Association of Family Income With Morbidity and Mortality Among US Lower-Income Children and Adolescents

Key Points

Question

Among children and adolescents from lower-income families in the US, is family income associated with health?

Findings

This cross-sectional study included 795 000 US participants aged 5 to 17 years with family income under 200% of the federal poverty threshold who accessed health care through Medicaid or the Children’s Health Insurance Program. Higher family income was significantly associated with a lower prevalence of diagnosed infections, mental health disorders, injury, asthma, anemia, and substance use disorders and lower 10-year mortality.

Meaning

Higher family income was associated with lower claims-based measures of morbidity and mortality among children and adolescents in lower-income families in the US.

Abstract

Importance

Family income is known to be associated with children’s health; the association may be particularly pronounced among lower-income children in the US, who tend to have more limited access to health resources than their higher-income peers.

Objective

To investigate the association of family income with claims-based measures of morbidity and mortality among children and adolescents in lower-income families in the US enrolled in Medicaid or the Children’s Health Insurance Program.

Design, Setting, and Participants

This cross-sectional analysis included 795 000 participants aged 5 to 17 years enrolled in Medicaid (Medicaid Analytic eXtract claims, 2011-2012) living in families with income below 200% of the federal poverty threshold (American Community Survey, 2008-2013). Follow-up ended in December 2021.

Exposures

Family income relative to the federal poverty threshold.

Main Outcomes and Measures

Record of International Classification of Diseases, Ninth Revision codes for an infection, mental health disorder, injury, asthma, anemia, or substance use disorder and death record within 10 years of observation (Social Security Administration death records through 2021).

Results

Among 795 000 individuals in the sample (all statistics weighted: mean [SD] income-to-poverty ratio, 90% [53%]; mean [SD] age, 10.6 [3.9] years; 56% aged 10 to 17 years), 33% had a diagnosed infection, 13% had a mental health disorder, 6% had an injury, 5% had asthma, 2% had anemia, 1% had a substance use disorder, and 0.6% died between 2011 and 2021, with the mean (SD) age at death of 19.8 (4.2) years. For those aged 5 to 9 years, higher family income was associated with lower adjusted prevalence of all outcomes, except mortality: children in families with an additional 100% income relative to the federal poverty threshold had 2.3 (95% CI, 1.8-2.9) percentage points fewer infections, 1.9 (95% CI, 1.5-2.2) percentage points fewer mental health diagnoses, 0.7 (95% CI, 0.5-0.8) percentage points fewer injuries, 0.3 (95% CI, 0.09-0.5) percentage points less asthma, 0.2 (95% CI, 0.08-0.3) percentage points less anemia, and 0.06 (95% CI, 0.03-0.09) percentage points fewer substance use disorder diagnoses. Except for injury and anemia, the associations were more pronounced among those aged 10 to 17 years than those 5 to 9 years (P for interaction <.05). For those aged 10 to 17 years, an additional 100% income relative to the federal poverty threshold was associated with a lower 10-year mortality rate by 0.18 (95% CI, 0.12-0.25) percentage points.

Conclusions and Relevance

Among children and adolescents in the US aged 5 to 17 years with family income under 200% of the federal poverty threshold who accessed health care through Medicaid or the Children’s Health Insurance Program, higher family income was significantly associated with a lower prevalence of diagnosed infections, mental health disorders, injury, asthma, anemia, and substance use disorders and lower 10-year mortality. Further research is needed to understand whether these associations are causal.

This study examines the association between family income and the prevalence of claims-based diagnoses for 6 types of chronic and acute conditions, along with 10-year mortality, among children and adolescents in lower-income families.

