Association Between Medicaid Coverage Continuity and Survival in Patients with Newly Diagnosed Pediatric and Adolescent Cancers

Abstract

Introduction:

Many patients with cancer do not gain Medicaid coverage until a cancer diagnosis, which can reduce access to early cancer detection and timely treatment, potentially driving inferior survival. Little is known about whether continuous Medicaid coverage pre- through post-diagnosis (versus gaining Medicaid at/after diagnosis) provides survival benefits for pediatric/adolescent oncology patients.

Methods:

We identified patients newly diagnosed with cancer at age 21 or younger in a large pediatric health system between 2007–2016. Electronic medical records (EMR) were linked to Medicaid administrative data to differentiate insurance continuity patterns during the six months preceding through the six months following cancer diagnosis (assessment window): continuous Medicaid, newly gained Medicaid (at or after diagnosis), and other Medicaid enrollment patterns. For patients not linked to Medicaid data, we used EMR-reported insurance types at diagnosis. We followed patients from six months post-diagnosis up to five years, death, or December 2020, whichever came first. Multivariable regressions estimated all-cause and cancer-specific survival, controlling for sociodemographic and cancer-related factors.

Results:

Over two-thirds of the 1,800 patients (1,293, 71.8%) had some Medicaid enrollment during the assessment window; among these, 47.6% had continuous Medicaid and 36.3% had newly gained Medicaid. Patients not linked with Medicaid data had private (26.9%) or other/no insurance (1.2%) at diagnosis. Compared to patients with continuous Medicaid, those with newly gained Medicaid had higher risks of all-cause death (hazard ratio [HR]=1.41, 95% CI=1.10–1.81, p=0.008) and cancer-specific death (HR=1.46, 95% CI=1.12–1.90, p=0.005).

Conclusion:

Continuous Medicaid coverage throughout cancer diagnosis is associated with survival benefits for pediatric/adolescent patients. This finding has critical implications as millions of Americans have been losing coverage under the unwinding of the Medicaid Continuous Enrollment Provision.

Précis:

This study examined the association of Medicaid enrollment patterns with overall and cancer-specific 5-year survival in a cohort of patients newly diagnosed with pediatric and adolescent cancer in a large pediatric health system in Georgia. Continuous Medicaid coverage throughout a cancer diagnosis is significantly associated with survival benefits for pediatric and adolescent patients; however, less than half of Medicaid-insured patients had continuous coverage.

INTRODUCTION

In the United States, Medicaid is the single largest insurer for children, covering 40 million children under age 19 years and approximately one in three children with a newly diagnosed cancer.^1,2 Cancer is a leading cause of death among U.S. children,³ with estimated 15,190 new cases ages 0–23 years diagnosed in 2023.^3,4 In many states, including Georgia, income-eligible children may join Medicaid due to a diagnosis of or treatment for cancer or parental financial hardship secondary to cancer diagnosis.⁵ Compared to children with continuous coverage before and throughout cancer diagnosis, those without Medicaid coverage until the point of, or after, a diagnosis may face barriers to accessing care timely; these can include skipped primary care visits at the onset of symptoms, or delayed diagnostic testing and initiation of life-saving therapies.^6,7 Missing these opportunities can result in high disease severity at presentation or late-stage diagnoses, and ultimately, poor prognosis.^8–13

To date, a few studies have leveraged Medicaid administrative records to assess insurance enrollment over time, yet the vast majority focused on adult-onset cancers.^14,15 We are aware of only one study examining the role of enrollment dynamics specifically in pediatric oncology patients;¹⁶ however, this study did not distinguish children who newly enrolled onto Medicaid at the point of or after cancer diagnosis from other enrollment patterns, and also lacked clinical characteristics particularly relevant when assessing pediatric cancer survival, such as second malignancy and recurrence. Furthermore, research comparing Medicaid enrollment rates estimated from Medicaid administrative records versus cancer registry data is scarce. To fill these literature gaps, we provided a detailed characterization of Medicaid enrollment patterns among children newly diagnosed with cancer in a large pediatric health system in Georgia, a state that has not expanded Medicaid eligibility under the Affordable Care Act (ACA),¹⁷ and examined the association of these patterns with overall and cancer-specific survival.

