Sociodemographic and Socioeconomic Factors Correlate with Late-Stage Pediatric Hodgkin Lymphoma and Rhabdomyosarcoma: A Report from the Children’s Oncology Group Registries

Abstract

Background:

We examined association between late stage diagnosis and individual- and community-level characteristics among pediatric Hodgkin lymphoma (HL) and rhabdomyosarcoma (RMS) patients.

Methods:

We obtained Children’s Oncology Group (COG) data from 1999–2021 including summary stage (local (L), regional (R), distant (D)), tumor subtype, demographics, and ZIP code at diagnosis. We linked ZIP codes to county-level redlining scores (C,D=greatest redlining), Child Opportunity Index (COI), and measures of segregation (racial dissimilarity indices (DI)). Logistic regressions calculated odds ratios for late stage diagnosis, and by race within tumor subtype.

Results:

5,933 HL and 2,800 RMS patients were included. Late stage diagnosis of HL was correlated with Black race (OR_{Distant(D) vs regional/local(R&L)}=1.38 [1.13–1.68]), being uninsured (OR_{D vs R&L}=1.38 [1.09–1.75]), and subtype (Nodular sclerosis vs Other HL: OR_{D vs R&L}=1.64 [1.34–2.01], Untyped: OR_{D vs R&L}=1.30 [1.04–1.63]). Late stage rhabdomyosarcoma was correlated with bilingual households (OR_{Distant/regional(D&R) vs local(L)}=2.66 [1.03–6.91]) and tumor type (Alveolar vs Embryonal OR_{D vs R&L}=6.16 [5.00–7.58]. Community-level factors associated with late stage HL were greater Black (OR_80–100%=1.83; 95% CI=1.11–3.02) and Hispanic (OR_60–79%=1.30; 95%CI=1.05–1.60) DI. Late stage diagnosis for RMS was associated with more redlined census tracts within counties (OR=1.54; 95% CI =1.02–2.35) and low/very low COI (OR=1.21; 95% CI=1.02–1.45).

Conclusion:

Novel markers of community deprivation, such as redlining and racial segregation, likely affect cancer outcomes for children with HL and RMS in this first disparities study using COG registries.

Impact:

The interplay of multilevel risk factors provides important consideration for efforts to improve early detection of pediatric cancer diagnosis.

Introduction

Improved survival of childhood cancers is one of the major successes in cancer.(1) However, these advancements have not been equitable among racial and ethnic minority patients.(2–5) The extent of disease at diagnosis – i.e., stage in solid tumors – has crucial implications on disease survival. Advanced disease at diagnosis is more prevalent among racial and ethnic minorities.(6,7) Tumor subtype has been proposed as an explanatory factor for racial and ethnic disparities in cancer outcomes,(8) but individual socioeconomic risk factors (e.g. low income, lack of health insurance) also increase risk for late stage diagnosis.(1,5,8) While these individual risk factors are important, the effect of community level risk factors on racial/ethnic disparities in pediatric cancer outcomes is not well understood.

Communities are shaped by policies and/or practices that can deprive certain populations of access to material, economic, educational, healthcare, or social resources. Systematic resource deprivation is particularly apparent in minority neighborhoods that were denied financial services because of having “socially undesirable” characteristics (i.e., “redlining”). Redlining led to racial and economic segregation that continues to impact neighborhoods today and has been associated with multiple adverse health outcomes, including increased risk of cancer and cancer mortality among adults.(9–11) In addition to redlining, multidimensional indices built from census data can identify small geographic areas at risk for poor health outcomes stemming from a lack of social, economic, and material resources.(8,12)

Yet, communities can also exert protective effects and promote positive health outcomes. For example, ethnic enclaves are geographic areas where ethnic groups are spatially clustered and socially and economically distinct from the majority group.(13) The shared culture and language within ethnic enclaves can nurture social ties that facilitate knowledge sharing about health systems, contact with community health navigators, and general health information. We are not aware of any studies that have considered the impact of historic redlining or ethnic enclaves on risk for late stage pediatric cancer diagnosis.

Hodgkin lymphoma (HL) and rhabdomyosarcoma (RMS) are two common childhood cancers with earlier evidence that socioeconomic factors are associated with both delays in diagnosis and advanced stage at diagnosis, with less known about community effects. (14–21) To date, most of the community-level analyses with pediatric HL and RMS focus on SEER data, which provides limited ability to link to novel area level socioeconomic status data. Thus, this cross-sectional study uses cohorts of pediatric HL and RMS patients created from the national Children’s Oncology Group registries to examine the effect of multiple sociodemographic risk factors on late stage diagnosis for the first time using these national resources. We expand the literature by investigating multiple individual and community-level factors (e.g. redlining, health professional shortage areas). Further, we examine the interplay of patient race/ethnicity with tumor subtype and community-level factors.

Materials and Methods

The National Cancer Institute-sponsored Children’s Oncology Group (COG) is the world’s largest organization devoted exclusively to pediatric and adolescent cancer research (22). COG conducts clinical and translational research trials for pediatric cancers at over 200 centers in the United States (U.S.) and globally. Starting in 2007, all patients at COG-affiliated institutions were approached and given the opportunity to participate in the Childhood Cancer Research Network (CCRN; ACCRN07) patient registry. The CCRN registry had a 96% patient participation rate. In 2015, Project:EveryChild (PEC; APEC14B1) replaced CCRN as the primary COG registry with voluntary enrollment with participation ongoing. Together, the CCRN and PEC registries represent the largest studies of childhood cancer patients in the nation.(23,24)

COG provided data on pediatric HL and RMS, as well as retinoblastoma and germ cell patients, enrolled in CCRN, PEC, and COG trials for those diseases (Supplemental Figure 1). All data sources utilized patient-specific identifiers which enabled us to link across registry and trial. The data included patient characteristics, stage, tumor classification information (e.g. ICD site, histology code), state, city, and ZIP code. We included patients with residence in the U.S. including Puerto Rico by validating if a patient’s reported ZIP code or city was located in their state of residence.

