The second-generation antipsychotic clozapine is the only antipsychotic drug approved by the U.S. Food and Drug Administration for the management of treatment-resistant symptoms and recurrent suicidal behavior, two severe complications of schizophrenia (1,2). Furthermore, among all antipsychotics, clozapine has the lowest rate of suicide-related and all-cause mortality (3,4). Despite these advantages, clozapine is underused in the routine care of U.S. patients with schizophrenia (5–7). Consistent with a body of evidence suggesting that patterns of schizophrenia care are influenced by patients’ race-ethnicity (7–9), studies have shown that use of clozapine is lower among patients from minority groups than among white patients (10).
Although clozapine’s superior efficacy has been long established (11,12), evidence about its superior effectiveness had lagged behind until publication in the mid-2000s of results from the publicly funded Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study (13). The study’s primary measure of effectiveness was time to drug discontinuation for any cause—“a synthesis of clinician and patient judgments” on a drug’s efficacy and tolerability (14). Use of time to discontinuation for any cause as a measure of effectiveness is supported by evidence that has shown that longer time to discontinuation is associated with greater clinical and functional improvement (15,16). Unfortunately, like efficacy researchers before them, CATIE researchers did not report on whether the comparative effectiveness of clozapine and the other antipsychotics varies by racial-ethnic group. We know of two inpatient studies that assessed whether black and white patients differed in their response to clozapine; one study found no differences, and the other had mixed findings (17,18). However, we are not aware of any study that has assessed clozapine treatment response by Latino ethnicity.
The purpose of this study was to fill this evidentiary gap. Using time to discontinuation for any cause as our measure of effectiveness, we examined the effect of race-ethnicity on the comparative effectiveness of clozapine and other antipsychotics among patients in maintenance antipsychotic treatment. The paucity of evidence on racial-ethnic differences in antipsychotic effectiveness precluded us from formulating hypotheses.
To construct the study cohort, we used enrollment files and medical and pharmacy claims from the Florida Medicaid program for fiscal years 2001–2005 (July 1, 2000, through June 30, 2005). Cohort members were adults aged 18–64 years who had at least two claims with a diagnosis of schizophrenia (ICD-9 diagnostic codes 295.xx) recorded on two different service dates during each fiscal year (FY) and who had filled at least one prescription for an antipsychotic drug during the study period.
We selected patients receiving maintenance antipsychotic treatment with clozapine or any of seven antipsychotic drugs (olanzapine, quetiapine, risperidone, ziprasidone, aripiprazole, haloperidol, and perphenazine). We focused on patients in maintenance antipsychotic treatment because drug discontinuation during maintenance-phase treatment is less likely than discontinuation during the acute treatment phase to be influenced by intolerability and patient factors associated with adherence behavior. We defined maintenance treatment as prescriptions filled for any of the eight study drugs beyond an initial three-month period (representing acute-phase treatment). Episodes were excluded if during the three-month period preceding the date when the antipsychotic prescription was first filled, the beneficiary had fewer than three months of continuous fee-for-service enrollment or any health maintenance organization (HMO) or Medicare coverage. When more than one qualifying maintenance antipsychotic treatment episode was observed for a beneficiary, we selected the first episode. Episodes that were ongoing at the end of the study period were observed for one additional year.
Our primary outcome variable was time to discontinuation of medication for any cause among beneficiaries receiving maintenance treatment with clozapine versus other antipsychotics. We operationalized discontinuation as a break in prescription fills of 45 or more days. We selected this interval to allow for dose reductions or short-lived treatment interruptions attributable to refill delays stemming from less than optimal adherence or other treatment barriers commonly encountered by this population (7). We sought to capture a broader set of poor outcomes through a secondary outcome variable that measured either time to antipsychotic discontinuation or time to hospitalization for schizophrenia.
Our main explanatory variables were antipsychotic drug treatment, race-ethnicity, and their interaction. Antipsychotic drug treatment categories were clozapine versus a group of seven antipsychotic drugs (other antipsychotics). The group of other antipsychotics comprised all then-available second-generation antipsychotics (olanzapine, quetiapine, risperidone, ziprasidone, and aripiprazole) and two first-generation antipsychotics (haloperidol and perphenazine). We included haloperidol because of its dominance in the first-generation antipsychotic market and its prominent role in antipsychotic outcomes research, and we included perphenazine because it was the first-generation drug included in the CATIE trial. Because fewer studies have assessed the comparative effectiveness of clozapine versus other second-generation agents, we conducted additional analyses that compared only clozapine to the second-generation agents in the “other antipsychotic” group.
