Comparison of Outcomes for African Americans, Hispanics, and Non-Hispanic Whites in the CATIE Study

The prevalence of schizophrenia among African Americans and Hispanics is similar to that for non-Hispanic whites (1), but there are minimal data on response to antipsychotic treatment by racial-ethnic group. Two studies of African Americans, one a review of four double-blind studies (2) and the other an analysis of data from the Texas Medication Algorithm Project (TMAP) on algorithm-guided care (3), found no outcome differences from non-Hispanic whites on a variety of medications. Dosing studies report that similar levels (4) or higher doses (5) of antipsychotics are prescribed to African Americans. In another study, African Americans were three times more likely than non-Hispanic whites to receive depot antipsychotic medications and were more likely to receive anticholinergics (6).

Among Hispanics with schizophrenia, a small double-blind risperidone dosing trial showed a faster rate of symptom improvement among Hispanics (7). In TMAP, psychotic symptoms declined significantly more among Hispanics (8) and they had lower depression symptoms (3) compared with non-Hispanic whites. For Hispanics, lower dosages of first-generation antipsychotics have been prescribed (4), perhaps related to concern about more frequent extrapyramidal side effects in this ethnic group (7).

The Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) (9), carried out in clinical settings in the United States, had 35% African-American and 12% Hispanic enrollment. It provided the only publicly available data set from a double-blind randomized controlled trial in schizophrenia with sufficient numbers of African Americans and Hispanics for analysis. The breadth of CATIE allowed research of important questions: Do schizophrenia illness characteristics differ at baseline among these three groups? In a controlled clinical trial with a standardized protocol, are there racial-ethnic differences in medication efficacy, side effects, or the patient’s decision to discontinue antipsychotic medication treatment? Are there outcome differences between groups on psychotic symptoms, depression, or self-reported quality of life? In this article we address these important questions and report on secondary analyses of CATIE data.

Methods

Study overview

CATIE studied five antipsychotic medications (olanzapine, quetiapine, risperidone, ziprasidone, and perphenazine) for treatment of chronic schizophrenia at 57 U.S. clinical sites (N=1,460). We analyzed data from phase 1, which could last up to 18 months. Patients with tardive dyskinesia could enroll but did not have the option for random assignment to perphenazine. Ziprasidone was added to the trial after 40% of the patients had been enrolled. At study end, 336 patients were assigned to olanzapine, 337 to quetiapine, 341 to risperidone, 261 to perphenazine, and 185 to ziprasidone. Patients had an equal chance of being assigned to a medication with exceptions noted above for perphenazine and ziprasidone. The dose of medications was flexible and based on the study physician’s judgment. A full description of the study design has been published elsewhere (10).

Patient characteristics

Eligible patients were 18 to 65 years of age and had a diagnosis of schizophrenia based on the Structured Clinical Interview for DSM-IV (SCID; 11), a clinician-administered semistructured interview for DSM-IV axis I diagnoses (12). Patients were able to take oral antipsychotic medication, as determined by the study physician. Patients were excluded if they had received a diagnosis of schizoaffective disorder, mental retardation, or other cognitive disorders; had a history of serious adverse reactions to the proposed treatments; had experienced only one prior schizophrenic episode; or had a history of treatment resistance or prior treatment with clozapine. Patients pregnant or breastfeeding or with a serious and unstable medical condition were excluded.

The primary independent variable was race-ethnicity. Participants were asked, “What is your race? Choose one or more: white, black or African American, American Indian or Alaskan Native, Asian or Native Hawaiian/other Pacific Islander.” Separately, they were asked if they were “Spanish, Hispanic or Latino” (country of origin was not asked). Those reporting any Hispanic background were classified as Hispanic regardless of race. Those noting white (not Hispanic) were classified as non-Hispanic white, or black or African American (not Hispanic). Persons who were not Hispanic and who reported more than one race were excluded from analyses because of small sample sizes, as were any other ethnic groups, again because of small sample sizes. Those who were not included in the analyses comprised 4% (N=62) of the CATIE sample.

