Treated Incidence of Psychotic Disorders in the Multinational EU-GEI Study

1The World Health Organization Ten-Country Study, the most recent multicenter international study of the incidence of psychotic disorders, was widely interpreted as demonstrating worldwide homogeneity in rates of schizophrenia and other psychotic disorders. In fact, there was a 2.5-fold variation in broadly defined nonaffective psychoses.

2Subsequent studies showed that psychotic disorders vary considerably across replicable social and environmental gradients, including increased rates among men, younger adults, racial/ethnic minority groups, and with urban birth and upbringing. This finding built on earlier epidemiologic studies from the United States and Europe, which revealed strong associations between neighborhood social deprivation and greater rates of psychosis. Nevertheless, to our knowledge, there has been no international comparison of the incidence of psychotic disorders since the World Health Organization study. We estimated the incidence of psychotic disorders across 17 catchment areas in 6 countries using comparable methods as part of the European Network of National Schizophrenia Networks Studying Gene-Environment Interactions (EU-GEI) study. Specifically, we tested whether differences in incidence could be attributed to putative social and environmental factors, including individual age, sex, and race/ethnic minority status, catchment area–level latitude, population density, unemployment, and proportion of single-person households and owner-occupied homes as markers of social disadvantage.

3The EU-GEI study is a multicenter incidence and case-sibling-control study of genetic and environmental determinants of psychotic disorders. Centers for the incidence study were England (n = 2; southeast London, Cambridgeshire and Peterborough), France (n = 3; 20th arrondissement of Paris, Val-de-Marne, Puy-de-Dôme), the Netherlands (n = 2; central Amsterdam, Gouda and Voorhout), Italy (n = 3; part of the Veneto region, Bologna municipality, and the city of Palermo), Spain (n = 6; Madrid [Vallecas], Barcelona, Valencia, Oviedo, Santiago, and Cuenca), and Brazil (n = 1; Ribeirão Preto). Catchment areas ranged from rural (Cuenca, 11 people per square kilometer) to urban (Paris, 33 260 people per square kilometer). Written informed consent was obtained from those who agreed to participate in the case-control study; otherwise, ethical approval was obtained to extract basic demographic and clinical details from patient records from local research ethics committees in each catchment area: South London and Maudsley and Institute of Psychiatry Research Ethics Committee; National Research Ethics Service Committee East of England–East Cambridge; Medisch-Ethische Toetsingscommissie van het Academisch Centrum te Amsterdam; Comité Ético de Investigación Clínica Hospital Gregorio Marañón; Comité Ético de Investigación Clínica del Hospital Clinic de Barcelona; Comité Ético de Investigación Clínica del Hospital Clinic Universitari de Valencia; Comité Ética de la Investigación Clínica del Principado de Asturias; Comité Ético de Investigación Clínica de Galicia; Comité Ético de Investigación Clínica del Hospital Virgen de la Luz de Cuenca; Comité de Protéction des Personnes–CPP Île de France IX; Comitato Etico Policlinico S Orsola Malpighi; Comitato Etico Azienda Ospedaleria Universitaria di Verona; Comitato Etico Palermo 1, Azienda Ospedaliera Policlinico “Paolo Giaccone”; and Research Ethics Committee of the clinical Hospital of Ribeirão Preto Medical School, University of São Paulo, Brazil.

4We identified all individuals who contacted mental health services in our catchment areas for a suspected first episode of psychosis (FEP). Case ascertainment varied from 12 months (London, England) to 48 months (Val-de-Marne, France) (Table 1), with a median of 25 months (interquartile range [IQR], 24-36 months). Case ascertainment predominantly took place between May 1, 2010, and April 1, 2015 (eTable 1 in the Supplement), and involved trained researchers making regular contact with all secondary and tertiary mental health care professionals to identify potential cases. In all countries, it was uncommon for people to be treated for FEP in primary care; instead, a patient with a suspicion of psychosis would typically be referred to specialist mental health services. Research teams were overseen by a psychiatrist with experience in epidemiologic research and included trained research nurses and clinical psychologists. Teams received training in epidemiologic principles and incidence study design to minimize nondifferential ascertainment bias across different local and national health care systems.