Outcome: Late stage diagnosis

Summary stage (local, regional, distant) was operationalized as an ordinal outcome for the regression models and is described further in the Statistical Methods.

Patient and family level risk factors

COG provided patient sex, Hispanic ethnicity, race, birthdate, enrollment date, and insurance status at enrollment. Race and ethnicity were obtained from each institutional electronic health record. Based on the data distribution, we categorized race as White, Black, Asian, and other races (American Indian/Alaska Native, multiracial, Pacific Islander, Race NOS, Other). We also classified persons as Hispanic or Not Hispanic. For the stratified analyses, race was classified as Non-Hispanic White, Non-Hispanic Black, Asian, and Hispanic (all races). Diagnosis age was computed as the difference between birthdate and diagnosis date.

We identified tumor subtype using histology codes and text string searches. We defined HL subtypes as nodular sclerosis (9661/3, 9662/3, 9663/3, 9664/3, 9663/3, 9665/3, 9666/3, 9667/3) or Other defined as lymphocyte depleted, enriched, and nodular lymphocyte-predominant (9651/3, 9651/3, 9653/3, 9654/3, 9655/3, 9653/3). All other tumors were classified as “not typed”. RMS were defined as alveolar (8920/3), embryonal (8910/3, 8912/3), or “not typed.”

Language(s) spoken at home was reported at enrollment (grouped as patients whose caregivers reported speaking English only, another language along with English, and no record of English as a preferred language). Health insurance was classified as: private, military, or employer-based insurance; public or national insurance (e.g. Medicaid, California Children’s services); no insurance, self-pay, or other/unknown insurance.

Community level socioeconomic risk factors

We verified patient-reported ZIP codes using data from the U.S. Department of Housing and Urban Development (HUD); 83% of ZIP codes were verified by HUD. The unverified ZIP codes were missing (n=1,421, 16%) or incomplete (n=84, 1%). We linked the HUD-verified ZIP codes with ZIP Code Tabulation Areas (ZCTAs) from the 2010 Census, which was the temporal midpoint of the enrollment period. The ZCTAs were then classified into counties based on where the majority of the ZCTA population was located (we used ≥71% of ZCTA population in one county as a threshold); 97% of ZCTAs were overwhelmingly located in a single county. We excluded patients diagnosed from 1990 to 2004 from the community level analyses based on data availability and to reduce error from misclassification due to changes in ZIP codes over time.

We used multiple unidimensional measures to describe community level socioeconomic risk factors. We identified patients as living in metropolitan (primary commute of residents within an urban area, ≥50,000 residents, or with >10% of commuters traveling to urban areas), micropolitan (primary commute within an urban cluster,10,000 to 49,999 individuals, or with >10% of commuters traveling to urban clusters), or rural areas and small towns (primary commute to an area with <10,000 individuals) by linking ZIP codes to Rural-Urban Commuting Area (RUCA) codes. We estimated access to COG centers by querying Google Maps for the driving time between the population-weighted centroid of each ZCTA and the nearest COG center. Driving time was categorized into 2–59 minutes, 60–119 minutes, and 120–1,900 minutes; persons who required flights were excluded. Access to health care identified which patients lived in an area that was ever designated as a health professional shortage area.

We obtained redlining data from the original 1930’s Home Owners’ Loan Corporation (HOLC) redlining maps for 2010 census tracts (https://ncrc.org/redlining-score/). The HOLC scoring system rated census tracts with >20% of the land area that was graded on a 1 to 4 scale, with 1 indicating the most “desirable tracts” and 4 as the “least desirable.” Tracts in metro areas that did not meet the 20% threshold for HOLC-graded areas were assigned a “0” value. Less than 10% of HL (n=610) and RMS (n=281) patients had ZCTAs that linked to a graded census tract. Because of this, we determined that averaging redlining scores by county was necessary because of the questions about the validity of ZCTA-linkage and to ensure adequate power for the analysis. We aggregated the redlining scores from census tracts into counties, calculated the population-weighted county average redlining score, and linked the counties to the ZCTAs. We classified counties with rounded averages of 1–2 as “A or B” indicating the most desirable counties, and counties with rounded averages of 3–4 as “C or D” indicating counties with the greatest degree of redlining. “Ungraded” counties had a score of 0.

We identified ethnic enclaves using racial segregation indices from 2010 for metropolitan statistical areas (MSA) with populations greater than 500,000 persons. (25) Segregation was measured using a “dissimilarity index” (DI) ranging from 0% (complete integration) to 100% (complete segregation).(26) The value indicates the percentage of the minority group that needs to move to be evenly distributed across an entire MSA. We linked the metropolitan indices to our data by ZIP code. We obtained separate segregation indices for Asian, Black, and Hispanic racial and ethnic groups and divided them into quintiles.

