Introduction

Significant improvements in the probability of survival in children with cancer have resulted in a burgeoning population of long-term survivors. Many of the almost 400,000 childhood cancer survivors alive in the USA [1] are at risk for chronic morbidity and premature mortality as a consequence of their cancer treatment; 80 % of survivors will develop one or more severe or life-threatening chronic health conditions by the age of 45 years [2]. Consequently, it has been broadly accepted that survivors require life-long risk-adapted health care aimed at health promotion and periodic surveillance in order to prevent or mitigate the development of late effects [3, 4].

Several models have been proposed for the care of survivors [5]. Among these, dedicated care at a specialized survivor clinic (usually in a cancer center) or shared care between a survivor clinic and a primary care physician (PCP) [6] has been advocated, particularly for survivors at higher risk for developing late effects. In Ontario, Canada’s most populous province, the provincial government has funded a network of childhood cancer survivor clinics since 1999. Seven specialized clinics operate in the five cities that house the province’s pediatric cancer centers. In three cities, a single clinic provides care to both children and adults who have survived childhood cancer. In two cities, survivors are transferred at age 18 years from a clinic situated in a pediatric hospital to a clinic situated in a separate adult cancer center. These clinics are accessible to survivors at no cost through Ontario’s publically funded health care system. Although almost every eligible survivor attends such a clinic as a child or adolescent, many do not continue to attend these clinics once they reach adulthood. In the present study, we assessed the patterns and predictors of specialized survivor clinic attendance by adult survivors of childhood cancer.

Methods

Study population

After obtaining Research Ethics Board approval, we conducted a retrospective study of a population-based cohort of adult survivors of childhood cancer in Ontario. Eligible participants were identified in the Pediatric Oncology Group of Ontario Networked Information System (POGONIS), an active registry of all children and adolescents treated for cancer at any of the province’s five pediatric cancer centers. Patients were eligible for inclusion if they were an Ontario resident diagnosed with cancer prior to age 18 years between January 1st, 1986, and December 31, 2005, had survived ≥5 years from their most recent pediatric cancer event, had at least 1 year of follow-up as an adult (≥18 years), and had contributed follow-up time after January 1, 1999 (when the clinics were launched). The most recent pediatric cancer event was defined as the latest of the initial cancer diagnosis or any relapse or subsequent malignant neoplasm [SMN] that occurred prior to age 18 years. This allowed survivors of relapse or SMN during childhood to be included in the cohort. Subsequently, each survivor’s index date was defined as the latest of the date at which they reached 5 years from the most recent pediatric event and their 18th birthday. Patients were followed from their index date until relapse, SMN, death, or end of study on December 31, 2012 (whichever occurred first).

Demographic, diagnostic, and treatment data in POGONIS were linked deterministically (using unique provincial health card numbers which were then encoded) to demographic and health administrative databases (Registered Persons Database [RPDB], Canadian Institution for Health Information Discharge Abstract Database, Ontario Health Insurance Plan [OHIP], and the National Ambulatory Care Reporting System) housed at the Institute for Clinical Evaluative Sciences. These databases capture information on ambulatory and hospital-based medical care received in the province, vital statistics, and current patient addresses which are updated at the time of receipt of any hospital-based care.

Outcome definition

We used the OHIP physician billings database to capture outpatient medical visits. We identified visits to survivor clinics by compiling a list of the unique OHIP billing numbers of all physicians who provided care at a survivor clinic during the study period. Since many of these physicians also provide acute oncology or radiation oncology care, we created an algorithm to differentiate visits for cancer therapy (e.g., in the case of relapse or SMN as an adult) from survivor clinic visits. Survivors who had evidence of an administrative code for a hospitalization or outpatient visit related to receipt of chemotherapy, radiation or palliative care or who had had greater than four visits to a regional cancer center within a 60-day period were designated as having had a relapse or SMN and not classified as having had a survivor clinic visit if seen by one of the designated physicians within that time period.

