Comparing Neighborhoods of Children With Autism Spectrum Disorder in a Medicaid Waiver Program and a State Population, 2007–2015

In 2007, South Carolina became one of the first states to offer a Medicaid waiver specifically for the statewide provision of early intensive behavioral intervention. For 10 years, the waiver provided up to 3 years of intervention to children between the ages of 3 and 10 who received a diagnosis of ASD from a professional psychologist by age 8. Based on need determined by provider assessment, each child received a maximum of $50,000 per year and up to 40 hours of direct line therapy per week, of which at least 50% must have taken place inside the child’s home (https://www.medicaid.gov/medicaid/section-1115-demo/demonstration-and-waiver-list/?entry=8378). We created a comprehensive data set of all children diagnosed as having ASD who enrolled in the waiver program between February 6, 2007 (when the first child was enrolled), and March 31, 2015 (the end of the first quarter of calendar year 2015). Detailed information about the waiver and data integration are provided elsewhere (unpublished manuscript, Yingling ME, 2017). Briefly, we integrated data from the South Carolina Department of Disabilities and Special Needs (DDSN) and the Office of Revenue and Fiscal Affairs (RFA). We included children in this study if the RFA assigned a census-tract ID to the residential address available in the administrative data set (N=2,111). For families that had two or more children with ASD in the program, we randomly selected one child. This reduced the data set by 50 children. We received approval from a university institutional review board.

To answer our research question, we based our comparisons on geocoded family street addresses and demographic characteristics from census tract–level data provided by the U.S. Census Bureau. We relied on neighborhood-level data because the socioeconomic indicators available at the child level were not comparable in quality to state-level demographic data collected by the U.S. Census. However, census tract–level characteristics are highly correlated with individual-level characteristics (25, 26), and we controlled for three available child-level demographic variables, which reduced measurement error. Thus, given evidence that child race-ethnicity, parent socioeconomic status, family structure, and urbanicity influence access to health care among children in general and access to services among children with ASD in particular (8, 11, 21, 22, 27, 28), we examined equity in enrollment by comparing children in the waiver program with children statewide on six characteristics for which data are collected at the census tract–level.

Specifically, we examined equity in the enrollment of children with ASD in an early intensive behavioral intervention Medicaid waiver in terms of race-ethnicity (percentage of individuals who identified as “white alone”), socioeconomic status (percentage of individuals ages 25 years or older with at least a bachelor’s degree, percentage of households living in poverty, and median annual income), family structure (percentage of female-headed households, no husband present), and geography (proportion of people living in urban versus nonurban neighborhoods). For race-ethnicity, socioeconomic status, and family structure, we used data from the 2006–2010 American Community Survey (29) in years 2007–2010 and data from the 2011–2015 American Community Survey (30) in years 2011–2015. For geography, we used the most recent rural-urban commuting area codes based on the 2010 decennial Census and the 2006–2010 American Community Survey.

Control variables include age (measured in months on the date of enrollment), sex (1=female, 0=male), and child race-ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, non-Hispanic other, and unknown). Both DDSN and RFA data included the category of unknown. Hispanic ethnicity includes children identified as Hispanic white and Hispanic black; the latter category included too few children to analyze separately, and ethnicity was prioritized. We derived the category of non-Hispanic other from a range of categories in the original data that were too small to analyze (i.e., Asian, Hawaiian/Pacific Islander, and American Indian). There were 690 children (32.7%) with unknown or missing data for race-ethnicity. Whereas the amount of missing data for child race-ethnicity is high, it is actually a much lower percentage than reported in a previous study with a similar administrative data set (22). Therefore, in an effort to preserve the integrity of the data, we included the group with unknown race-ethnicity in our analyses.

We conducted all analyses using SAS, version 9.4. For the five continuous variables, we calculated the difference between children’s neighborhood characteristics and corresponding state characteristics for each of the 9 years in which children were enrolled (2007–2015) by using geocoded demographic data at the census-tract level. We conducted ordinary least squares (OLS) regression using PROC REG to explore differences between each enrollment cohort and population values for all of South Carolina. In the OLS models, the dependent variable was the difference between South Carolina population values and the values of the children’s neighborhood. A significant F test in the regression model indicated a significant difference between neighborhood and state characteristics. We controlled for child age, sex, and race-ethnicity in all regression analyses, except for the model comparing the percentage of white residents in a neighborhood, in which we controlled only for child age and sex. To compare the proportion of children in the sample who lived in urban neighborhoods with the proportion of South Carolina residents who live in urban areas, we estimated a chi-square goodness-of-fit test. Because there is no statistical procedure for binary variables (i.e., each child’s neighborhood was coded 1=urban, 0=nonurban) comparable to the one we used with the continuous OLS models, these comparisons were bivariate and did not include any control variables. Because we estimated multiple models for all variables, we followed guidelines of the Bonferroni correction and used an adjusted alpha (α=.006).

In the sample, average age at enrollment was 66.9 months; 17.7% were female (N=374); and 40.6% identified as non-Hispanic white, 18.1% as non-Hispanic black, 5.6% as Hispanic, and 2.4% as other non-Hispanic, including Asian (1.1%) and Native American (0.1%). Table 1 shows the neighborhood values and state values for each characteristic for each year, along with information on which comparisons were statistically significant on the basis of the OLS models.