The Child Opportunity Index (COI) is a multidimensional measure of the quality of neighborhood resources tailored to children’s developmental needs (https://www.diversitydatakids.org/child-opportunity-index).(27) COI measured in 2010 was available by census tracts while COI measured in 2015 was available by ZIP code. For patients enrolled between 2005 and 2012, we linked ZIP codes to the 2010 census tract COI and calculated population-weighted averages for each ZIP code. For patients enrolled from 2013–2021, we directly linked to the 2015 COI data by ZIP code. We categorized national COI overall based on predetermined cutoff points (Very low/Low, Moderate, Very High/High).(27)

Statistical Methods

We explored percent differences in late stage diagnosis by socioeconomic risk factors. As some tumor subtypes may be more prevalent in certain racial groups, we identified significant differences in the distribution of HL and RMS subtypes by race using the Chi-square test.

We computed odds of late stage diagnosis and individual and family level risk factors using ordinal logistic regression, with local tumors as the referent outcome and distant tumors as the most severe outcome. To account for non-proportionality of the odds for individual level sociodemographic variables, we compared the model fit across general ordinal models, fully nonproportional models, and partial proportional odds models (PPO; https://documentation.sas.com/doc/en/pgmsascdc/v_023/statug/statug_logistic_examples22.htm) (Supplemental Table 1).(28) For HL, sex, enrollment year, race, age at diagnosis, insurance status, and subtype violated the proportional odds assumption. For RMS, enrollment year, language spoken at home, and subtype violated the proportional odds assumption. For these models, the outcomes were reported as logistic models comparing Distant vs Regional & Local stage (referent) and Distant & Regional vs Local. Sex and enrollment year were run in a single multivariable model. Up to 35% of data were missing for individual demographic variables; thus, a single multivariable model that included all individual demographic variables would exclude the missing observations and leave the model subject to bias from missing data. To avoid this, we ran individual models for Hispanic ethnicity, race, age at enrollment, insurance status, language, and tumor subtype while controlling for sex and enrollment year.

We compared the slopes of the community level risk factors in ordinal, PPO, and fully nonproportional logistic models and found no violations of the proportional odds assumption. To calculate the area-level effect of community level risk factors on late stage diagnosis, we used multilevel ordinal models with generalized estimating equations and robust standard errors. We used an independent correlation matrix with a random intercept for each ZCTA, controlling for sex and enrollment year. We explored race and age at enrollment as confounders; race did not alter the effect estimates ≥10%. We determined statistical significance based on the confidence interval overlap with the null value.

We examined the association of late stage diagnosis and race, ethnicity, and insurance within strata of tumor subtype using PPO models where appropriate. We excluded untyped tumors from this analysis as we wanted to focus on tumors with known subtypes. The stratified models controlled for sex and enrollment year. We ran models for the effect of race on late stage diagnosis while controlling for health insurance status. Effect modification was determined using an interaction term between tumor subtype and the variable of interest, which was significant if p<0.05.

Last, we examined the association of community level risk factors on late stage diagnosis within strata of race. We tested for linearity of the effect of each community level factor. This was done by inputting each category of community level factor as a numeric variable, with greater numeric values representing worsening neighborhood redlining, dissimilarity index, or COI. The numeric variable was input into a regression model to determine if the slope of the line was equal to 1. Due to small numbers for both HL and RMS, Asian patients are presented in Supplemental Table 2.

Consent from the participants’ approval was waived. The University of Utah Institutional Review Board approved this study. This study was performed in accordance with the Declaration of Helsinki.

Data availability

Data supporting the results reported in the article can be found through request and review from the COG data committee (https://childrensoncologygroup.org/childrens-oncology-group).

Results

Our cohorts were more male (Overall=55.6%, HL=54.2%, RMS=58.4%) and primarily White (Overall=61.2%, HL=63.1%, RMS=57.3%), not Hispanic (Overall=78.7%, HL=79.2%, RMS=77.4%, Table 1). The majority (62.7%) were enrolled in a trial or registry between 2010 and 2021. The most common age at diagnosis was 15–19 years (37.6%) for HL and 0–4 years (29.4%) for RMS. English was the primary language at home (HL=41.9%, RMS=45.0%). Private insurance was most common (HL=38.0%; RMS=36.5%). Most lived in Metropolitan areas (HL=69.2%, RMS=69.0%). Only 19.2% of HL and 34.3% of RMS patients were diagnosed at local stage.

Table 1.

Characteristics of Hodgkin Lymphoma and Rhabdomyosarcoma Patients in Children’s Oncology Group Trials and Registries

