Introduction

The societal burden of schizophrenia is substantial despite its low prevalence [1, 2]. In addition to the suffering it causes on patients and their families, schizophrenia also results in huge economic losses [3]. Currently, there is no cure for this chronic condition, although the symptoms are usually treated and respond well to medications. Most existing clinical practice guidelines recommend the use of antipsychotic medication as first-line treatment for schizophrenia [4]. Specifically, the newer types of antipsychotics are indicated because of its arguably better clinical and risk profiles [5].

A variety of studies have been conducted to determine whether there are ethnic disparities in antipsychotic treatment, with some indicating significant differences in access and treatment type across ethnic groups [6–9]. To our knowledge, however, no study has synthesized the evidence that has accumulated. To address this gap, we systematically reviewed and meta-analyzed published studies that report ethnic and racial disparities in antipsychotic treatment and in the type of antipsychotic medication (newer versus older) used among those who were treated.

Examining the consistency of evidence in the disparities in antipsychotic medication use is important as the results can highlight areas in the mental health system that can be improved. In many instances, the presence of disparities can reveal a mismatch between need and appropriate care that could cause or further exacerbate existing inequalities in health outcomes [10].

Methods

Search strategy

The search strategy comprised electronic database searching, scanning reference lists of key articles, and checking the personal libraries and networks of the authors, for peer-reviewed articles. The electronic databases searched include CINAHL (Ebsco), EMBASE (OvidSP), International Pharmaceutical Abstracts (OvidSP), MEDLINE (OvidSP), PsycINFO (Ebsco), and Web of Science (Thompson Reuters). Search strategies aimed to capture the intersection of ethnicity-related keywords and subject headings, and subject headings and keywords for typical (older) and atypical (newer) antipsychotic medications (see “Appendix” for details of actual search parameters). A master’s-trained information specialist performed the electronic database searches. To obtain further data or clarification, we contacted eight authors, three of whom replied.

Selection criteria

We sought peer-reviewed studies published in English between January 1, 1980 and December 30, 2010. To be included, studies must have focused on ethnic or racial disparities; assessed antipsychotic medication use in outpatient settings; and reported odds ratios adjusted for at least some of the other major factors that may influence medicine use (e.g., age, sex, health status, income, insurance, and related diagnosis) [11]. The full review protocol is available upon request from the corresponding author.

Study selection process

Study selection was completed by two independent reviewers in three stages: (1) a title review (2) an abstract review, and (3) a full-text review. At the title and abstract review stages, bibliographic entries deemed relevant by at least one reviewer were selected. At the final stage (full-text review), only those studies considered potentially relevant by both reviewers were selected and included for data extraction and analysis. Differences in inclusion assessment were discussed and resolved by consensus. We calculated kappa coefficients to determine reviewer agreement.

Data extraction

Using a standardized form, we extracted the following information from each included study: sample size, sampling frame, year(s) of data collection, source of drug utilization data, covariates used for adjustment, type of antipsychotic (old versus new), crude prevalence of use, adjusted prevalence and adjusted odds ratios. Two independent reviewers performed the data extraction and the results were merged into one dataset. Discrepancies between the reviewers’ extractions were discussed and resolved by consensus.

Data analysis

For studies conducted in the US, we categorized “Blacks” as African Americans and “Hispanics” as Latinos. In the two non-US studies included in the review, we retained the racial/ethnic categories used by the study authors.

We performed meta-analyses to measure ethnic/racial disparities in (a) the use versus non-use, and (b) use of newer vs older antipsychotics. We expressed all odds ratios using non-minorities (i.e., non-African Americans and non-Latinos) as the reference group. That is, in studies comparing whether ethnic minorities were more likely to receive or not receive any antipsychotic treatment, odds ratios lower than one indicate that ethnic minorities were less likely to receive antipsychotic treatment. Similarly, in studies investigating whether individuals were more likely to receive newer versus older antipsychotics, odds ratios of less than one indicate that minorities are less likely to receive newer antipsychotics.

