Neighborhood socioeconomic status and overall survival among children with acute lymphoblastic leukemia Open Access

Survival rates among children with acute lymphoblastic leukemia (ALL) have improved substantially in the last 5 decades, with present-day 5-year survival estimates exceeding 90% in the United States.^1,2 These gains have not been shared equitably among all children with ALL.^3-5 There is some indication that both individual-level and area-level disparities in socioeconomic status (SES) influence outcomes in the first 3 to 5 years after the diagnosis.^3,4,6 However, the impact of SES on children who have survived in the initial 2 to 5 years after diagnosis (ie, conditional survival) has not been evaluated. Childhood ALL has a long treatment duration (>2 years) to ensure sustained remissions,^2,7 making it imperative to understand the role of SES in influencing not only survival from diagnosis onward but also after the children have completed therapy. We addressed this gap using a US-based hospital registry from the National Cancer Database (NCDB)⁸ to evaluate the association between SES (as defined by neighborhood income and education) and overall survival among children with ALL from diagnosis and conditional on surviving 2 and 5 years after diagnosis. We hypothesized that lower SES would adversely influence both early and late mortality.

We leveraged the resources offered by the NCDB, a hospital-based, national data set maintained jointly by the American Cancer Society and the American College of Surgeons.⁸ Children and adolescents diagnosed with ALL between 2004 and 2019 at age 1 to 17 years were included. NCDB sites complete active follow-up (eg, contacting other facilities, reviewing obituaries) and passive follow-up (eg, state and federal data linkages) to determine vital status with ≥90% follow-up required for reporting sites. For the current analysis, a diagnosis of ALL was determined based on histology in accordance with the International Classification of Childhood Cancer.⁹ Those diagnosed at age <1 year (n = 557 [2.6%]) were excluded given the unique cancer biology, treatment regimens, and poor prognosis among patients diagnosed in this age group.¹⁰ Patients not receiving treatment at the reporting center (n = 542 [2.5%]) were excluded consistent with NCDB data analysis guidelines. Finally, 3454 children with ALL (16.9%) lacking key variables were also excluded. These included the following non-mutually exclusive categories: race (n = 335), time to treatment (n = 605), distance to care (n = 1975), rural/urban residence (n = 760), and information regarding area-level education (n = 2029) or area-level income (n = 2050). Figure 1 illustrates the CONSORT diagram and key exclusion criteria. Patients were followed for a minimum of 1 year, and follow-up ended in 2020.

Patient-level variables included age at ALL diagnosis, sex, race/ethnicity (non-Hispanic Black, non-Hispanic White, Hispanic, American Indian and Alaska Native, Asian, and other races/ethnicities), primary payer (private insurance, Medicaid, other/uninsured, and unknown), Charlson-Deyo comorbidity score (0 and ≥1), distance to care (treating facility, in miles), and rural/urban residence (using rural-urban continuum codes to categorize as metropolitan [metro] and non-metro/rural).¹¹ 

Area-level variables included median income and education from the American Community Surveys (ACSs; ACS 2012, ACS 2016, and ACS 2020) based on the values for the residential zip code at cancer diagnosis and divided into quartiles based on data from all US zip codes. Thus, between 2016 and 2020, the lowest median income quartile corresponded to a household income of <$46 277, quartile 2 ranged between $46 277 and $57 856, quartile 3 ranged between $57 857 and $74 062, and the highest median income quartile was >$74 063. Education quartiles measured the proportion of adults ≥25 years in the designated zip code lacking a high school degree. We applied ACS education and income quartiles based on year of cancer diagnosis. Area-level income and education quartiles were combined to more accurately reflect area-level SES as a composite metric.¹² 

