4 | DISCUSSION
Nearly a third of survivors were not seen in a subspecialty clinic five to seven years after their initial diagnosis and there was insufficient evidence to support an association between risk strata for serious adverse health outcomes and likelihood of follow-up. This retrospective cohort study of survivors demonstrates the feasibility and utility of cancer registry data to construct a single-institution childhood cancer survivorship cohort. Furthermore, the integration of EHR and spatial data to expand analysis to include outcomes beyond mortality and disease recurrence, such as appropriate follow-up care, holds great promise to serve as a platform to enhance clinical oncology research with real-world data. The similar proportions of survivors seen in a subspecialty clinic between five and seven years, regardless of their risk stratification, is striking. Both the high proportion of patients not seen in clinic and the implementation of a validated risk-stratification system for survivors highlight the need for improved systems to enhance retention of patients for follow-up care.
While literature to support the benefits of long-term follow-up for survivors is abundant2,24,41 and guidelines inform clinicians on surveillance of late effect,11 a uniform definition of “appropriate” follow-up and “lost to” follow-up remains elusive. One limitation with the measurement of the primary “lost to follow-up” outcome variable in our cohort is whether or not patients seen in clinic were truly lost to follow-up or if they had moved or transferred care to another tertiary care center. On manual review of a random sample of thirty-seven patients lost to follow-up revealed only five had documentation of transfer to another institution. A CCSS analysis of 6,176 survivors showed approximately 40.3% self-reported survivor-focused care within the preceding two years at the baseline questionnaire, which then declined to 30.2% within at the most recent follow-up questionnaire.42 This markedly differs from the findings in our cohort with 67% of survivors with a subspecialty clinic visit between five and seven years after the initial date of diagnosis. The difference in these results may be due to a longer length of time since initial diagnosis at the time of the baseline questionnaire, with a mean of 17.5 years from diagnosis,43 and the exclusion of survivors <18 years in the CCSS compared to our cohort. Other single-institution25 and regional44studies for predictors of follow-up care reported higher rates of follow-up for younger patients and leukemia survivors, similar to our cohort, though definitions of appropriate follow-up were not uniform.
The opportunity to implement findings from larger cohort studies, such as the risk stratification system developed by the BCCSS,4-6,45 helps to frame the concept of “appropriate” follow-up care. The heterogeneity of survivors, based on their primary diagnosis and treatment exposures, merits careful consideration clinically on an individual level as well as a health systems level to optimize the follow-up care for this diverse population. Risk-adapted long-term follow-up care from recent European guidelines3 and the role of onco-primary care to facilitate transition of care for low, and potentially intermediate, risk survivors to primary care providers through the use of survivorship care plans offers a strategy to improve care. Indeed, the initial association between low risk survivors and increased likelihood of follow-up was attenuated by adjustment for primary diagnosis of ALL. Prioritization of efforts to target high risk survivors to re-engage them in subspecialty survivorship care is essential.
Routine follow-up in a survivorship clinic serves as an initial step; however, refinement of therapy-related exposures and ascertainment of adherence to guideline recommendations is critical to ensure timely delivery of appropriate care. Dichotomous exposures for chemotherapy, radiation, and surgery may lead to misclassification of patients into lower risk strata, despite receipt of high doses of chemotherapy known to significantly increase risk for late effects (i.e. cumulative anthracycline exposure and risks for cardiotoxicity).46 Over the past twenty years, significant progress in the risk prediction of late effects based on cumulative doses of specific chemotherapy agents and radiation20-23 calls for the inclusion of more granular exposure data to further sharpen risk stratification models. Successful de-escalation of therapy in the last several decades with sustained improvement in survival, such as leukemia and lymphoma,47 the observed reduction in the burden of late effects based on decade of treatment48, and the emergence of novel agents with unknown late effects7,8necessitate an adaptable system to incorporate more than simply dichotomous treatment exposures for risk stratification.
Potential disparities in appropriate follow-up care were identified in this analysis by race/ethnicity and SES. Specifically survivors who identified as black or Hispanic, were from locations with a higher area deprivation index, were of older age or lived further away from the primary treatment center or COG-affiliated site were less likely to receive follow-up care five to seven years after the initial diagnosis. Multivariable analyses were not executed for these associations, as the primary predictor focused on risk stratification, thus interpretation of these observed associations must be made with caution. Furthermore, the complex interactions of race/ethnicity and SES in general,49,50 and in childhood cancer survivorship research,51 are often difficult to disentangle. The development of potential interventions to reengage at-risk patients through an integrated implementation science and community-based participatory research approach could help ensure health equity for all survivors.
Future applications for biomedical informatics tools, such as the integration of cancer registry and EHR data, include extraction of cumulative chemotherapy exposures from the EHR as well as discrete data elements to assess adherence to guideline recommendations based on specific treatment exposures. Validation of new risk stratification models based on more granular data with large cohorts, such as the CCSS, may help bolster new models of care. Single-institution cohorts, particularly with lack of sufficient follow-up time since the widespread adoption of the modern EHR, may display limited power to detect rare events, yet may be sufficiently versatile to translate research through an implementation science approach. Furthermore, collaboration with other institutions within the NCDB with uniform cancer registry data and similar EHR platforms is feasible to maximize the impact of real-world data. The Cancer Moonshot prioritized the creation of innovative oncology data-sharing, the reduction of health disparities, and the identification of health-care delivery models to optimize care for survivors.52 The National Cancer Institute also recently launched the Childhood Cancer Data Initiative to foster collaboration between researchers and build an infrastructure to integrate data from multiple sources.53 This study provides a reproducible model to integrate cancer registry and EHR data to construct risk-adapted survivorship cohorts to assess follow-up care and aligns with national efforts to apply biomedical informatics methods to revolutionize clinical research and improve survivorship care.