Race-ethnicity was defined as black, Latino, and non-Latino white, and it was assessed as of the date when the study drug prescription was first filled. The Florida Medicaid program uses a racial classification that describes beneficiaries as white, black, Hispanic, Oriental, American Indian, or “other.” We excluded the small proportion of beneficiaries (less than 1%) otherwise meeting inclusion criteria who during the study period were classified as Oriental or American Indian. While the percentage of beneficiaries who were classified as black or white varied little during the study period, the percentage classified as “other” and Hispanic varied dramatically because of changes in data recording. For example, an analysis of beneficiaries present in the data in both FY 2005 and FY 2006 showed that 92% of those who were classified as Hispanic in FY 2006 had been classified as “other” in FY 2005. In addition, individuals classified as “other” in FY 2005 who were also present in the FY 2006 data indicated that 71% were reclassified as Hispanic in FY 2006. Most of the remaining 29% remained “other” in 2006, which may indicate that that they too were Hispanic. Therefore, because most beneficiaries who were classified as “other” during our study period were classified as Hispanic in previous or subsequent years, we reclassified the “other” group as “ever categorized as Hispanic” and we included these patients and all those classified as Hispanic in a group that we labeled “Latinos.” As a result of this decision, we have some minor misclassification in our racial-ethnic groups.
Other explanatory variables included several need-related variables that may have an impact on the decision to use clozapine over other antipsychotics. These were age (a continuous variable), sex, and four measures of illness severity: psychiatric disorder comorbidity, substance use disorder comorbidity, intensity of use of inpatient services for schizophrenia, and Supplemental Security Income (SSI) status (further described below). Because patients’ general medical status and geographic location may influence prescriber decision making, additional explanatory variables were metabolic comorbidity and other general medical comorbidity and a geographic variable indicating 11 multicounty areas used by the Florida Medicaid program to administer benefits. [A table listing the counties in each of the 11 areas is available in an online data supplement to this article.] Age, sex, and area of residence were assessed as of the date when the antipsychotic drug prescription was first filled. The measures of comorbidity, intensity of inpatient use, and SSI status were constructed with data observed during the three-month period before the date when the antipsychotic prescription was first filled (baseline period). The binary comorbidity measures required the observation of one or more claims with selected ICD-9 diagnoses. These diagnoses were major depression, dysthymia, and anxiety disorders (psychiatric comorbidity); abuse of and dependence on alcohol or drugs other than tobacco (substance use disorder comorbidity); diabetes, dyslipidemias, and obesity (metabolic comorbidity); and all chronic general medical disorders (other general medical comorbidity). Intensity of inpatient service use was defined as the total number of inpatient days with a primary diagnosis of schizophrenia. SSI status, a severity indicator because SSI eligibility suggests a more disabling illness, was operationalized as SSI, Temporary Assistance for Needy Families (TANF), or other.
To account for possible secular trends, we also created a categorical variable indicating the quarter in which the antipsychotic prescription was first filled.
Propensity score models for clozapine use.
We used propensity score matching to address potential selection bias in medications received (19). We modeled the log-odds of receiving clozapine versus other antipsychotics as a function of the explanatory variables (age, sex, four comorbidity variables, intensity of inpatient use, SSI status, geographic area, and quarter of the year). Our propensity score models were estimated separately within each race-ethnicity group. We estimated log-odds of receiving clozapine versus other antipsychotics and compared probabilities between patients who received clozapine and those who received other antipsychotics. We evaluated the comparability of clozapine-treated patients and those treated with other antipsychotic drugs by assessing the extent to which the patients overlapped on each variable. Adequate comparability was defined a priori as standardized differences of 10% or smaller between the groups (20). For each clozapine-treated patient, we identified two similar persons of the same race-ethnicity who were treated with other antipsychotics. This resulted in matched sets for each racial-ethnic group of one clozapine-treated person and two persons treated with other antipsychotics. We assumed that use of clozapine and race-ethnicity might vary widely by geographic area and that medication discontinuation rates might vary over time; therefore, we used fine balancing (21) to exactly match clozapine and other antipsychotic users on the basis of when they filled their index prescription (quarter) and where they lived (area) at that time.
For each matched set of patients, we assessed number of days from the start of the antipsychotic episode to four mutually exclusive outcomes, whichever was observed first: antipsychotic discontinuation as defined above, hospitalization for schizophrenia, withdrawal (defined as any discontinuity in Medicaid enrollment or more than one month of HMO or Medicare coverage), and administrative censoring (defined as ongoing use of the study drug by June 1, 2006, the end of the additional period of observation).