Covariate definitions

Cancers were classified into eight groups: acute lymphoblastic leukemia (ALL), acute myeloid leukemia (AML) and other leukemias, lymphoma, brain tumors, neuroblastoma, renal tumors, bone and soft tissue sarcomas, and other tumors. The overall intensity of cancer therapy was classified using the Intensity of Treatment Rating Scale version 3 (ITR-3) [7] on a four-level scale with 4 representing the most intensive therapy. Cumulative alkylating agent doses were converted to cyclophosphamide-equivalent doses (CED) [8] and divided into four groups. Similarly, anthracycline doses were expressed as doxorubicin-equivalent doses [9] and divided into three groups. Radiation therapy was classified as none, brain only, chest only, brain and chest, or other. We created a binary variable to describe patients who had suffered a second event (relapse, SMN) prior to age 18 years. We also created a binary variable to distinguish between patients diagnosed before and after the launch of the survivor clinics in 1999. We examined socioeconomic status (SES) [10] and distance to a specialized clinic [11], both variables that have been demonstrated to impact access to specialized health care services. The postal code corresponding to each survivor’s primary residence was updated annually over the follow-up period. SES was defined using the Ontario Marginalization Index which provides neighborhood-level measures of deprivation and marginalization based on the 2006 Canadian census measures of residential stability, material deprivation, ethnic concentration, and dependency [12]. This variable was classified into quintiles, with survivors living in neighborhoods in the first quintile considered to be the least deprived. We calculated the straight-line distance from current place of residence to the nearest survivor clinic. We examined the impact of the receipt of care from a PCP by determining whether each survivor had visited a PCP for an annual assessment in each 1-year period. Finally, we classified the model of survivor clinic at each subject’s treating institution (combined pediatric/adult survivor clinic vs. separate pediatric and adult clinics).

Statistical analysis

We calculated proportions for categorical variables and medians for continuous variables. To investigate factors associated with the rate of clinic visits, we implemented an Andersen-Gill recurrent event multivariable regression model [13–15]. This model can be viewed as an extension of the Cox model that allows for repeated occurrences of a single type of event over time. Event times for each survivor were defined as the times from their index date to the dates of each clinic visit. A counting process data structure was created to implement the recurrent event model [15, 16]. Since the index date for some individuals was prior to the launch of the survivor clinics, discontinuous risk intervals were used to exclude the time from index to January 1, 1999, from the analysis [16]. A robust variance estimation approach was used to handle multiple visits by each survivor, and the Breslow method [15] was implemented to accommodate any ties in the clinic visit times across survivors. The p values for the regression estimates were obtained based on a Wald test using robust sandwich variance estimates [15].

The recurrent event regression model adjusted for all covariates listed above was conducted under a complete case analysis. Neighborhood SES, distance to the closest clinic, and annual physical exam were treated as time-varying covariates. All other covariates were treated as baseline (time-fixed) measures.

To illustrate the relationship between distance and clinic visits, we used a nonparametric estimation approach to compute the mean cumulative function [17]. We plotted the mean cumulative number of clinic visits over time for each category of distance to closest survivor clinic. Variables with p values less than 0.05 were considered to be significant. All analyses were completed using SAS 9.2 (Cary, NC).

Results

Of the 7842 children and adolescents diagnosed with cancer and treated at one of the province’s pediatric cancer centers between 1986 and 2005, 3972 met the criteria for study inclusion and 3912 (98.5 %) were successfully linked to their administrative health data (Fig. 1). Their demographic, disease, and treatment characteristics are described in Table 1.

Fig. 1
figure 1

Flow diagram of derivation of study cohort

Full size image
Table 1 Baseline characteristics (at the index date) of survivors who did and did not attend at least one survivor clinic over the follow-up period
Full size table

Survivors were followed for a median of 7.8 years (range 0.2–14.0) after their index date and contributed 32,029 person-years of follow-up. Only 1695/3912 (43.3 %) attended at least one survivor clinic visit as an adult. Among attendees, the median number of visits over the study period was 3 (range 1–30).

Table 2 displays the recurrent event multivariate regression model for clinic attendance. Statistically significant increased rates of attendance were associated with female gender; higher treatment intensity; radiation to the brain, chest, or other sites (compared to no radiation); higher alkylating agent exposure; lower SES; and an annual exam by a PCP. Decreased attendance was associated with a diagnosis of a brain tumor or other cancer (compared to ALL), diagnosis prior to the launch of the survivor clinics in 1999, treatment at a center with a combined pediatric/adult survivor program, and increasing distance from a survivor clinic. Survivors who lived >50 km from a clinic were significantly less likely to attend (Fig. 2). The mean straight-line distance to the nearest clinic was 84 km (standard deviation [SD] 294) among survivors who attended at least one clinic compared to 120 km (SD 384) among those that did not attend any clinics.