Introduction

The positive relationship between income and health has been well established in many different countries and contexts.^1,2,3,4 The association has been found to be strongest in adulthood, but to originate in early childhood.^5,6,7 In the US, where the health insurance system for people younger than 65 years is fragmented and health care access varies substantially by socioeconomic circumstances, the association between higher income and lower adult mortality is particularly strong at lower income levels.⁸ This finding may reflect the lifelong accumulation of steeper gradients in health among lower-income children.⁹ Knowing the shape of the childhood health-income gradient at different points in the income distribution for morbidity and mortality measures may be important for understanding the nature of the health-income patterns observed among adults. Extensive literature has examined the health-income gradients among children.^5,6,7,10,11 In the context of the US, nationally representative studies of the association between family income and child health traditionally relied on self-reported measures of health and relatively coarse measures of income due to the challenge of connecting health care encounter data and socioeconomic data at the individual level.^5,6,7

The current study used novel linkages between national Medicaid claims data (2011-2012), the American Community Survey (ACS; 2008-2013), and the US Census Bureau version of the Social Security Administration death records (through 2021) to measure the association between family income and the prevalence of claims-based diagnoses for 6 types of chronic and acute conditions, along with 10-year mortality, among children and adolescents in lower-income families enrolled in Medicaid or the Children’s Health Insurance Program (CHIP). The study hypothesis was that children and adolescents living in families with moderately higher income would have a lower prevalence of these conditions as well as lower incidence of 10-year mortality, with any associations more pronounced among older children.

Methods

Institutional review board determination for this analysis was obtained through Stanford University; this research was determined to not involve human subjects as defined in federal regulations 45 CFR 46.102 or 21 CFR 50.3. The analysis used preexisting deidentified data. The US Census Bureau reviewed this data product for unauthorized disclosure of confidential information and approved the disclosure avoidance practices applied to this release under authorization numbers CBDRB-FY22-POP001-0088 and CBDRB-FY23-POP001-0003.

Participants

Study participants were identified using 3 data sources. First, Medicaid Analytic eXtract (MAX) data from the Centers for Medicare & Medicaid Services, which is a collection of nationally harmonized enrollment and claims data for persons enrolled in Medicaid and CHIP, were obtained for 2011 to 2012. Our data extract did not include MAX files for Florida for any years and was missing a subset of claim files for Idaho in 2011, thus these states were excluded from analysis. All other US states and the District of Columbia were included in the analysis. Second, we obtained records from the 2008 to 2013 1-year American Community Survey (ACS). The ACS is a cross-sectional nationally representative household survey of 1% of the US population administered by the US Census Bureau. Approximately 3 million to 3.5 million addresses are selected annually and response rates range from 90% to 97%; person weights are provided in the survey to account for sampling design. Third, we obtained the US Census Bureau’s version of the Social Security Administration’s Numerical Identification (Census Numident) database with records through the end of 2021. The Numident file is a collection of information from applications for Social Security numbers and cards. The data set is cumulative, adding individuals as they receive Social Security numbers. Deceased individuals are not removed from the data and their date of death is recorded. Under the Enhancing Health Data (EHealth) program of the US Census Bureau, we merged these 3 data sources. The merge used a unique individual-level anonymous identifier that was common across all data sources, as assigned by the US Census Bureau, referred to as a Protected Identification Key, which is created using personally identifiable information and probabilistic record linkage and has been described elsewhere.^12,13,14 Our analysis required observing a valid Protected Identification Key for the individual to be included in the study.

From the merged data, we selected a subsample of individuals who were aged 5 to 17 years in 2011 to 2012 based on the date of birth as recorded in the Census Numident; eligible for Medicaid or CHIP in the same state for 12 consecutive months in 2011, 2012, or both (based on MAX Person Summary files); eligible for Medicaid or CHIP coverage for reasons other than disability (excluding eligibility codes 12, 22, 32, 42, and 52 from the MAX Person Summary files); and had family income under 200% of the federal poverty threshold when observed in an ACS sample from 2008 to 2013. The last criterion restricted income to be measured at most 5 years before and 2 years after our measurement of Medicaid or CHIP eligibility. The age was restricted to 5 years or older because ACS as well as other records of the US Census Bureau may undercount children aged 0 to 4 years.¹⁵