Education level was categorized into less than high school, high school, and technical degree and above. Insurance status was categorized into none, public (Medicaid, Medicare, and veterans benefits), and private insurance.

All-cause treatment discontinuation

The primary measure of outcome effectiveness was all-cause treatment discontinuation—the number of days the patient remained on the randomly assigned phase 1 antipsychotic. This outcome considers the efficacy of the medication and its safety and tolerability. All-cause treatment discontinuation was further categorized by reasons for discontinuation: lack of efficacy, intolerability owing to side effects, and patient decision (including patient advocate decisions). Time in phase 1 was determined by the date the clinician changed the medication (moved to phase 2) or the patient decided to withdraw from the study. Patients could not discontinue the study medication altogether and remain in phase 1. Patients self-reported how many days they took the study medication, which determined adherence.

Other outcomes

Data were collected at one, three, six, nine, 12, 15, and 18 months for all of the measures listed next, except the second quality-of-life measure, which was assessed at six, 12, and 18 months.

Symptoms.

Symptoms of schizophrenia were assessed by the Positive and Negative Syndrome Scale (PANSS), which assesses positive symptoms, negative symptoms, and general psychopathology (13,14). The 30-item PANSS scores range from 30 to 210, and a higher score indicates more severe psychopathology.

The Clinical Global Impression scale is clinician rated and was used to rate overall psychiatric symptom severity. Possible scores range from 1 to 7, and higher scores indicate greater illness severity.

Depression.

We used the Calgary Depression Rating Scale (CDRS) to assess depression. The CDRS is a nine-item scale designed to assess severity of depressive symptoms of patients with schizophrenia. For CATIE, analyses were ranked on a scale from 0, absent, to 3, severe, with possible scores ranging from 0 to 27 (15,16).

Quality of life.

Psychosocial functioning and quality of life were assessed with two self-report items: “Please rate your current state of mental or emotional health. A score of 100 means perfect emotional health, and a score of 1 means worst possible emotional health.” The second item was the question, “How do you feel about your life in general?” Answers are rated on a scale that ranges from 1, terrible, to 7, delighted.

Neurological side effects.

The Simpson-Angus Extrapyramidal Side Effect Scale (SAEPS) assesses drug-induced parkinsonism and contains ten items rated on a scale of 0 to 4; for CATIE outcome analyses, the mean of items 1 through 6 was calculated. The SAEPS is widely used, reliable, and valid (17). The four-item Barnes Akathisia Scale assesses objective akathisia, subjective awareness of restlessness, and subjective distress related to restlessness. The first three items were used for CATIE outcome analyses to provide a global assessment of akathisia, each rated on a 0–3 scale, with possible scores ranging from 0 to 9 (18). The 12-item Abnormal Involuntary Movement Scale (AIMS) of tardive dyskinesia rates dyskinetic movements in various body parts and overall severity on a 0–4 scale. For CATIE outcome analyses, the first seven items on the scale were used, so possible scores range from 0 to 28. For each of the neurological side effect scales, a higher score indicates more severe side effects.

Hospitalization.

Time to hospitalization and the number of days hospitalized were assessed by asking participants if they had been hospitalized in the prior month and if so, for how many days.

Statistical analysis

Descriptive characteristics were compared with analysis of variance or the chi square test, as appropriate. We ran two sets of analyses: one covarying only baseline scores and a second adding age, gender, education level, and insurance status. Analyses with covariates did not affect statistical significance except for the conditions noted below. Discontinuations were analyzed with the Kaplan-Meier method and accelerated failure time survival regression analysis. Categorical outcomes were analyzed with chi square statistics. Analyses of continuous outcome variables over time were done with mixed-effects regression models, with race-ethnicity as the primary independent variable. Time was treated as a continuous variable, modeling an intercept and slope for each participant. Only postbaseline assessment data were analyzed.