MATERIALS & METHODS

Data Sources and Linkages

Data came from electronic medical records (EMR) of Children’s Healthcare of Atlanta (hereafter Children’s), Georgia Cancer Registry (GCR), and Medicaid administrative enrollment data. Patient demographics, addresses, insurance type recorded at diagnosis date, and cancer-related information were abstracted from EMR. Patients’ residential addresses at cancer diagnosis were linked with the county-level Rural-Urban Continuum Codes to measure rurality of residence,¹⁸ and with the census tract-level Social Deprivation Index (SDI) files to measure neighborhood deprivation level.^19,20 Vital status including date and cause of death from state vital records and the National Death Index was obtained via the GCR, available through December 31, 2020.

Patients’ EMR were linked with Medicaid administrative data from the Centers for Medicare and Medicaid Services (CMS), specifically the 2007–2015 Medicaid Analytic eXtract (MAX) files and 2014–2016 Transformed Medicaid Statistical Information System Analytic Files (TAF).^21,22 These files provide information on monthly Medicaid enrollment status, which was used to trace Medicaid enrollment patterns preceding, during, and following diagnosis. The linkage with Medicaid data used a deterministic matching approach based on social security number (SSN), date of birth, and sex.²³ If multiple Medicaid records were linked to a patient (largely due to Medicaid enrollment in multiple states), all records were aggregated when determining if the patient was enrolled in Medicaid in a given month. For linkage validation, a matching score 1–4 was assigned to each linked Medicaid record (4=records matched on all three identifiers, 3=matched on SSN and date of birth, 2=matched on SSN and sex, and 1=matched on SSN only). We conducted a sensitivity analysis excluding patients with a score of 1–2 from the Medicaid-insured groups;^24,25 which showed consistent results (available upon request).

Analytic Sample

Our retrospective cohort included patients diagnosed with an initial cancer at age 21 or younger who received treatment at Children’s between July 1, 2007, and June 30, 2016.²⁶ Inclusion required complete data on SSN, date of birth, and sex to allow the linkage with Medicaid data. Patients also had to have survived ≥6 months after cancer diagnosis to allow full assessment of insurance continuity patterns over a 13-month window (detailed below); this alleviates any potential bias from a reverse causal pathway (i.e., people who died within six months from diagnosis would not have continuous Medicaid coverage because of their death). Our primary sample included 1,800 patients (Appendix Figure 1). In analyses of cancer-specific death, we narrowed to patients with known causes of death (N=1,792).

Insurance Continuity Patterns

Insurance continuity was assessed over a 13-month assessment window: the six months preceding, the month of, and the six months following cancer diagnosis. We classified patients into the following mutually exclusive groups: 1) enrollment in Medicaid for 12–13 months of this window (“continuous Medicaid” hereafter), 2) joined Medicaid within two months before, in the month of, or within six months after diagnosis (“newly gained Medicaid”), 3) other patterns of Medicaid enrollment (“noncontinuous Medicaid”); 4) EMR-reported private (including military) insurance at diagnosis date, and 5) EMR-reported “other” or no (including unknown) insurance at diagnosis date (Appendix Figure 2).^6,7 Of note, individuals with “noncontinuous Medicaid” may have appeared on and off Medicaid throughout the assessment window, and the enrollment patterns of this group are illustrated in Figure 1 and Appendix Table 1.

Figure 1.

Example visualization of the three groups of Medicaid-insured patients

Regarding the timeframe classified as “newly gained Medicaid” (group 2), the Georgia Medicaid program allows “retroactive coverage” up to three months prior to the application for Medicaid coverage if medical expenses have incurred.²⁷ Thus, to fully identify patients with newly gained Medicaid, we traced Medicaid enrollment up to two months before diagnosis to capture patients who had utilized medical services (e.g., diagnostic testing) leading up to their diagnosis under the “retroactive coverage” provision of Medicaid. In a sensitivity analysis, we traced enrollment up to three months, instead of two months, prior to cancer diagnosis to define this group and showed consistent results (Appendix Table 2). Groups 4 and 5 comprised patients not linked to Medicaid data; our analysis used insurance type recorded at diagnosis date in Children’s EMR for these groups.

