Rural, Large Town, and Urban Differences in Optimal Subspecialty Follow-up and Survivorship Care Plan Documentation among Childhood Cancer Survivors

This article is featured in Highlights of This Issue, p. 583

An estimated 500,000 adult survivors of childhood cancer live in the United States and, with therapeutic advances pushing 5-year survival rates in recent years to 85%; the number will undoubtedly continue to grow (1). The recognition of late effects and risk-adapted treatment strategies, notably for leukemia and lymphoma therapy, in recent decades resulted in a decreased burden of late chronic conditions (2, 3). Nevertheless, the cure comes at a cost with severe chronic health conditions occurring in over a quarter of survivors and considerable excess mortality from subsequent malignancies and cardiovascular disease (4–9). The heterogeneity of pediatric malignancies, associated treatment exposures, and subsequent comorbidities for survivors call for risk-adapted therapy to optimize lifelong health for survivors (10). Moreover, survivors from nonurban areas may face additional unique challenges and barriers to equitable survivorship care.

Guidelines and survivorship care plans offer comprehensive tools for clinicians and survivors to support optimal care delivery. The Children's Oncology Group Long-Term Follow-up Guidelines are regularly updated and efforts such as the International Guideline Harmonization Group help synthesize the growing body of evidence on late effects with the goal to improve care (11–13). Survivorship care plans, such as Passport for Care (PFC), serve as tools to promote patient education and the diverse care delivery models reflect the creative approaches to meet the needs of this burgeoning population (14). Nevertheless, nearly a third of survivors receive suboptimal follow-up in the early survivorship period, which worsens dramatically as 70% of adult survivors of childhood cancer report no survivorship-focused care (15, 16). This decline in optimal care coincides with the progressive cumulative incidence of life-threatening late effects, such as heart failure and breast cancer, during the young adult years (17, 18).

Oklahoma is a highly rural state with a high burden of chronic health conditions and survivors from nonurban areas face unique challenges for optimal survivorship-focused care. Thirty-four percent of residents live in rural areas in 2019, more than twice the national average of 14% (19). Rural disparities in the general population represent an important public health challenge (20). For survivors, hypertension and other cardiovascular risk factors potentiate risk for cardiomyopathy and coronary artery disease from anthracyclines and radiation, thus highlighting the critical need to engage high-risk survivors in care to mitigate rural health inequities later in life (21). Rural-urban differences in the general population for diabetes, hypertension, heart disease, stroke, and cardiovascular mortality portend even graver consequences for childhood cancer survivors at risk for late effects (22–24). In the adult literature, rurality has been associated with poor health status and inadequate care among survivors of adult cancers (25–27). Transportation barriers and distance to tertiary care centers, which contribute to disparities while on active treatment in pediatric oncology, may also adversely impact optimal survivorship care (28, 29). Follow-up care patterns and potential disparities for those from nonurban areas represent key gaps to care for this vulnerable population.

This study aimed to assess the impact of rurality on suboptimal subspecialty follow-up during the early survivorship period through the integration of cancer registry, electronic health records, and geospatial data. We hypothesized that survivors from nonurban areas would be more likely to receive suboptimal subspecialty follow-up care. Secondary outcomes included the proportion of survivors with a survivorship care plan, generated through PFC, by rurality.