Because medication effects were very significant in CATIE, preliminary models included medication and the medication × race-ethnicity interactions, including the three-way medication × race-ethnicity × time interaction in the longitudinal analyses. Patients with tardive dyskinesia were not randomly assigned to perphenazine, and ziprasidone was added to CATIE after about 40% of the participants had been recruited. CATIE’s focus was on medication comparisons, so the CATIE investigators dealt with the restriction of perphenazine and the delayed availability of ziprasidone by performing separate analyses within strata depending on which drugs were being compared. Our focus was on race-ethnicity, so we dealt with these confounded effects in initial models that included medication by including two covariates, one indicating the presence or absence of tardive dyskinesia and the other indicating whether the participant was recruited before or after the ziprasidone arm was added to the trial. When the interaction effects of medication with race-ethnicity were not significant, they were dropped from the analyses, and the final models were reestimated with only main effects. When interaction effects were significant, the relevant simple effects of race-ethnicity were explored.

CATIE’s convention was to assign the end of phase 1 and phase 1A and unscheduled interim assessments to the next regular assessment (9). This approach allowed mean estimates at the fixed assessment months but could involve displacing an assessment by over five months. At these unscheduled assessments, patients were more symptomatic. Patients had varying numbers of assessments, and spacing of unscheduled visits was irregular. Effects of time were modeled with a df=1 linear trend effect to include these visits without reassigning the month at which they occurred. This permitted use of all assessments with no artificial displacement of the timing of measures.

Analyses were conducted according to the procedures in the SAS version 9.2 statistical library. Unadjusted, two-tailed tests were considered significant at p≤.05. Pairwise comparisons between racial-ethnic groups were done by t test (least significant difference) only if the overall race-ethnicity effect was significant at p≤.05.

Results

African Americans were more likely to be female, and Hispanics were younger than the rest of the sample (Table 1). Non-Hispanic whites were more likely to have at least a high school education (80%) compared with African Americans (70%) and Hispanics (63%), were more likely to have employment income, and were less likely to have no insurance (non-Hispanic whites, 36%; African Americans, 43%; and Hispanics, 57%). Hispanics were more likely to live with family. Of the Hispanics, 11% (N=18) were predominantly Spanish speaking.

At baseline, Hispanics reported fewer years of treatment (14 years, versus 17 years for non-Hispanic whites and African Americans) and had higher baseline PANSS total scores. African Americans were more likely to have had a psychiatric hospitalization in the month prior and were more likely to meet criteria for drug dependence in the past five years (Table 2). The prescribed daily dose in phase 1 did not differ by racial-ethnic group for olanzapine, quetiapine, perphenazine, or ziprasidone but differed for risperidone (F=3.21, df=2 and 293, p=.04). Non-Hispanic whites had the highest average risperidone dose (4.4 mg), followed by African Americans (4 mg) and Hispanics (3.9 mg).

All-cause discontinuation

The risk of all-cause discontinuation did not differ by race-ethnicity, and the race-ethnicity × medication interaction was also not significant. However, there were racial-ethnic differences when the reason for discontinuation was considered.

Secondary outcomes

Lack of efficacy.

The survival analysis yielded a highly significant interaction of race-ethnicity × medication (χ²=21.9, df=8, p=.005). Effects involving tardive dyskinesia or timing of recruitment relative to ziprasidone were not close to significant, so these variables were dropped from the final model. Table 3 presents the estimated failure rates from separate Kaplan-Meier survival analyses run within each of the drug comparison strata. Below we consider each of the strata.

Efficacy discontinuations of olanzapine, quetiapine, and risperidone.

The first analysis compared olanzapine, quetiapine, and risperidone. This analysis used data for all patients randomly assigned to one of these medications. The interaction of race-ethnicity and drug was not significant, so it was dropped and the model was reestimated with only main effects. The medication effect was highly significant (χ²=24.1, df=2, p<.001), and the race-ethnicity effect was significant (χ²=6.6, df=2, p=.037). That is, non-Hispanic whites were more likely to discontinue for lack of efficacy in general in this stratum.

