Persistent Child Poverty and Mortality in a Cohort of Children with Cancer in Alabama

This article is featured in Highlights of This Issue, p. 287

Survival rates for children with cancer have improved dramatically over the last four decades; currently 84% of children diagnosed with cancer will survive 5 years and beyond (1). These improvements are due to risk-based treatment in the setting of clinical trials, accompanied by effective supportive care (2, 3). However, these improvements have not been shared equitably among all; social determinants of health (including individual-level poverty) contribute to the outcome disparities observed by children with cancer (4, 5).

Both individual-level and area-level poverty are associated with adverse outcomes among adults with cancer (6, 7). One consideration with regards to area-level poverty is its persistence over time (8). Counties with persistent poverty face unique risk factors when compared with counties without persistent poverty, such as greater unemployment, lower educational attainment, and a much younger population (9). In the United States, counties with persistent poverty have significantly higher adult cancer mortality rates when compared with counties without poverty, or with current (but not persistent) poverty (6). Observed disparities associated with persistent poverty are likely multifactorial including structural-, institutional-, and individual-level risk factors (6).

The United States Department of Agriculture (USDA) has an area-level persistent poverty measure specific to children, termed “persistent child poverty” which is defined as counties in which ≥20% of children in the county were poor since 1980 (8). This designation applies to 708 counties in the United States (one-fifth of all U.S. counties, but 49% of counties in the state of Alabama; ref. 8). Although this exposure is prevalent, the impact of a persistent child poverty environment on the outcomes experienced by children with cancer is unknown. Furthermore, possible mechanisms (e.g., access to care) through which persistent child poverty may impact outcomes in children with cancer have not been explored. In this study, we test the a priori hypothesis that exposure to a persistent child poverty environment is associated with inferior overall survival in children diagnosed with cancer in the state of Alabama.

Alabama is home to three Children's Hospitals, all Children's Oncology Group sites (Children's of Alabama, Huntsville Hospital for Women & Children, University of South Alabama Children's and Women's Hospital), and all available sites participated in the current study. Patients ≤19 years and diagnosed with cancer between 2000 and 2016 were included. All sites obtained local Institutional Review Board approval to submit their data as maintained for the state cancer registry. Patients lacking any of the following (not mutually exclusive) were excluded: cancer diagnosis code in accordance with the International Classification of Childhood Cancer (refs. 10, 11; N = 18, 0.9%), those lacking a vital status (N = 7, 0.3%), and those with residence outside of Alabama (N = 20, 1%).

Persistent child poverty was defined as outlined by the USDA as counties where ≥20% of related children <18y were poor as measured by three consecutive decennial censuses (1980, 1990, 2000) and the American Community Survey 5-year estimates (2007–2011; refs. 8, 12). Patients were classified as exposed to a persistent child poverty county based on the county of residence at cancer diagnosis using Federal Information Processing Standard county codes.

Sociodemographic (sex, race/ethnicity, insurance status) and clinical (date of diagnosis, age at diagnosis, cancer site, histology, grade and stage) characteristics were reported at the patient level by participating institutions. Primary payer at diagnosis was reported by participating institutions and categorized as: no insurance/public (uninsured, self-pay, and Medicaid), private, and insured-type unknown (those who had insurance at diagnosis but did not report the specific payer). Cancer type was determined using the International Classification of Childhood Cancer guidelines (relying on histology, primary site, and grade; refs. 10, 11). Cancer types were grouped together as appropriate when the sample size per individual cancer type was inadequate (Supplementary Table S1).

Because most persistent child poverty counties are in the southern half of Alabama (distant from the largest children's hospital in Jefferson County; Supplementary Fig. S1), distance to treating center (as a measure of health care access) was evaluated. Distance to treating center was calculated as a sphere distance (closest distance between two points on earth) measured between the center of the patient's county of residence and the center of the treating hospital's county. Distance was treated as a continuous variable and a categorical variable in quintiles (distance quintile in our cohort was ∼20 miles). Those living >100 miles had an increased risk of mortality in our preliminary analyses so we further dichotomized distance as >100 versus ≤100 miles.