Statistical analysis used the SAS program (version 9.4; SAS Institute Inc). The study outcome was overall survival and the primary exposure was area-level SES. Descriptive statistics were evaluated overall and by vital status. Kaplan-Meier and log-rank tests assessed overall survival by area-level SES. Cox proportional hazard modeling was used to quantify the association between area-level SES and all-cause mortality. The following variables were evaluated for inclusion in the multivariable Cox models: race/ethnicity, sex, age at ALL diagnosis, primary payer, rurality, distance to care, Charlson-Deyo score, and time to treatment initiation. Distance to care was evaluated as a 3-level variable (0-50 miles, 51-100 miles, and >100 miles). Owing to the relatively small proportion of patients who were covered with Medicare or other government health insurance plans, or were uninsured, these categories were combined, resulting in a 4-level primary payer variable (private insurance, Medicaid, other/uninsured, and unknown insurance status). Medicaid was maintained as a discrete category because of the income-based eligibility. All multivariable Cox proportional hazard models were adjusted for age at ALL diagnosis, year of diagnosis, and insurance status.^13-15 We further applied a backward stepwise elimination approach removing the variable with the largest P value >.05 sequentially. We further evaluated the models conditional on surviving the first 2 years and first 5 years from ALL diagnosis. Kaplan-Meier curves, log of negative log plots, Schoenfeld residuals, and Martingale-based residuals were used to assess the proportionality assumption.

Institutional review board approval is in place for this study at Prisma Health.

The cohort included 17 044 eligible children and adolescents with ALL. Results are reported consistent with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for cohort studies.¹⁶ Descriptive statistics for the entire cohort and by vital status are displayed in supplemental Table 1. The median age at diagnosis was 5 years (range, 1-17); 57% were males, 61% were non-Hispanic White, 56% were covered with private health insurance, and 36% with Medicaid. Only 5% of the cohort had a Charlson-Deyo score of ≥1. Most children (91%) started treatment within 1 week of diagnosis. Most patients (80%) lived within 50 miles of treatment, and only 7% lived >100 miles of treatment. Furthermore, most patients (84%) resided in metropolitan areas at diagnosis. Overall, 12% of the cohort lived in neighborhoods with the lowest SES and 18% lived in neighborhoods with the highest SES. Table 1 illustrates descriptive statistics by neighborhood SES. Compared with those in the highest SES neighborhoods, those in the lowest SES neighborhoods were more likely to be Hispanic (44.2% vs 7.7%) or non-Hispanic Black (16.4% vs 3.5%), more likely to have Medicaid (56.6% vs 15.2%), have Charlson-Deyo score ≥1 (6.8% vs 4.7%), and have delayed treatment start (11.4% vs 8.5%). Low neighborhood SES was also associated with living farther to treatment (51-100 miles, 18% vs 2.8%; >100 miles, 10.5% vs 2.7%) and in rural areas (25.7% vs 1.9%).

Overall survival varied significantly by area-level SES (Figure 2; P < .001), with those living in the highest SES neighborhoods having a significantly superior survival (91.0% at 14 years from diagnosis) when compared with those living in the lowest SES neighborhoods (83.6%). As found in Table 2, multivariable analysis revealed that when compared with those living in the highest SES neighborhood category, the hazard of all-cause mortality was 1.55-fold higher (95% confidence interval [CI], 1.29-1.88) for those in the lowest SES neighborhoods and 1.24-fold higher (95% CI, 1.07-1.44) for those living in the middle SES category. Each increase in age category at ALL diagnosis was associated with increased risk of death, following a clear monotonic trend (5-9 years [adjusted hazard ratio (aHR), 1.51; 95% CI, 1.31-1.73]; 10-14 years [aHR, 3.15; 95% CI, 2.76-3.59]; 15-17 years [aHR, 4.61; 95% CI, 4.01-5.29]; reference: 1-4 years). Other factors significantly associated with increased risk were nonprivate health insurance particularly for those insured by Medicaid (aHR, 2.28; 95% CI, 1.06-1.31; reference, private insurance) and non-White race/ethnicity (non-Hispanic Black [aHR, 1.37; 95% CI, 1.17-1.62]; Hispanic [aHR, 1.20; 95% CI, 1.06-1.35]; other race/ethnicities [aHR, 1.49; 95% CI, 1.04-2.12]; reference, non-Hispanic White). Presence of a comorbidity was associated with a 1.45-fold greater hazard of death (95% CI, 1.21-1.73; reference, none). Increasing calendar year of ALL diagnosis was associated with a lower hazard of all-cause mortality (aHR_{per_1_year}, 0.96; 95% CI, 0.94-0.97). Sex, time to treatment, rurality, and distance to care were not associated with overall survival.