All-cause and Cancer-specific Survival

Patients were followed from 6-month after cancer diagnosis (i.e., index date) until the date of death, five years post-index date, or December 31, 2020, whichever occurred first (Appendix Figure 3). Notably, patients diagnosed in 2015–2016 with <5 years follow-up were censored by December 31, 2020, if no death event occurred. Cause of death codes were reviewed to identify cancer-specific death, defined as death as direct consequence of initial cancer diagnosis (e.g., recurrent or progressive disease, acute toxicity while on therapy for the original disease), or due to cancer treatment sequelae where non-acute treatment side effects (e.g. cardiac toxicity, subsequent neoplasm) were the major contributing factor to the death.^28,29 Deaths due to other medical conditions, external causes (e.g. suicide, car accident), or unknown causes were excluded from cancer-specific deaths.

Statistical Analysis

Sample characteristics overall and by insurance continuity patterns are presented in Table 1. Chi-square tests were used for comparison across insurance continuity patterns. We generated cumulative incidence curves for all-cause death and cancer-related death, respectively, and used Gray’s test to compare the incidence of death across insurance continuity patterns.³⁰ We used the extended Cox model³¹ that can accommodate time-varying independent variables to assess the association between insurance continuity and all-cause survival, controlling for patient demographics (age at diagnosis, sex, race/ethnicity), cancer-related factors (cancer type, year of initial cancer diagnosis, cancer relapse or second malignancy), and socioeconomic status (SES; rurality and SDI).³² Notably, cancer relapse or second malignancy was included as a time-varying independent variable, which indicated each patient’s earliest event of relapse or second malignancy. Patients known to be alive were censored at December 31, 2020, or at five years post-index date if occurred earlier. Additionally, we assessed cancer-specific survival using the Fine-Gray sub-distribution hazard model among patients with known cause of death, where non-cancer death was accounted for as competing risks.³³ Data analyses were performed in SAS v.9.4 (Cary, NC). Statistical significance was evaluated at 0.05 threshold using 2-sided tests.

Table 1.

Characteristics of children and adolescents newly diagnosed with cancer between 2007 and 2016

	Total		Insurance Continuity Patterns
			Continuous Medicaid		Newly gained Medicaid		Noncontinuous Medicaid		Private insurance (EMR-reported)		Other/no insurance (EMR-reported)		P-Value a
	No.	Col. %	No.	Col. %	No.	Col. %	No.	Col %	No.	Col. %	No.	Col. %
Total	1800	100.0	615	100.0	470	100.0	208	100.0	485	100.0	22	100.0
Age at initial diagnosis (years)													<.001
0–14	1253	69.6	474	77.1	325	69.1	150	72.1	292	60.2	12	54.5
15 or older	547	30.4	141	22.9	145	30.9	58	27.9	193	39.8	10	45.5
Sex													0.57
Male	966	53.7	333	54.1	260	55.3	112	53.8	247	50.9	14	63.6
Female	834	46.3	282	45.9	210	44.7	96	46.2	238	49.1	8	36.4
Race/ethnicity													<.001
Non-Hispanic White	908	50.4	216	35.1	274	58.3	77	37.0	332	68.5	9	40.9
Non-Hispanic Black	558	31.0	256	41.6	116	24.7	73	35.1	107	22.1	6	27.3
Hispanic	233	12.9	108	17.6	53	11.3	45	21.6	21	4.3	6	27.3
Non-Hispanic other or unknown race/ethnicity	101	5.6	35	5.7	27	5.7	13	6.3	25	5.2	1	4.5
Cancer type													<.001
Leukemia	380	21.1	148	24.1	142	30.2	47	22.6	40	8.2	3	13.6
Lymphoma	243	13.5	62	10.1	54	11.5	26	12.5	96	19.8	5	22.7
CNS tumor	519	28.8	174	28.3	105	22.3	52	25.0	184	37.9	4	18.2
Non-CNS solid tumor	658	36.6	231	37.6	169	36.0	83	39.9	165	34.0	10	45.5
Year of initial diagnosis													<.001
2007–2011	987	54.8	315	51.2	229	48.7	116	55.8	310	63.9	17	77.3
2012–2016	813	45.2	300	48.8	241	51.3	92	44.2	175	36.1	5	22.7
Had cancer relapse or second cancerb	645	35.8	234	38.0	171	36.4	79	38.0	156	32.2	5	22.7	0.18
Rurality of residencec													<.001
Non-metropolitan areas	280	15.5	125	20.1	63	14.1	37	17.6	53	10.9	2	8.7
Metropolitan areas	1520	84.4	490	79.7	407	86.6	171	82.2	432	89.1	20	90.9
SDId													<.001
Below 75th percentile	1267	70.4	354	57.6	353	75.1	121	58.2	424	87.4	15	68.2
75th percentile or higher	533	29.6	261	42.4	117	24.9	87	41.8	61	12.6	7	31.8