We pooled the adjusted odds ratios following the inverse-variance method of weighting effect sizes [12]. This procedure required standard errors that were calculated from the reported 95 % confidence intervals. In studies where confidence intervals were not available, we used the reported point-estimates and p-values to estimate the standard errors.

I 2 statistics were calculated to measure the degree of potential heterogeneity between studies. Because of heterogeneity observed in some of the results, we pooled the odds ratios using a random effects model and stratified the analyses by ethnic/racial groups.

To assess potential publication bias, we produced a funnel plot and tested its asymmetry using the Egger statistic [13]. To supplement the results of Egger’s test, which is biased toward indicating asymmetry when data points consist of odds ratios and standard errors, we added contours of statistical significance in our funnel plots [14]. In these graphs, a noticeable clustering of odds ratios with large standard errors in statistically significant regions is suggestive of potential publication bias. This is because the clustering indicates that potentially missing studies are likely to be located in the statistically non-significant regions of the plot [14]. All statistical analyses were performed using version 10.1 of Stata for Windows [15].

Results

Our electronic database search produced a total of 1,825 unique citations. At the title review stage, we selected 532 potentially relevant titles. Abstracts for these titles were reviewed against the inclusion criteria and lead to the selection of 121 potentially relevant abstracts. In the last stage of the selection process (full-text review), we identified 12 studies that met inclusion criteria (Agreement: 96.7 %; Kappa: 0.92, 95 % CI = 0.83–0.99). A detailed description of the selection process can be found in Fig. 1.

Fig. 1
figure 1

Study selection and review process

Full size image

Description of included studies

Table 1 summarizes the key features of each study. Of the 12 included studies, 10 were from the US [6–9, 16–21], one was from New Zealand [22] and one was from the United Kingdom (UK) [23]. Years of data collection ranged from 1992 to 2006. Prescription drug utilization data for these studies were obtained from administrative databases (n = 6), medical chart reviews (n = 4), survey (n = 1) and a combination of survey and administrative data (n = 1). Many of the studies (n = 8) used the entire population of their sampling frame (e.g., insurance plan, outpatient list); the rest used purposive (n = 1), random (n = 2) and convenience (n = 1) samples. Study sample sizes ranged from 344 to 69,787.

Table 1 Design and sample/population characteristics of the studies included in the systematic review
Full size table

Most of the studies (n = 7) reported results based on the adult population (individuals between the ages of 15 and older). One study examined children and youth exclusively and another focused on individuals 60 years and older. Mean age for three studies where age was not reported ranged from 38.2 to 43.7. We found wide variation in the number and type of covariates used for adjustment, though age, sex and psychiatric morbidity were the most frequently mentioned.

The minority ethnic/racial groups examined in the 11 US-based studies were African Americans and Latinos. Asians, Maoris and Pacific Islanders were the ethnic groups examined in the NZ study. The UK study compared “Black” and “Black British” with “White”.

Racial or ethnic disparities in the use versus non-use of antipsychotic drugs

Disparities in the use versus non-use of any antipsychotics were investigated in six studies. In most of the studies, comparisons were made across three or more ethnic groups; hence, we were able to extract a total of 12 adjusted odds ratios from all six included studies. The odds ratios range from 0.64 to 1.19.

Figure 2 presents the contour-enhanced funnel plot of the log odds ratios for the included studies. There appeared to be more log odds ratios with lower precision on the left side of the plot and odds ratios with relatively higher precision on the right-hand side, but overall, the data points were scattered in regions of low statistical significance. Similarly, Egger’s test statistic failed to detect significant funnel asymmetry (bias coefficient = 0.85; se = 1.51; p = 0.59). Together, these assessments do not indicate a bias toward publishing statistically significant findings.

Fig. 2
figure 2

Funnel plot of receiving versus not receiving any antipsychotics

Full size image

Pooled odds ratios (Fig. 3) stratified by ethnicity showed neither large nor statistically significant disparities in the use versus non-use of antipsychotics. In US studies, the odds of receiving any antipsychotic medication among African Americans (OR = 1.01, CI = 0.99–1.02) and Latinos (OR = 0.98, CI = 0.86–1.13) were not significantly different from those of non-African Americans and non-Latinos.