Excluding 1637 patients due to death (n = 888) or censoring (n = 749) in the first 2 years from ALL diagnosis yielded a cohort of 15 407 children who had survived 2 or more years. Conditional on surviving the first 2 years after ALL diagnosis, area-level SES remained significantly associated with overall survival (Figure 3A; P < .001). Those in the highest SES neighborhoods had a 14-year overall survival of 94.9% compared with 90.4% in the lowest SES neighborhoods. As found in Table 2, multivariable analysis revealed that conditional on surviving 2 years, the hazard of all-cause mortality was 1.70-fold higher (95% CI, 1.28-2.26) and 1.37-fold higher for those living in the middle SES neighborhoods (95% CI, 1.10-1.70) when compared with those living in the highest SES neighborhoods.

Excluding 5972 patients (1411 deaths and 4561 censored during the first 5 years) yielded a cohort of 11 072 patients who had survived ≥5 years from ALL diagnosis. Conditional on surviving the first 5 years after ALL diagnosis, area-level SES remained significantly associated with overall survival (Figure 3B; P < .001). Among those in the highest SES neighborhoods, 14-year overall survival was 97.2% compared with 95.7% among those in the lowest SES neighborhoods. As found in Table 2, the hazard of all-cause mortality was 1.87-fold higher among those living in the lowest SES neighborhoods (95% CI, 1.17-3.00) and 1.37-fold higher (95% CI, 0.96-1.97) for those in the middle SES neighborhoods, when compared with those living in the highest SES neighborhoods. Figure 4 summarizes the increased hazard of mortality associated with low neighborhood SES overall, for the ≥ 2 year survivors, and the ≥ 5 year survivors. Univariable and multivariable models for all cohorts are found in supplemental Tables 2-4.

As found in Table 3, children from non-White racial/ethnic backgrounds (non-Hispanic Black [odds ratio (OR), 10.44; 95% CI, 8.27-13.19]; American Indian/Alaska Native [OR, 15.13; 95% CI, 6.20-36.90]; Hispanic [OR, 12.84; 95% CI, 10.87-15.17]; reference. non-Hispanic White), those with non-private insurance (Medicaid [OR, 8.98; 95% CI, 7.82-10.31]; other/uninsured [OR, 4.94; 95% CI, 3.77-6.48]; unknown [OR, 5.42; 95% CI, 3.71-7.92]; reference, private insurance), and those with comorbidities (Charlson-Deyo score of ≥1 [OR, 1.48; 95% CI, 1.16-1.88]; reference, none) and longer time to treatment (>1 week [OR, 1.40; 95% CI, 1.16-1.68]; reference, within 1 week) were more likely to reside in low-SES neighborhoods. Furthermore, those in low-SES neighborhoods were more likely to live further from care (51-100 miles [OR, 8.49; 95% CI, 6.65-10.83]; >100 miles [OR, 5.14; 95% CI, 3.95-6.69]) and in non-metro/rural areas (OR, 18.02; 95% CI, 13.64-23.81; reference, metro).

In this hospital-based cohort of 17 044 children and adolescents with ALL, we evaluated the association between area-level SES and overall survival from diagnosis to last follow-up and conditional on surviving the first 2 and 5 years from diagnosis. Consistent with previous reports in the United States and globally,⁵ we found that lower area-level SES was significantly associated with worse overall survival after childhood ALL diagnosis. A key novel finding in this study is that lower area-level SES remains a predictor of all-cause mortality conditional on surviving the first 2 and 5 years after ALL diagnosis.

Previous studies have evaluated socioeconomic disparities in individuals with adult-onset cancer. Unger et al¹⁷ evaluated the impact of socioeconomic deprivation (as measured by the area deprivation index [ADI]) among adults enrolled in clinical trials through the SWOG Cancer Research Network. This study identified disparities in overall survival, progression-free survival, and cancer-specific survival at 5 years with increasing risk of death as the ADI quintile increased. Because of socioeconomic disparities in survival among individuals who received uniform cancer treatment, the authors concluded that efforts to alleviate these disparities must include support beyond frontline treatment, including the early off-protocol time period and beyond.¹⁷ In childhood ALL, a US state–based cohort studied census-tract ADI measured at cancer diagnosis and found that mortality increased with increasing deprivation in a dose-response fashion; these survival disparities extended beyond 5 years from diagnosis.¹⁸ 