Overall, across the six characteristics that we examined, for most years, there was no evidence that children’s neighborhood characteristics differed from state characteristics (i.e., no evidence of inequitable enrollment). However, for some variables there were a few years in which there were statistically significant differences, including the percentage of individuals older than 25 years with a bachelor’s degree (2012), median household income (2011–2013), percentage of female-headed households (2009, 2011–2014), and percentage of individuals in poverty (2010, 2012, 2013). Likewise, the chi-square goodness-of-fit tests resulted in a significant difference in urbanicity between enrollees and the statewide population in 2011, 2012, and 2013. For each of these significant comparisons, the differences between enrollees’ neighborhoods and the state population favored children who lived in neighborhoods of privilege (i.e., children who lived in neighborhoods with higher median incomes, lower poverty levels, and fewer female-headed households and children who lived in urban neighborhoods). As shown in Table 2, the key elements of the OLS output used to determine whether there were statistical differences between neighborhood and state characteristics after control for child demographic characteristics generally supported the bivariate results presented in Table 1. (Information on the significance of each of the control variables in each of the OLS models is provided in the table in the online supplement to this article.)

Although researchers consistently document inequities in early identification and medical services for children with ASD, comparatively little is known about children’s enrollment in Medicaid-funded early intensive behavioral intervention. Results from this study suggest that in general, enrollment in South Carolina was equitable. Findings also suggest that children with ASD who live in neighborhoods characterized by greater privilege could be more likely to get enrolled in interventions. Depending on the year, the neighborhood socioeconomic status, family structure, and geography of families enrolled in South Carolina’s Medicaid waiver differed significantly from those of the state population. However, this was the exception and not the rule.

After the first 3 years of the program’s existence, enrollees lived in neighborhoods with higher median household incomes (2011–2013) and a lower percentage of children who lived in poverty (2010, 2012, 2013) than the statewide population. These results are consistent with the finding that children who were enrolled in Wisconsin’s early intensive behavioral intervention program lived in census tracts with a higher percentage of families with incomes greater than or equal to 200% of the federal poverty level (22). Findings demonstrate the importance of including measures of both poverty and affluence in research on health care inequities among children with ASD. Notably, there were no significant differences in any of the variables studied in 2007, 2008, or 2015. Although unclear, it is possible that the parents of children who live in neighborhoods with higher median incomes become aware of new programs that are available for children with ASD earlier and enroll their children in services sooner, before a waitlist is established. Approximately 6 months after the waiver’s inception, South Carolina established a waitlist to manage the overwhelming numbers of eligible children. If parents who lived in neighborhoods with lower median incomes and higher poverty levels learned about the waiver program later than parents who lived in neighborhoods with higher median incomes and lower poverty levels, children’s enrollment would be delayed. This hypothesis aligns with current research on delays in enrollment in state early intervention services (e.g., speech therapy, occupational therapy, audiology services) made available through part C of the Individuals with Disabilities and Education Act for children younger than age 3. Specifically, low-income mothers report that lack of substantial information, conflicting information, and busy schedules prevent them from seeking these services for children with developmental delays (31).

In years 2009 and 2011–2014, enrollees lived in neighborhoods with a lower percentage of female-headed households. This finding conflicts with a recent study in which being a single parent was associated with a lesser degree of perceived challenges to utilization of allotted intervention hours among children in the same Medicaid waiver program examined in this study (32). Yet it aligns with reports that children of single parents experience greater challenges to health care utilization (27, 28, 33, 34). It is possible, therefore, that children of female-headed households experience difficulty with enrolling in intervention, but once enrolled, parents perceive fewer barriers to utilization. Furthermore, evidence of equity in 2015 suggests that these parents take longer to become aware of available resources. This population may be an important target for outreach efforts.

Children who do not live in urban areas may also benefit from outreach, given that fewer children who live in nonurban areas were enrolled in the program in 2011, 2012, and 2013. This difference supports the experience of children who live in rural areas whose parents report greater difficulty in accessing ASD treatment providers than their urban counterparts (8). Results also add credence to the hypothesis that urbanicity is not related to utilization of early intensive behavioral intervention because children in nonurban areas are less likely to enroll and therefore make up a smaller percentage of samples (35).

The examination of neighborhood characteristics and enrollment in one of the largest statewide early intensive behavioral intervention programs yields important insights for program administrators and state policy makers. However, study results must be interpreted in the context of several limitations. Despite the relevance of study results to the provision of publicly funded early intensive behavioral intervention, generalization is limited given that the study examined enrollment in one state. The use of census tract–level data, rather than individual-level data, limits conclusions that can be drawn regarding children’s and families’ demographic characteristics. However, because this study reduced measurement error by controlling for child age, sex, and race-ethnicity, it is more robust than similar analyses (22). Additionally, although census tracts are more precise than counties or zip codes, they are an imperfect measure of neighborhood. Because of the small number of children living in rural neighborhoods, it was necessary to combine rural and suburban neighborhoods into one category. This may have diminished the relationship between urbanicity and enrollment. Moreover, the small number of children in the program from rural tracts is telling, given that South Carolina is one of the most rural states in the country (36). In a study on participants in the same waiver program, once placed on a waitlist, children waited an average of 587 days before being enrolled (37). It is possible, therefore, that children from rural areas were placed on the program waitlist but were never enrolled.

As the provision of early intensive behavioral intervention expands across the United States, results underscore the value in monitoring the equity of enrollment over time. This study detected equity in the South Carolina waiver program, and states may turn to South Carolina for insight on how to facilitate equitable enrollment, although the reasons for the program’s equity are unknown. At minimum, it is important for states to conduct targeted outreach efforts to underserved populations to promote equitable enrollment. Moreover, as states shift early intensive behavioral intervention provision from Medicaid waivers to early and periodic screening, diagnostic, and treatment services—or begin providing it for the first time—additional research on equitable enrollment at both state and national levels will be prudent.

The authors acknowledge the South Carolina Department of Disabilities and Special Needs (DDSN) and the University of South Carolina’s Institute for African American Research for their support of this work.