Fig. 3
figure 3

Odds of receiving versus not receiving any type of antipsychotic treatment

Full size image

Small to moderate but statistically non-significant disparities were also noted when odds ratios for other ethnic minorities were pooled: Asians (OR = 1.10, CI = 0.88–1.36), Maoris (OR = 0.78, CI = 0.53–1.13) and Pacific Islanders (OR = 0.97, CI = 0.84–1.11). Odds ratios for these minority groups were derived from the single NZ study and the pairs of odds ratios that were pooled represent measurements taken from two separate time points (2000 and 2004). Of all the ethnic groups examined in the NZ study, only the pooled results for the Maori group exhibited significant heterogeneity (98.4 %, p = 0.00).

Finally, the single study from the UK comparing “Black” and “Black British” against “Whites” reported slightly higher but statistically non-significant odds of antipsychotic use among “Black” and “Black British” (OR = 1.16, 95 % CI = 0.88–1.53).

Racial or ethnic disparities in the use of newer versus older antipsychotics

Seven studies, all conducted in the US, examined the odds of being treated with newer versus older types of antipsychotics. A total of 14 odds ratios were extracted and they were all less than one, ranging from 0.24 to 0.97, suggesting lower use of newer antipsychotics among ethnic minorities.

Figure 4 presents the contour-enhanced funnel plot of the log odds ratios for these results. There is a noticeable clustering of data points on the left side of the funnel plot. However, the odds ratios appeared to be evenly dispersed across regions of low and high statistical significance, suggesting that if there are missing data points on the right side of the plot, they should be evenly dispersed as well in regions of low and high statistical significance. Egger’s test did not indicate statistically significant funnel asymmetry (bias coefficient = −0.89; se = 0.45; p = 0.07). Based on these assessments, we did not conclude that the reported results were influenced by potential publication bias.

Fig. 4
figure 4

Funnel plot of receiving newer versus older antipsychotics

Full size image

The results of pooling the odds ratios (Fig. 5) suggest that ethnic minorities who were treated with antipsychotics, compared with non-minorities who were also treated with antipsychotics, were less likely to receive the newer type of antipsychotics and therefore more likely to receive older antipsychotics. African Americans, compared with non-African Americans, appeared to have 38 % lower odds of receiving newer antipsychotics (OR = 0.62, 95 % CI = 0.50–0.78). Similarly Latinos, compared with non-Latinos, have a 23 % lower odds of receiving newer antipsychotics (OR = 0.77, 95 % CI = 0.73–0.81). No substantial heterogeneity was observed (I 2 = 0 %, p = 0.58) among the studies that examined Latinos, but in those that studied African Americans, a high degree of heterogeneity was noted (I 2 = 73.2 %, p = 0.00).

Fig. 5
figure 5

Odds of receiving newer versus older type of antipsychotics among individuals who received antipsychotic treatment

Full size image

Discussion

This study sought to systematically review and synthesize findings from published studies that examined racial and ethnic disparities in the receipt and type of antipsychotic medication used in outpatient settings. Our results are that published evidence does not establish significant ethnic disparities in likelihood of receiving any antipsychotic medication. However, among those treated with antipsychotics, results of published studies suggest that African Americans and Latinos were less likely than non-minorities to be treated with the newer type of antipsychotics.

The evidence for lack of ethnic disparities in the receipt of any antipsychotic treatment seems reliable. In the two largest US studies that are likely to have adequate power to detect existing disparities [17, 24], the reported odds ratios were close to one. In the other two US studies [16, 20] where some degree of disparities were observed, the results were based on smaller sample sizes that produced less reliable estimates as reflected in the wide confidence intervals reported. In the NZ study, no disparities were noted with respect to Asians; Maoris, however, were found in a previous time period to have significantly lower odds of use. This difference disappeared in subsequent assessment (by the same authors in the same study), and our pooling of the odds ratios for both time periods yielded an effect size that although < 1, was not statistically significant. Likewise, in the NZ study, Pacific Islanders were found to have significantly lower odds of use but the effect size was small and the pooled odds ratio was not statistically significant. Lastly, the reported odds ratio from the UK study was small and not statistically significant.