Notes: EMR: electronic medical records. CNS: Central Nervous System. SDI: Social Deprivation Index.

^aChi-square test was used to compare the distribution of a covariate by insurance continuity patterns.

^bCancer relapse includes disease progression.

^cRurality of residence was measured by the Rural-Urban Continuum Codes (RUCC), with metropolitan areas classified by RUCC 1–3 and non-metropolitan (including non-metropolitan urban and rural) areas classified by RUCC 4–9.

^dThe SDI is a census tract-level composite measure of deprivation developed and validated by prior research based on the following seven characteristics from the American Community Survey: percent of residents <100% federal poverty level, percent of residents with <12 years education, percent of unemployed residents, percent of households living in renter-occupied housing, percent of households living in crowded housing, percent of households with no vehicle, and percent of single-parent families with dependents <18 years. Consistent with prior research, this analysis dichotomized patients based on whether they resided in census tracts with the highest SDI quartile (most deprived) – also known as “cold spots” associated with worse health outcomes. This measure was merged with the study cohort using patients’ residence addresses and the ArcGIS Desktop: Release 10.

RESULTS

Sample Characteristics

Among 1,800 children included, the average age at diagnosis was 10.0 years (standard deviation: 5.9); 46.3% were female; 31.0% and 12.9% were non-Hispanic Black and Hispanic patients, respectively (Table 1). Non-central nervous system (non-CNS) solid tumor was the most common cancer type (36.6%), followed by CNS tumor (28.8%) and leukemia (21.1%). Most children resided in metropolitan areas (84.4%).

Over two-thirds (n=1,293; 71.8%) were linked to Medicaid data (i.e., had any [≥1 month] Medicaid enrollment) during the 13-month assessment window; among these, 47.6% were classified as having continuous Medicaid, 36.3% as newly gained Medicaid, and 16.1% as noncontinuous Medicaid (Figure 2). For those not linked to Medicaid data, 485 (26.9%) had EMR-reported private insurance at diagnosis, and 22 (1.2%) had EMR-reported other or no insurance.

Figure 2.

Distribution of insurance continuity patterns during 13-month assessment window among children and adolescents newly diagnosed with cancer

Notes: EMR: electronic medical records. N=1,800 patients in the analytic sample.

^a Patients with Medicaid insurance (n=1,293, 71.8%) are defined as those who had any (≥1 month of) Medicaid enrollment during the 13-month assessment window.

^b Continuous Medicaid is defined as enrolling in Medicaid for at least 12 months during the 13-month assessment window.

^c Newly gained Medicaid is defined as enrolling onto Medicaid only within the two months before diagnosis, in the month of diagnosis, or within the six months after diagnosis.

^d Other patterns of noncontinuous Medicaid is defined as 1) enrolling in Medicaid for some time during the 13-month window but not classified as continuous Medicaid or newly gained Medicaid or 2) being recorded as EMR-reported Medicaid at diagnosis date but not linked to the Medicaid administrative data.

^e EMR-reported private insurance and EMR-reported other/unknown/no insurance include patients not linked to Medicaid administrative data; our analysis used insurance type recorded at the diagnosis date in Children’s EMR for these patients. See more details in Appendix Figure 2.

Sample characteristics differed across insurance continuity patterns, with Medicaid-insured groups more likely than those privately insured to be younger, non-Hispanic Black or Hispanic, and to reside in non-metropolitan or most deprived areas (p-values<0.001, Table 1).