	Overall		Hodgkin Lymphoma		Rhabdomyosarcoma
	N	%	N	%	N	%
All	8,756	100	5,956	100	2,800	100
Sex
Female	3,892	44.4	2,726	45.8	1,166	41.6
Male	4,864	55.6	3,230	54.2	1,634	58.4
Enrollment year
1990 to 2004	412	4.7	406	6.8	6	0.2
2005 to 2010	2,866	32.7	2,040	34.3	826	29.5
2010 to 2014	2,405	27.5	1,420	23.8	985	35.2
2015 to 2021	3,073	35.1	2,090	35.1	983	35.1
Age at enrollment
.	2,461	28.1	1,809	30.4	652	23.3
00–04 years	899	10.3	75	1.3	824	29.4
05–09 years	842	9.6	364	6.1	478	17.1
10–14 years	1,707	19.5	1,293	21.7	414	14.8
15–19 years	2,596	29.7	2,228	37.4	368	13.1
20–24 years	234	2.7	176	3	58	2.1
25–30 years	17	0.2	11	0.2	6	0.2
Birth year
.	2,461	28.1	1,809	30.3	652	23.3
1978 to 1989	95	1.1	60	1	35	1.3
1990 to 1999	2,517	28.8	2,066	34.7	451	16.1
2000 to 2009	2,876	32.9	1,894	31.8	982	35.1
2010 to 2020	807	9.2	127	2.1	680	24.3
Hispanic ethnicity
.	396	4.5	241	4.1	155	5.5
Not Hispanic	6,890	78.7	4,722	79.3	2,168	77.4
Hispanic	1,470	16.8	993	16.7	477	17
Race
.	571	6.5	350	5.9	221	7.9
Asian	216	2.5	147	2.5	69	2.5
Black	1,045	11.9	656	11	389	13.9
Other races	1,562	17.8	1,045	17.6	517	18.5
White	5,362	61.2	3,758	63.1	1,604	57.3
Language(s) spoken at home
.	4,676	53.4	3,253	54.6	1,423	50.8
Bilingual non-Spanish, French, or other language	93	1.1	56	0.9	37	1.3
Spanish only, Spanish bilingual	239	2.7	159	2.7	80	2.9
English only	3,748	42.8	2,488	41.8	1,260	45
Health insurance at enrollment
.	3,085	35.2	2,194	36.8	891	31.8
No insurance, self-pay	856	9.8	521	8.8	335	12
Private, military, or employer	3,286	37.5	2,264	38.0	1,022	36.5
Public or national	1,529	17.5	977	16.4	552	19.7
Urban-rural ZIP code
.	1466	16.7	990	16.6	476	17.0
Metropolitan area	6055	69.2	4124	69.2	1931	69.0
Micropolitan area	658	7.5	466	7.8	192	6.9
Small town/rural	577	6.6	376	6.3	201	7.2
Tumor subtype
HG: Nodular sclerosis HL	3,479	39.7	3,479	58.4	.	.
HG: Other Hodgkin’s Lymphoma	807	9.2	807	13.6	.	.
RH: Alveolar	796	9.1	.	.	796	28.4
RH: Embryonal	1,430	16.4	.	.	1,430	51.1
Unknown	2,244	25.6	1,670	28.0	574	20.5
Stage at diagnosis
Local	2,104	24.0	1,144	19.2	960	34.3
Regional	4,669	53.3	3,603	60.5	1,066	38.1
Distant	1,983	22.7	1,209	20.3	774	27.6

. Indicates missing data

We found significant differences in the distribution of tumor subtype by race (Figure 1). For HL, this was driven by a higher percent of White patients being diagnosed with nodular sclerosis HL (61.0%) relative to patients of other races (p<0.0001). For RMS, this was driven by a higher percent of tumors that were not typed among Black patients (23.1%) and patients of other races (24.6%) relative to Asian and White patients (p<0.05).

Figure 1.

Percent Differences in Tumor Subtype by Race Among Hodgkin Lymphoma and Rhabdomyosarcoma Patients in Children’s Oncology Group Trials and Registries

Individual Sociodemographic Models

The effect estimates between individual sociodemographic factors varied depending on the cancer type (Table 2). Female HL and RMS patients had higher odds of being diagnosed with later stage tumors (HL D&R vs L OR=1.47, 95% CI=1.29–1.68; RMS OR=1.29, 95% CI=1.12–1.48). HL and RMS patients enrolled between 2015 and 2021 also had a higher odds of late stage diagnoses (HL D vs R&L OR=1.82 (1.57–2.12); RMS D vs R&L OR=1.81, 95% CI=1.46–2.25; RMS D&R vs L OR=3.86, 95% CI=3.14–4.76). For HL, patients who were Black (D vs R&L OR=1.38, 95% CI=1.13–1.68) or did not have health insurance (D vs R&L OR=1.38, 95% CI=1.09–1.75) had significant increases in their odds of being diagnosed with late stage tumors relative to their comparison groups. Compared to HL patients with private/employer insurance, patients with no insurance had higher odds of late stage diagnosis (D vs R&L OR=1.38, 95% CI=1.09–1.75), but patients with public insurance had lower odds of late stage diagnosis (D&R vs L OR=0.75, 95% CI=0.57–0.98). Older age at enrollment was associated with greater odds of late stage diagnosis for both HL and RMS.

Table 2.

Late Stage Diagnosis and Individual Sociodemographic Risk Factors Among Hodgkin Lymphoma and Rhabdomyosarcoma Patients in Children’s Oncology Group Trials and Registries