Our findings build on prior studies using a national cohort and illustrating the adverse impact of neighborhood SES even after completion of planned frontline treatment and conditional on surviving the first 2 and 5 years after ALL diagnosis. Conditional on surviving the first 2 years from ALL diagnosis, those in the lowest area-level SES neighborhoods had a 70% increased risk of death. Furthermore, and expanding on the work by Unger et al¹⁷ we identified survival disparities extending beyond the initial 5 years of follow-up. Among children who survived the first 5 years from ALL diagnosis, those living in the lowest SES neighborhoods had an 87% increased risk of late mortality when compared with those living in the highest SES neighborhoods. These associations were adjusted for known individual-level prognosticators (age at ALL diagnosis, sex, race/ethnicity, and payer status).

A single-center study found that children with ALL from high-poverty areas (based on residential zip code at cancer diagnosis) had an increased risk of early relapse (<36 months from ALL diagnosis).³ In the Children’s Oncology Group study AALL03N1, we found that household-level poverty (analyzed as 120% of the federal poverty level) was associated with an increased hazard of disease relapse.⁴ Relapse may have contributed to the observed disparities in this study. However, the association persisted among children who survived the first 2 and 5 years from ALL diagnosis; most disease recurrences would have occurred before the 5-year time point. Our findings highlight the relationship between area-level SES and early and late mortality for children with ALL.

In addition to neighborhood-level metrics, this study evaluated individual-level social determinants of health (SDoH), including insurance status and distance to care. Medicaid enrollment at cancer diagnosis was associated with an increased risk of death as reported previously among children with ALL in the population-based Surveillance, Epidemiology, and End Results Program,¹⁴ though the prior study did not evaluate area-level risk factors. Enrollment in Medicaid at cancer diagnosis may serve as a proxy measure of household income, reflecting household-level poverty at diagnosis which has been associated with increased risk of relapse for childhood ALL.⁴ Our current findings illustrate that both health insurance status and area-level SES were independently and significantly contributing to an increased risk of all-cause mortality among children with ALL. However, the association between insurance status and overall survival did not persist in the conditional survival analysis. This may be because insurance status was measured only at cancer diagnosis; we are, thus, unable to evaluate how insurance change or loss may affect outcomes among survivors of ALL.

State-based and national cohorts have described distance to care as an adverse risk factor for survival in children with cancer. Our previous study from Alabama had identified distance to care (per 20-mile increase) to be associated with increased risk of death among children with cancer, including ALL.¹⁹ Rotz et al²⁰ used NCDB (2004-2015) to evaluate children and young adults ≤39 years of age at diagnosis of ALL and found distance >50 miles to treatment and area-level income to be independently associated with inferior survival. Furthermore, older patients (≥18 years of age at diagnosis) were at an increased risk of all-cause mortality. In this study, children from low-SES neighborhoods were more likely to live farther from treatment. However, we did not identify an independent association between distance and overall survival in multivariable analysis when adjusting for area-level SES. Distance to care may confer the greatest risk for adolescents and young adults with ALL, which are underrepresented in this study (upper age limit 17 years) as compared with the previously published cohort by Rotz et al (upper age limit of 39 years). Other differences in this study include adjustment for area-level educational attainment and time to treatment. The mixed results regarding distance to care as an adverse risk factor suggest that the impact of distance to care on survival outcomes may be context specific and likely varies by patient characteristics, including age at cancer diagnosis.

The mechanisms driving excess mortality among ALL survivors from low-SES neighborhoods are likely multifactorial with possible mechanisms, including poor access to care, low engagement with survivorship programs, lack of focus on health promotion, and a higher burden of treatment-related chronic health conditions. Though cause of death is not available in this study, we hypothesize that comorbid conditions may contribute to an increased risk of death among survivors of childhood ALL. In our cohort, children with ALL from lower SES neighborhoods had 48% higher odds of having comorbidities present at diagnosis. The heightened risk of obesity specifically is well known among survivors of ALL.²¹ The presence of obesity can drive adverse cardiometabolic outcomes and has been linked to increase ALL relapse risk.²² A hospital-based cohort study in Texas identified more than twofold odds of obesity/overweight among ALL survivors by neighborhood deprivation, highlighting the role of community context in obesity risk for this population.²³ However, area-level deprivation is also associated with increased risk of child morality due to noncancer causes (eg, accidents, homicide).²⁴ Future research with respect to SES is warranted, specifically to prospectively evaluate cause of death and to include measurement of comorbid conditions by type at diagnosis and longitudinally.

