Although schizophrenia affects less than 1% of the U.S. population (1), it has a much larger societal impact because of its early age of onset, its chronicity, and the resultant significant long-term disability and premature mortality (2). Underuse of recommended interventions is widespread (3), and quality of care is modest at best (4–6). Moreover, persons with schizophrenia who are from racial-ethnic minority groups are less likely than whites to receive recommended interventions (7,8). Although much has been learned about factors associated with adoption of health care innovations (9,10), little is known about whether race-ethnicity influences access to innovative treatments for schizophrenia in the period that follows their release to market (11–13).
Several factors are associated with the existence and persistence of racial-ethnic health service disparities in the United States. The Institute of Medicine conceptualized service disparities as the outcome of both direct race-ethnicity effects and effects mediated by socioeconomic status, insurance coverage, and geographic location of minority groups (14). Geography is treated as a mediator of disparities because it is assumed that for minority groups, geographic location is the result of discrimination and lack of opportunities and not a personal choice (15). As demonstrated by Fisher and colleagues (16), the characteristics of the communities where patients live are associated with the volume and quality of care they receive (www.dartmouthatlas.com/index.shtm). Multiple factors are likely to be implicated in these unwarranted geographic variations in care; key among them are differences in clinicians’ treatment practices and other characteristics of the health care system (17). Because racial-ethnic minority groups are not homogeneously distributed across the United States, geographic disparities can confound the estimation of service disparities when assessed over large and diverse geographic areas (18,19).
We sought to assess the effect of race-ethnicity and geography on the adoption of an evidence-based innovation among Medicaid beneficiaries with schizophrenia who were residing in Florida. In addition, we sought to assess the contribution of geographic location to observed racial-ethnic disparities. We focused on Medicaid because of its primary role as payer of health services for people with schizophrenia (20) and selected Florida because of its size and its diverse population. The evidence-based innovation was the long-acting injectable formulation of the second-generation antipsychotic risperidone, approved by the U.S. Food and Drug Administration (FDA) on October 29, 2003.
Three characteristics of the antipsychotic prescribing practices prevalent in the United States during the period that preceded the launch of long-acting injectable risperidone (LAIR) render this drug an interesting case study. First, this form of risperidone is the only second-generation long-acting injectable antipsychotic available in the United States. This is significant because in the early 2000s, second-generation antipsychotics were generally thought to be more effective and safer than first-generation antipsychotics (21). Second, despite the recommendation by key clinical guidelines to consider long-acting injectable antipsychotics for persons with poor medication adherence (22–24), multiple studies conducted before the launch of LAIR showed low use of long-acting injectable antipsychotics in this country (25–27), perhaps because providers perceive these agents as coercive (28,29). Hence, as of its market launch, LAIR held the promise of ensuring greater adherence at a time when second-generation antipsychotics were regarded as a therapeutically superior medication class (30,31). Third, disparities research on these antipsychotic practices (use of second-generation antipsychotics and use of long-acting injectables) suggests the existence of racial disparities working in different directions. Although studies have shown that use of the purportedly superior second-generation drugs is lower among blacks than whites (8,32), some studies (32), but not all (26), conducted before the launch of LAIR showed that use of the purportedly coercive long-acting injectable antipsychotics was higher among blacks than whites.
We studied racial-ethnic disparities in adoption of LAIR among black, Latino, and non-Latino white Florida Medicaid beneficiaries with comparable need for LAIR. In our primary model, we used the disparities definition proposed by the Institute of Medicine and adjusted only for need variables. We estimated a second model that included geographic location to evaluate whether geographic disparities exist and whether geography affects the racial-ethnic disparity estimates.
Data sources and study population
We used enrollment files and medical and pharmacy claims from the Florida state Medicaid program for the period January 1, 2004, to June 30, 2006. Our study population was a cohort of continuously enrolled adults ages 18–64 years who during the study period had at least two claims recorded on two different dates with a diagnosis of schizophrenia or schizoaffective disorder (ICD-9 diagnostic code 295.xx) and who had filled at least one antipsychotic prescription. We defined continuous enrollment as having no more than two consecutive months of lapsed Medicaid enrollment and at least 23 months of enrollment over the 30-month study period. We excluded beneficiaries who had less than three months of enrollment before the first filled antipsychotic prescription. We also excluded beneficiaries with Medicare coverage or with more than two months of health maintenance organization coverage during the study period because we could not observe all their care.
