Selection of studies and participants

Data were obtained from the double-blind, randomised, placebo-controlled short-term efficacy trials with antipsychotics for the treatment of psychotic episodes in patients with a DSM-III-R or DSM-IV diagnosis of schizophrenia, identified from documentation submitted by pharmaceutical companies to the Dutch regulatory authority for the purpose of marketing authorisation application. These studies (n = 22, including 5233 patients; 3727 on active medication and 1506 on placebo) were initiated between 1991 and 2004.^{Reference Mattila, Wohlfarth, Koeter, Storosum, van den Brink and de Haan21} A study period of 6 weeks was chosen for the analysis cut-off point because this is the duration of short-term schizophrenia trials recommended in the European Medicines Agency's Committee for Medicinal Products for Human Use (CHMP) guideline on clinical investigation of medicinal products in the treatment of schizophrenia.²² The Yale University Open Data Access (YODA) Project was consulted for relevant additional individual patient data from high-quality trials, which yielded no additional data (Fig. 1).

Fig. 1 Interaction of ethnicity × treatment (with main effects) β-coefficients for (a) BPRS change and (b) response. BPRS, Brief Psychiatric Rating Scale.

For the current study, the data were compiled with a specific focus on ethnicity. Adults (age 18 years or over) with schizophrenia receiving antipsychotic medications were included as participants and adults with schizophrenia receiving placebo were included as controls. Availability of data on ethnicity was a prerequisite for inclusion. In the database, the following predefined ethnic groups were available: Caucasian, Black, Asian, Oriental, Hispanic, Native American, other. The original individual patient studies were identified for the purpose of collecting manuscripts and corresponding authors were contacted in case of unclarities or missing information (e.g. on the definition of the ethnicity subgroups). In addition, the ethnic subgroups were examined and redefined to the terms American Indian or Alaska Native, Asian, Black, Native Hawaiian or other Pacific Islander, White and some other race, according to current JAMA Network guidelines.^{Reference Flanagin, Frey and Christiansen23} In the original studies, the term ‘race’ was used, or may have been used interchangeably with ‘ethnicity’. In the current report, however, only the term ethnicity is used because of the controversial implications regarding the term ‘race’ and because the concept of ethnicity encompasses a broader definition.^{Reference Flanagin, Frey and Christiansen23}

To ensure adequately powered statistical analyses, individual studies were excluded if the ethnicity subgroups consisted of fewer than ten patients per study arm (placebo or active medication).

The five atypical antipsychotic medications that are currently most frequently prescribed in clinical practice were included. Because of agreements between the Medicines Evaluation Board and the pharmaceutical companies that provided their individual patient data, the names of the compounds cannot be disclosed in this report. To avoid bias from ineffective dosing, only studies with potential effective doses (according to current Summaries of Product Characteristics) of antipsychotic medications were included. We did not perform a formal systematic literature search, because in the design of the current study, only available individual patient data were used. Regarding background information, PubMed, EMBASE and PsycINFO were consulted. Additionally, quoted articles were checked for relevant references. To provide transparent and accurate reporting, the EQUATOR network was consulted for guidelines.^{Reference Stewart, Clarke, Rovers, Riley, Simmonds and Stewart24} To determine the risk of bias of the individual studies, the individual patient data-specific extension to the guideline for the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines was followed. A Cochrane Risk of Bias tool was used, and the study protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO; identifier CRD42022327122).

Outcome measures

The main efficacy measure was the Brief Psychiatric Rating Scale (BPRS). The scale consists of 18 items with potential scores ranging from 1 (not present) to 7 (extremely severe), resulting in a total score between 18 and 126. In the original individual patient data, either the BPRS or the PANSS was used for measuring symptoms. When no BPRS data were available, PANSS data were converted into BPRS scores, following the method previously described in the literature.^{Reference Mattila, Wohlfarth, Koeter, Storosum, van den Brink and de Haan21} When individual BPRS/PANNS item scores were missing, the average of the other subgroup scores of that visit were used. The main outcome measure was the difference between total BPRS scores at baseline and week 6 (defined as BPRS change). As an additional outcome measure, a minimum of 30% reduction of BPRS scores from baseline to week 6 was used (defined as response). The response criterion of 30% is in accordance with the EMA's CHMP guideline.²² When patients dropped out earlier than 6 weeks, the last observation was carried forward to week 6.