The University of Oklahoma institutional cancer registry was used as the base cohort for this analysis. The Cancer Registries Amendment Act of 1992 established federal support and standards for cancer registries across the United States (30). Over the last several decades, the National Cancer Database, in conjunction with the Commission on Cancer and the American Cancer Society, provides rigorous data standards and captures greater than 70% of new diagnoses of cancer annually (31). Age at diagnosis, sex, primary diagnosis, the first course of treatment, demographic data, including the address at diagnosis, mortality, and recurrence data are all documented by trained cancer registrars. At Oklahoma Children's Hospital, the cancer registry reliably includes new diagnoses of pediatric cancer since 2005, thus patients ≤ 18 years old diagnosed after January 1, 2005, were considered for the full cohort (n = 1,421). Patients diagnosed after September 24, 2014, those not seen in the pediatric oncology clinic at the Jimmy Everest Center (JEC) within a year of diagnosis, nonanalytic cases (those not diagnosed or receiving the first course of treatment at the institution), and those with documented death within seven years of their initial diagnosis were excluded from the analysis. Survivors with documented relapse were also excluded to focus on those off therapy or the early survivorship period. Subsequent malignant neoplasms were not excluded, as this was not routinely captured by the cancer registry. The age range of survivors at the end of the 7-year follow-up period was 7 to 25 years old. For the secondary analysis focused on the documentation of a survivorship care plan, we also excluded patients with a history of a primary central nervous system (CNS) tumor as they were not routinely entered into PFC. Documentation of a survivorship care plan was validated through the institutional log-in for the PFC website. This research was submitted to and approved by the University of Oklahoma Health Sciences Institutional Review Board (IRB#: 13687) on September 22, 2021.

The cancer registry defines the primary malignancy by the International Classification of Diseases Oncology, third edition (ICD-O-3), which facilitates additional categorization based on the International Classification of Childhood Cancer, third edition (ICCC-3; ref. 32). Major pediatric oncology disease categories were used to implement late effects risk stratification developed by the British Childhood Cancer Survivor Study (33). Primary diagnosis and dichotomous treatment exposures, as captured in the cancer registry, were used to identify survivors as low, moderate, and high risk for late effects. The full methodology for this was documented previously (15).

The University of Oklahoma Clinical Research Data Warehouse integrates data from multiple clinical sources across the University of Oklahoma Health Sciences Center. For patients meeting the inclusion criteria, details were extracted using standard query language for all clinic visits at the JEC and the Stephenson Cancer Center (SCC) for each patient in the cancer registry as the base cohort. A valid completed visit was defined as a visit that was scheduled, the patient arrived, and the visit was subsequently closed at the end of the encounter. In addition, we limited to provider visits (e.g., laboratory-only or nurse-only visits were not considered a valid completed visit for the purposes of this analysis). These dates were landmarked to the initial date of diagnosis, as documented in the cancer registry. Optimal follow-up was defined as a completed clinic visit at the JEC or the SCC during a 2-year period between 5 and 7 years (60–84 months) after the initial diagnosis. Race/ethnicity was categorized as non-Hispanic White, non-Hispanic Black, Hispanic, American Indian, or Other. For each patient, follow-up was assessed with informatics methods on structured database fields; valid completed clinic visits were identified in this first pass. If this approach did not recognize an oncology-related visit between 5 and 7 years, a manual chart review was performed. Of the 152 survivors without a valid clinic visit in the follow-up window, 51 had a documented transfer of care to another institution and three were not recommended to follow-up.

The primary association of interest was rurality, defined as urban, large town, small town/isolated rural based on the 2010 Rural-Urban Commuting Area (RUCA) codes, which consider population density and distance to nearest urban centers (Supplementary Table S1; refs. 34, 35). The primary urban centers in Oklahoma are the Oklahoma City area, Tulsa, and Lawton. We geocoded patient addresses at the time of diagnosis using the Texas A&M Geocoding Service ESRI Toolbox (36). We joined the addresses to the 2010 US census tract TIGER/Line shapefiles (37), which were then joined to the appropriate census tract-level RUCA codes (35).

Additional geographic variables included the Area Deprivation Index (ADI), distance to the medical center, and travel time to the medical center. We obtained the 2019 ADI from the Neighborhood Atlas at the census block group level and used the national rankings to classify patients into quartiles (<25%, 25%–49%, 50%–74%, and 75%–100%; refs. 38, 39). To calculate the distance and travel time from the patient's residence to the medical center, we used the Origin-Destination Cost Matrix Tool in ArcMap, accounting for road networks. All geospatial analyses were conducted in ArcMap 10.8.1 (ESRI, Redlands, CA).