Model		Hodgkin Lymphoma (N=5,956)		Rhabdomyosarcoma (N=2,800)
		Cancer Stage (PPO)	OR (95% CI)	Cancer Stage (PPO)	OR (95% CI)
Multivariable model
	Sex
	Female vs Male	D vs R&L	1.10 (0.97–1.25)		1.29 (1.12–1.48) *
		D&R vs L	1.47 (1.29–1.68) *
	Enrollment year
	2015 to 2021 vs 1990 to 2010	D vs R&L	1.82 (1.57–2.12) *	D vs R&L	1.81 (1.46–2.25) *
		D&R vs L	0.96 (0.82–1.13)	D&R vs L	3.86 (3.14–4.76) *
	2010 to 2014 vs 1990 to 2010	D vs R&L	2.18 (1.85–2.56) *	D vs R&L	1.65 (1.33–2.05) *
		D&R vs L	0.40 (0.34–0.46) *	D&R vs L	1.63 (1.35–1.97) *
Individual associations adjusted for sex and enrollment year
1	Hispanic
	Yes vs No		0.91 (0.80–1.05)		1.11 (0.92–1.33)
2	Race
	Asian vs White	D vs R&L	1.60 (1.10–2.33) *		1.45 (0.94–2.25)
		D&R vs L	1.00 (0.65–1.53)
	Black vs White	D vs R&L	1.38 (1.13–1.68) *		1.08 (0.88–1.33)
		D&R vs L	1.12 (0.90–1.39)
	Other race vs White	D vs R&L	1.07 (0.90–1.27)		1.15 (0.96–1.38)
		D&R vs L	0.82 (0.69–0.97)
3	Age at enrollment
	10–19 yrs vs 0–9 yrs	D vs R&L	1.31 (1.02–1.68) *		2.28 (1.92–2.70) *
		D&R vs L	1.92 (1.55–2.39) *
	20–30 yrs vs 0–9 yrs	D vs R&L	0.97 (0.63–1.51)		2.33 (1.46–3.73) *
		D&R vs L	1.54 (1.03–2.32) *
4	Insurance status
	No insurance or self-pay vs Private/Employer	D vs R&L	1.38 (1.09–1.75) *		1.23 (0.97–1.56)
		D&R vs L	1.04 (0.72–1.51)
	Public or national insurance vs Private/Employer	D vs R&L	1.07 (0.88–1.30)		1.09 (0.89–1.33)
		D&R vs L	0.75 (0.57–0.98) *
5	Language spoken in home
	Bilingual (non-Spanish), French, or other language vs English only		1.46 (0.87–2.45)	D vs R&L	0.77 (0.37–1.61)
				D&R vs L	2.66 (1.03–6.91) *
	Any Spanish vs English only		0.79 (0.58–1.08)	D vs R&L	1.19 (0.73–1.93)
				D&R vs L	1.59 (0.92–2.73)
6	Tumor subtype
	Hodgkin Lymphoma
	Nodular sclerosis vs Other HL	D vs R&L	1.64 (1.34–2.01) *
		D&R vs L	3.05 (2.55–3.65) *
	Untyped vs Other HL	D vs R&L	1.30 (1.04–1.63) *
		D&R vs L	1.79 (1.48–2.17) *
	Rhabdomyosarcoma
	Alveolar vs Embryonal	D vs R&L			6.16 (5.00–7.58) *
		D&R vs L			4.87 (3.89–6.10) *
	Untyped vs Embryonal	D vs R&L			2.81 (2.23–3.55) *
		D&R vs L			1.69 (1.38–2.09) *

If there is entry for the cancer stage column, it means the results are from a partial proportional odds model. Otherwise, it is the proportional odds model where coefficients are same across local (L), regional (R), and distant (D) stage

HL Model Ns: Hispanic=5,715; Race=5,606; Age at enrollment=4,147; Insurance status=3,762; Language spoken at home=2,703; Tumor subtype=5,956

RMS Model Ns: Hispanic=2,645; Race=2,579; Age at enrollment=2,148; Insurance status=1,909; Language spoken at home= 1,377; Tumor subtype= 2,800

^*Indicates statistical significance

Tumor subtype had a significant association with increased odds of late stage tumor diagnosis for both HL and RMS. Nodular sclerosis HL and untyped HL had higher odds of late stage diagnosis relative to patients with Other HL (Nodular HL D&R vs L OR=3.05, 95% CI=2.55–3.65; Untyped HL D&R vs L (OR=1.79, 95% CI=1.48–2.17). Patients with alveolar RMS and untyped RMS had greater odds of late stage diagnosis compared with patients with embryonal RMS (Alveolar D&R vs L OR=4.87, 95% CI=3.89–6.10; Untyped OR=1.69, 95 CI%=1.38–2.09).

Community-Level Models

Certain community level risk factors were associated with late stage diagnosis (Table 3). Counties with greater redlining or counties that were ungraded had higher odds of late stage RMS diagnosis relative to counties considered more desirable (Redlining C,D score OR=1.53, 95% CI=1.02–2.28; Ungraded OR=1.54, 95% CI=1.02–2.35). Metropolitan areas with a Black DI of 80–100% or a Hispanic DI of 60–79% had higher odds of late stage HL diagnosis relative to neighborhoods with greater integration (Black DI OR=1.83, 95% CI=1.11–3.02; Hispanic DI OR=1.30, 95% CI=1.05–1.60). ZIP codes with very low/low COI had increased odds of late stage RMS relative to patients residing in ZIP codes with Very high/High COI (OR=1.21, 95% CI=1.02–1.45).

Table 3.

Late Stage Diagnosis and Community Level Socioeconomic Risk Factors Among Hodgkin Lymphoma and Rhabdomyosarcoma Patients in Children’s Oncology Group Trials and Registries