Efficacy discontinuations on perphenazine.

As in CATIE, comparisons of perphenazine with the other medications omitted patients with tardive dyskinesia, and the comparison with ziprasidone omitted patients with tardive dyskinesia and those recruited before ziprasidone was added. The interaction of medication × race-ethnicity across four groups (olanzapine, quetiapine, risperidone, and perphenazine) was significant (χ²=15.3, df=6, p=.018). In the much smaller stratum including only perphenazine and ziprasidone, the main effect of race-ethnicity was significant (χ²=3.8, df=2, p=.001). As shown in Table 3, perphenazine outcomes showed a very high proportion of efficacy discontinuations among Hispanics (62%–64%).

Efficacy discontinuations on ziprasidone.

Comparisons with olanzapine, quetiapine, and risperidone were made after omitting the patients recruited before the addition of ziprasidone. The resultant sample sizes were about 60% as large, and there was a consequent reduction in statistical power. For example, the numbers of Hispanics ranged from 18 taking ziprasidone to 26 taking quetiapine. The interaction of medication × race-ethnicity for all four medication groups (olanzapine, quetiapine, risperidone, and ziprasidone) was significant (χ²=13.2, df=6, p=.039) despite the much smaller N from omitting patients who had participated before ziprasidone was introduced in the trials. The comparison of perphenazine and ziprasidone, presented in the preceding paragraph, excluded patients with tardive dyskinesia and those recruited in the pre-ziprasidone phase. The results with ziprasidone were very similar to those with perphenazine, with an even larger number of efficacy discontinuations among Hispanics (69%–71%).

Patient decisions to discontinue.

Risk of patient decisions to discontinue differed by ethnicity-race (χ²=20.2, df=2, p≤.001) and medication (χ²=21.1, df=4, p≤.001) (tardive dyskinesia, pre-post ziprasidone, and the race-ethnicity × medication interaction were all dropped from the final model). African Americans were most likely to decide to discontinue (39%, N=195 of 506), followed by Hispanics (30%, N=51 of 170) and non-Hispanic whites (25%, N=181 of 722). A simple chi square test of these proportions yielded χ²=25.5, df=2, p<.001. Patients’ decision to discontinue was very highly related to continuing into CATIE’s phase 2. Among those who decided to discontinue, only 20% (90 of 445) entered phase 2. Among those discontinuing for other reasons, 70% (453 of 643) entered phase 2 (χ²=265.4, df=1, p<.001). Consistent with this pattern, significantly fewer African Americans (32%, N=162 of 506) entered phase 2 compared with non-Hispanic whites (40%, N=292 of 722) and Hispanics (41%, N=70 of 170) (χ²=10.14, df=2, p=.006).

Intolerability.

In the proportional hazard regression for tolerability, neither the main effect of race-ethnicity nor the race-ethnicity × medication interaction was significant.

Symptom severity.

The pattern of results was similar for the PANSS total score or positive, negative, and general psychopathology subscales, with no main effect of race-ethnicity (Table 4). There was significant improvement over time on all PANSS measures, with one exception. Hispanics improved from their baseline scores on the PANSS positive subscale, but this improvement was not significant after adjustment for covariates.

Depression.

On the CDRS, non-Hispanic whites had worse scores than African Americans and Hispanics on follow-up (main effect F=7.76, p≤.001; Table 4). Non-Hispanic whites also did not improve significantly during the follow-up compared with African Americans (p≤.001) and Hispanics (p≤.001).

Quality of life.

There were no main racial-ethnic between-group differences in self-reported emotional health or in overall feelings about life in general during the follow-up period. Feelings about life in general improved significantly among non-Hispanic whites (p=.004) and African Americans (p≤.001); feelings among Hispanics also improved, but less so and not significantly. There was a significant three-way interaction between time, race-ethnicity, and medication (F=3.26, df=8 and 591, p=.001).