Association between Insurance Continuity and Risk of Death

Cumulative incidence of all-cause death and cancer-specific death five years post-index date was 20.0% (95% confidence interval [CI]=18.2–21.9%) and 19.6% (95% CI=17.8–21.5%), respectively, and differed by insurance continuity patterns (Gray’s test p<0.001). The cumulative incidence of all-cause death five years post-index was lower among patients with private insurance (10.5%, 95% CI=8.0–13.5%) than those with continuous Medicaid (22.8%, 95% CI=19.6–26.2%) or newly gained Medicaid (25.3%, 95% CI=21.5–29.3%, Figure 3A). Cumulative incidence of cancer-specific death by insurance continuity patterns was similar to that for all-cause death (Figure 3B).

Figure 3.

Cumulative incidence of death by insurance continuity patterns among children and adolescents newly diagnosed with cancer

A. All-cause death^a

B. Cancer-specific death^a

Notes: ^a Our analytic sample was restricted to patients who survived at least 6 months after diagnosis to allow full assessment of insurance continuity over the 13-month window; this alleviates any potential bias from the reverse causal relationship (i.e., patients who died within six months from diagnosis would not have continuous Medicaid coverage). In our analysis, each patient was followed from six months after diagnosis for five years (i.e., up to 5.5 years after diagnosis).

Multivariable models suggested significantly higher risk of all-cause death (adjusted hazard ratio [aHR]=1.41, 95% CI=1.10–1.81, p=0.008) and cancer-specific death (aHR=1.46, 95% CI=1.12–1.90, p=0.005; Table 2) among patients with newly gained Medicaid, compared to those with continuous Medicaid. Patients with private insurance had lower risk of all-cause death (aHR=0.52, 95% CI=0.37–0.73, p<0.001) and cancer-specific death (aHR=0.52, 95% CI=0.36–0.74, p<0.001), relative to those with continuous Medicaid.

Table 2.

Adjusted association between insurance continuity patterns and risk of death among children and adolescents newly diagnosed with cancer

Characteristics	All-Cause Death (n=1,800)			Cancer-Specific Death (n=1,792)a
	aHR	95% CI	P-value	aHR	95% CI	P-value
Insurance continuity patterns
Continuous Medicaid	Ref			Ref
Newly gained Medicaid	1.41	(1.10, 1.81)	.008	1.46	(1.12, 1.90)	.005
Noncontinuous Medicaid	1.06	(0.76, 1.48)	.73	0.97	(0.66, 1.43)	.89
Private insurance (EMR-reported)	0.52	(0.37, 0.73)	<.001	0.52	(0.36, 0.74)	<.001
Other, unknown, or no insurance (EMR-reported)	1.15	(0.42, 3.14)	.78	1.17	(0.43, 3.21)	.76
Age at initial diagnosis (years)
0–14	Ref			Ref
15 or older	0.81	(0.63, 1.04)	.09	0.85	(0.66, 1.10)	.21
Sex
Male	Ref			Ref
Female	1.02	(0.83, 1.26)	.84	1.02	(0.82, 1.28)	.85
Race/ethnicity
Non-Hispanic White	Ref			Ref
Non-Hispanic Black	1.15	(0.89, 1.48)	.28	1.19	(0.91, 1.56)	.20
Hispanic	0.86	(0.61, 1.22)	.40	0.91	(0.63, 1.32)	.61
Non-Hispanic other or unknown race/ethnicity	1.22	(0.80, 1.87)	.35	1.30	(0.88, 1.92)	.19
Cancer type
Lymphoma	Ref			Ref
Leukemia	3.03	(1.78, 5.17)	<.001	4.01	(2.13, 7.56)	<.001
CNS tumor	1.43	(0.85, 2.39)	.18	1.92	(1.03, 3.59)	.04
Non-CNS solid tumor	1.74	(1.04, 2.91)	.03	2.36	(1.28, 4.34)	.006
Year of initial diagnosis
2007–2011	Ref			Ref
2012–2016	1.21	(0.98, 1.49)	.08	1.18	(0.94, 1.47)	.15
Cancer relapse or second cancer b	22.98	(17.30, 30.54)	<.001	23.51	(17.11, 32.29)	<.001
Rurality of residence
Non-Metropolitan areas	Ref			Ref
Metropolitan areas	0.90	(0.68, 1.19)	.47	0.99	(0.72, 1.37)	.97
SDI
Below 75th percentile	Ref			Ref
75th percentile or higher	1.06	(0.84, 1.34)	.64	1.06	(0.82, 1.37)	.63

Note: aHR: adjusted hazard ratio. CI: confidence interval. Ref: reference. EMR: electronic medical records. CNS: Central Nervous System. SDI: Social Deprivation Index.