Cause of death was provided by one participating center and categorized as follows: death from primary cancer, health-related cause of death (secondary malignant neoplasm, infection, cardiovascular, pulmonary, endocrine, neurologic, gastrointestinal, and liver causes), other/unknown, and external causes of death (accidents, homicides, suicides, etc.).

For our statistical analysis, we used SAS Version 9.4 (SAS Institute Inc.).

Our primary aim was to analyze the difference in all-cause mortality among children with cancer exposed to persistent child poverty environments versus those residing in counties not experiencing persistent child poverty (mutually exclusive groups) in the state of Alabama. Descriptive statistics were examined overall and by persistent child poverty county status. Overall survival by persistent child poverty status was estimated using the Kaplan–Meier method. Univariable Cox proportional hazards modeling was used to evaluate the association between all-cause mortality and patient characteristics (age, sex, race/ethnicity, insurance, diagnosis year, cancer type, and stage at diagnosis) including our exposure of interest (persistent child poverty county). Multivariable analysis was conducted by censoring the cohort at three different survival times (survival times of 1, 5, and 10 years from cancer diagnosis). When examining the association between persistent child poverty and hazard of mortality, Model 1 included sex, race/ethnicity, diagnosis year, stage, regardless of their significance level on univariate analysis. In addition, Model 1 retained any of the following variables with P < 0.001 on univariate analysis: age at cancer diagnosis, cancer type, and insurance. Model 2 added distance in miles to treating center (per 20-mile increase) to Model 1. Model 3 added a three-level combined variable of persistent child poverty and prolonged distance to care (>100 miles) with patients being exposed to both, either, or neither risk factor.

In subanalysis, cause-specific death was modeled using fine-gray subdistribution hazards models adjusted for race, sex, age, diagnosis year, insurance status, cancer type, and stage. We analyzed cause-specific mortality for “death from primary cancer” and “death from health-related causes.”

Data collected for this work are not public due to inclusion of protected health information. Data requests for additional evaluation of childhood cancer outcomes can be addressed to the corresponding author. A data use agreement would be required due to protected health information.

Our cohort included 2,089 children with cancer. On the basis of aggregate data from the Alabama Statewide Cancer Registry over the same time period, our cohort captured 70% of all children (N = 2,984) diagnosed with cancer in the state of Alabama. Data are reported as an aggregate (of the three sites) with Children's of Alabama contributing 86% of the patients; Huntsville Hospital for Women & Children and University of South Alabama Children's and Women's Hospital contributed 9% and 5%, respectively. Clinical and sociodemographic characteristics are described in Table 1. The median age at cancer diagnosis was 7 years (range, 0–20 years). The cohort was comprised of 54% males, and 68% non-Hispanic Whites, 27% African Americans, 4% Hispanics, and 2% from other racial/ethnic backgrounds. The most common insurance type was private (49%), followed by public/no insurance (43%) and insured type unknown (8.5%). The cancer diagnoses included central nervous system (CNS) tumors (23%), precursor cell leukemias (23%), bone tumors/sarcomas (12%), and other diagnoses (42%). Approximately, one-third (32%) of the sample had regional or metastatic disease at diagnosis. With a median follow-up was 6 years (range, 0–21 years), we observed 493 deaths; 183 in the first year, 423 by 5 years, 469 by 10 years from diagnosis, and 24 beyond 10 years. The median distance to treating center was 60 miles (range, 0–260); 23% of the cohort lived >100 miles to the respective treating center.

Forty-two percent of children were exposed to a persistent child poverty environment at cancer diagnosis. As shown in Table 1, a higher proportion of African Americans (42% vs. 15%, P < 0.0001) and lower proportion of Hispanic (2.7% vs. 5.1%) and non-Hispanic White patients (54% vs. 78%) lived in persistent child poverty counties. There was no difference in the proportion of nonmetro counties by poverty designation (P = 0.06), though persistent child poverty counties were more likely to be designated as a completely rural county (4.4% vs. 2.1%, P = 0.002). Those in persistent child poverty counties were more likely to have public/no insurance (48% vs. 39%, P = 0.0009). Furthermore, a higher proportion of children in persistent child poverty counties lived >100 miles to treating center (38% vs. 12%, P < 0.0001). There was no difference in sex, age at cancer diagnosis, year of diagnosis, or cancer type by county poverty status.