With perphenazine the race-ethnicity × time interaction was highly significant (p=.008), with ratings about life in general declining significantly among Hispanics (t=–2.8, df=157, p=.006). Their quality of life also declined if they had taken ziprasidone. The rate of decline was only slightly slower than if they had taken perphenazine and was not statistically significant. The difference in these two statistical results was affected by low power because of the smaller number of Hispanics taking ziprasidone (N=18) than taking perphenazine (N=24). Results with perphenazine and ziprasidone were consistent with increased rates of discontinuation by Hispanics for the medication’s lack of efficacy.

Side effects.

African Americans had lower scores on measures of akathisia (Barnes Akathisia Scale) and parkinsonism side effects (SAEPS) during the follow-up period compared with non-Hispanic whites and Hispanics (p ≤.001 and p=.007, respectively). There were no racial-ethnic differences in ratings of tardive dyskinesia as rated on the AIMS.

Hospitalization.

There were no significant differences in time to hospitalization or number of days hospitalized during the follow-up period. Table 4 provides further detail on secondary outcomes.

Discussion

We examined outcomes in a well-designed clinical trial with large samples of three racial-ethnic groups with schizophrenia. All-cause discontinuations, CATIE’s primary outcome, did not differ by race-ethnicity. However, in trials with perphenazine and ziprasidone, Hispanics were much more likely to discontinue for lack of efficacy, and quality of life for this group declined over time. The average dose in phase 1 and 1A did not differ across racial-ethnic groups for either of these medications. These findings suggest a promising area of research in examining medication outcome differences. Because of medication efficacy differences for Hispanics versus non-Hispanic whites, we recommend clearly defining separate white samples by Hispanic or non-Hispanic ethnicity (19).

We found that non-Hispanic whites in general had more discontinuations from lack of efficacy when considering discontinuations while taking olanzapine, quetiapine, or risperidone. In a review of olanzapine trials to treat psychosis, non-Hispanic whites were more likely than African Americans to discontinue when taking olanzapine (20), consistent with our general finding and a pilot study that reported that non-Hispanic whites had worse responses to risperidone (7). Thus medication outcomes can differ by race-ethnicity, and the above findings should be tested by replication.

Both pharmacokinetic and pharmacodynamic hypotheses are plausible explanations for racial-ethnic group differences in treatment response. The most studied reason is that genetic differences across racial-ethnic groups affect drug metabolism (21,22). For example, in a subset of CATIE participants, African Americans had 17% greater clearance of olanzapine than Caucasians, which was moderated by genotype differences (22). Other hypotheses for racial-ethnic differences in medication efficacy include medical comorbidity, other medications, and lifestyle habits that affect drug metabolism, such as cigarette smoking (23) or diet. Racial-ethnic groups may have unique responses to side-effect profiles of medications for psychosocial reasons, for example, concern about sexual side effects and weight gain. Racial-ethnic groups also may have unique concerns about how the illness is experienced, for example whether a medication targets positive or negative symptoms. Research in these areas is sparse.

Dosing differences can also affect the beneficial medication effects; however, of the medication differences we report here, the only dosing difference we found was for risperidone, which showed that non-Hispanic whites were prescribed higher doses during phase 1. However, if this increased dose led to more side effects and worse tolerability, the relevant CATIE outcome would have been discontinuation due to intolerability. It may be that before concluding that risperidone was ineffective, clinician researchers increased doses to optimize medication response.

Non-Hispanic whites were more depressed than the other groups in our study, consistent with TMAP (3). Other studies of psychosis have found higher depression rates or antidepressant prescribing among non-Hispanic whites compared with African Americans (24,25). One hypothesis is that antipsychotic medications have greater antidepressant effects for African Americans and Hispanics.

