Disparities in Receipt of Specialty Services Among Children With Mental Health Need Enrolled in the CMHI

To date, the CMHI program has collected intake data on six cohorts of participating children (corresponding to distinct funding phases), with the cohort defined by the year funding began for the site. In addition, extensive follow-up data are also collected at six-month intervals for a subsample of these children—the sample for the longitudinal outcomes study. Enrollment in the outcomes study was variable; smaller sites recruited all willing families, whereas larger sites could opt for sampling strategies. The mix of strategies ensured adequate numbers of participating children at all sites. All information was obtained through in-person interviews with each child and his or her primary caregiver.

We used data from phases 4 (years 2002–2004) and 5 (years 2005–2006) of the CMHI. Because data collection quality improved over time, earlier phases could not be combined. We used baseline and follow-up data collected at six and 12 months for children ages 2–18 enrolled in the longitudinal outcomes study (N=8,848). We excluded children with missing data on age (N=179) and children who did not have 12 months of follow-up data because they were recruited too late (N=952) or because they were missing follow-up surveys (N=3,797). The final sample consisted of 3,920 children, including 2,228 children at 26 phase 4 sites and 1,692 children at 29 phase 5 sites. In bivariate comparisons, baseline characteristics of the children retained versus those excluded from the sample were for the most part similar. However, the final sample had a larger proportion of African Americans (27% versus 24%), a smaller proportion of Latinos (18% versus 21%), a larger proportion of children ages 6–13 (52% versus 46%), a smaller proportion of children ages 14–18 (34% versus 41%), a larger proportion of caregivers with depression (41% versus 38%), and a larger proportion of caregivers who had at least a high school education (45% versus 42%). The final sample also had slightly worse scores on the behavioral checklist used in the study (1.31 and 1.41 points higher than the excluded children on the internalizing and externalizing behavioral scales, respectively). These differences were not considered clinically significant.

The main study outcomes were the total number of days over the 12 months after enrollment during which the child received services in each of the following categories: individual psychotherapy, family and group psychotherapy, medication monitoring, assessment and evaluation, case management, residential, and inpatient.

Children’s race-ethnicity was categorized on the basis of caregiver report as white (non-Latino), African American, Latino, Native American, Asian, Pacific Islander, and multiracial. Other child-level predictors of service use were gender, age group (2–5, 6–10, 11–13, and 14–18), living situation (at home with parents or relatives, with nonrelatives, in foster care, or not living at home at any time during the 12 months), and insurance coverage (public, private, both public and private, and no insurance). The child’s behavioral problems were measured by the internalizing and externalizing scales of the Child Behavior Checklist (CBCL) (24).

The models also adjusted for caregiver-level predictors, including depression, other mental health problems, household income, number of children in the household, and education (less than high school, high school completed, and at least some college).

Unadjusted racial-ethnic differences in utilization could be attributable to differences in site, child, or caregiver characteristics. To address the clustering of children within sites, we considered several options, including generalized estimating equations, multilevel models, random-effects models (a special case of multilevel models), and fixed-effects models (25). [A detailed discussion of the strengths and limitations of each model is included in an online supplement to this article.] Our choice of fixed-effects models was based on several considerations, including a focus on the main effects of race-ethnicity, rather than on profiling of sites, and limited sample sizes by race-ethnicity at the site level. Most important, we considered the biggest threat to the validity of our findings to be potential confounding of the child’s race with unobserved site-level heterogeneity. For example, children from minority groups might be more likely to live in areas with a more limited provider supply or worse quality of care (both unobserved). In this case, fixed-effects models would still yield unbiased estimates of the effects of race-ethnicity, because they allow for correlation between observable individual characteristics and site-level heterogeneity (26,27). Other models might not. Therefore, the race-ethnicity effect could pick up not only differences in children’s experience at a given location but also the effect of being in a location with worse access and quality. Given that our primary concern was to avoid confounding by unmeasured site characteristics, fixed-effects models appeared to be the most conservative choice.

Because our utilization outcomes were variables with large numbers of zero values and skewed conditional distributions, we employed two-part regression models (28) to evaluate the associations of each outcome with race-ethnicity. We first estimated conditional fixed-effects logistic regressions to predict the probability of any use of the service type being examined. Estimates reported from these regressions are (adjusted) odds ratios (ORs), standard errors, and p values. The OR associated with each race-ethnicity group reflects the odds of any use of the service among that group compared with white non-Latino children.

We then estimated conditional fixed-effects negative binomial models for the number of days the service was used among the subset of children who used that type of service. Estimates reported from these models are incidence rate ratios (IRRs), standard errors, and p values. IRRs reflect the relative rate of use among each racial-ethnic group compared with white non-Latino children; for example, if African-American children had an IRR of .7, it means that they received only .7 times the number of service days per year that the white children did.

In initial analyses, regression coefficients from multiply imputed data and complete-case data were similar. Therefore, only the complete-case results are presented. For brevity, tables show estimates for individual psychotherapy, family and group psychotherapy, and medication monitoring; results for other outcomes are summarized in the text.

All analyses were performed with Stata MP 12.1. Prior to conducting analyses, we obtained institutional review board approval and a waiver of consent from the University of California, Los Angeles.

Sample characteristics are summarized in Table 1. Mean CBCL scores were 65.7 for externalizing behaviors and 70.0 for internalizing behaviors; scores above 63 are in the clinical range. The most common race-ethnicity was white non-Latino, followed by African American and Latino. Roughly two-thirds of the children were boys. Most children resided in their own homes with parents or other relatives.

More than half of caregivers reported household incomes of less than $20,000 per year; less than 5% reported household incomes over $75,000 per year. Forty-five percent of caregivers reported having attended or graduated from college. Most children had insurance coverage, predominantly public.

Table 2 provides descriptive information on utilization patterns by service type, including the proportion of the sample using each type of service and, among those using it, the mean and standard deviation of the number of days on which the service was used. Most children used individual psychotherapy, assessment and evaluation, and case management services. Roughly half used medication monitoring and family or group psychotherapy services. Residential and inpatient care were less frequently used. Among users, the average number of days of service use varied widely.

Table 3 shows the regression-adjusted ORs and IRRs associated with each racial-ethnic group. For any use of services, African-American children had significantly lower odds than white non-Latino children of using individual psychotherapy (OR=.73, p=.019), family and group psychotherapy (OR=.79, p=.043), and psychotropic medication monitoring (OR=.51, p<.001). Latino children also had lower odds of receiving medication monitoring (OR=.70, p=.007) and assessment and evaluation services (OR=.75, p=.027; results not shown). No other significant associations were found of race-ethnicity with the probability of any specific service use.

In the negative binomial regressions of the number of days on which services were used (Table 3), among children who received individual psychotherapy, African-American and Latino children had lower rates of utilization than white non-Latino children. For African Americans, IRRs were .79 (p<.001) for individual psychotherapy and .86 (p=.011) for family and group psychotherapy. For Latinos, IRRs were .91 (p=.044) for individual psychotherapy and .88 (p=.044) for family and group psychotherapy. Among children receiving medication monitoring, the only minority group for which a difference was found in service use rate was Pacific Islanders, for whom the IRR was .60 (p=.009). Finally, among children who were hospitalized (data not shown), African-American children had only three-quarters as many inpatient days per year as white non-Latino children (IRR=.75, p=.026). Associations of racial-ethnic group with all other outcomes (assessment and evaluation, case management, and residential treatment) were not significant.