5Potential participants with FEP were included if they met the following criteria: resident within the catchment area at first presentation; 18 to 64 years of age; and presentation with a clinical diagnosis for an untreated FEP, even if longstanding (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10] codes F20-F33). We excluded individuals who had previous contact with mental health services for psychosis, evidence of psychotic symptoms precipitated by an organic cause, and transient psychotic symptoms resulting from acute intoxication, as defined by the ICD-10 (codes F1X.5)

6For participants who met these criteria, we obtained research-based diagnoses using the Operational Criteria Checklist algorithm (OPCRIT) to ensure comparability of diagnoses across catchment areas. The OPCRIT has high interrater reliability generally, and in our study after training (κ = 0.7). Assessment with OPCRIT was based on a semistructured clinical interview or review of case notes and other relevant information. The clinical interview schedule used at each site followed local expertise, including the Schedules for Clinical Assessment in Neuropsychiatry (United Kingdom and Italy), the Comprehensive Assessment of Symptoms and History (the Netherlands), the Structured Interview for DSM-IV (Brazil), and the Diagnostic Interview for Genetic Studies (France). Where OPCRIT assessment was not possible, we relied on clinical diagnoses.

7We estimated the population at risk, those 18 to 64 years of age, in each catchment area from the most accurate local or national routine demographic data available (eTable 2 in the Supplement), stratified by age (18-24 years, then 5-year bands), sex, and racial/ethnic minority status. We multiplied the population by case ascertainment duration (in years) to estimate person-years at risk.

8Our primary outcome was an OPCRIT-confirmed ICD-10 diagnosis of any clinically relevant psychotic disorder (ICD-10 codes F20-F33). This broad phenotype was considered alongside 2 secondary outcomes: nonaffective psychoses (ICD-10 codes F20-F29) and affective psychoses (ICD-10 codes F30-F33).

9Data on age group (as above), sex, racial/ethnic minority status, and country of birth were collected at baseline for all participants using the Medical Research Council Sociodemographic Questionnaire and case notes. We defined a binary variable to distinguish between the racial/ethnic majority population in each catchment area, and all other racial/ethnic minority groups. In each country, the racial/ethnic majority population was classified as the majority (white) racial/ethnic group, following national conventions (eAppendix 1 and eTable 2 in the Supplement), with all other groups classified as the racial/ethnic minority.

10Latitude was estimated in degrees from the equator. Population density was derived as number of inhabitants per square kilometer, based on official total population estimates. We derived 3 measures of the social environment (unemployment, owner-occupied housing, and single-person households) from the 2011 European Household and Population Census, a decennial census that provides comparable data at a provincial level (NUTS-2 [Nomenclature of Territorial Units for Statistics–2] regions). Equivalent data for Ribeirão Preto were derived from the 2010 National Census of Brazil. Duration of untreated psychosis (in weeks) was estimated for descriptive purposes, assessed via the Nottingham Onset Schedule, and based on time from onset of symptoms to first contact with secondary mental health services for suspected psychosis. For deviations from the protocol, see eAppendix 1 in the Supplement.

11Seven of 2774 cases (0.3%) were missing data on age or sex, and were excluded from direct standardization and statistical modeling, but retained for crude incidence rate estimation. Except for Puy-de-Dôme, France (eAppendix 1 in the Supplement), we coded any participants missing data on racial/ethnic minority status (n = 5 [0.2%]) to the racial/ethnic majority group.