Our main outcome variable was any use of LAIR, a binary-valued variable defined as one or more LAIR fills observed during the study period. We identified LAIR prescriptions of all strengths through national drug codes, which are unique product identifiers for all human drugs for commercial use in the United States.
Our main explanatory variable was race-ethnicity, defined as black, Latino, and non-Latino white. The Florida Medicaid program uses a racial classification that describes beneficiaries as white, black, Hispanic, Oriental, American Indian, or “other.” Because less than 1% of people in our cohort were classified as Oriental or American Indian during the study period, these groups were excluded from the analyses. Although the percentage of people who were classified as black or white varied little during the study period, the percentage of people classified as “other” and Hispanic varied dramatically because of changes in data recording. Analyses of beneficiaries present in the data in fiscal years (FYs) 2005 and 2006 (specifically, between July 1, 2004, and June 30, 2005) showed that 92% of beneficiaries classified as Hispanic in FY 2006 had been classified as “other” in FY 2005. Conversely, analyses of individuals classified as “other” in FY 2005 who were also observed in FY 2006 revealed that 71% were reclassified as Hispanic in FY 2006. Most of the remaining 29% retained the “other” designation, suggesting that they too were Hispanic. Because a majority of those classified as “other” during our study period were classified as Hispanic in previous or subsequent years, we reclassified the “other” group as “ever Hispanic.” As a result of this decision, we have some minor misclassification in our racial-ethnic groups. We refer to individuals classified as Hispanic or “ever Hispanic” as Latinos.
Our second primary explanatory variable was geographic location, defined by 11 geographic units used by the Florida Medicaid program to administer benefits (www.fdhc.state.fl.us/Medicaid/Areas). Because of their geographic, cultural, and socioeconomic differences, these geographic units—henceforth referred to as areas—have some latitude to discharge their administrative and quality management functions. Florida Medicaid areas encompass from one to 16 contiguous counties (median=5). Beneficiaries were assigned to the area in which they resided at the date of the first filled prescription for an antipsychotic drug. [Florida counties are listed by area in an online data supplement to this article.]
Our model included several variables found to be associated with adherence to antipsychotic medication, the main indicator of need for LAIR (33–36): age (continuous), sex, substance use disorder comorbidity, and three measures of illness severity (psychiatric comorbidity, intensity of use of inpatient services for schizophrenia, and benefit mechanism). Benefit mechanism is considered an indicator of illness severity because receipt of Supplemental Security Income (SSI) suggests a more chronic and disabling illness. Because patients’ general health status may influence the decision to prescribe LAIR, we also included two measures of medical comorbidity: metabolic comorbidity and other general medical comorbidity.
Race-ethnicity, age, sex, and geographic location were assessed at the date of the first filled antipsychotic prescription. All other variables were constructed with data observed during the three-month period before the first filled antipsychotic prescription. The comorbidity measures required the observation of one or more claims with selected ICD-9 diagnoses. Diagnoses used to construct the psychiatric comorbidity variable were major depression, dysthymia, panic disorder, obsessive-compulsive disorder, and specific personality disorders, among others. Diagnoses used to construct the substance use disorder comorbidity variable were abuse or dependence on drugs or alcohol. Diagnoses used to construct the metabolic comorbidity variable were diabetes, dyslipidemias, and obesity. Diagnoses used to construct the other medical comorbidity variable included cardiovascular and cerebrovascular disorders, neurologic disorders, and hypertension, among others. ICD-9 codes used are available from the authors on request. Intensity of inpatient service use was defined as the total number of inpatient days for schizophrenia. Benefit mechanism was a categorical variable reflecting the mechanism most frequently observed (SSI versus Temporary Assistance for Needy Families).