^aExcluded 8 patients with unknown cause of death.

^bCancer relapse or second cancer was included as a time-varying covariate. Of our analytic sample (N=1,800), only 27 patients experienced both relapse and second cancer; for these patients, we used their earliest event of relapse or second cancer.

DISCUSSION

The linkage of Medicaid administrative data to a large cohort of patients with pediatric and adolescent cancers from Children’s Healthcare of Atlanta indicates that 71.8% of patients were enrolled in Medicaid at some point in the period from six months before to six months after their initial cancer diagnosis. However, less than half (47.6%) of Medicaid-insured patients had continuous coverage. Continuous coverage of care is critical for early cancer detection, prompt diagnosis, and timely treatment initiation. We demonstrate that having continuous Medicaid prior to and throughout diagnosis (vs. gaining Medicaid coverage only at, or after, diagnosis) was significantly associated with an improvement in 5-year survival.

After adjusting for patients’ clinical and sociodemographic characteristics, children who gained Medicaid coverage only at or after cancer diagnosis faced a 41% higher risk of all-cause death and 46% higher risk of cancer-specific death at five years post-index date, while those with private insurance had 48% lower risk of all-cause and cancer-specific death, compared to those with continuous Medicaid coverage. These findings align with previous research focusing on pediatric oncology patients.^14,16 Importantly, our study contributes to this literature by identifying children who gained Medicaid coverage only at the point of or after their cancer diagnosis. Further, we enriched the analysis with clinical characteristics, including recurrence of primary malignancy and second cancer, that were critical in driving pediatric cancer survival, utilizing the rich data from institutional EMRs.

There are potential explanations for the observed survival differences. First, continuous coverage may increase access to care for detecting cancer at earlier stages or with less severe morbidity. There are no screening guidelines for early cancer detection for most cancers occurring in children and adolescents. Pediatric and adolescent cancers can have a range of warning signs that patients, caregivers, and/or primary care physicians can detect; survival may improve if patients have access to care in a timely fashion when cancer can be detected and treated at a point of lower acuity or with less advanced disease.^34,35 Detecting cancer through physical examination or symptom evaluation requires access to and contact with the healthcare system.^36,37 Insurance coverage gaps, even for a short period, may delay access and healthcare contact. In adult-onset cancers, such delays have been shown to be associated with late detection and metastatic disease at diagnosis.^15,38 Our analysis adds to this literature by providing new evidence for pediatric and adolescent cancers.

Second, seamless insurance coverage following a diagnosis may improve survival by allowing timely initiation of needed treatment. Children who are not covered by Medicaid at the initial point of diagnosis may require additional time for caregivers to navigate the insurance system, including applying for and being accepted by Medicaid even with eligibility criteria having been met; this can delay treatment initiation.³⁹ Prolonged time to treatment initiation has been associated with worse prognosis in adult cancers and with increased risk of death for some pediatric and adolescent cancers.⁴⁰

More research is needed to better understand the root causes of the observed survival differences by insurance continuity patterns among pediatric and adolescent oncology patients, including assessing the role of disease acuity and severity at initial presentation, stage at diagnosis, time to treatment initiation, and receipt of guideline care. While our analysis is limited to early mortality, patients without continuous Medicaid coverage before diagnosis may be also at risk of lacking continuous insurance following completion of initial cancer treatment, and therefore, having reduced financial access to surveillance to detect disease recurrence, second cancer, and late effects of treatment. Understanding the trajectory and continuity of insurance coverage post-treatment and through long-term survivorship is a critical area meriting future investigation.