Income-to-poverty ratios compare family income with an appropriate poverty threshold that depends on the size of family and the number of related children and is updated every year for inflation. The US Census Bureau reported that for a family of 3 with 1 related child in the family, the poverty threshold corresponded to an annual income of $18 106 in 2011 and $21 811 in 2021. Our analyses included families with income up to 200% of the federal poverty threshold, which corresponded to $36 212 in 2011 and $43 622 in 2021 for a family of 3 individuals. US Census Bureau poverty statistics suggest that nationally among children and adolescents aged 5 to 17 years, 43.0% lived in families with income at or below 200% of the federal poverty threshold in 2011 (34.4% in 2021).¹⁶ In the same year, among children and adolescents aged 0 to 18 years in families with income at or below 200% of the federal poverty threshold, 65.3% were enrolled in Medicaid.¹⁷

Exposure

The exposure of interest was family income, measured using the income-to-poverty ratio as reported in the ACS. This measure expressed the total income of each family over the previous 12-month period in percent of the (family size–adjusted) federal poverty threshold. Most families were observed only once in 2008 to 2013 ACS survey years, but in the rare cases of multiple observations, we used income as recorded in the most recent year of data.

Outcomes

Outcome variables of interest were indicators for a diagnosis of an infection, mental health disorder, injury, asthma, anemia, or a substance use disorder and death by December 31, 2021.

An individual was considered to have a mental health disorder if at least 1 of the following 8 conditions was present in their Medicaid claims according to the CMS Chronic Conditions Data Warehouse (CCW) aggregation of International Classification of Diseases, Ninth Revision (ICD-9) codes and claim frequency: ADHD, conduct disorders, and hyperkinetic syndrome; anxiety disorders; autism spectrum disorder; bipolar disorders; depressive disorders; personality disorders; posttraumatic stress disorder; and schizophrenia and other psychotic disorders. An individual was considered to have a substance use disorder if at least 1 of the following 3 conditions were present according to CMS Chronic Conditions Data Warehouse: alcohol use disorder, tobacco use disorder, or drug use disorder. Indicators for asthma and anemia were directly available as CMS Chronic Conditions Data Warehouse indicators.

ICD-9, Clinical Modification variables in MAX Inpatient and MAX Other Services files that cover claims for physician services, laboratories and imaging, outpatient care, clinic services, and home services were used to construct indicators for having had an infection or an injury. We used the 2011 to 2012 “short list” of ICD-9 codes for family medicine provided by the American Academy of Family Physicians (AAFP) to define infections and injuries. An individual was considered to have an infection diagnosis if there was a claim record with an ICD-9 code from section I “infectious & parasitic diseases” of the AAFP short list, an ICD-9 code corresponding to all diagnoses with “otitis” in the short list description, or an ICD-9 code from section VIII “respiratory system” other than those for asthma. An individual was considered to have an injury diagnosis if there was a claim record observed with an ICD-9 code from section XVII “injuries & adverse effects” of the AAFP short list. The exact list of ICD-9 codes is provided in the eAppendix in Supplement 1.

Date of death in the Census Numident record was used to create an indicator for the individual dying between 2011 and 2021.

Statistical Analysis

The individual-year data were collapsed into a data set with 1 observation per individual, inputting the individual’s age in 2012 if the person was observed in both 2011 and 2012. We took the maximum of each diagnosis indicator between 2011 and 2012, thus classifying an individual as having a condition if it was observed in either 2011 or 2012. There were no missing data in our sample because to construct the sample we required observing individuals continuously enrolled in the Medicaid program or CHIP, year of birth, and income. Conditional on observing this information, we had no missing data for any outcomes or demographic adjustment variables.