12For each outcome, we estimated crude incidence rates per 100 000 person-years and 95% CIs by catchment area and sociodemographic characteristics. Next, we used direct standardization for age-band and sex and for age-band, sex, and racial/ethnic minority status to investigate variation in rates between catchment areas. We used the total population of England and Wales (2011 Census) as our standard population, and estimated standardized incidence ratios using the overall sample incidence rate as the reference category. Finally, we used random-effects (intercepts) Poisson regression to investigate variance in incidence by sociodemographic and environmental factors, accounting for the hierarchical structure of the data set. Age, sex, their interaction, and racial/ethnic minority status were treated as a priori confounders. We entered catchment area–level variables into our models one at a time based on the strength of association with incidence in univariable analyses, assessed via Akaike Information Criterion (lower scores indicate better model fit). Model building was assessed via likelihood ratio test. Analyses were carried out in Stata, version 13 (StataCorp). Results for secondary outcomes, as well as sensitivity analyses, are reported in eAppendix 2 in the Supplement. P <.05 (2-sided) was considered significant.

13We identified 2774 people presenting with a first episode of psychotic disorder, as defined by ICD-10 criteria, during 12.94 million person-years, corresponding to a crude incidence of 21.4 (95% CI, 19.4-23.4) per 100 000 person-years. A total of 1578 participants with FEP were men (56.9%), varying from 48.8% (40 of 82; Oviedo, Spain) to 77.8% (21 of 27; Cuenca, Spain) (χ2 = 34.3; P =.005; Table 1). A total of 1091 participants with FEP were from a racial/ethnic minority background (39.8%), varying from 2.8% (1 of 36; Santiago, Spain) to 75.6% (198 of 262; Southeast London, England) (χ2 = 455.8; P <.001). By comparison, almost 49.5% of the population at risk were men (6 401 911 of 12 933 665), and 22.9% were from a racial/ethnic minority group (2 962 395 of 12 933 665).

14Median age at first contact was 30.5 years (IQR, 23.0-41.0 years), varying from 26.0 years (IQR, 21.0-37.0 years) in Cuenca, Spain, to 35.5 years (IQR, 28.0-42.0 years) in Veneto, Italy (Kruskal-Wallis χ216 = 51.3; P <.001). First contact was earlier in men (28.0 years; IQR, 22.0-38.0 years) than in women (34.0 years; IQR, 26.0-45.0 years; Mann-Whitney test = –11.1; P <.001), but did not differ by racial/ethnic minority status (Mann-Whitney test = 1.0; P =.31). Median duration of untreated psychosis was 8.0 weeks (IQR, 2.0-35.0 weeks), varying from 2.5 weeks (IQR, 1.0-7.0 weeks) in Madrid, Spain, to 26.0 weeks (IQR, 2.0-77.0 weeks) in Cuenca, Spain (eTable 1 in the Supplement; Kruskal-Wallis χ215 = 119.7; P <.001).

15The age pattern of the incidence of FEP differed between men and women (Figure 1; likelihood ratio test χ28 = 119.3; P <.001). Crude rates of FEP peaked for men between 18 and 24 years of age (61.0 per 100 000 person-years; 95% CI, 59.0-63.1 per 100 000 person-years) and decreased steeply thereafter. For women, the incidence of FEP also peaked in the youngest age group (18-24 years) at 27.0 per 100 000 person-years (95% CI, 24.9-29.1 per 100 000 person-years), but decreased more gradually thereafter, with a small secondary peak between 50 and 54 years of age. Rates were higher in racial/ethnic minority groups (incidence rate ratio [IRR], 1.59; 95% CI, 1.46-1.72) after multivariable adjustment for age, sex, their interaction, and relevant catchment area–level characteristics.

16We observed 10-fold variation in crude incidence of FEP across our catchment areas (Figure 2 and Table 2), from 6.3 (95% CI, 3.9-8.6) per 100 000 person-years in Santiago, Spain, to 61.4 (95% CI, 59.4-63.5) per 100 000 person-years in southeast London, England. Age-sex standardization had a negligible effect on this variation (Figure 2). Additional standardization for racial/ethnic minority status attenuated variance, although an almost 8-fold variation remained; standardized incidence ratios varied from 0.29 (95% CI, 0.21-0.40) in Santiago, Spain, to 2.21 (95% CI, 1.84-2.65) in Paris, France.

