appear to affect the ability to locate African American or white women (Supplementary Fig. S2D). The success of GCR contact varied by cancer type from 45% for cervical cancer to 70% for melanoma. This variability persisted for study participation from 16% for cervical cancer to 32% for breast cancer. Low participation of cervical cancer survivors was driven by greater difficulty locating them compared with all cancers combined (Supplementary Fig. S4).

Factors most strongly associated with nonparticipation included race other than African-American or white compared with white [odds ratio (OR) 5.5, 95% confidence interval (CI): 2.4–12.8] and Hispanic compared with non-Hispanic (OR 5.1; 95% CI 3.0, 8.6), but these estimates were imprecise (Fig. 2; Supplementary Table S1). Weaker but more precise estimates were observed for African-American versus white (OR 1.5; 95% CI: 1.3–1.8), living in a rural area versus metropolitan Atlanta (OR 1.8; 95% CI: 1.4–2.2), and cervical cancer versus the less common cancers combined (OR 1.7; 95% CI: 1.3–2.2).

In the polytomous model, race other than African-American or white compared with white was strongly associated with all nonparticipation outcomes (refusal, unable to contact, unresolved), but the estimates were imprecise (Fig. 3; Supplementary Table S1). The pattern was less consistent for other characteristics.

The odds of refusing to participate compared with being interviewed and of being unresolved versus being interviewed decreased with increasing age at contact, but the associations for unable to contact versus being interviewed did not vary by age. Being African-American compared with being white was associated with unable to contact (OR 1.7; 95% CI: 1.5–2.1), but not with refusal or unresolved. Being Hispanic compared with non-Hispanic was strongly associated with unable to contact (OR 8.3; 95% CI: 4.9–14.2), but the estimate was imprecise. Being Hispanic was also moderately associated with refusal and unresolved. The odds of all nonparticipation outcomes were elevated for women living in a rural area versus a large metropolitan area; the strongest association was for unable to contact (OR 2.1; 95% CI: 1.7–2.7).

Age at diagnosis was not associated with nonparticipation. Women who were diagnosed 16–21 years before attempted contact were the most likely to refuse (OR 2.5; 95% CI: 1.1–5.7) and to be unresolved (OR 1.7; 95% CI: 0.7–4.5) compared with women diagnosed 2–5 years before, but these results were imprecise. Women with cervical cancer were less likely to be contacted than women with the less common cancers (OR 2.1; 95% CI: 1.6–2.8). However, they were similar to other cancers with respect to refusal and unresolved.

The sensitivity analyses were not meaningfully different from the primary analyses.

The overarching goal of the FUCHSIA Women's Study was to assess women's health issues in a population-based cohort of female survivors of young adult cancers. We recruited a cohort of survivors that closely resembled eligible women in a population-based cancer registry with respect to demographic and cancer-related characteristics, but doing so required substantial resources. The greatest challenge was contacting women rather than participant refusal. The GCR was unable to reach approximately 40% of the population, despite using tracing services. Comparing nonresponse levels across studies is difficult because definitions of unable to contact differ (e.g., invalid addresses only or invalid addresses and nonresponse) and because of differences in the denominator (e.g., only eligible or eligible and unknown eligibility). Nevertheless, our results are consistent with other studies recruiting long-term (≥1 year) survivors of adult cancers from population-based registries in the United States. These studies report being unable to contact from 24% to 66% of those presumed eligible (8, 13–16).

The population interviewed for the FUCHSIA Women's Study was more likely to be white, non-Hispanic, and live in metropolitan Atlanta than the eligible population in the GCR. In addition, women with cervical cancer were less likely to participate. Counter to expectations, the percent of survivors who were unreachable did not increase with years since diagnosis suggesting that survivors can be traced long after diagnosis.

Hispanic women were substantially less likely to participate than non-Hispanic women, which might be, in part, because study materials were only available in English. Women whose primary language was Spanish may have screened their calls (and may not have been eligible if reached). However, for all women, we could not distinguish passive refusal from incorrect contact information.

The overall difference in recruitment between African American and white women was small, but similar to other registry-based studies (8, 11, 15, 17, 18). Although African American women were more likely to participate than Hispanic women, African American women were a larger proportion of the eligible population and therefore, more African American women did not participate. Similar to other registry-based studies (8, 16, 17), participation differences between white and African American women were driven by greater difficulty contacting African American women. Greater mobility and less use of resources monitored by contact tracing companies may affect the ability to contact some women. According to the 2000 US Census, both African-Americans and Hispanics were more likely to move between 1995 and 2000 than non-Hispanic whites (19). Furthermore, in a study of workers compensation claimants, African American claimants required more intense tracing than whites (20). In addition, workers requiring more intense tracing reported more financial problems. Financial difficulties might be associated with less use of resources tracked by tracing companies. Thus, investigators might improve recruitment by devoting additional resources to contact tracing of African-Americans and Hispanics.

In our study and a hospital-based cancer registry study (21), women living in rural areas were less likely to be contacted. Although a higher proportion of women from rural areas were not interviewed, a greater number of women from metropolitan Atlanta were not interviewed. If urban–rural differences are of interest, then identifying alternative approaches to trace women in rural areas is important. However, if these differences are not critical, greater sample size might be achieved by improving recruitment in metropolitan areas.

Participation by cervical cancer survivors, the second most common cancer in our study, was low. Although half of those contacted participated, only 16.7% of those who were eligible participated. These results are similar to a California registry–based study where 51% of those contacted and 22% of those eligible participated (13). Low participation in our study was driven by the inability to contact cervical cancer survivors. Because the incidence of cervical cancer can be reduced through routine screening (22, 23), we hypothesize that women with cervical cancer were more likely to be from economically disadvantaged populations with less access to care and lower utilization of services audited by tracing companies. This hypothesis gains some support from behavioral surveillance data that indicate women without health insurance are less likely to have been screened for cervical cancer in the past 5 years compared with all women aged 21–65 years (24). This raises concerns about selection bias for registry-based studies of cervical cancer survivors.