We first estimated Kaplan-Meier curves of time to antipsychotic discontinuation and considered the other three outcomes (hospitalization for schizophrenia, withdrawal, and administrative censoring) as censored events. We also estimated Kaplan-Meier curves of time to antipsychotic discontinuation or to hospitalization for schizophrenia and considered withdrawal and administrative censoring as censored events. Final model estimates were based on Cox proportional hazard regression models that accounted for the matched sets and included a clozapine dummy variable, a set of race-ethnicity dummy variables, and a set of interaction terms between antipsychotic drug treatment and race-ethnicity. We assessed differences in risk of discontinuation among the groups (that is, the proportionality assumption) through graphical summaries of Schoenfeld residuals plotted against log(time), chi square tests for each covariate (drug treatment and race-ethnicity), and a chi square global test. Statistical models were estimated using the S-PLUS software system, version 8.0. Statistical significance was set at p≤.05. We report comparisons in the form of relative risks (RRs) and 95% confidence intervals (CIs).
Our study was granted exempt status by the University of Pittsburgh Institutional Review Board because we used data free of personal identifiers.
Characteristics of the study cohort
Over the entire study period, 41,262 persons met diagnostic, age, and racial-ethnic criteria and also filled an antipsychotic prescription. Most of them filled prescriptions for one or more of the eight study drugs (99%, N=40,725). Further application of our inclusion and exclusion criteria reduced the size of the study cohort to an analytic sample of 20,122 persons, each contributing one episode of maintenance antipsychotic treatment. [A flowchart depicting the cohort-building steps is available in the online data supplement to this article.] As shown in Table 1, 23% of the patients were black, 36% were Latino, and 41% were white. The mean age was 42 years, 52% of the patients were female, and 87% were receiving SSI. Comorbidity rates ranged between 5% for substance use disorder comorbidity and 29% for other general medical comorbidity. Most episodes were observed in area 11, which includes Miami-Dade and Monroe counties (41%) and occurred in the first six quarters of the study period (50%, N=10,125).
Table 1Characteristics of adult Medicaid beneficiaries with schizophrenia, by drug treatment
| Add to My POL
| Substance use||1,027||5.1||20||2.7||1,007||5.2||−13.0|
| Other general medical||5,816||28.9||75||10.0||5,741||29.6||−37.5|
| 1 (July–Sept. 2000)||5,188||25.8||475||63.4||4,713||24.3||85.7|
| 2 (Oct.–Dec. 2000)||1,351||6.7||39||5.2||1,312||6.8||−6.6|
| 3 (Jan.–March 2001)||1,055||5.2||38||5.1||1,017||5.3||–.8|
| 4 (April–June 2001)||828||4.1||24||3.2||804||4.2||−5.0|
| 5 (July–Sept. 2001)||919||4.6||12||1.6||907||4.7||−17.7|
| 6 (Oct.–Dec. 2001)||784||3.9||13||1.7||771||4.0||−13.5|
| 7 (Jan.–March 2002)||771||3.8||14||1.9||757||3.9||−12.2|
| 8 (April–June 2002)||765||3.8||13||1.7||752||3.9||−13.0|
| 9 (July–Sept. 2002)||765||3.8||21||1.9||744||3.8||−5.8|
| 10 (Oct.–Dec. 2002)||748||3.7||8||1.1||740||3.8||−17.9|
| 11 (Jan.–March 2003)||862||4.3||11||1.5||850||4.4||−17.4|
| 12 (April–June 2003)||785||3.9||12||1.6||773||4.0||−14.5|
| 13 (July–Sept. 2003)||752||3.7||16||2.1||736||3.8||−9.8|
| 14 (Oct.–Dec. 2003)||690||3.4||12||1.6||678||3.5||−12.1|
| 15 (Jan.–March 2004)||721||3.6||10||1.3||711||3.7||−15.0|
| 16 (April–June 2004)||646||3.2||6||.8||640||3.3||−17.7|
| 17 (July–Sept. 2004)||679||3.4||8||1.1||671||3.5||−16.1|
| 18 (Oct.–Dec. 2004)||632||3.1||8||1.1||624||3.2||−14.9|
| 19 (Jan.–March 2005)||649||3.2||5||.7||644||3.3||−19.1|
| 20 (April–June 2005)||533||2.6||4||.5||529||2.7||−17.4|
Clozapine was used in 3.7% of the person-episodes (N=749). The proportions of blacks (2.3%) and Latinos (2.1%) using clozapine were lower than the proportion of whites (5.9%). For the other 19,373 episodes, the index antipsychotic drug was a second-generation antipsychotic other than clozapine in 87% of the cases (N=16,903); risperidone (28%, N=5,398) and olanzapine (35%, N=6,692) were the most frequently used drugs. On the basis of standardized differences, clozapine users differed from users of other antipsychotics on most of the variables included in our multivariate models. With few exceptions, these differences were also observed within each racial-ethnic group.