17The association between crude incidence of FEP and catchment area–level exposures are shown in the eFigure in the Supplement; univariable random intercepts Poisson regression showed that greater owner-occupancy (IRR for a 10% increase, 0.73; 95% CI, 0.65-0.81) and unemployment (IRR for a 10% increase, 0.54; 95% CI, 0.34-0.84) were associated with a lower incidence of FEP, while percentage of single-person households (IRR for a 10% increase, 1.68; 95% CI, 1.24-2.27) was associated with a higher incidence of FEP (Table 3). A null random intercepts Poisson model confirmed substantial variation in incidence by catchment area (σ = 0.32; P =.006), which persisted after adjustment for age, sex, their interaction, and racial/ethnic minority status (σ = 0.23; P =.007). In multivariable analyses, incidence of FEP was 1.59 (95% CI, 1.46-1.72) times higher in racial/ethnic minority groups compared with the racial/ethnic majority population, and lower in catchment areas with more owner-occupied homes (IRR for a 10% increase in owner-occupancy, 0.76; 95% CI, 0.70-0.83). No other setting-level variables, including latitude (IRR, 0.99; 95% CI 0.97-1.01), improved our final model (Table 3), where residual variance by catchment area remained, albeit attenuated (σ = 0.06; P =.02). Similar results were observed for nonaffective and affective psychoses separately (eAppendix 2 in the Supplement).

18We observed substantial variation in the incidence of FEP across 17 catchment areas in 6 countries, confirming differential risk by place and person. In line with previous studies, we observed higher rates of all psychotic disorders in racial/ethnic minority groups and among young people, particularly for men. We confirmed a small but robust secondary peak in the risk of all FEPs for women older than 45 years. Catchment areas with higher owner-occupancy levels were associated with lower incidence rates of FEPs, implicating socioeconomic factors in the presentation of psychotic disorders, in line with findings of previous research.

19Overall, our incidence rates were consistent with those in the literature, although between-study heterogeneity in methods, inclusion criteria, and diagnoses studied make direct comparisons difficult. For example, the incidence of broadly defined schizophrenia in the World Health Organization Ten-Country Study varied from 15 to 42 per 100 000 person-years, although that study used a different age range (15-54 years) and did not consider affective psychoses. In our study, comparable rates of nonaffective psychoses varied from 5 to 41 per 100 000 person-years after standardization for age, sex, and racial/ethnic minority status. A systematic review limited to England observed a pooled crude incidence of all psychotic disorders of 32.7 per 100 000 person-years, somewhat higher than the overall crude incidence rate we observed (21.4 per 100 000 person-years). Such comparisons should be interpreted with caution, given heterogeneity in estimation methods and setting; few incidence studies have been conducted in southern Europe until recently, where rates appeared to be uniformly low, despite inclusion of urban catchment areas.

20The higher rates of psychotic disorders we observed in men, younger age groups, and racial/ethnic minorities, as well as for nonaffective psychoses, are also frequently reported in the literature. Our study provided further robust evidence of a secondary peak in the risk of psychosis for women older than 45 years, building on previous observations. Our findings add further evidence to the observation that early intervention services with an upper age limit of 35 years (or lower) may lead to sex-related mental health inequalities: only 50.8% of women (605 of 1190) with psychosis were identified before 35 years of age in our settings, compared with 67.9% of men (1070 of 1577) (Figure 1).