The overall survival was significantly greater among children living in counties without a persistent child poverty designation (Fig. 1). Table 2 reports the univariable and multivariable associations between covariates and all-cause mortality. Missing data values (listed below Table 1) were considered as a separate level for that variable in the models. There was no significant difference in survival by persistent child poverty designation at a survival time of 1 year [HR = 1.05; 95% confidence interval (CI) = 0.77–1.43; P = 0.78]. Children exposed to a persistent child poverty environment were 30% more likely to die by 5 years (HR = 1.30; 95% CI = 1.06–1.59; P = 0.012) after adjusting for race/ethnicity, sex and age at cancer diagnosis, diagnosis year, insurance, cancer type, and stage (Model 1). The risk of mortality at 10 years was similar to the risk seen at the 5 years (HR = 1.33; 95% CI = 1.10–1.62; P = 0.003).

Model 2 (Table 3) added distance (per 20-mile increase) to Model 1. There was no interaction between distance and persistent child poverty county status (P = 0.93) and patients were evenly distributed over the distance quintiles (Supplementary Fig. S2). At all survival times, longer distance to treatment was associated with an increased risk of mortality (5 years: HR = 1.09; 95% CI = 1.05–1.14; P < 0.0001). There was no difference in the risk of mortality by distance to treating center at the 10-year survival time as compared with the 5-year survival time. Adding distance to care into the model (per 20-mile increase to cancer treatment) reduced the HR of living in a persistent child poverty county from 1.30 to 1.27 (95% CI = 1.04–1.56; P = 0.02).

As shown in Model 3 (Table 3), patients who had the dual exposure of living in a persistent child poverty environment and distance to treating center >100 miles had the highest risk of mortality at 5 years (HR = 1.80; 95% CI = 1.39–2.33; P < 0.0001) compared with patients with neither exposure. Patients with either exposure (distance or persistent child poverty) had a 30% increased risk of mortality at 5 years (HR = 1.30; 95% CI = 1.04–1.63; P = 0.021) compared with patients with neither exposure. At 1 year, there was not a significantly increased risk of mortality for children with either or both exposure (as compared with neither), while at 10 years there was a statistically significant increased risk of mortality for both groups, similar in magnitude as compared to the 5-year survival time.

With regards to cause-specific mortality, our largest center contributed 1,361 patients and 438 deaths (317 from primary cancer, 53 from health related causes, 58 from other/unknown, and 10 from external causes). Those in persistent child poverty counties were more likely to die from their primary cancer at 5 years (HR = 1.33; 95% CI = 1.05–1.70; P = 0.02) and 10 years (HR = 1.37; 95% CI, 1.09–1.73; P = 0.008). Those in persistent child poverty counties were as likely to die of health-related causes as those in nonpoverty counties (HR = 1.07; 95% CI = 0.59–1.94; P = 0.83).

Unlike exposure to poverty at the individual level, persistent poverty examines poverty in the context of the community in which the individual lives. Persistent poverty is a longitudinal marker that reflects enduring community level poverty for over 30 years. The mechanisms underlying observed cancer disparities in persistent poverty communities are not well understood and likely reflect multilevel risk factors at the structural, institutional, and individual levels.

In our cohort of children with cancer from the state of Alabama, exposure to a persistent child poverty environment was prevalent (>40% of our sample), and children in these areas were more likely to be African American, have public or no health insurance, and lived >100 miles to their cancer treatment center. We found that children in persistent child poverty counties were 30% more likely to die by 5 years from their cancer diagnosis as compared with children from nonpoverty counties. In subanalysis, of cause-specific mortality, children from persistent child poverty counties were more likely to die from their primary cancer but as likely to die from health-related causes as those not living in persistent poverty counties. Children with the dual exposure of distance >100 miles and persistent child poverty environment were 80% more likely to die by 5 years from their cancer diagnosis than children with neither exposure.