African Americans were more likely to discontinue CATIE at the end of phase 1 and were more likely to decide to discontinue, regardless of medication assignment. Qualitative reasons noted included lack of efficacy, side effects, research burden, nonadherence to medication, research team’s loss of contact with patient for further study visits, or patient’s withdrawal of consent without further explanation. Earlier discontinuation compared with other racial-ethnic groups is consistent with findings from two serious mental illness trials (20,26), including discontinuing for “other” reasons (20), but is not consistent with another (27).

We adjusted for age, gender, education level, and insurance status—important variables to consider in health disparities research because of their confounding with ethnicity and race. That is, in some cases when health disparities have been reported to show more favorable outcomes for non-Hispanic whites, consideration of socioeconomic factors renders race-ethnicity effects nonsignificant (28). Although we controlled for important covariates, other socioeconomic and psychosocial factors may have affected study participation (family responsibility and hours worked per week). Early dropout by African Americans may be affected by mistrust of professional mental health treatment (29,30) or a mistrust of research (31). One study hypothesized that greater initial improvement contributed to African-American dropout (32). Unless research is developed with a priori hypotheses regarding racial-ethnic dropout, the reason for the earlier dropout is likely to remain uncertain.

African Americans improved on measures of akathisia and parkinsonism but not in tardive dyskinesia during the follow-up period, with no interactions by medication. Prior findings have been inconsistent. Lawson and colleagues (2) reported worse scores on akathisia and parkinsonism (but not tardive dyskinesia) for non-Hispanic whites but not for African Americans on haloperidol compared with placebo. For ziprasidone, significant increases in akathisia and parkinsonism occurred for African Americans but not for non-Hispanic whites (2).

Baseline racial-ethnic differences were consistent with U.S. sociodemographic characteristics (education level, income, and insurance status were more favorable among non-Hispanic whites). Consistent with other studies, Hispanics had higher PANSS scores (3) and were more likely to live with family (33), and African Americans reported more hospitalizations (34) and higher rates of drug dependence (35,36).

There were limitations to this study. These findings represent secondary analyses. Sites were confounded by ethnicity and race, and thus we were not able to adjust for site. However, typically if there are site differences, sites with more racial-ethnic diversity have greater resource limitations, and we did not find outcome disparities favoring non-Hispanic whites. We were not able to make group comparisons on the basis of Hispanics’ country of origin. We do not think our findings were compromised by the issue of multiple statistical comparisons. There were relatively few measures (typically three or four at most) within each of a relatively small number of distinct outcome domains. Comparisons between racial-ethnic groups or analyses stratified by medication were only done when either an overall main effect or interaction effect was significant. Because schizophrenia subtype was not collected in CATIE, we were unable to examine whether there were any race-ethnicity effects of subtype on outcomes. The two quality-of-life questions were not validated. Also, it is important to note that self-report of hospitalizations can be unreliable. Replication of the findings is necessary in ascertaining whether these medication effects are consistent across studies.

Conclusions

In treatment as usual, ethnicity and race disparities are often seen (37,38). There are disparities in the diagnosis of schizophrenia (39,40) in nonresearch settings and in use of antipsychotics to treat it (41) that were not addressed by CATIE, which was a large-scale clinical trial. Research addressing the effects of race-ethnicity on access to mental health specialists, medication costs, treatment preferences, and adherence should be conducted as well. We did not find disparities per se for medication outcomes, and we saw rather unique patterns of outcomes. Perhaps CATIE offered a higher quality of care across racial-ethnic groups, thereby leveling those factors hypothesized to negatively affect quality of care for patients in racial-ethnic minority groups. Miranda and colleagues (42) have noted that high-quality care can improve outcomes for African Americans and Hispanics as well as non-Hispanic whites.

Better prediction of treatment response and personalization of medicine could shorten time to recovery, decrease the number and intensity of adverse events, and improve overall outcomes for individuals with schizophrenia. In addition, more targeted treatment approaches can reduce health care costs.