The primary outcome of interest was a completed oncology-related clinic visit at either the JEC or SCC between 5 and 7 years after the initial diagnosis, thus categorized as a dichotomous variable. Continuous variables were presented as means or medians, as appropriate based on distribution, and differences between those with and without optimal follow-up were analyzed using the t test. Categorical variables, such as rurality, race/ethnicity, sex, adolescent at diagnosis, late effects risk strata, primary diagnosis, and PFC, were presented as proportions and differences were assessed using the χ² test.

To calculate risk ratios (RR) and 95% confidence intervals (CI), we used modified Poisson regression due to challenges with convergence in the log-binomial model. In this model, we assumed that the data follow an underlying Poisson distribution, although our outcome was binomial. Standard Poisson regression can lead to overdispersion in binomial data, which can result in the overestimation of the standard error. Modified Poisson regression uses the sandwich estimator to correct for any overestimation of the standard error (40–42). We used a directed acyclic graph (DAG) to guide the selection of confounders, with race/ethnicity being the only variable necessary to include in the multivariable models to remove confounding (refs. 43, 44; Supplementary Fig. S1). Using a DAG allowed us to visualize the causal relationships between variables to determine which variables may confound the relationship between rurality and optimal follow-up without introducing selection bias (i.e., collider stratification bias), over adjustment (i.e., adjusting for a mediator), or unnecessary adjustment (i.e., adjusting for variables that do not introduce bias, but impact precision; refs. 45, 46). We considered race/ethnicity, adolescent at diagnosis, ADI, sex, primary diagnosis, late effects risk stratification, and year of diagnosis as potential confounders or mediators on the pathway to suboptimal follow-up, using subject matter knowledge to guide DAG development. This guided statistical modeling for the association between rurality and the risk of suboptimal subspecialty follow-up care among survivors of childhood cancer in Oklahoma. As a secondary outcome and sensitivity analysis, we repeated analyses for the association between rurality and the documentation of a survivorship care plan in PFC for general pediatric oncology survivors (n = 369; Supplementary Fig. S2). All statistical analyses were conducted using Stata/SE version 17.0 (Stata Corp LLC, College Station, TX), SAS v. 9.4 (Cary, NC), and R (RStudio, 2022.07.1, Build 554).

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

The final analytic cohort included 449 childhood cancer survivors (Fig. 1), of whom 78.5% (n = 358) had a completed clinic visit at the JEC or the SCC between 5 and 7 years after their initial diagnosis (Table 1). Among survivors in Oklahoma (n = 440), those from large towns were most likely to receive suboptimal follow-up at 36% (n = 28/78) compared with 21% (n = 20/95) and 17% (n = 46/276) of survivors from small town/isolated rural and urban areas, respectively (P = 0.01). The median distance for survivors with suboptimal follow-up was 46.7 miles compared with 24.3 miles for those with optimal follow-up (P < 0.01). There was significant heterogeneity in follow-up by primary diagnosis (P < 0.01) and survivors at intermediate risk for late effects were more likely to receive suboptimal follow-up at 28% in contrast to 15% of survivors at low and high risk for late effects (P < 0.01). Forty-five percent of survivors who were adolescents at the time of diagnosis received suboptimal follow-up compared with 17% of survivors who were younger than thirteen at the time of diagnosis (P < 0.01) There were no significant differences in optimal follow-up by sex (P = 0.81), race/ethnicity (P = 0.28), or ADI (P = 0.06).

A heat map of Oklahoma provides a visualization of optimal subspecialty follow-up by county (Fig. 2). We observed a higher percentage of survivors with optimal follow-up in the central part of the state, including the counties surrounding the medical center in Oklahoma City. The adjusted RR for survivors from large towns was 2.17 (95% CI, 1.46, 3.23) compared with survivors from urban areas as the referent population (Table 2). For survivors from small town/isolated rural areas, the risk was increased but the 95% CI encompassed one, with survivors from urban areas as the referent population. When checking the interaction between rurality and the distance from the cancer center in the modified Poisson regression model, the algorithm would not converge and therefore distance was not included in the final model. Consequently, we conducted a sub-analysis to separate out non-Oklahoma City urban areas compared with Oklahoma City-urban as the referent population (Supplementary Fig. S3 and Supplementary Table S2).

