In the United States, significant racial-ethnic disparities exist in mental health outcomes (
1). In many instances, these disparities are related to delays in treatment, different patterns of service use, and uneven quality of care. Racial-ethnic disparities have also been reported in service use among individuals with psychotic disorder. Compared with white patients, African Americans are less likely to continuously use outpatient services (
2) or to be prescribed second-generation antipsychotic medication (
3,
4), and they are more likely to be admitted to the emergency department (
3). In the public mental health system, African-American and Hispanic patients with schizophrenia are less likely than whites with similar diagnoses to receive case management (
5). Most previous findings concerning racial-ethnic differences in the treatment of serious mental illnesses rely on adult samples of chronic psychotic disorders receiving public health insurance (
2–
5). Little is known about racial-ethnic disparities in service use among commercially insured persons immediately after an initial diagnosis of psychotic disorder.
Racial-ethnic groups differ in sociodemographic and other contextual factors that affect mental health care use, including socioeconomic status (SES) (
7). SES represents an important determinant of health care use because it reflects economic resources, vocational opportunities, health literacy, and other characteristics that affect use of mental health care (
8). Additionally, race-ethnicity may affect mental health care use through several other influences, including stigma toward mental illness and treatment seeking in racial-ethnic minority communities, cultural distance between patients and service providers, and negative beliefs about and discrimination within the health care system (
9,
10).
In this study, we examined patterns of use of inpatient and outpatient mental health services and antipsychotic medication in a large, racially and ethnically diverse cohort of young, commercially insured individuals during 1 year after a new diagnosis of psychotic disorder. Based on previous studies (
2,
5,
11), we hypothesized that African-American and Hispanic individuals would be less likely than non-Hispanic white patients to use mental health services, especially outpatient care. We also explored whether any significant differences in service use among racial-ethnic groups were dependent on variation in household income, a proxy for SES. We anticipated that household income would not entirely account for racial-ethnic differences in service use following a new diagnosis of psychotic disorder.
Methods
This study involved a retrospective analysis of deidentified administrative claims data from the OptumLabs Data Warehouse (OLDW), including medical claims and eligibility information, from more than 200 million enrollees of a population representing a diverse mixture of ages, ethnicities and geographical regions across the United States (
12). The claims are dated January 1, 2009, to December 31, 2016. The household income distribution of commercial enrollees included in this study was similar to that of the commercially insured population in the entire country, although the U.S. commercially insured population includes a slightly lower proportion of patients in the highest income bracket and a slightly higher proportion in the lowest income bracket (
12). Every individual in the OLDW is assigned a unique, lifetime, and encrypted identifier to ensure that counts are unduplicated. The OLDW extract includes individuals ages 14 to 30 with comprehensive commercial insurance coverage for hospital, physician, behavioral health, and prescription drug services. We selected this age range because it coincides with the onset of a large proportion of psychotic disorders and corresponds to the age selection for many early intervention programs for first-episode psychosis (
13,
14).
Individuals with a new diagnosis of psychotic disorder were identified through their claims histories from January 1, 2011, through December 31, 2015. An initial diagnosis of psychosis was identified hierarchically by using the following ICD-9-CM and ICD-10-CM diagnosis codes: schizophrenia and schizophreniform disorder (295, F20, or F25), bipolar disorder with psychotic features (296.04, 296.14, 296.44, 296.54, 296.64, F30.2, F31.2, F31.5, and F31.64), major depressive disorder with psychotic features (296.24, 296.34, F32.3, and F33.3), and other psychotic disorders (297.1, 297.3, 298.8, 298.9, F22, F23, F24, F28, and F29). Diagnosis codes were also used to identify comorbid substance use disorder. In order to increase the likelihood of identifying the first episode of psychosis, we limited the sample to persons who had at least 2 years of continuous insurance coverage prior to the index diagnosis. We selected a 2-year cutoff to maximize the likelihood of capturing the first diagnosis while maintaining sufficient power to generate meaningful results. To support the analyses described below, we also limited the sample to persons with 1 year of full insurance coverage following the index diagnosis, because this permitted tracking of mental health care service use for the entire year.