A linear multivariable regression model was estimated for each outcome comparing the prevalence across children and adolescents with different levels of family income (which entered the model as a continuous income-to-poverty ratio), allowing for a differential effect of income among children aged 5 to 9 years (reference group) compared with children and adolescents aged 10 to 17 years. The relationship between each outcome and income was estimated adjusting for race and ethnicity as fixed effects (as self-reported in the ACS based on detailed fixed categories and a write-in option), female sex as a fixed effect (as self-reported in ACS), fixed effects for state of residence (as reported in MAX Person Summary files), whether age was measured in 2011 or 2012, and fixed effects for the year of the ACS that provided income information. We thus measured the association between income and the prevalence of conditions (and mortality) net of variation in morbidity and mortality that can be attributed to other observable demographic (such as sex and race and ethnicity) and geographic differences.

The coefficient multiplying the income term was the coefficient of interest. It measured, in each age group when the interaction term was added, what the absolute difference in the probability of the outcome was at 100% of poverty higher income (corresponding to $18 106 more income in 2011 and $21 811 more in 2021 for a family of 3). Tests of statistical significance were based on 95% CIs for a 2-tailed t test (P value < .05 used as significance threshold), accounting for ACS sampling survey weights and clustering of standard errors among individuals who resided in the same state. To aid in the interpretation of magnitudes we also computed the size of the association coefficient relative to adjusted prevalence of each outcome within each age group at mean income in the sample. Because of the potential for type I error due to multiple comparisons, findings should be interpreted as exploratory. All analyses were performed in Stata for Unix 16.1/MP.

To illustrate the shape of the relationship between health and income graphically, we plotted the adjusted and weighted prevalence of each condition (in percentages) within each ventile of the in-sample income distribution for children aged 5 to 9 years and for children and adolescents aged 10 to 17 years. These data points were plotted together with the age group–specific linear trends from the regression model described above.

The main analysis used aggregate measures of mental health and substance use disorders as outcomes. In the eFigure in Supplement 1, a graphical illustration of the association between adjusted prevalence and income is shown for 8 underlying mental health disorders and 3 underlying substance use disorders individually.

Results

Study Sample Characteristics

The cross-sectional analytic sample included 795 000 (weighted N = 59 390 000) unique children and adolescents (Table). Mean (SD) weighted family income in the sample was 90% (53%) of the federal poverty threshold. Mean (SD) weighted family income among children and adolescents in families with below-median (in-sample weighted) income was 46% (26%) of the federal poverty threshold; among children and adolescents in families with above-median (in-sample weighted) income, mean (SD) income was 134% (32%) of the federal poverty threshold. Mean (SD) weighted age was 10.6 (3.9) years; 44% of participants were children aged 5 to 9 years and 56% were children and adolescents aged 10 to 17 years. The mean weighted age among children and adolescents in families with above-median income was 0.2 years higher (P < .001) than in families with below-median income. In the weighted sample, 50% of children and adolescents were female and 50% were male; families with above-median income had 0.8 percentage points (weighted) fewer female children (P < .001). The weighted race and ethnicity composition of the sample was 34% Hispanic, 25% non-Hispanic Black, 35% non-Hispanic White, and 5% non-Hispanic other race or ethnicity. The race and ethnicity composition differed by income. In families with above-median income, participants were 8.2 percentage points (weighted) less likely to be non-Hispanic Black (P < .001), 7.2 percentage points (weighted) more likely to be non-Hispanic White (P < .001), and 0.3 percentage points (weighted) more likely to be Hispanic (P = .02). The 5 states with the highest percentage of participants in the sample were California (14%), Texas (10%), Illinois (6%), New York (6%), and Ohio (5%) (all weighted percentages). Overall, 15% of study participants (weighted) were observed to be continuously enrolled in Medicaid only in 2011 and thus their age and outcomes were measured in 2011 only. The number of study participants observed in each wave (2008-2013) of ACS varied between 15% and 18% (weighted).