Documentation of a survivorship care plan in PFC was strongly associated with subspecialty follow-up with a RR of 0.20 (95% CI, 0.14–0.28). By rurality, 67% of survivors from large town had a survivorship care plan compared with 79% of survivors from urban areas and 70% of survivors from small town/isolated rural areas (P < 0.01). Our secondary analyses yielded similar results to the analysis of rurality and suboptimal subspecialty care follow-up care (Tables 3 and 4; Supplementary Fig. S4). The notable exception for this was that late effects risk stratification was not associated with a survivorship care plan (P = 0.40).

Survivors of childhood cancer from nonurban areas face complex challenges in optimal oncology-related follow-up care. Those from large towns were least likely to see an oncologist in the early survivorship period. This observed disparity persisted despite adjustment for race/ethnicity as a potential confounder. Furthermore, a strong association between optimal follow-up and a documented survivorship care plan with similar differences by rurality reinforces the importance to integrate late effects education into routine cancer care. The geographic distribution of suboptimal follow-up informs potential strategies to reengage survivors at greatest risk for late effects, particularly during the transition to adult care when the burden of serious adverse events increases.

Surprisingly, survivors from small town/rural areas had a similar likelihood as those from urban areas of receiving optimal subspecialty follow-up. Among survivors from large towns, no clear pattern emerged in the geographical distribution of survivors with suboptimal care, though the small population size over a large area was a limitation. This is an area for future work to include survivors of young adult cancers. Another limitation of this study was the lack of data on primary care visits in the community and, if care was delivered, whether this constituted survivorship-focused care. The shortage of primary care providers across the United States is well documented and evidence suggests worsening inequities in the workforce, especially in rural areas (47, 48). Survivors from rural areas may have a differential level of access to primary care or prefer to return to their primary oncologist for their survivorship care needs whereas those from large towns may prefer exclusive care from their primary care provider. Nevertheless, a lack of clarity in provider roles portends suboptimal survivorship-focused care as primary care providers often report unmet information needs on the late effects of cancer treatment (49). Indeed, the transition to primary care requires clear, ongoing communication and additional training of community providers concerning specific treatment-related late effects (50).

A team-based approach to survivorship-focused care integrates oncology, primary care, subspecialists, patients, and caregivers with the goal to optimize long-term health and outcomes for survivors (51). Shared care with oncology and primary care offers improved follow-up for survivors (52). Survivorship care plans, particularly those available digitally such as PFC, are an initial step to bridge gaps in guideline-adherent care, whether by subspecialists or primary care. Seventy-three percent of pediatric oncology survivors had a survivorship care plan in the Oklahoma Childhood Cancer Survivor Cohort, in contrast to other reports in the literature of only a minority receiving a survivorship care plan, which could facilitate care coordination (53). While an important starting point, optimal care is not guaranteed since primary care providers report additional barriers to the implementation of survivorship care plans (54, 55). Therefore, evidence-based interventions to promote team-based care in partnership with primary care as standard practice for survivors are essential to ensure the long-term quality of life for this vulnerable and unique population.

The COVID-19 pandemic spurred the explosion of telehealth applications across the medical field. Within pediatric oncology, the early disruption in long-term follow-up care of survivors and the high acceptability of telehealth during the pandemic illustrate the need for adaptable health systems (56, 57). Digital technologies offer innovative, evidence-based strategies to combat cancer care inequities, such as those observed among survivors from large towns in Oklahoma (58). The success of telementoring models, such as Extension for Community Healthcare Outcomes (ECHO), to increase local capacity for subspecialty care in limited resource settings serve as a template for survivorship care delivery among survivors far from tertiary care centers (59). Distance was strongly associated with optimal follow-up care in this study. Consequently, states with a large geographic area and limited subspecialty locations, such as Oklahoma, would benefit from approaches such as ECHO. Other digital interventions, such as mobile health, text messaging, and web-based platforms, challenge traditional models to engage survivors in their care (60).