Information on age, gender, and race-ethnicity was extracted from health insurance enrollment data. A supplier of consumer marketing data provided information on race-ethnicity derived from public and predictive data. These data were generated on the basis of a structured, rule-based system that combines analysis of first names, middle names, surnames, and surname prefixes and suffixes, with, in case of doubt, geographic reference files (
15). The supplier used imputation methods for race-ethnicity that had moderate sensitivity (48%), high specificity (97%), and moderate positive predictive value (71%) for identifying race-ethnicity (
16).
Values were categorized in the OLDW to comply with data deidentification requirements. In OLDW, race and ethnicity are defined as African American, Hispanic, Asian, or non-Hispanic white. We did not include the Asian group because of its relatively small sample (N=224) and because the socioeconomic indicators of the Asian group more closely resembled those of the non-Hispanic white group than those of other racial-ethnic minority groups. We also excluded individuals with other or unknown race-ethnicity. Place-of-service codes, procedure codes, and revenue codes were used to identify the site of index diagnosis, according to the following hierarchy: inpatient, emergency department, hospital-based outpatient department, community-based outpatient visit, or other. For example, if a person had an inpatient and an outpatient hospital claim on the same day that he or she received an index diagnosis of psychosis, the site of the index diagnosis would be identified as inpatient. The rationale was to assign the most intensive treatment setting.
Data on annual household income, a proxy for SES, was provided by OLDW and was measured by using data from both public and private consumer information (credit card statements, loan amounts, and loan payments). This variable was assigned at the household level, and a household was defined as individuals with the same surname living at the same street address. Categories included an annual household income of less than $50,000, $50,000–$99,000, more than $99,000, or “missing.”
Health Service Use
Mental health services included inpatient mental health days, outpatient mental health visits, and antipsychotic pharmacy fills. Inpatient mental health days were identified from inpatient admissions with a primary diagnosis related to mental health. Outpatient mental health services were identified as outpatient services with either a specialty mental health procedure code or a primary diagnosis related to mental health. Fills for antipsychotic medications were identified by using therapeutic class codes.
Study Design and Statistical Analysis
We used two-part models to analyze the number of inpatient mental health days, outpatient mental health visits, and antipsychotic pharmacy fills by race-ethnicity (
17–
19). These models are useful in their ability to discriminate between the decision to seek out services and the amount of services received (
20). Both of these processes, although distinct, can determine health service use and might be a source of disparities. We used logistic regression to estimate the probability of any use of services and negative binomial regression models to estimate the number of services conditional on some use. We selected these specific distributions on the basis of standard tests for assessing alternative transformed models (
21–
23). We assessed goodness of fit using a modified Hosmer-Lemeshow test and a Pregibon’s link test (
24,
25).
We examined mental health service use in the year following the index diagnosis by race-ethnicity, including additional covariates for age, gender, type of initial psychosis, comorbid substance use disorder, and site of index diagnosis. We then adjusted the results of service use for household income (<$50,000, $50,000–$99,000, >$99,000). We estimated the probability of any use and mean use among users. These estimates were standardized by using recycled predictions. Standard errors were calculated by using the nonparametric bootstrap, and p values were computed by using the percentile method from the empirical distributions of the results from 1,000 replicates (
26). All analyses were conducted in Stata, version 14 (
27).
The University of California, San Diego, Human Research Protections Program as well as the State of California Committee for the Protection of Human Subjects approved this study in accordance with the Privacy Rule of HIPAA.
Results
Sample Selection
We identified 21,003 patients between the ages of 14 and 30 years with a new diagnosis of psychotic disorder. We excluded 6,106 persons who did not have 2 years of comprehensive insurance coverage prior to the index diagnosis, 6,253 persons who did not have 365 days of coverage after the index diagnosis, 224 persons with Asian race-ethnicity, and 399 persons with unknown or other race-ethnicity.