Exposure to a persistent poverty environment has recently been shown to be associated with higher cancer mortality among adults (6). Possible mechanisms for this finding among adults are likely multifactorial and may include risk factors at the structural level (reduced access to care including lower rates of cancer screening), institutional level, and the individual level (higher rates of cancer risk behaviors including smoking, obesity (13–15). Compared with persistent poverty, persistent child poverty is a much more prevalent exposure affecting 23% of all counties in the United States, mostly in the South (8). However, the impact of persistent child poverty on children with cancer has not been examined.

Persistence is an important dimension because it describes a community level transgenerational exposure of poverty that has endured over time (8). In contrast to communities that may go into and out of poverty overtime, these areas have had a high proportion of residents that remain in poverty over the last 30 years. Persistent poverty counties (adult measure) have unique characteristics as compared with nonpoverty counties including having greater unemployment, lower educational attainment, and a higher proportion of the population <18 years of age (9). Though persistent poverty counties have a higher adult cancer mortality than nonpoverty counties (6), the mechanisms behind this finding remain poorly understood.

Even less is known about persistent child poverty counties (8), though the characteristics of these communities warrant further exploration, particularly given that one-fifth of all U.S. counties are classified as having persistent child poverty. Our analyses show that in Alabama, children with cancer in persistent child poverty counties are more likely to have public/no insurance, and live distant to their cancer treatment center. As such, children living in persistent child poverty counties may be more likely to have multiple simultaneous risk factors for poor cancer outcomes.

Children from persistent child poverty counties were also more likely to be African American compared with children in nonpoverty counties. There are several publications showing that among African American adults with cancer, living areas with high residential segregation (a marker for structural racism), is associated with inferior outcomes (16–18). On the basis of our findings, specific structural barriers and challenges faced by African American children in persistent child poverty counties warrant further exploration.

By censoring our cohort at different survival times (1, 5, and 10 years from cancer diagnosis), we are able to generate hypotheses for possible mechanisms underlying the observed disparity in overall survival. At 1 year from cancer diagnosis, there was no difference in survival between those in persistent child poverty counties versus nonpoverty counties, suggesting that acute treatment-related toxicities are not the mechanism for the association. After adjusting for other prognosticators, children in our cohort were 30% more likely to die by 5 years after their cancer diagnosis if they resided in a persistent child poverty county. In subanalysis of one center, children from persistent child poverty counties were more likely to die from their primary cancer. One hypothesis for these findings is reduced access to cancer care among children in persistent child poverty environments, which we explored by examining distance to cancer treatment center. Prolonged distances to cancer treatment centers may place a repetitive burden on families and compromise treatment, follow up, and appropriate cancer surveillance over time.

Most persistent child poverty counties in Alabama are located in the southern part of the state, distant from the largest children's hospital in Jefferson County. Though there was no interaction between persistent child poverty and distance to care, a higher proportion of children in persistent child poverty counties lived >100 miles to their cancer treatment center. Distances to treatment center of greater than 50 miles have been previously described as a risk factor for mortality among children and young adults with acute lymphoblastic leukemia (19), though the impact of distance to care on other cancer types and the impact of multiple overlapping exposures (i.e., distance and poverty) among children with cancer are not well understood. In our cohort, distance (per 20-mile increase to treatment center) was independently associated with an increased risk of mortality, and those with the dual exposure of a persistent child poverty environment and distance >100 miles to treatment were at highest risk of death.

Other pathways through which a persistent child poverty environment may negatively impact children at the individual level warrant further exploration. Numerous studies have described the negative physiologic impact of living in a low socioeconomic status (SES) environment on child health. Lower levels of SES are associated with higher levels of obesity, sedentary lifestyle, and increased levels of chronic stress among children (20, 21). Among children with cancer, there may be physiologic differences secondary to the stress of living in a high poverty environment that could impact their response and resiliency to cancer treatment.