To examine possible confounding effects, baseline severity, baseline negative symptoms, age and gender were used as covariates in the various statistical models. To compute baseline negative symptoms, three subscale items from the BPRS were used: emotional withdrawal (item 3), motor retardation (item 13) and blunted affect (item 16), as per previous research.^{Reference Mattila, Wohlfarth, Koeter, Storosum, van den Brink and de Haan21} To provide an impression of the possible influence of the publication date of the individual studies on the results, the studies were ranked in chronological order.

Statistical analyses

A two-step, random-effects, individual patient data meta-analysis was performed. To explore participant-level variations and to control for potential confounders, individual patient data meta-analysis was chosen over study aggregate meta-analysis. Because of existing heterogeneity between studies (e.g. different patient populations, different types of medications and different companies), random-effect instead of fixed-effect models were used.^{Reference Riley, Lambert and Abo-Zaid25}

First, basic characteristics and outcome measures (BPRS change from baseline and response rates) were calculated. Subsequently, multivariate linear regression analyses were performed with mean BPRS change from baseline as the dependent variable and treatment condition, ethnicity and treatment condition×ethnicity as the independent variables. Similarly, a multivariate logistic regression analysis was performed with response as dependent variable. Thus, in both analyses, the interaction of ethnicity×treatment condition (active medication versus placebo) was added to the main effects (ethnicity and treatment condition) as independent variable as an indicator for a modifier effect of ethnicity on treatment effect. Subsequently, to examine the effect of baseline severity, baseline negative symptoms, age and gender, these variables were cumulatively added as independent variables to the main effects and the interaction of ethnicity by treatment. For these analyses, SPSS version 26 for Windows was used.

Subsequently, Comprehensive Meta-Analysis (CMA for Windows, Biostat Inc., USA; https://www.meta-analysis.com/) version 2 software was used to perform random-effect meta-analyses, with outcomes created in the abovementioned linear and logistic regression analyses, i.e. the regression coefficients and odds ratios for the treatment condition×ethnicity interactions. The 95% confidence interval indicates the scope of uncertainty in the effect estimate of the treatment condition×ethnicity interactions, considering heterogeneity between studies.

Finally, the treatment effect in the ethnicity subgroups (Black and White) was examined separately. A conventional individual patient data meta-analysis was performed, yielding an overall pooled mean difference in outcomes (BPRS change from baseline and response rates) between participants receiving antipsychotic medication and participants receiving placebo.

Definitions

In the literature, the terms ‘race’ and ‘ethnicity’ are often used interchangeably. According to the Oxford English dictionary, early use of the word ‘race’ was applied to groups of people with obviously distinct physical characteristics such as skin colour. An influential early system dating from 1775 (De Generis Humani Varietati Nativa by J. F. Blumenbach) divided the human species into five races: American, Caucasian, Ethiopian, Malay and Mongolian. This system was based on conformation of the head and skin colour, and assigned the races in qualitative ranking. Because of such early theories and ideologies, the use of the word ‘race’ in reference to specific ethnic groups is avoided in the current literature. Instead, the term ‘ethnicity’ is used for human groups that entertain a subjective belief in their common descent because of similarities of physical type or of customs or both, or because of memories of colonisation and migration. This indicates that the concept of ethnicity encompasses various characteristics, such as genetic profile, culture, migration history, ethnic identity, socioeconomic factors and discrimination. In the current study, the term ‘ethnicity’ will be used rather than ‘race’, even though ‘race’ was used in certain quoted literature and in the databases. Current guidelines of describing specific ethnic groups are followed (e.g. capitalising and using adjectival forms instead of nouns): White and Black are used instead of Caucasian and African American.