Table 2 Recurrent event multivariable regression model examining the predictors of survivor clinic attendance
Full size table
Fig. 2
figure 2

Cumulative mean survivor clinic visits over time according to distance from nearest survivor clinic

Full size image

Discussion

In this population-based cohort of childhood cancer survivors treated over a 20-year period, less than half attended a specialized survivor clinic as an adult despite access to these services at no cost as a function of Canada’s universal health care system. A survey of US survivors reported that with the introduction of the Patient Protection and Affordable Care Act, almost 90 % of respondents have health insurance [18]. Thus, lack of insurance should not be a barrier to clinic attendance for many survivors, although co-pays for medical visits and surveillance tests may be a barrier to low-income patients. Further, indirect costs (e.g., time off work, travel, etc.) may impact attendance, particularly since specialized survivor clinics are usually located in large urban centers. In the present study, distance from the closest clinic significantly impacted attendance—48 % of survivors living within 50 km of a clinic had at least one visit compared with only 35 % of those who lived more than 50 km away.

Several experts have advocated a tiered model of survivor care based on the intensity of cancer therapy and the consequent risk for late effects [4, 19, 20]. McCabe proposed a risk-stratified shared care model in which the survivor clinic provides direct care to the highest-risk survivors (e.g., those who have received high doses of alkylating or anthracycline agents, high-dose radiation, and allogeneic stem cell transplant and those who have persistent multi-organ toxicities of therapy) [20]. A PCP assumes primary responsibility for moderate- and low-risk patients, with the survivor program playing a supportive role. Although patients in the present study treated with more intensive therapy or with modalities known to increase the risk for late effects were more likely to attend a survivor clinic, only 41 % of survivors in the most intensively treated quartile had even a single clinic visit. Of particular concern was the observation that only 26 % of survivors of CNS tumors attended even a single appointment. Many of these survivors are at elevated risk for late effects (neurocognitive, endocrine, SMN, among others) as a consequence of cranial radiation or surgery, and some are unable to live or function independently as adults. Our findings mirror those observed in a recent publication describing specialized clinic attendance among survivors treated at single center in the USA [21]—in that cohort, only 7.2 % of CNS tumor survivors attended a specialized clinic, a proportion significantly lower that observed in other diagnostic groups. The reasons for this poor compliance with follow-up cannot be identified in the present analysis but require further investigation given the vulnerability of this cohort of survivors.

It is plausible that some survivors continue to receive care in the cancer center in an acute care oncology clinic (e.g., a stem cell transplant or brain tumor clinic), but these clinics frequently focus on surveillance for recurrence of the original cancer rather than the risks for late effects. Most survivors receive care only from a PCP. Unfortunately, surveys of North American family physicians [22] and US general internists [23] revealed that although approximately 85 % of PCPs are willing to care for childhood cancer survivors in collaboration with a cancer center, 63–77 % of respondents express discomfort with caring for this population. Only 1 % of family physicians and 6 % of general internists are willing to care for survivors independently. Knowledge regarding recommended surveillance is quite limited, with only 2 % of family physicians and 5 % of general internists being able to identify guideline-recommended cardiac, breast cancer, and thyroid function surveillance in a hypothetical survivor of childhood Hodgkin lymphoma. Survivors at increased risk for therapy-related cardiac dysfunction or breast cancer have been demonstrated to be less likely to receive guideline-recommended mammograms or echocardiograms if they are followed outside of a cancer center [24].

Data on the availability of specialized programs for adult survivors of childhood cancer in the USA is limited. A 2007 survey of 179 centers affiliated with the Children’s Oncology Group revealed that only half offered care to cancer survivors in a specialized program during their pediatric years. Forty-four percent of programs retained their survivors as adults without transition to a PCP. Similar to the survivors in our study, these adults are at risk for attrition from follow-up. Consequently, the barriers to providing appropriate care to adult survivors of childhood cancer appear to be threefold: (1) there is a paucity of specialized survivor clinics [25–27]; (2) even when such clinics exist, many survivors do not attend; and (3) most PCPs lack the comfort and knowledge to provide care to survivors, particularly in the absence of cancer center support [22, 23]. Efforts to improve PCP knowledge and comfort have included the creation of electronic survivor care plans that can be accessed by survivors and their PCP [28] and the development of a systematic approach to alternating visits between survivor clinics and PCPs [6].