The unit of observation for our analyses was the beneficiary. We used logistic regression to model the log-odds of the probability of LAIR use as a function of race-ethnicity (blacks and Latinos versus non-Latino whites) and explanatory variables. The primary model included age, sex, all the comorbidity variables, intensity of inpatient service utilization, and benefit mechanism. The secondary model included all primary model variables as well as random effects for each geographic region to account for within-region correlation. This strategy reflected our assumption that, all else being equal, the probability of use of LAIR for two beneficiaries living in the same area would be more alike than the probability for two beneficiaries living in two different areas. We assumed that the random effects arose from a normal distribution. To assess for racial-ethnic effects that were uncorrelated with geographic effects, we also ran the secondary model and included geographic region as a fixed effect.
Data were analyzed with SAS version 9.1 (20). We fit a random effects model with Proc NLMixed and used a critical value of .05 to evaluate statistical significance of the p values. We report our adjusted findings as odds ratios (ORs) and 95% confidence intervals (CIs).
Our study was granted exempt status by the University of Pittsburgh Institutional Review Board because we used previously collected data that had no personal identifiers (the study was initiated when MH-L was at the University of Pittsburgh).
Study sample characteristics
We observed the care received by 13,992 Medicaid beneficiaries during the study period. Our sample included 25% blacks, 37% Latinos, and 38% whites. The mean±SD age was 44±11.4 years, and 52% of the beneficiaries were female. Three percent and 11% had any substance use disorder and any psychiatric comorbidity, respectively, and the mean number of schizophrenia-related inpatient days was .6±3.4, with a range of 0–67 days. Eight percent and 18% had any metabolic and any other medical comorbidity, respectively. Medicaid eligibility was mediated by SSI for 89% of the sample. As shown in Table 1, the racial-ethnic groups differed with regard to all need variables.
Table 1Sample characteristics of Florida Medicaid beneficiaries with schizophrenia, by racial-ethnic groupaa
| Add to My POL
|Variablebb||All (N=13,992)||Blacks (N=3,523)||Latinos (N=5,126)||Whites (N=5,343)|
|Substance use disorder comorbidity||466||3.3||182||5.2||114||2.2||170||3.2|
|Inpatient days (mean±SD)||.6±3.4||.9±4.0||.6±3.5||.5±2.9|
|Other medical comorbidity||2,443||17.5||662||18.8||992||19.4||789||14.8|
|Social Security Disability Insurance||12,428||88.8||3,275||93.0||4,753||93.0||4,400||82.4|
The overall unadjusted probability of LAIR use was 6.5%. Probability of LAIR use was comparable for blacks and whites (8.2% versus 7.2%; OR=1.66, CI=.99–1.36). Latinos, however, had a lower probability of LAIR use than whites (4.6% versus 7.2%; OR=.62, CI=.53–.73).
The unadjusted probability of LAIR use varied across the Medicaid areas. Probability of LAIR use was lowest in areas 2 and 11 (3.1% and 3.7%, respectively) and highest in areas 1 and 9 (12.3% and 14.2%, respectively) (Table 2). Whereas all racial-ethnic groups were unequally distributed across the state (Table 2), Latinos were the most concentrated, with 73% of Latino beneficiaries residing in area 11 (Miami Dade and Monroe counties). Although blacks were as likely as whites to reside in the two areas with the highest rates of LAIR use (10.9% and 10.8%, respectively), a higher proportion of them resided in the two areas with the lowest rates of LAIR use (38% and 28%, blacks and whites, respectively).
Table 2Racial-ethnic distribution of adult Medicaid beneficiaries with schizophrenia across 11 Florida Medicaid areas, by unadjusted probability (Pr) of LAIR useaa
| Add to My POL
|2 (Pr=3.1%)||11 (Pr=3.7%)||8 (Pr=6.5%)||7 (Pr=6.8%)||4 (Pr=6.8%)||6 (Pr=7.8%)||10 (Pr=8.7%)||5 (Pr=9.4%)||3 (Pr=9.8%)||1 (Pr=12.3%)||9 (Pr=14.2%)|
Multivariate analyses with the primary model.