Table. Characteristics of Study Participants Overall and by Family Income^a.

Characteristic	Weighted No.
	Full sample	Below-median income	Above-median income	Absolute difference between above- and below-median income	P value
Unweighted individuals	795 000	387 000	408 000
Individuals (weighted)	59 390 000	29 695 000	29 695 000
Income-to-poverty ratio, mean (SD), %b	89.8 (53.1)	45.5 (25.9)	134.4 (31.8)	88.9	<.001
Age, mean (SD), y	10.6 (3.9)	10.5 (3.9)	10.7 (3.9)	0.2	<.001
Aged 5-9 y, %	44.1	45.0	43.2	–1.8	<.001
Aged 10-17 y, %	55.9	55.1	56.8	1.8	<.001
Female	49.7	50.1	49.3	–0.8	<.001
Male	50.3	49.9	50.7	0.8	<.001
Race and ethnicity, %
Hispanic	34.4	34.3	34.6	0.3	.02
Non-Hispanic Black	25.3	29.3	21.2	–8.2	<.001
Non-Hispanic otherc	5.3	4.9	5.6	0.7	<.001
Non-Hispanic White	35.0	31.5	38.6	7.2	<.001
Geography (5 states with the largest sample), %
California	13.8	13.8	13.8	–0.1	.60
Texas	10.0	11.2	8.7	–2.5	<.001
Illinois	6.3	5.6	7.1	1.5	<.001
New York	6.1	6.8	5.4	–1.4	<.001
Ohio	5.3	5.5	5.0	–0.5	<.001
Year individual age was measured, %
2011	15.4	13.9	17.0	3.1	<.001
2012	84.6	86.2	83.0	–3.1	<.001
Year income was observed in ACS, %
2008	15.0	14.2	15.7	1.5	<.001
2009	15.7	15.5	15.9	0.4	<.001
2010	17.9	18.1	17.7	–0.3	.002
2011	18.3	19.0	17.7	–1.3	<.001
2012	17.1	17.5	16.7	–0.8	<.001
2013	16.0	15.8	16.3	0.4	<.001
Individuals with outcome conditions, %
Infection	32.6	33.7	31.6	–2.1	<.001
Mental health disorder	13.1	13.8	12.4	–1.4	<.001
Injury	5.9	6.1	5.6	–0.5	<.001
Asthma	5.1	5.4	4.7	–0.7	<.001
Anemia	2.3	2.4	2.1	–0.3	<.001
Substance use disorder	1.0	1.0	0.9	–0.1	.07
10-y mortality, %	0.6	0.6	0.5	–0.1	<.001
Age at death, mean (SD), y	19.8 (4.2)	19.6 (4.2)	20.0 (4.1)	0.4	.02

^a Source: 2011-2012 CMS MAX, 2008-2013 1-Year American Community Survey, Census Numident. All results were approved for release by the Census Bureau, authorization CBDRB-FY22-POP001-0088 and CBDRB-FY23-POP001-0003.

^b Poverty is defined as the federal poverty threshold, which corresponded to an annual income of $21 811 for a family of 3, with 1 child, in 2021.

^c Other race category includes American Indian and Alaska Native, Asian, Native Hawaiian and Other Pacific Islander, other race, and 2 or more races.

In the full sample, 33% (all weighted) of children and adolescents had an infection in 2011 and/or 2012, 13% of children and adolescents had a mental health disorder diagnosis, 6% had an injury, 5% had an asthma diagnosis, 2% had anemia, and 1% had a diagnosed substance use disorder; 0.6% of children and adolescents (weighted) died by the end of 2021, with a mean (SD) weighted age of death of 19.8 (4.2) years (Table). Except for substance use disorder, the prevalence of each condition and mortality was lower among children and adolescents in families with above-median income (P < .001). In proportional terms, the largest difference was observed for mortality, which was 15% (weighted) lower among children and adolescents in families with above-median income compared with children and adolescents in families with below-median income, followed by anemia and asthma (both 13% lower), mental health disorders (10% lower), injuries (8% lower), and infections (6% lower) (all weighted).