Participation is also affected by the protocol for tracing eligible survivors and following-up with nonresponders. Our study was complicated by the requirement that the GCR contact women before the study. Follow-up calls after the initial GCR mailing were critical for contacting survivors and required substantial GCR staff effort. A 2004 survey of United States cancer registries reported that 21% did not allow study contact with cancer survivors, and 36% of those allowing contact required the registry to make first contact (25). Furthermore, some registries reported changing their policies to decrease study-initiated contact in response to HIPAA policies. Thus, additional registries may have made similar changes after 2004. Some registries can send study questionnaires to eligible individuals directly on behalf of the study, which might increase participation. However, this protocol is not possible for more complex studies, such as ours, which require access to identifying information (e.g., medical records) or have study protocols that require more resources than the registry has available (e.g., extensive interviews).

Another challenge was call screening. Call screening is often considered passive refusal, but some women interested in participating contacted us to ask why we had not called them. Typically, we had already left messages at the correct phone number, indicating that call screening may prevent even desired contact. Some studies define cancer survivors as inaccessible after seven phone calls or fewer (11, 15, 16), which may affect participation levels. In our study, the median number of calls to eventual study participants was 10 (IQR: 4–21), and another registry-based study concluded that up to 15 calls was productive (26). An alternative approach would be to text women who have not been reached. We hypothesize that some women would be more responsive to texts if only to refuse actively instead of passively. Texting has been used to communicate with study participants after enrollment but less often for recruitment (27), presumably because it requires a cell phone number. Nevertheless, registries could facilitate this recruitment approach by collecting cell phone numbers during prescreening. Collecting e-mail addresses may be of value as well.

Even after prescreening by the GCR, 7% of women agreeing to study contact could not be reached for the interview at the phone number they provided and could not be traced. We began recruitment after GCR outreach was complete. Participation would likely improve if study recruitment began during registry outreach because women would have less time to change their contact information and would be more likely to remember hearing about the study.

Some women incorrectly reported being ineligible during screening, including women unaware of their cancer diagnosis. This discrepancy between registry data and self-report has been observed in other studies (17, 21, 26). Thus, it is important that registry and study staff are prepared to address this situation sensitively.

To participate in our study, women had to consent to have cancer diagnosis and treatment information abstracted from their medical records. Concerns related to the privacy of their medical records may have caused some women to refuse to participate. Furthermore, although all interviewed women agreed to release their medical records orally, obtaining medical records was affected by numerous factors other than willingness to participate in the study such as failure to return signed medical record release forms, length of time facilities store medical records, and facilities' willingness to process requests.

Generally, our recruitment results are reassuring. However, one of the main limitations of this study is that participation may differ by unmeasured factors. The degree to which this introduces bias would depend on the research question and how strongly the factors are related to the exposure and outcome of interest. A Norwegian cancer survivor study examined bias due to nonresponse and concluded that low participation introduced minimal biased for many research questions (10). Another limitation is that women were 2 to 21 years postcancer diagnosis at recruitment. Survival is associated with some of the participant characteristics we considered, which affected the number of eligible women with different characteristics. This may limit generalizability of our results to studies restricted to other times since cancer diagnosis. Furthermore, because this study focused on women's health in survivors of young adult cancers, our results may not be generalizable to male survivors or survivors of childhood or older adult cancers.

Identifying survivors from a cancer registry provides an opportunity to recruit a broad spectrum of cancer survivors including survivors of different cancer types, survivors living in different geographic areas, and survivors who might not attend a specific clinic for other reasons, such as limitations of their insurance coverage. It also provides an opportunity to collect information not available in administrative records (10). The success of registry-based recruitment compared with other methods is affected by whether current patients or long-term survivors are approached, the study burden (e.g., mail-in questionnaires, medical records release, clinical interventions), and recruitment protocols. The biggest challenge to recruitment in our study was not active refusals but contacting women because of incorrect contact information and unanswered calls. Given the loss of potential participants due to incorrect contact information, contact-tracing services are critical, but there is potential for selection bias. Some studies may be able to perform quantitative bias analyses by estimating participation probabilities based on data from their registries or based on our study. Generally, maximizing the success of population-based recruiting requires substantial resources, which is challenging when funding is limited and rapid results are desired. Nevertheless, the value gained is the opportunity to address research questions that cannot be answered with administrative data in populations that might be underrepresented based on other recruitment schemes.

P.P. Howards reports grants from NIH/NICHD during the conduct of the study. P.J. Mink reports grants from NIH during the conduct of the study. A.C. Mertens reported grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

Part of this work was previously presented as a poster at the 2013 Annual Meeting of the Society for Epidemiologic Research (Boston, MA).

P.P. Howards: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. P.J. Mink: Conceptualization, writing–review and editing, advising on key decisions related to the study. K.H. Kim: Conceptualization, resources, data curation, validation, writing–review and editing. J.J. Woodard: Conceptualization, resources, data curation, supervision, validation, project administration, writing–review and editing. A.C. Mertens: Conceptualization, writing–review and editing, advising on key decisions related to the study.

The authors thank Amy Fothergill for verifying the reported results, Helen Chin for providing feedback on the manuscript, and the Helen Riaboff Whiteley Center, University of Washington (Friday Harbor, Washington), for providing a work environment conducive to writing.

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.