Adjustment for need variables did not fundamentally change our unadjusted findings (Table 3). Blacks did not differ significantly from whites in their probability of LAIR use (OR=.98, CI=.82–1.16), yet Latinos’ probability of LAIR use was lower than that of whites (OR=.58, CI=.49–.70).
Table 3Adjusted estimated probabilities of using long-acting injectable risperidone among Florida Medicaid beneficiaries with schizophrenia, January 1, 2004, through June 30, 2006
| Add to My POL
|Beneficiary characteristic||Primary model||Secondary modelaa|
|Log-OR||SE||p||OR||95% CI||Log-OR||SE||p||OR||95% CI|
|Black (reference: white)||–.023||.087||ns||.98||.82–1.16||.084||.089||ns||1.09||.89–1.33|
|Latino (reference: white)||–.542||.092||<.001||.58||.49–.70||–.108||.101||ns||.90||.72–1.12|
|Female (reference: male)||–.083||.075||ns||.92||.80–1.07||–.126||.076||ns||.88||.74–1.04|
|Substance use disorder comorbidity (reference: none)||.081||.158||ns||1.08||.80–1.48||.023||.160||ns||1.02||.72–1.46|
|Psychiatric comorbidity (reference: none)||.202||.105||ns||1.22||.99–1.50||.214||.107||ns||1.24||.98–1.57|
|Intensity of inpatient service use||.131||.007||<.001||1.14||1.12–1.16||.142||.008||<.001||1.15||1.13–1.17|
|TANF (reference: SSI)bb||−1.275||.373||.001||.27||.13–.58||−1.234||.379||.009||.29||.13–.68|
|Metabolic comorbidity (reference: none)||.632||.109||<.001||1.88||1.52–2.33||.625||.111||<.001||1.87||1.46–2.39|
|Other medical comorbidity (reference: none)||.706||.087||<.001||2.03||1.71–2.40||.771||.088||<.001||2.16||1.78–2.63|
Multivariate analyses with the secondary model.
Inclusion of the geographic random effects in the model eliminated the Latino-white disparity (Table 3), a result that remained unchanged when area was included as a fixed effect (results not shown). Further, the odds of LAIR use varied substantially across areas, ranging from a low of .44 in area 11, to a high of 1.91 in area 1. Compared with the state average, odds of LAIR use were significantly lower for areas 2 and 11 and significantly higher for areas 1 and 9.
In both models, odds of LAIR use were higher for younger beneficiaries or those with SSI benefits and for those with heavier use of inpatient services for treatment of schizophrenia or general medical comorbidities (Table 3).
Our Florida-wide study of LAIR use among Medicaid beneficiaries during the 30-month period that followed its FDA approval offers a unique window into the process of adoption of mental health treatment innovations within a large and diverse state Medicaid program. Whether racial-ethnic disparities exist in the adoption of LAIR depends on whether geographic effects are accounted for. When we assessed for adoption disparities within the entire Florida Medicaid program, we found that Latinos had a lower probability of LAIR use than whites. Once we accounted for beneficiaries’ geographic location, the Latino-white adoption disparity evaporated. The explanation for the powerful effect of geography on the estimation of ethnic disparities lies in two phenomena: the geographic concentration of Latino Medicaid beneficiaries and geographic disparities in LAIR adoption within the state of Florida. Because choice of geographic area of residence depends on unobserved subject and geographic factors that may be related to medication use, our results are only associative and not causative.
We are aware of only three previous studies that have produced evidence of disparities in the adoption of antipsychotic medications in the United States. A study conducted in the Veterans Health Administration (VHA) found that black and Latino veterans with schizophrenia were less likely than whites to use ziprasidone immediately after its launch into the U.S. market (11). However, the authors found little evidence of disparities across administrative VHA regions. Opolka and colleagues (13) assessed for racial-ethnic disparities in receipt of olanzapine versus a first-generation oral antipsychotic among Texas Medicaid adult beneficiaries with schizophrenia during a 20-month period that covered the market launch of olanzapine. After adjusting for health status and geographic region, the investigators found that blacks but not Mexican-Americans were less likely than whites to receive olanzapine. The authors also found geographic variation in adjusted probability of olanzapine use. In a similarly designed study, Opolka and colleagues (12) assessed for racial differences in receipt of olanzapine versus an established second-generation antipsychotic; they did not find racial disparities but found geographic variations.