Last, we identified a constellation of factors that were more prevalent among children with cancer residing in low area-level SES neighborhoods, including increased prevalence of comorbid condition at cancer diagnosis and of delayed treatment initiation (>1 week). Because SES-associated survival disparities persisted among those who survived the first 5 years from cancer diagnosis, it will be important to understand the types of comorbidities present at diagnosis and their role in long-term health outcomes in survivorship. Childhood ALL is generally treated promptly. The reasons for treatment delay among children from low-SES neighborhoods are unknown and warrant further study.

Our study needs to be considered in the context of its limitations. Although we include age at ALL diagnosis, other clinical prognosticators including cytogenetics and response to induction chemotherapy are lacking.¹⁵ However, we created subcohorts of children who survived the first 2 years and 5 years from diagnosis as an alternative approach to exclude those with refractory/relapsed disease leading to early death. The cause of death is lacking in the NCDB data set; hence, cause-specific mortality could not be evaluated. The upper age limit of our sample is 17 years yielding an underrepresentation of adolescents and young adults who may face significant challenges related to SDoH.²⁵ Area-level indicators smaller than zip code level (eg, census tract) are not provided by the NCDB, precluding the ability to evaluate smaller area-level metrics. Though the zip code level SES metrics used in this study are often more accessible for evaluation, they are less granular than individual or household measures of SES. Because of the clear associations between SES and outcomes among childhood ALL and other childhood cancers, ongoing prospective studies evaluating SES exposures in more detail are warranted. Last, we lack individual-level measurements for some variables, including income and education, which have been associated with ALL relapse,⁴ though Medicaid enrollment at ALL diagnosis was included as a proxy measure of individual-level SES.

Notwithstanding these limitations, this study evaluated a large, nationally representative sample of children with ALL using era-specific SES measures to evaluate all-cause mortality in the entire cohort from ALL diagnosis to last follow-up and conditional on surviving the early off-treatment period. The NCDB includes all US Commission on Cancer facilities. As a result, the data quality reflects Commission on Cancer standards of certification. Furthermore, as a US hospital–based registry, the NCDB has a larger sample size for childhood cancers as compared with other population-based registries, such as the Surveillance, Epidemiology, and End Results database.²⁶ Last, not all children with ALL are treated on clinical trials, and leveraging the hospital-based NCDB allowed evaluation of all patients, including those not enrolled or eligible for clinical trials.

Lower area-level SES was significantly associated with worse overall survival in this cohort of children with ALL in the United States, an association that persisted when conditioned on surviving the first 2 years and 5 years from cancer diagnosis. There has been increasing recognition of SES as a driver of persistent inequities within pediatric oncology and the need for ongoing efforts to address them.^27-29 The present findings illustrate that area-level SES-associated vulnerabilities persist during off-therapy follow-up among children with ALL. Longitudinal strategies to understand the underlying causes of these vulnerabilities and efforts to support at-risk patients beyond frontline therapy are warranted.

This work is supported, in part, by awards from Hyundai Hope On Wheels (A.L.H.) and St. Baldrick's Foundation (A.L.H.).

Contribution: A.L.H. contributed to the conceptualization, investigation, methodology, and writing of the original manuscript draft; D.M.H. contributed for formal analysis, methodology, and data visualization; R.G. and M.J.B. contributed to methodology and conceptualization; S.C. contributed to methodology and resources; A.J.A. contributed to methodology, investigation, and provided supervision; S.B. provided guidance on conceptualization, investigation, methodology, visualization, and provided supervision; and all authors participated in reviewing and editing the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Anna L. Hoppmann, Pediatrics, Prisma Health Children's Hospital, 7 Medical Park Dr, Columbia, SC 29203; email: anna.hoppmann@prismahealth.org.