Model		Hodgkin Lymphoma		Rhabdomyosarcoma
		N	OR (95% CI)	N	OR (95% CI)
1	Health professional shortage area	3,527		1,824
	Yes	1,122	1.02 (0.89–1.17)	606	1.15 (0.96–1.38)
	No	2,405	1	1,218	1
2	Driving time to nearest COG center (minutes)	4,605		2,316
	120–1,900	420	1.07 (0.88–1.30)	267	0.99 (0.78–1.27)
	60–119	949	0.93 (0.81–1.07)	460	1.09 (0.90–1.31)
	02–59	3,236	1	1,589	1
3	RUCA code	4,617		2,318
	Small town/rural	345	1.12 (0.90–1.39)	201	0.85 (0.65–1.12)
	Micropolitan	432	1.05 (0.87–1.27)	191	1.10 (0.83–1.44)
	Metropolitan	3,840	1	1,926	1
4	Redlined county	4,575		2,303
	Ungraded	3,349	1.06 (0.78–1.44)	1,616	1.53 (1.02–2.28) *
	C, D (Least desirable)	1,094	1.00 (0.73–1.39)	619	1.54 (1.02–2.35) *
	A, B (Most desirable)	132	1	68	1
5	Black dissimilarity index (DI)	3,122		1,553
	80–100%	553	1.83 (1.11–3.02) *	218	0.53 (0.28–1.02)
	60–79%	1,515	1.12 (0.69–1.80)	763	0.67 (0.36–1.24)
	40–59%	973	1.09 (0.67–1.76)	526	0.69 (0.37–1.28)
	20–39%	81	1	46	1
6	Hispanic dissimilarity index (DI)	3,122		1,553
	60–79%	628	1.30 (1.05–1.60) *	274	1.00 (0.75–1.33)
	40–59%	1,667	1.16 (0.99–1.37)	858	0.92 (0.74–1.15)
	20–39%	827	1	421	1
7	Asian dissimilarity index (DI)	3,122		1,553
	40–59%	1,975	1.01 (0.88–1.17)	998	0.89 (0.73–1.08)
	20–39%	1,147	1	555	1
8	Child Opportunity Index (National)	3,809		1,867
	Very low/Low	957	1.02 (0.89–1.17)	521	1.21 (1.02–1.45) *
	Moderate	1,058	1.02 (0.88–1.19)	515	1.13 (0.92–1.38)
	Very high/high	1,794	1	831	1

Ordinal logistic regression models with local stage disease as referent compared to regional and distant.

All GEE models control for sex and enrollment year

^*Indicates statistical significance

Within strata of HL tumor subtype, we found significant differences by race and insurance (Figure 2). Of patients diagnosed with nodular sclerosis HL, Black (OR=1.42, 95% CI=1.09–1.85) or uninsured patients had a higher odds of late stage diagnosis (D vs R&L OR=1.67, 95% CI=1.26–2.22). Within strata of Other HL, we found significant elevation in the odds for late stage diagnosis among Asian and Black patients relative to White patients (Asian D vs R&L OR=3.56, 95% CI=1.52–8.35; Black D&R vs L OR=1.86, 95% CI=1.13–3.07). The racial disparity between Black and White Other HL patients remained significant after adjustment for health insurance at enrollment (OR=2.37, 95% CI=1.02–5.59), although a large number of observations in this model were excluded due to missing insurance data.

Figure 2.

Racial and Ethnic Disparities in Late Stage Diagnosis Within Strata of Tumor Subtype with Partial Proportional Odds Models

Community-Level Models among Socioeconomic and Racial/Ethnic Groups

Community level socioeconomic factors had different effect estimates within racial and ethnic strata (Table 4). We found increased odds for late stage HL diagnosis in White patients living in areas with the greatest Black DI (80–100% OR=1.84, 95% CI=1.39–2.43). These effect estimates showed a significant increasing linear trend (p=0.001). We also found increased odds of late stage HL diagnosis among Whites living in areas with greater Hispanic DI (40–59% OR=1.24, 95% CI=1.01–1.52; 60–79% OR=1.65, 95% CI=1.25–2.17). These odds ratios also showed a positive linear trend (p=0.0004).

Table 4.

Community Level Sociodemographic Factors and Late Stage Diagnosis Within Strata of Race/Ethnicity

	Hodgkin Lymphoma			Rhabdomyosarcoma
	N	OR (95% CI)	p-value for linear trend	N	OR (95% CI)	p-value for linear trend
Black
Black dissimilarity index	399			258
20–59%	120	1	0.39	74	1	0.13
60–79%	199	0.90 (0.58–1.4)		138	0.80 (0.45–1.41)
80–100%	80	1.27 (0.71–2.28)		46	0.62 (0.33–1.16)
Hispanic dissimilarity index	399			258
20–39%	115	1	0.76	84	1	0.11
40–59%	209	1.05 (0.69–1.60)		133	0.77 (0.45–1.34)
60–79%	75	1.09 (0.61–1.97)		41	0.58 (0.31–1.10)
Child Opportunity Index	302			201
Very high/High	71	1	0.41	47	1	0.82
Moderate	84	1.29 (0.69–2.41)		58	1.54 (0.75–3.15)
Very low/Low	147	0.84 (0.47–1.52)		96	1.16 (0.61–2.19)
Hispanic
Black dissimilarity index	640			339
20–59%	257	1	0.04*	147	1	0.30
60–79%	281	0.87(0.62–1.22)		148	1.57 (1.03–2.41)
80–100%	102	1.60 (1.06–2.41) *		44	1.16 (0.63–2.12)
Hispanic dissimilarity index	640			339
20–39%	88	1	0.95	44	1	0.35
40–59%	373	1.06 (0.67–1.70)		205	0.90 (0.50–1.63)
60–79%	179	1.01 (0.60–1.70)		90	1.25 (0.62–2.49)
Child Opportunity Index	622			316
Very high/High	175	1	0.51	113	1	0.63
Moderate	194	0.90 (0.60–1.35)		82	0.91 (0.53–1.56)
Very low/Low	253	1.12 (0.77–1.63)		121	0.89 (0.55–1.43)
White
Black dissimilarity index	1,862			829
20–59%	595	1	0.001*	303	1	0.09
60–79%	945	1.12(0.91–1.36)		414	0.81(0.61–1.07)
80–100%	322	1.84(1.39–2.43) *		112	0.65(0.43–0.98) *
Hispanic dissimilarity index	1,862			829
20–39%	572	1	0.0004*	259	1	0.69
40–59%	971	1.24 (1.01–1.52) *		450	0.95 (0.72–1.27)
60–79%	319	1.65 (1.25–2.17) *		120	0.93 (0.61–1.41)
Child Opportunity Index	2,641			1,200
Very high/High	1,422	1	0.9	603	1	0.04*
Moderate	710	1.08 (0.90–1.29)		315	1.03 (0.80–1.32)
Very low/Low	509	0.96 (0.79–1.16)		282	1.35 (1.03–1.76) *