Our study differs from these studies in two respects. First, we framed our study both conceptually and methodologically as an investigation of disparities in the adoption of a new medical technology. Because policy remedies differ depending on whether the main issue is racial-ethnic versus geographic variations (37), we explicitly sought to disentangle the effects of race-ethnicity and geography in the estimation of racial-ethnic disparities. Second, we contrasted probability of use of the innovation versus use of all other antipsychotics and systematically assessed for geographic disparities through a random-effects approach that accounted for within-region correlation.
Our finding that the Florida Medicaid areas differed in their LAIR adoption rates adds to a small body of research suggesting geographic variation in adoption of mental health innovations within state Medicaid programs (12,13,38,39). Why would geographic variations exist for fee-for-service Medicaid beneficiaries subject to the same drug coverage and utilization management policies and unaffected by potential differences in managed care contracts? Although we were unable to generate evidence on the drivers of these variations, the literature points to possible explanations. The rate at which an innovation spreads through the health care system is associated with factors related to patients (race-ethnicity, socioeconomic status, and preferences), clinicians (knowledge and attitudes), the system of care (policies, organizational structure, and culture), and private-sector initiatives (advocacy and pharmaceutical promotion) (3,9,10,40–46). Evidence of differences in use of long-acting injectable antipsychotics across U.S. facilities subject to the same regulatory and financial constraints (47,48) suggests that the culture and the organizational structure in which prescribers operate may have played an important role in the observed geographic variations in adoption of LAIR. Although cultural factors of relevance include opinion leaders and attitudes toward and exposure to pharmaceutical promotion, a key structural factor is the availability of nursing staff to administer injections (10,49).
Our finding that geographic disparities confounded the estimation of racial-ethnic disparities when these were assessed for the entire state is in keeping with studies of other populations (19,50). However, as far as we are aware, ours is the first such finding for a Medicaid population with schizophrenia.
In the absence of extraneous dynamics, rate of adoption of new health care technologies should be similar for all those who stand to benefit from their use. Although the field no longer regards second-generation antipsychotics as the standard of care (51) and recent studies have produced mixed evidence on the effectiveness of long-acting injectable antipsychotics and LAIR (52,53), our study peers into a time when LAIR held the promise of delivering the advantages of both second-generation antipsychotics and long-acting injectable antipsychotics.
Our study had some limitations. First, because of the observational nature of our design, our study may not have compared racial-ethnic groups that were entirely balanced with regard to history of poor medication adherence and other factors affecting need for long-acting injectable antipsychotics. Second, the generalizability of our study may be limited because of our focus on Florida, a state that differs from many others because of its racial-ethnic diversity and its restrictive requirements for Medicaid eligibility.
Our results indicate that within a state Medicaid program, the initial finding of a Latino-white disparity in the adoption of a novel treatment for schizophrenia was driven by geographic disparities in adoption rates and the geographic concentration of Latinos in a low-adoption area. When we accounted for place of residence, the ethnic disparity disappeared. Our study adds to an expanding body of evidence suggesting that as a result of the heterogeneous distribution of racial-ethnic groups, racial-ethnic disparity estimates that represent average effects over large geographic areas may be confounded by the unaccounted effects of geographic variations. This evidence has important implications for efforts to eradicate disparities and improve quality of care for all.
This research was supported by grants R01MH087488 and P50 MHO 73469 from the National Institute of Mental Health and by grant P60 MD0 02261 from the National Center for Minority Health and Health Disparities. The authors are grateful to Roger Boothroyd, Ph.D., and Mark Engelhardt, M.S., M.S.W., for their comments on a draft of this article.
The authors report no competing interests.