Our findings have important implications for the state and national dialogue surrounding Medicaid programs, particularly addressing challenges related to coverage gaps. In Georgia, children age 18 or younger from households with incomes up to 252% of the federal poverty level (FPL) are eligible for Medicaid or the Children’s Health Insurance Program (CHIP).⁴¹ Children from higher-income households may obtain coverage through other Medicaid eligibility pathways, such as the Katie Beckett waiver, which does not consider parental income.⁴² Yet, the Katie Beckett waiver generally covers only a period of active cancer treatment; many families face challenges navigating the complicated renewal procedures to maintain coverage during and after their first-line treatment. For young adults ages >18 who are no longer eligible for CHIP, options to enroll and maintain on Medicaid become even more limited especially in Georgia, a non-expansion state. Medicaid income eligibility threshold for those with dependents is as low as 31% of FPL (approximately $8,000 annual income for a family of three in 2023), and those without dependents are largely ineligible for Medicaid in Georgia.⁴³ Consequently, adolescent patients turning 18 amidst therapy or during follow-up care may lose Medicaid coverage entirely. Nationally, the continuous Medicaid enrollment mandate, implemented under the Families First Coronavirus Response Act (FFCRA) during the COVID-19 pandemic, was terminated in early 2023.⁴⁴ This termination was estimated to result in 7.2 million children losing Medicaid coverage.⁴⁵

Together, these challenges underscore the importance of our finding that Medicaid-eligible children and adolescents lacking continuous Medicaid coverage are at elevated risk of poorer cancer survival. Clearly, additional efforts are needed at both the institution/community and state levels to provide seamless Medicaid coverage for children and adolescents and for survivors of pediatric/adolescent cancers. Such efforts may include the adoption of state policies aimed at simplifying the procedures for Medicaid enrollment and renewal, initiation of community outreach and education programs to enhance knowledge about Medicaid eligibility and benefits, and institutional investments into social workers, insurance navigators, and other healthcare workforce that can help children and their families navigate the Medicaid system. These policy and infrastructure efforts are critical to current gaps underlying the continuity of Medicaid coverage and care for children with cancer.

Interestingly, a group of patients (n=408) recorded in EMR as privately insured at the diagnosis encounter was linked to Medicaid administrative data and subsequently classified into one of the Medicaid groups in our analysis; specifically, the majority (n=316, 77.5%) were classified as newly gained Medicaid, followed by continuous Medicaid (n=54, 13.2%) and noncontinuous Medicaid (n=38, 9.3%; Appendix Table 3). This finding may be explained by two reasons. First, in states like Georgia, children with a new cancer diagnosis can become eligible for Medicaid, even if they already have private insurance, through waiver programs such as Katie Beckett Medicaid.^46,47 Second, this finding potentially highlights “Medicaid undercount” (i.e., discrepancies between EMR-based estimates of Medicaid enrollment and the number of enrollees actually reported in state/national administrative data).^48,49 This is consistent with previous evidence that registry- or EMR-reported payor information involved varying levels of misclassification biases across clinics.⁵⁰ Our linkage of EMR with Medicaid administrative data goes beyond the EMR measurement of patients’ insurance type at a one-time point by characterizing enrollment dynamics and coverage continuity over time.

Limitations

Our study has limitations. First, our cohort was based at a single large cancer care system, which covers >90% of pediatric cancer care in the state of Georgia, which may limit the generalizability of our findings. Patterns of Medicaid continuity in Georgia, to date a non-expansion state, may not reflect those in states that have expanded Medicaid under the ACA. Nevertheless, because our institutional experience captures the majority of pediatric and adolescent cancer cases treated in Georgia,^26,51,52 we have essentially generated population-based, statewide estimates on the survival impact of insurance continuity. Second, our sample sizes for certain insurance groups, particularly patients with noncontinuous Medicaid and those with other or no insurance, were relatively small; this may limit the statistical power to detect survival differences and prevent us from distinguishing the effect of having other insurance and being uninsured.

Third, since SSN were required for Medicaid data linkage, those missing SSN were excluded. A comparison of patient characteristics between those with and without SSN showed that patients missing SSN were more likely to be younger (age group 0–14 years: 70% among patients with SSNs vs. 96% among patients without SSNs; Appendix Table 4), which may reflect parental delays in SSN application and administrative delays or errors in processing SSN applications for a period after birth.⁵³ Additionally, children missing SSNs may be undocumented immigrants, a population that our institution serves. Reassuringly, other characteristics were generally similar between the two groups, particularly the distribution of EMR-based insurance type. Additionally, we lacked data on insurance continuity for patients not insured with Medicaid; future studies may address this gap by linking pediatric/adolescent oncology cohorts to other insurance payor data (e.g., all-payer claims).