Our study has many strengths, including the availability of an institutional cancer registry at a Commission on Cancer institution. This allowed us to use high-quality cancer data, coupled with EHR data, to evaluate factors related to childhood cancer survivorship. However, limitations include only having a single address at cancer diagnosis, which may not reflect residence during survivorship care, resulting in potential misclassification of rurality status. In a separate analysis of children diagnosed with leukemia between 2005 and 2019 at the Oklahoma Children's Hospital, approximately 9% of children moved during this time period, of whom 73% moved further away from the medical center (61). The use of the 5- to 7-year window after initial diagnosis as the early survivorship period may overlap with ongoing surveillance for disease recurrence by the primary oncologist and not represent survivorship-focused care. While 73% of general pediatric oncology survivors had a documented survivorship plan (n = 271/369), survivors from CNS tumors may have different recommendations for follow-up care and this merits further investigation. Moreover, because there are two children's hospitals providing treatment and survivorship care in the state, those residing in the northeastern part of the state may seek survivorship care at another institution. A manual chart review of all survivors without a completed clinic visit within the follow-up window excluded survivors with a documented transfer of care, including survivorship care, which helps ameliorate this bias. The use of a single institutional cancer registry, rather than a statewide cancer registry, represents another limitation and prompts additional investigation as to the representativeness of these data for the state of Oklahoma.

With regards to external validity, these observations in follow-up care may not be generalizable to rural states with earlier Medicaid expansion. This particularly impacts the identified disparities among adolescents at diagnosis who transitioned to young adulthood during the follow-up window, as insurance status is critical and has been shown to improve access to care for young adults (62). Oklahoma was a late adopter of Medicaid expansion with implementation in July 2021, which may have exacerbated these observed disparities (63). Fragmented healthcare systems across the United States highlight the need for access to services, data interoperability, and health information exchange to improve care, particularly for survivors from nonurban areas (64).

Approximately one fifth of survivors were not seen in an oncology-related subspecialty clinic during the early survivorship period 5 to 7 years after their initial diagnosis. Survivors from large towns, adolescents at diagnosis, and those further from the cancer center were significantly less likely to receive optimal follow-up. Primary diagnosis and late effects risk strata were also associated with optimal follow-up, which raises questions for further research to target at-risk groups. This analysis focused primarily on rurality. There is a clear need for continued, longitudinal surveillance of survivor cohorts from nonurban communities to assess inequities and develop health systems-based interventions tailored to the unique needs of this at-risk community. Clinical research informatics methods, such as those presented here, to integrate cancer registry, EHR, geospatial, and other data sources provide a framework for population health management of survivors at high risk for late effects.

No disclosures were reported.

D.H. Noyd: Conceptualization, data curation, formal analysis, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing. A.E. Janitz: Conceptualization, resources, software, formal analysis, investigation, visualization, methodology, writing–review and editing. A.A. Baker: Conceptualization, resources, data curation, writing–review and editing. W.H. Beasley: Data curation, project administration, writing–review and editing. N.C. Etzold: Resources, data curation, writing–review and editing. D.C. Kendrick: Conceptualization, resources, supervision, writing–review and editing. K.C. Oeffinger: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing.

D.H. Noyd was supported by the 2021 American Society of Clinical Oncology Conquer Cancer Foundation Young Investigator Award.

Each person listed as an author is aware of the content of the manuscript and has participated in the study to a significant extent. D.H. Noyd, A.E. Janitz, W.H. Beasley, and K.C. Oeffinger, developed the concept and drafted the manuscript.  D.H. Noyd, A.E. Janitz, W.H. Beasley, N.C. Etzold contributed to the data preparation, modeling, and prepared tables and figures. All other authors provided guidance on the methodology, reviewed the manuscript, and provided critical revisions.

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.