The overall rate of discontinuation was 72.7% (N=14,634). Mean and median times to discontinuation were 794 days and 548 days, respectively. As shown in Table 2, clozapine users had lower rates of discontinuation than other antipsychotic users regardless of race-ethnicity. Latinos had lower rates of discontinuation from clozapine treatment than blacks and whites, but the differences were not statistically significant. Whites had lower rates of discontinuation from other antipsychotics than blacks and Latinos. Results for time to discontinuation mirrored the results for rates.
Table 2Unadjusted analysis of antipsychotic treatment discontinuation among Medicaid beneficiaries with schizophrenia (N=20,122), by race-ethnicity and drug treatment
| Add to My POL
|Drug and outcomea||Discontinuation for any cause||p|
| Sample N||749||107||149||493|
| Median days to discontinuation||740||646||789||736|
| Rate (%)||53.3||55.1||48.3||54.4||.397|
| Sample N||19,373||4,516||7,057||7,800|
| Median days to discontinuation||282||248||281||304|
| Rate (%)b||73.5||75.5||74.0||71.9||<.001|
We created 749 matched sets that included all clozapine users (N=749) and twice as many users of other antipsychotics (N=1,498). Our propensity score model generated well-balanced groups, as evidenced by the fact that with just one exception, standardized differences for all variables were smaller than 10%. The exception was time and only for the 20th quarter (standardized difference=10.4%).
Adjusted analyses confirmed most of the unadjusted results (Table 3). [A figure in the online data supplement presents results of this analysis.] Clozapine users had a lower risk of discontinuation than users of other antipsychotics (RR=.45). This effect was not moderated by race-ethnicity. Overall risk of discontinuation, however, was higher for blacks than for whites (RR=1.56) and higher for Latinos than for whites (RR=1.23). Although assessment of the proportionality assumption suggested some nonconstant hazards (global χ2=9.53; p=.021), with blacks having a slightly higher overall risk of earlier discontinuation than whites (χ2=6.06, p=.014), the (log) hazard ratio was small (<1.0). In keeping with the Cox model results, clozapine users had longer times to discontinuation than users of other antipsychotics (Table 4).
Table 3Adjusted risk of antipsychotic treatment discontinuation among Medicaid beneficiaries with schizophrenia, by drug treatment and race-ethnicitya
| Add to My POL
|Model covariate||Risk ratio||95% CI|
| Clozapine treatment (reference: other antipsychotics)||.45||.39–.52|
| Race-ethnicity (reference: whites)|
|Interactions (reference: whites)c|
Table 4Days to antipsychotic treatment discontinuation among Medicaid beneficiaries with schizophrenia, by drug treatment and race-ethnicitya
| Add to My POL
|Drug and group||N||Median days||95% CI|
Results did not change when we examined time to discontinuation or to hospitalization or when clozapine was compared only with second-generation antipsychotics. [Three figures in the online data supplement present results of these analyses.]
In this five-year study of a Medicaid cohort receiving maintenance treatment for schizophrenia, we did not observe any significant differences by race-ethnicity in the effectiveness of clozapine relative to that of other antipsychotics. These findings did not change when we assessed risk of discontinuation or hospitalization or when the comparator group included only second-generation antipsychotics other than clozapine. This is an important set of results given the paucity of evidence on racial-ethnic differences in responses to clozapine and on whether findings of comparative effectiveness research can be generalized to racial-ethnic minority groups.
Our study replicated the findings of the CATIE study in a Medicaid population comprising only patients in maintenance antipsychotic treatment. Our study expanded on the CATIE findings by evaluating the influence of race-ethnicity on outcomes of clozapine treatment and treatment with other antipsychotics. We know of only two studies that have assessed whether clozapine outcomes vary by race, and we know of no study that has assessed whether outcomes vary by Latino ethnicity. A randomized controlled trial conducted among hospitalized veterans with treatment-refractory symptoms of schizophrenia found that black race was not predictive of differential response to clozapine versus haloperidol, as assessed by standardized assessments and number of hospital days (18). A study that used a longitudinal statewide data set comprising all inpatient episodes during which patients received clozapine found that blacks experienced more robust early improvements on some but not all the standardized scales that were used to assess response (17).