Finally, although our models adjusted for SES measures, including rurality (a proxy for distance to healthcare resources) and SDI (a census tract level, multidimensional measure of SES), it is possible that unmeasured demographic factors and social determinants of health (SDOH) such as household material hardship⁵⁴, which go beyond insurance, may also influence the survival differences observed between patients with private insurance and continuous Medicaid enrollees.⁵⁵ These confounders could exist at both the system level (e.g., challenges in finding specialists who accept Medicaid) and the patient level (e.g., less social support, lower health literacy, preexisting conditions before a cancer diagnosis, and logistic barriers to care despite having insurance), which may be more prevalent among Medicaid-insured families.^56–58 With the introduction of new International Classification of Diseases, Tenth Revision (ICD-10) codes for collecting data on SDOH⁵⁹ and the upcoming CMS SDOH reporting requirements starting in 2024,⁶⁰ future research should investigate the prevalence of various SDOH aspects, such as housing insecurity and transportation barriers. This line of work is essential to understanding the interplay between SDOH, insurance coverage continuity, and pediatric cancer survival.

CONCLUSION

In this cohort study of pediatric and adolescent patients newly diagnosed with cancer at a comprehensive cancer center in Georgia, we found that lacking continuous Medicaid coverage prior to and post-diagnosis was significantly associated with higher risks of all-cause and cancer-specific death. Future analyses will follow this cohort to examine patterns of Medicaid enrollment post-treatment through survivorship, assess their impact on cancer outcomes, identify salient reasons for insurance coverage gaps along the pediatric and adolescent cancer continuum, and evaluate the impact of unwinding the FFCRA continuous Medicaid mandate on this vulnerable young population.

Supplementary Material

PV Appendix Figure 1 OP.24.00268R1

Appendix Figure 1. Sample derivation flowchart

Notes: Of the 73 patients who died within six months after diagnosis, 53 were linked to Medicaid enrollment data (i.e., with ≥1 month of Medicaid enrollment during the 13-month assessment window), 19 had private insurance at diagnosis (electronic medical records [EMR]-reported), and one had other insurance, unknown insurance status, or no insurance at diagnosis (EMR-reported).

PV Appendix Figure 2 OP.24.00268R1

Appendix Figure 2. Classification of insurance continuity patterns among children and adolescents newly diagnosed with cancer

Notes: EMR: electronic medical record.

The “other patterns of noncontinuous Medicaid” category included 64 patients not linked to Medicaid administrative data but recorded in Children’s EMR as having Medicaid on the date of diagnosis (i.e., EMR-reported Medicaid).

PV Appendix Figure 3 OP.24.00268R1

Appendix Figure 3. Patients’ follow-up timeline

Context Summary.

Key Objective:

Many patients with cancer do not gain Medicaid coverage until diagnosis, which can reduce access to early cancer detection and timely treatment, potentially driving inferior survival. We examined whether continuous Medicaid coverage pre- through post-diagnosis (versus gaining Medicaid at/after diagnosis) provides survival benefits for pediatric/adolescent oncology patients.

Knowledge Generated:

Using a retrospective cohort from a large pediatric health system in Georgia, a state that opted out of Medicaid expansion, we showed that patients who gained Medicaid coverage only at or after cancer diagnosis had a significantly higher risk of all-cause death (adjusted hazard ratio [aHR]=1.41, 95% CI=1.10–1.81) and cancer-specific death (aHR=1.46, 95% CI=1.12–1.90) compared to those having continuous Medicaid coverage.

Relevance:

Our findings highlight the importance of continuous insurance coverage for patients newly diagnosed with pediatric/adolescent cancers and have critical implications as millions of Americans have been losing coverage under the unwinding of the Medicaid Continuous Enrollment Provision.

Data availability:

The Medicaid administrative data used in this study cannot be shared publicly per the Data Use Agreement with the Centers for Medicare and Medicaid Services. The electronic medical records from the Children’s Healthcare of Atlanta (Children’s) cannot be shared publicly per the Data Use Agreement with the Children’s.