Sample Characteristics
Study sample characteristics are shown in
Table 1. Of 8,021 persons with a new diagnosis of psychotic disorder, 6,211 (77.4%) were non-Hispanic white, 921 (11.5%) were African American, and 889 (11.0%) were Hispanic.
Compared with non-Hispanic whites, African Americans were more likely to receive their index diagnosis at the emergency department (44.1% versus 40.4%, respectively; p=0.035) and to receive a diagnosis of schizophrenia (12.8% versus 9.5%, respectively; p<0.001). Both racial-ethnic minority groups were less likely than non-Hispanic whites to receive a diagnosis of bipolar disorder with psychotic features (p<0.001). The proportion of individuals receiving a diagnosis of major depressive disorder with psychotic features was highest among Hispanics (44.5%) followed by non-Hispanic whites (41.5%) and African Americans (38.8%) (p<0.001). The proportion of individuals with a comorbid substance use disorder ranged from 12.1% among African Americans to 15.0% among Hispanics and 17.8% among non-Hispanic whites (p<0.001).
We found substantial racial-ethnic differences in household income. Larger proportions of African Americans (28.3%) and Hispanics (21.6%) were in the lower category (<$50,000 in annual income) compared with non-Hispanic whites, (10.8%) (p<0.001). The proportion of individuals with missing data on household income was similar across racial-ethnic groups, ranging from 29.2% among non-Hispanic whites and African Americans to 31.5% among Hispanics.
Mental Health Service Use
Table 2 shows standardized estimates of service use by race-ethnicity in the year after index diagnosis. Compared with non-Hispanic white individuals, Hispanics were less likely to use inpatient mental health services (32.4%±1.5% versus 36.0%±0.6%, respectively; p<0.05). Among people with inpatient use, Hispanics also had fewer inpatient days than non-Hispanic whites (10.3±0.9 versus 12.4±0.4, respectively; p<0.05).
The probability of having any outpatient mental health service use was lower among African-American (67.4%±1.4%) and Hispanic individuals (66.5%±1.5%) compared with non-Hispanic white patients (72.3%±0.6%, p<0.01 for each comparison). Among those using services, African Americans (9.7±0.7) and Hispanics (10.2±0.7) had fewer outpatient mental health visits compared with non-Hispanic whites (14.3±0.5, p<0.001 for each comparison).
The mean proportion of persons using antipsychotic medication was similar across racial-ethnic groups. Among those using antipsychotic medication, African Americans had a significantly lower mean number of fills compared with non-Hispanic white individuals (5.8±0.4 versus 6.6±0.2, respectively, p=0.042).
Additional analyses that included a time variable (index year) did not provide evidence that racial-ethnic differences in mental health service use changed during the study period.
Adjustment for Annual Household Income
Table 3 shows service use among racial-ethnic groups adjusted for household income. Many of the patterns in service use observed in the initial analyses were also observed when the association between mental health service use and race-ethnicity was adjusted for household income. Among users of inpatient services, the Hispanic group had a lower mean number of days compared with non-Hispanic white patients (10.4±0.9 versus 12.3±0.4, respectively; p=.014). Compared with the non-Hispanic white group, African-American and Hispanic individuals had a lower mean probability of using outpatient services (72.0%±0.6% versus 68.6%±1.4% and 67.3%±1.4%, respectively; p<0.05 for each comparison). Among those using outpatient mental health services, African Americans and Hispanics had fewer visits compared with non-Hispanic white patients (9.9±0.6 and 10.2±0.7 versus 14.2±0.4, respectively; p<0.001 for each comparison). When adjusted for household income, the mean probability of antipsychotic use and mean number of fills among users was similar across racial-ethnic groups.