Time to discontinuation for any cause is influenced by a drug’s efficacy and tolerability, by patient adherence behavior, and by prescriber behavior (13). It is therefore notable that we did not find racial-ethnic differences in the effectiveness of clozapine relative to that of other antipsychotics given evidence that black patients treated with clozapine receive lower doses (17), experience more weight gain (22), and are more likely to have clozapine unnecessarily discontinued by their physicians (17,23,24).
Our finding that blacks and Latinos had a higher risk than whites of discontinuing maintenance antipsychotic treatment is in keeping with a study of California Medicaid beneficiaries that found that black race was associated with time to antipsychotic discontinuation; the study looked at use of three second-generation antipsychotics other than clozapine and a first-generation antipsychotic (25). What may underlie this result? Although it has been conjectured that racial-ethnic differences in antipsychotic efficacy may exist (26), there is no evidence from a randomized controlled trial to support this possibility. There is evidence that persons from racial-ethnic minority groups may be more prone to certain side effects (27), a phenomenon that may be driven by inadequate matching of drugs to patients or excessive dosing (28). Finally, some adherence studies (29–32) but not all (33) have found that blacks and Latinos are more likely to exhibit poor adherence to antipsychotics.
Only a small fraction of patients in our study were on maintenance treatment with clozapine. Although all major schizophrenia treatment guidelines call for the use of clozapine to manage severe presentations (34), rates of any clozapine use remain low in the United States (5–7). Clozapine treatment requires periodic blood draws to monitor the patient for agranulocytosis, a life-threatening condition. It has been posited that underuse stems from patients’ aversion to needles, psychiatrists’ safety concerns, or the higher burden associated with clozapine prescribing; however, these factors are unlikely to fully explain the underuse phenomenon (35,36). Multiple studies have found that the likelihood of any use of clozapine is even lower for blacks (7,17) and Latinos (10,37). The factors driving lower use of clozapine among racial-ethnic minority groups are not well understood. It has been suggested that lower use of clozapine among blacks may stem from providers’ expectation that blacks will have lower response rates, their perception of blacks as less treatment adherent, and their unfounded concern that agranulocytosis may be more likely to occur among blacks (24).
Our findings should be considered in light of some limitations. First, because of the nonexperimental nature of our study design, unmeasured differences may have existed between users of clozapine and users of other antipsychotics. In particular, we were unable to select patients on the basis of their treatment response status, an important confounder. However, this limitation may have worked to attenuate our findings because we compared clozapine-treated persons, all of whom were very likely to be nonresponsive to other drugs, with persons receiving treatment with other antipsychotics who were likely to exhibit greater degrees of treatment responsiveness. Second, unlike CATIE researchers, we were unable to differentiate between discontinuation due to lack of efficacy, intolerability, or patient factors associated with adherence behavior. However, because we assessed discontinuation among patients in maintenance treatment—a population that had already shed many persons most likely to discontinue because of intolerability or treatment nonadherence—our any-cause discontinuation results are more likely to reflect discontinuation due to lack of efficacy than to other causes. In addition, we evaluated incidence of concurrent antipsychotic polypharmacy because polypharmacy can lead to discontinuation of the entire drug regimen due to a higher incidence of side effects. Our analyses showed that rates of polypharmacy were similar for clozapine users and other antipsychotic users for all racial-ethnic groups combined and for whites (results not shown). Among blacks and Latinos, rates of polypharmacy were higher for clozapine users than for users of other antipsychotics. However, if differences in polypharmacy had any impact, they would have worked to attenuate our findings by increasing the likelihood of clozapine discontinuation among minority groups. Third, our analyses may have been underpowered to detect the racial-ethnic differences in the comparative effectiveness of clozapine and other antipsychotics that were suggested by the unadjusted analyses. Last, the generalizability of our findings is limited by the fact that Florida differs from other state Medicaid programs because of its racial-ethnic diversity and restrictive eligibility requirements.
We found that race-ethnicity did not modify the superior effectiveness of clozapine compared with that of other antipsychotics in a population of Medicaid beneficiaries with schizophrenia. We have thus confirmed and expanded CATIE’s findings, highlighting the need for efforts to understand factors that explain the underuse of clozapine and to increase its use, particularly among minority groups.
This work was supported by grants R01MH087488 and R34MH082682 from the National Institute of Mental Health and grant R01HS017695 from the Agency for Healthcare Research and Quality. The authors are grateful to Christina Fu, Ph.D., for programming assistance and to Frank Yoon, Ph.D., for statistical expertise with matching.
The authors report no competing interests.