Except for anemia, the adjusted prevalence of all conditions and the incidence of 10-year mortality differed between those aged 5 to 9 years compared with those aged 10 to 17 years (eTable in Supplement 1) (Figure). Children and adolescents aged 10 to 17 years were 1.6 (95% CI, 1.0-2.2) percentage points (weighted) less likely to have an infection and 1.2 (95% CI, 1.0-1.5) percentage points less likely to have asthma (a difference of –5% for infections and –21% for asthma relative to the adjusted prevalence in those aged 5-9 years at mean income). Older children and adolescents aged 10 to 17 years were more likely to have a diagnosis of a mental health disorder (weighted difference, 6.6 [95% CI, 6.0 to 7.3] percentage points), an injury (weighted difference, 3.9 [95% CI, 3.6-4.3] percentage points), or a substance use disorder (weighted difference, 1.5 [95% CI, 1.4-1.6] percentage points). These absolute (weighted) differences corresponded to 69% times higher prevalence of mental health disorder diagnosis, 105% times higher prevalence of injury, and 573% times higher prevalence of a substance use disorder relative to the adjusted prevalence in those aged 5 to 9 years at mean income. Those aged 10 to 17 years also had 0.8 (95% CI, 0.7 to 0.9) percentage points (or more than 3-fold relative to those aged 5-9 years) higher 10-year mortality.

Figure. Adjusted Morbidity and Mortality by Family Income.

Prevalence and income measured as income-to-poverty ratio (in percentage) are regression-adjusted for sex, race and ethnicity, state of residence, year that age was measured, and year that income was observed. The trend lines and their slopes correspond to multivariable regression estimates reported in the eTable in Supplement 1. The federal poverty threshold corresponded to an annual income of $21 811 for a family of 3 with 1 child in 2021. Source: 2011-2012 CMS MAX, 2008-2013 1-Year American Community Survey, Census Numident.

Associations of Income With Morbidity and Mortality

Among children aged 5 to 9 years, living in a family with an additional 100% income relative to the federal poverty threshold was associated with a difference of –2.3 (95% CI, –2.9 to –1.8) percentage points in the rate of infections, –1.9 (95% CI, –2.2 to –1.5) percentage points in the rate of a mental health disorder diagnosis, –0.7 (95% CI, –0.8 to –0.5) percentage points in the rate of injury, –0.3 (95% CI, –0.5 to –0.1) percentage points in the rate of asthma, –0.2 (95% CI, –0.3 to –0.1) percentage points in the rate of anemia, and –0.06 (95% CI, –0.09 to –0.03) percentage points in the rate of a diagnosed substance use disorder (eTable in Supplement 1) (Figure). For this age group, family income was not significantly associated with the probability of dying within 10 years.

Except for injury and anemia, the slope of the associations between income and morbidity was steeper among children and adolescents aged 10 to 17 years (eTable in Supplement 1) (Figure). An additional 100% income relative to the federal poverty threshold in this age group was associated with a difference of –3.7 (95% CI, –4.3 to –3.2) percentage points in the rate of infections (difference in slope vs younger age group, –1.4; P < .001), a difference of –2.3 (95% CI, –2.7 to –1.9) percentage points in the likelihood of a mental health diagnosis (difference in slope vs younger age group, –0.45; P = .02), a difference of –0.5 (95% CI, –0.7 to –0.4) percentage points in the rate of asthma (difference in slope vs younger age group, –0.25; P = .02), and a difference of –0.3 (95% CI, –0.4 to –0.2) percentage points in the rate of a substance use disorder diagnosis (difference in slope vs younger age group, –0.3; P < .001). In the older age group, the probability of dying within 10 years was significantly associated with income, with a difference of –0.2 (95% CI, –0.2 to –0.1) percentage points in 10-year mortality in families with 100% more income relative to the federal poverty threshold.