Subjects diagnosed 1990 to 2004 excluded from analysis

Models control for sex and enrollment year

^*Indicates statistical significance

Among White RMS patients, living in neighborhoods with the greatest Black DI (80–100% OR=0.65, 95% CI=0.43–0.98), with a non-significant downward linear trend (p=0.09) was associated with reduced odds of late stage diagnosis. Black patients living in areas with greater Black DI had potential decreases in their odds of late stage diagnosis for RMS although non-significant. We found no evidence of trends for Hispanic HL or RMS patients living in areas with greater Hispanic DI.

We found a significant elevation in the odds of late stage RMS diagnosis among patients living in ZIP codes with lower COI Moderate OR=1.03, 95% CI=0.80–1.32; Very low/Low OR=1.35, 95% CI=1.03–1.76); the odds ratios had a positive linear trend (p=0.04).

Discussion

In this national study of pediatric Hodgkin Lymphoma and rhabdomyosarcoma, both cancers with previous evidence of the role of socioeconomic status on delays in diagnosis and advanced stage disease, we found large and differential influences of both individual and community level socioeconomic status on late stage diagnosis among race, ethnicity, and cancer subtype. This represents the first study to use COG’s large, patient-level, national registries to study childhood cancer disparities, demonstrating the feasibility and power of using these resources. We found that a later stage of HL diagnosis was correlated with the individual factors of age ≥10 years at enrollment, Black race, and being uninsured; late stage RMS diagnosis was correlated with age ≥10 years at enrollment and tumor type. For both disease groups, female patients were more likely to present with later stage disease.

These findings complement and expand on earlier work typically done using SEER registry data to describe new influential factors, such as community deprivation and ethnic enclaves, that may play a substantive role in childhood cancer outcomes in addition to individual-level factors such as age, race, and ethnicity. In particular, we found that community level socioeconomic factors that were related to late stage HL were greater for Whites living in Black and Hispanic enclaves. For RMS, counties with greater redlining and low community-level socioeconomic status as measured by the Child Opportunity Index was correlated with late stage diagnosis.

Our results demonstrate that accessing appropriate diagnostic procedures and treatment for pediatric cancers could be affected by both individual level access (e.g., insurance) and community level barriers (e.g. economic deprivation).(29,30) Families with fewer resources or from disadvantaged backgrounds may face barriers due to factors such as the cost of provider visits and a lack of familiarity with the health care system.(31) We found that uninsured patients had increased odds of late stage diagnosis for HL relative to patients with private insurance. While conclusions about uninsured patients’ odds for late stage RMS were limited by sample size, our results suggest similar patterns for the association of health insurance in late stage RMS diagnosis as HL. We did not find significant differences in the correlation of health insurance and late stage diagnosis across racial groups. Yet, in models that were stratified by tumor-subtype and controlled for health insurance, Black patients still had higher odds of late stage diagnosis of Other HL (lymphocyte depleted, enriched, and nodular lymphocyte-predominant) than White patients with the same diagnosis.

One novel aspect is our ability to examine tumor subtype in associations with individual and community SES by stage among racial and ethnic strata. Tumor subtype is related to both genetic ancestry and the speed at which a cancer manifests itself, which may facilitate early stage disease detection.(32) While tumor subtype has been correlated with race, late stage diagnosis, and survival in multiple studies, these analyses have largely been limited to adult onset cancers.(1,7,9) Our results support that cancer subtype is correlated with race and highly relevant to late stage cancer diagnosis. However, subtype cannot fully explain the differences in risk for late stage diagnosis by race. Among HL patients with Other HL (lymphocyte depleted, enriched, and nodular lymphocyte-predominant), Asian and Black patients had greater odds of late stage diagnosis relative to White patients. This finding suggested that race and tumor subtype reflected different aspects of vulnerability and should be interpreted as such in the context of pediatric cancer.

The different symptoms experienced by patients prior to the diagnosis of HL and RMS may explain these complex patterns we found among racial/ethnic groups, tumor subtype, health insurance, and late stage diagnosis. For example, patients with undiagnosed HL display symptoms of fever, fatigue, achiness, and sweating that can be mistaken for repeated occurrence of acute childhood illnesses like the flu. These intermittent episodes may lead a caregiver to speculate that their child is prone to getting sick rather than having a disease like cancer. These beliefs may be amplified by cultural differences in parental beliefs of what causes childhood illness and a lack of knowledge about cancer symptoms. Further, parents who are members of certain marginalized groups may hesitate to request referrals from their primary care providers to see pediatric specialists or may lack access to pediatric specialists due to resource constraints.(33,34)