The lower odds of newer antipsychotic use among ethnic minorities appeared robust across the seven US studies included in this review. Though not all effect estimates were statistically significant, all were < 1. No significant heterogeneity was observed in the odds ratios for Latinos, but in the studies that reported odds ratios for African Americans, a high degree of heterogeneity was noted. We re-analyzed the data for these studies without including the study [8] that appeared to contribute highly to the observed heterogeneity. This sensitivity analysis substantially decreased but did not completely eliminate the observed heterogeneity (I 2 = 56.7 %, p = 0.031) and did not considerably changed the pooled effect estimate (OR = 0.69, 95 % CI = 0.57–0.84). As a result, we used the original results produced using a random effects model, which accounts for the heterogeneity present between studies.

Our finding of persistent ethnic differences in the receipt of newer versus older antipsychotics is consistent with findings of ethnic differences in the use of health services for other mental disorders [25–27]. With respect to the use of newer antipsychotics, the disparity could mean that members of ethnic minorities, compared with non-minorities, received poorer care quality since practice guidelines have recommended the use of newer antipsychotics as first-line treatment for psychotic disorders such as schizophrenia [28, 29]. Newer antipsychotics have generally been regarded as better at improving cognitive functions (i.e., verbal fluency, attention, memory for facts and events) and are also considered less likely to cause irreversible movement disorders (i.e., extrapyramidal syndromes and tardive dyskinesia) [5].

We note that data for all the included studies were collected at a time when newer antipsychotics were still considerably more expensive than older antipsychotics. During that period, low income and inadequate insurance coverage, which were reported to be associated with race/ethnicity in the United States [30], may have limited ethnic minorities’ access to newer antipsychotic treatment. It is possible that ethnic minorities who are uninsured or are perceived to be unable to afford the co-payment for the newer medication, have requested or were given the older antipsychotics to ensure that some form of pharmaceutical treatment is available.

Alternatively, it is also possible that the lower odds of use of newer antipsychotics among ethnic minorities may have been driven by concerns about the specific adverse effects of the newer type of antipsychotics. Previous research has linked the prolonged use of newer antipsychotics with increased risks of cardiovascular events and metabolic syndrome [5]. Since epidemiological studies have determined that hypertension and diabetes are more prevalent among African Americans and Latinos [31–33], it could be that the lower odds of use of newer antipsychotics among ethnic minorities were prompted by efforts to minimize those risks.

There is already substantial evidence about disparities in antipsychotic medication use pertaining to African Americans and Latinos in the US. The number of studies included in this review would have been greater, if we counted those that did not adjust for factors that influenced medicine use and those that examined antipsychotic use in inpatient settings. While there is still value in monitoring whether disparities persist over time, future studies could benefit from the examination of potential determinants and outcomes of observed treatment disparities. In addition, future research could also focus on measuring antipsychotic treatment disparities in understudied ethnic minorities in other countries (e.g., Canada and Australia) with rapidly changing ethnocultural make-up due to immigration.

The following limitations of our study need to be considered when interpreting the results. First, our search was limited to literature published in academic journals. Any findings of disparity or the lack thereof that were reported in the non-peer-reviewed literature or other types of reports and publications would therefore have been excluded. Second, the review was limited to journal articles that specifically reported on ethnic or racial disparities in antipsychotic use. We may therefore have missed some studies that included but did not discuss ethnicity or race as a control variable when examining more general determinants of antipsychotic use. Last, we were not able to calculate Peter’s statistic because our data points consisted of adjusted odds ratios and standard errors derived from confidence intervals. Previous research has demonstrated that Peter’s [34] statistic produce lower rates of false positives with respect to publication bias.

Conclusions

Our review found neither strong nor consistent ethnic disparities in the use versus non-use of any antipsychotics. However, among those who received antipsychotic treatment, ethnic minorities such as African Americans and Latinos were found to be consistently less likely than members of non-minorities to be treated with newer antipsychotics and were therefore more likely to have used older antipsychotics.