Relative to the level of adjusted prevalence at mean income within each age group in families with 100% more income relative to the federal poverty threshold, among children aged 5 to 9 years, the estimated associations corresponded to a 7% lower rate of infections, 19% lower rate of a mental health diagnosis, 18% lower probability of an injury, 5% lower probability of asthma, 9% lower rate of anemia, and 22% lower rate of substance use disorders. Among children and adolescents aged 10 to 17 years, the corresponding relative differences associated with an additional 100% of the federal poverty threshold in income were a 12% lower rate of infections, 15% lower probability of a mental health diagnosis, 12% lower rate of injury, 12% lower rate of asthma, 21% lower rate of a substance use disorder diagnosis, and 21% lower 10-year mortality.

Qualitatively similar patterns were observed for the conditions underlying the mental health disorders indicator and the substance use disorders indicator (eFigure in Supplement 1). The most prevalent among mental health disorders was ADHD. In both age groups, lower family income was significantly associated with a higher adjusted prevalence of ADHD and related disorders (P < .001 for both age groups), anxiety disorders (P < .001 for both age groups), depressive disorders (P < .001 for both age groups), bipolar disorder (P < .001 for both age groups), posttraumatic stress disorder (P = .004 for age 5-9 y and P < .001 for age 10-17 y), and schizophrenia and other psychotic disorders (P = .02 for age 5-9 y and P = .002 for age 10-17 y). In the younger age group, but not in the older age group, lower family income was also significantly associated with a higher prevalence of autism spectrum disorder (P = .02). The reverse was observed for the personality disorders outcome, with a statistically significant association between higher income and lower prevalence only in the older age group (P = .002). Substance use disorders were rare in the younger age group. In those aged 10 to 17 years, drug use disorder was the most prevalent condition. Higher family income was significantly associated with a lower prevalence of all 3 conditions underlying the substance use disorder indicator in the older age group (alcohol use disorder [P = .02], drug use disorder [P < .001], and tobacco use disorder [P < .001]).

Discussion

Among 795 000 children and adolescents aged 5 to 17 years living in families with income under 200% of the federal poverty threshold and enrolled in the Medicaid program or CHIP, higher income was associated with a lower probability of having a diagnosis record for infections, mental health disorders, injury, asthma, anemia, and substance use disorders and lower 10-year mortality.

These results were broadly consistent with many other studies that either used other sources of data from the US or examined childhood health-income gradients in other countries, but with some notable differences.^{5,6,18,19,20,21,22} Qualitatively, the results among children and adolescents who were all enrolled in Medicaid (at least at the time of measurement) were consistent with findings in non-US high-income contexts that children in lower-income families have worse health even in the presence of health insurance.^{22,23,24,25,26,27,28,29} For the US context, unlike the study by Evans et al⁷ but similar to the study by Case et al⁵ and other studies, injuries and a diagnosis of asthma were significantly associated with family income.^30,31,32 There was no statistically significant association between family income and 10-year mortality among children aged 5 to 9 years, which is in contrast to many studies that have found that an income-mortality gradient is present at any age.^5,33 A key distinguishing feature of the current study was that it only included children and adolescents with Medicaid or CHIP coverage, suggesting that it would be important to further investigate the ability of Medicaid to flatten mortality gradients in this age group.^34,35,36

The finding of a more pronounced association between most outcomes and income among older children and adolescents was consistent with several studies that did not find support for the “equalization in youth” hypothesis, which argues that childhood inequality in morbidity and mortality declines in adolescence and early adulthood in the US context.^{5,19,25,26,37,38,39} For injuries and anemia, there was not a steepening of the gradient, but there was also not a decline in the gradient as would have been consistent with the notion of equalization. Currie and Stabile²³ argued that the higher frequency of health shocks at lower income levels, rather than a worse ability to recover from these shocks, may explain the more pronounced gradient among older children and adolescents. In this study, the findings on the differential likelihood of experiencing an injury or an infection by income provided support for the hypothesis that children and adolescents in lower-income families experience more health shocks.