Discussion
This study examined patterns in use of mental health services by race-ethnicity following a new diagnosis of psychotic disorder among young individuals with commercial insurance. Among patients using outpatient services, African Americans and Hispanics received about 30% fewer outpatient visits compared with the non-Hispanic white group, also following adjustment for household income. Previously reported deficiencies in the quality of care for young people with early psychosis in the United States, therefore, appear to be a particular concern for members of racial-ethnic minority groups (
6,
28). Notably, higher intensity of outpatient treatment has previously been associated with fewer hospitalizations and lower all-cause mortality rates in the years following the onset of a psychotic disorder (
6,
29,
30). Despite this marked disparity, other differences between racial-ethnic groups in mental health service use were relatively small. As such, we found limited support for our hypothesis that compared with non-Hispanic white patients, young African Americans and Hispanics are less likely to use various types of mental health services following the onset of a psychotic disorder.
Racial-ethnic patterns of inpatient mental health care and antipsychotic medication use were inconsistent. Hispanics had a lower probability of an inpatient admission and fewer inpatient days among users of inpatient mental health services but average rates of antipsychotic medication use. African Americans had average rates of inpatient service use but a lower number of mean fills of antipsychotic medication use, a finding that dissolved when adjusted for household income. Overall, during the year after index diagnosis, only a very small proportion—less than one-fourth of individuals—used antipsychotic medication and less than three-fourths used outpatient services, indicating that there is a treatment gap in pharmacological management in this population, irrespective of racial-ethnic background.
Compared with non-Hispanic white individuals, a higher proportion of African-American, but not Hispanic individuals received their index diagnosis in the emergency department, which may be a marker of a more adverse pathway to care. Previous studies from the United Kingdom and Canada showed that there was more police involvement and fewer primary care providers in the pathway to care of black individuals with first-episode psychosis compared with the white majority group (
31). Another finding that is in line with previous work pertains to racial-ethnic differences in diagnosis. African Americans were more likely than non-Hispanic whites to receive a diagnosis on the schizophrenia spectrum as their index diagnosis and less likely to receive a diagnosis of mood disorder with psychotic features. Similar ethnic differences in diagnostic patterns of psychotic disorders, especially higher rates of schizophrenia among African Americans, have been reported in previous studies (
32,
33). As this study shows, this divergence is already present at initial diagnosis within a large, commercially insured sample.
Aspects of the social environment influence help seeking for a psychotic disorder. A few of these factors include health care affordability, mental health literacy, health beliefs, racial concordance between physician and patient, patient attitudes, and provider bias (
8). Even among this commercially insured group, we observed striking racial-ethnic differences in SES, as measured by household income. However, adjusting the analyses for household income did not have substantial effects on the association between race-ethnicity and patterns of mental health service use.
This study had several limitations. First, the classification of racial-ethnic groups masks large within-group heterogeneity, and we were unable to capture the full racial-ethnic experience of African-American, Hispanic, or non-Hispanic white individuals on the basis of available data. Second, individuals with a newly diagnosed psychotic disorder were identified by using diagnoses from claims data, and we were not able to verify these diagnoses using medical record review. Thus diagnoses were based on clinical judgments and were not subject to expert validation. However, previous work has found that claims data have high specificity in identifying psychosis among young individuals in inpatient, emergency, and outpatient settings (
34).
Third, it was not possible to determine mental health service use or diagnoses prior to the 2-year look-back period. As a result, some of the patients in our cohort may have had an earlier clinical diagnosis of psychotic disorder. Moreover, the selection of a 2-year cutoff was arbitrary. Fourth, data on household income was missing for about one-third of individuals (29.4%). We therefore chose to enter household income as a covariate in the model, which may diminish the total effect of race-ethnicity. Future studies should employ mediation analyses to assess the potentially mediating role of socioeconomic indicators in the relationship between race-ethnicity and mental health service use.
Fifth, the data used for defining race-ethnicity had imperfect sensitivity and moderate positive predictive value, suggesting possible misclassification. Sixth, the large sample size may lead to spurious findings of statistical significance but limited clinical significance. Finally, a substantial proportion of the cohort was excluded because of a lack of continuous insurance coverage. A more economically unstable group may, therefore, have been excluded from analyses, which likely diminished observed disparities in service use among this group.