Consistent with findings from other studies of access and use of health services [10], survivors with lower SES were less likely to attend specialized clinics. Among the least marginalized survivors, 48 % attended at least one clinic compared with 36 % of the most marginalized group. In adjusted models, females were 20 % more likely than males to attend clinics, a phenomenon also observed in other health care contexts [29]. As anticipated, survivors diagnosed prior to the launch of the clinic in 1999 were 27 % less likely to attend a clinic than those diagnosed more recently. Although survivors can “self-refer” to these clinics, direct transition from a pediatric survivor clinic or active oncology clinic understandably leads to better attendance. Further, survivors who self-refer as adults usually do so when they have developed a symptomatic late effect or when functional limitations interfere with normal activity or employment. At this point, the opportunity for the clinic to provide health promotion or preemptive surveillance and intervention has been lost. Strategies to reach survivors who graduated from their pediatric center before the launch of the survivor clinics or who attended a survivor clinic but subsequently dropped out have not been implemented in Ontario. Such strategies might include public service announcements, particularly those that target subgroups at high risk for late effects (e.g., breast cancer after chest radiation; cardiomyopathy after anthracycline therapy). Another approach would be to contact survivors directly using current address information available in provincial databases. Although this may raise concerns about privacy, a recent initiative in Canadian province of British Columbia aims to contact approximately 3400 adult survivors who have been lost to follow-up.

Given the inconsistency with which survivors attend these clinics and the general discomfort among PCPs, empowerment of the survivor to seek appropriate care (regardless of care location or provider) is critical. However, survivor knowledge about their diagnosis, treatment, and consequent risk for late effects is often inadequate [30, 31], creating a significant barrier to seeking appropriate care. Although provision of a treatment summary and survivor care plan that provides specific instructions regarding recommended long-term care and surveillance is now mandated by groups such as the American College of Surgeons Commission on Cancer [32], more focused interventions are necessary to ensure that survivors integrate these recommendations into their health-seeking behavior. For example, a randomized trial of the addition of a telephone encounter with an advanced practice nurse to a survivorship care plan demonstrated a twofold increase in compliance with cardiomyopathy screening in survivors treated with anthracycline chemotherapy [33]. Whether this improved compliance can be sustained over time and whether such interventions can be implemented in a cost-effective manner will require further research.

Several limitations must be considered when interpreting the results of this study. We did not assess survivor clinic attendance prior to age 18 years. However, attendance at specialized survivor clinics during childhood is almost 100 %, and consequently, we chose to focus on attendance as adults, when the risk of attrition rises. Since the outcomes were derived from administrative data and not direct patient assessment, we could not estimate the presence or absence of late effects that may drive care-seeking behavior. Knowledge of the types and intensity of cancer therapy allows for estimation of risk for late effects, but there is no way to know if survivors who visited the clinics had existing morbidities from their prior therapy. This study focuses on survivors in the most populous Canadian province where there is an existing network of survivor clinics and no cost for access, which may impact generalizability to other jurisdictions. However, the Ontario model serves as a paradigm for centering survivor care on a network of specialized clinics. Given that other countries have publicly funded health systems and many survivors in the USA have individual health insurance, the experience with attendance in the Ontario model should have direct relevance to survivor programs elsewhere. Finally, this study focuses on the patterns of attendance at specialized clinics and does not assess how clinic attendance impacts quality of care or long-term health outcomes. Ongoing research by our group is assessing the impact of clinic attendance on compliance with published surveillance guidelines and with the risk for requiring care in an emergency department.

In conclusion, a survivor care strategy that relies exclusively on attendance at a cancer center-based survivor clinic is unlikely to meet the needs of all adult survivors of childhood cancer. Models must be built that consider geographic distribution, the intensity of therapy and the risk for late effects, and access to PCPs willing to care for survivors. Developing tools and programs to facilitate the provision of care by PCPs and interventions to empower survivors to recognize their risks and to seek appropriate care will be critical for maximizing the long-term health outcomes of this growing and vulnerable population.