The existence of a statistically significant health-income gradient does not imply that the steepness of the gradient was clinically important. In relative terms, the beneficial association with higher income was the largest for substance use disorders, mental health, injuries among younger children, and mortality among older children and adolescents. In these instances, the probability of a substance use disorder, a mental health disorder, and mortality for older children decreased by approximately 20% for approximately $20 000 in additional annual income. The smallest relative difference was found for asthma (5%). For comparison, Larson and Halfon³² used data from the 2003 National Survey of Children's Health, where income was reported in terms of the poverty ratio, and found that moving from below 100% relative to the federal poverty threshold to between 100% to 200% relative to the federal poverty threshold corresponded to a decline in an asthma diagnosis probability from 13.8% to 13% (a 6% reduction), a decline in ADHD from 8.2% to 7.0% (15% reduction), and a decline in depression from 5.8% to 4.3% (26% reduction). These results were similar in that the mental health outcome was particularly strongly related to income. The analysis in this study found that the magnitudes of gradients for substance use disorder and mortality among older children and adolescents were similar to the steep gradient in mental health. Overall, more evidence is needed using consistent measures of outcomes and exposure at all points in the income distribution to further understand the clinical importance of the gradients for each condition.

Limitations

This study had several limitations. First, a cross-sectional design identifies correlations and inherently precludes reaching any causal conclusions. The focus on children and adolescents reduced, but did not eliminate, concern of reverse causality from health to income; reverse causality may still exist if having a child with poor health leads parents to reduce labor supply. Moreover, there is an important potential for unmeasured confounding that may have influenced both familial income and children’s health. Second, the sample was restricted to children and adolescents older than 5 years because ACS as well as other records of the US Census Bureau may undercount children aged 0 to 4 years.¹⁵ It was therefore not possible to measure the health-income gradient in the first months and years of life. Third, the analyses were restricted to a nonrandom sample of all children and adolescents from low-income families in the US. To link data sources, the analysis required individuals to have a valid Protected Identification Key. Previous literature shows that Protected Identification Key assignment may be nonrandom because immigrants, young people, and members of racial and ethnic minority groups are less likely to receive a Protected Identification Key. Participants had to have been continuously enrolled in the Medicaid program, which increased the accuracy of diagnosis measures, but further led to a nonrandom selection of children and adolescents. Fourth, the analysis was only able to capture those instances of disease for which an individual sought formal health care and received a diagnosis and for which Medicaid claims were generated. There could be underdiagnosed cases not captured by the data.¹⁰ Moreover, any changes in Medicaid policies related to reimbursement and billing, but unrelated to true underlying changes in health, would appear as changes in diagnosis prevalence in the data. Fifth, all known limitations (eg, variation in data availability and quality across states, delays in reporting, incomplete reporting of managed care encounter data) of reporting Medicaid fee-for-service claims data and Medicaid managed care encounter data into the MAX database apply to the study. Sixth, the analyses relied on 1 observation of family income, which may not be an accurate measure of permanent income. Seventh, the underlying measure of income in ACS was self-reported; although the quality of income reporting is considered high, this measure may not capture all economic resources available to families.⁴⁰

Conclusions

Among US children and adolescents aged 5 to 17 years with family income under 200% of the federal poverty threshold who accessed health care through Medicaid or CHIP, higher family income was significantly associated with a lower prevalence of diagnosed infections, mental health disorders, injury, asthma, anemia, and substance use disorders and lower 10-year mortality. Further research is needed to understand whether these associations are causal.

Supplement 1.

eAppendix

eTable

eFigure

Supplement 2.

Data sharing statement