It is also important to consider the compound effect of individual (household) risk factors combined with a persistent poverty environment in future studies. For example, housing insecurity, food insecurity, and financial toxicity are commonly described among families undergoing childhood cancer treatment and in survivorship (22, 23). Persistent poverty communities may lack the infrastructure and support systems to assist patients and families who face these individual issues while undergoing cancer treatment. Individual household poverty may pose a particular risk to those in persistent poverty communities. Parents and caregivers with household poverty may lack the resources needed to adhere to cancer treatment over time (24), and the burden of seeking cancer treatment (i.e., lost work, lost wages—particularly among those traveling long distances) may be insurmountable without additional community safety nets in place (22, 23).

Not all participating centers reported cause of death, which is a limitation of the current study. By censoring the cohort at various survival times, we can generate hypotheses regarding mechanisms for our findings, and we were able to complete a cause-specific death analysis for the largest contributing center. Though most children with cancer are treated on clinical trials (3), specific treatment information was not available for analyses, nor were patient-level comorbidities or functional status. Though stage at diagnosis was included as a prognosticator in our analysis, prognostic markers specific for childhood leukemia (such as the NCI risk group) were not able to be calculated with the available data, thus was not included. We did not examine cancer-specific models due to limited sample size of the individual cancer types. However, an examination by cancer type is warranted in future studies to understand how the exposure of persistent child poverty and distance to care may differentially impact different cancer types. Without residential address, more specific measures of healthcare access, including patient commute time, were not able to be analyzed. Because household level poverty is not routinely collected for cancer registry reporting, it was not available in the current study, though we were able to include Medicaid status (as a proxy measure of income) in our analysis. Given the smaller sample sizes from two contributing institutions, outcomes were not analyzed by individual treatment center. Finally, our sample mirrors the racial/ethnic make-up of the population in Alabama (25); however, compared with the United States as a whole, our sample has relatively fewer Hispanic and Asian patients, and a greater proportion of African Americans.

The current study has numerous strengths including a cohort that is large and racially diverse, with data from all children's hospitals in the state representing most children diagnosed with cancer in Alabama over the observation period. Our analyses report findings regarding an underrepresented group of children in the literature, children with cancer from the Deep South, and include a high proportion of children exposed to persistent child poverty environments and living physically distant from the treating center.

In conclusion, children with cancer living in persistent child poverty counties in Alabama face an increased risk of mortality at 5 and 10 years as compared with children in nonpoverty counties. This risk was present regardless of their distance to treatment (both children near to and far from treatment centers faced an increased risk of mortality if they lived in a persistent child poverty county). However, distance to care was independently associated with inferior survival, and children with the dual exposure of prolonged distance to care and a persistent child poverty environment faced the highest risk of mortality. In subanalysis, children in persistent child poverty counties were more likely to die from their primary cancer and at equal risk for mortality from health-related causes. Future research with longer follow-up (beyond 10 years) is warranted to investigate health-related mortality in this vulnerable group of children given the long latency of health-related complications following childhood cancer diagnosis.

In contrast to the concept of poverty, persistent poverty spans generations, and reflects enduring deprivation in a community for over 30 years. To mitigate disparities, further etiologic research is urgently needed to understand the multilevel causes of cancer disparities faced by children in persistent child poverty communities. Understanding structural issues, as well as institutional policies and practices, and individual risk factors that contribute to ongoing disparities in these communities will be critical to inform future interventions and policy changes. Improved access to care is of paramount importance in improving cancer-related outcomes for this vulnerable population in Alabama; strategies including expansion of telehealth and satellite clinic systems are under exploration.

A.L. Hoppmann reports grants from NCI during the conduct of the study. K. Kenzik reports grants from American Cancer Society during the conduct of the study. No disclosures were reported by the other authors.

A.L. Hoppmann: Conceptualization, investigation, methodology, writing–original draft. C. Dai: Formal analysis, methodology. M. Daves: Data curation, writing–review and editing. H. Imran: Data curation, writing–review and editing. K. Whelan: Writing-review and editing. K. Kenzik: Formal analysis, supervision, methodology, writing–review and editing. S. Bhatia: Conceptualization, resources, supervision, methodology, writing–review and editing.

This work was supported by the NIH, NCI (CA 047888, to A.L. Hoppmann); and by a Mentored Research Scholar Grants in Applied and Clinical Research from the American Cancer Society (MRSG-18-020-01-CPPB, to K. Kenzik).

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.