In our study, we used both unidimensional and multidimensional measures to describe community level socioeconomic deprivation. While multiple reviews report no consensus regarding community level socioeconomic indicators as predictors of childhood cancer disparities, findings do differ based on how socioeconomic status was measured.(35–38) Studies often rely on unidimensional measures, such as distance to pediatric cancer care facilities and ZIP code median income,(39) or they may utilize multidimensional descriptions of neighborhood socioeconomic such as the Yost index or COI.(8) The majority of the unidimensional measures in this current report, with the exception of redlining, were not correlated with late stage diagnosis of either HL or RMS. The multidimensional measure for socioeconomic status (COI) was correlated with late stage RMS diagnosis but not late stage HL diagnosis. Thus, our findings provide information that socioeconomic factors that affect cancer outcomes cannot be easily dissected into single components like rurality or distance to the nearest healthcare facility. In other words, the confluence of multiple components of socioeconomic status likely led to late stage diagnosis rather than a single contributing factor. Even redlining and the dissimilarity indices, which are technically unidimensional measures, are proxy indicators for multiple social and economic characteristics of specific geographic areas.

Ethnic enclaves may be protective against adverse cancer outcomes due to shared cultural history and language that can facilitate transfer of health information and access to healthcare providers.(40) We found that communities with greater percentages of Black and Hispanic residents had higher odds of late stage HL diagnosis. If examined alone, this result implied that Black and Hispanic enclaves were not protective against late stage pediatric cancers. Yet, in our analysis stratified by race, increased odds of late stage diagnosis in higher Black and Hispanic DI was only observed among White patients. This supports the idea that social isolation increases odds for late stage diagnosis. In contrast to the findings for White patients, Black patients had lower odds of late stage HL and RMS diagnosis if they lived in Black enclaves. Social connectivity and social isolation may be important to preventing late stage cancer diagnosis and need to be better understood in the context of pediatric cancer.

Together these results imply that individual factors, such as health insurance access, may attenuate the vulnerability of certain racial groups to late stage diagnosis, as public insured HL patients had a lower odds of distant/regional disease compared to those with private/employer insurance. At the same time, our report of novel community factors associated with stage among certain patient subgroups demonstrate that multiple influences, including community connectedness and economic resources, may play a role in outcomes. Extending health insurance coverage is often proposed to address disparities in late stage diagnosis among racial and ethnic minorities.(41) Policies such as the Affordable Care Act and Medicaid expansion have attempted to reduce the barriers to quality health insurance. Ensuring public insurance access as a path to reduce late stage cancer diagnoses is at risk due to Medicaid unwinding that substantially affected insurance coverage for many US children. Thus, while insurance plays an important role, without policies that address community-level barriers or ensure insurance continuity, effective strategies to reduce pediatric cancer disparities will not be feasible.

Limitations and strengths

Although we used a large national pediatric cancer data resource, we had missing data for age at enrollment, language spoken at home, and health insurance, which limited our ability to draw conclusions about these variables in the analysis or use them to control for confounding. While patients were asked to report their demographic information, it was unclear what percent of the race and Hispanic ethnicity information was determined through self-report or study recruitment team observation. This could lead to misclassification of race and ethnicity as self-report is the gold standard measure for race and ethnicity. Moreover, we had ZIP code data for measuring community level socioeconomic status. ZIP codes are imprecise methods of measuring geography as they can cover large geographic areas, change over time, and do not directly correspond with census tracts, which are more stable over time and are standardized to a specific population size. Further, participants with missing zip code tended to be diagnosed in more recent years, were more likely to have missing data on certain factors, such as race/ethnicity or language spoken in the home. Finally, due to our interest in considering multiple individual, family, and community-level factors of interest, we opted to focus solely on HL and RMS rather than include other cancers, as both HL and RMS have earlier evidence linking socioeconomic factors to stage of disease and provide context as common hematologic and solid tumors in pediatrics. Thus, future work in novel socioeconomic factors and pediatric cancer outcomes should include other disease groups.

Despite these limitations, using the COG registry data allowed us to consider patients from across the United States compared to earlier studies in this area that have been limited to SEER data, which covers approximately 30% of the US population. Thus, the national focus of these locations, coupled with valuable demographic information from the COG registries and trials data, provided novel insights into the role of socioeconomic status and race on late stage diagnosis of HL and RMS childhood cancer survivors in the United States. Furthermore, our analyses point to areas that require additional research. We found that female patients have later stage disease for both HL and RMS, which could be due to medical delays in diagnosis.(42) At the same time, more recent diagnoses had a higher odds of late stage diagnosis, potentially due to improvements in medical technology (e.g., higher-resolution CT/PET/MRI), which should be confirmed in contemporaneous samples. Finally, we found that unstaged tumors had a higher odds of late stage disease, which could reflect the underlying cancer type; nodular sclerosis is by far the most common type of HL and embryonal for RMS. However, untyped tumors have more atypical presentation which also complicates diagnosis; thus, subsequent investigations of unstaged HL and RMS are needed.

In summary, we found that both individual and community level socioeconomic factors have a complex relationship with late stage diagnosis for HL and RMS. Rather than attributing disparities in late stage diagnosis to single individual or community level risk factor, our study suggests that the interplay of these multilevel risk factors needs to be considered and that these risk factors impact racial and ethnic groups in different ways. Studies of SES that focus on individual factors alone may miss the profound experiences of patients in communities with substantial economic barriers and patients who are socially isolated from their larger communities. At the same time, studies that aggregate patients by single biological factors, without investigating differences by race and ethnicity, or that do not include examination of tumor subtypes, may miss eliciting some of the important individual differences that could be key to supporting patients and their families.