21Incidence of FEP varied not only by person, but importantly, by place, suggesting that the social environment may shape incidence patterns of FEP. Our best-fitting models of all FEP and nonaffective psychoses (eTable 3 in the Supplement) suggested that owner-occupancy levels were associated with incidence of FEP, although residual variation at the setting level was not explained by other catchment area–level measures. Acknowledging the potential for reverse causality, owner-occupancy may also be a proxy for a variety of social exposures, most obviously socioeconomic position, but extending to social stability and cohesiveness, which have previously been associated with psychosis. The incidence of FEP appeared to be lower in southern Europe, but we found no evidence of variation by latitude in our multivariable models. Nevertheless, settings were located within a narrow band (38°-53° north of the equator), except for Brazil (21° south of the equator). This location may have contributed to our null findings, and the absence of high rates of psychosis in our southern Europe settings, particularly in major urban centers, requires further investigation; incidence patterns with respect to population density in southern European settings appeared to diverge from those observed in northern Europe (eTable 4 in the Supplement). Variation in the incidence of affective disorders, with lower rates in catchment areas with higher levels of unemployment (eTable 3 in the Supplement), is counterintuitive and unexpected; further research is required to examine this finding.

22Our findings should be interpreted alongside the strengths and limitations of our study. Our large sample size allowed us to estimate 3 psychotic outcomes in 17 settings with a high degree of precision. To minimize ascertainment bias, all researchers received training via face-to-face epidemiologic training sessions, regular teleconferencing, online training manuals, and interrater reliability protocols. Nonetheless, some limitations of our multinational design need to be acknowledged.

23Detection of patients who never present to services is an issue for all epidemiologic studies, and our rate estimates should be interpreted as the treated incidence. Although our overarching case ascertainment method was similar across all settings, some adaptation to local health care systems was necessary. For example, primary care in each catchment area may have referred different proportions of patients with FEP to secondary mental health care services, but referral guidelines were very similar across national settings; these guidelines all urge prompt referral of anyone with FEP. That said, we did not assess whether referral practices were consistent within and between catchment areas. Difference in the average timing of referral may have affected the case mix within the FEP category, but not the overall number of referrals; each center was in a steady state.

24Differences in the organization of secondary mental health care services across localities may also have influenced detection of patients. In England and the Netherlands, for example, the widespread commissioning of early intervention in psychosis services may have led to improved detection of new cases of FEP. The leakage study in Brazil revealed a substantial number of new cases at this site (279 [49%]), while similar approaches in 2 French sites (Paris and Val-de-Marne) identified far fewer missed cases (7 [6%] in Paris and 28 [13%] in Val-de-Marne). Comprehensive, regular contact with mental health services should have helped minimize underascertainment, although some patients, including those treated privately, may have been missed; in general, we believe these biases are unlikely to account for the 8-fold variation between catchment areas.

25We used validated semistructured interview schedules, where possible, to obtain standardized research-based OPCRIT diagnoses close to the patient’s first presentation. We have no reason to believe the use of different schedules by setting biased our estimates; indeed, this choice was adapted to local expertise to minimize bias, which may have otherwise arisen from using unfamiliar interview schedules. We considered total incidence of FEP as our primary outcome, as this is useful for, and consistent with, contemporary practice in the management and treatment of psychosis, to allow symptoms to evolve at first presentation and minimize stigma. This practice is also consistent with some evidence of diagnostic instability in the early course of disorder, particularly for psychotic disorders other than schizophrenia. Although we relied on clinical diagnoses in a small proportion of patients, this did not alter the interpretation of our findings (eAppendix 2 and eTable 5 in the Supplement).

26We classified racial/ethnic minority status as a binary variable, following official definitions used in each country to distinguish racial/ethnic majority and minority groups. This approach may have led to some misclassification, particularly in France, which does not differentiate between people born in mainland France vs its overseas territories, nor is able to identify second-generation (French-born) migrants. This misclassification would have conservatively biased IRRs with respect to racial/ethnic minority status, as would have our decision to code participants with missing data on race/ethnicity (0.2%) to the racial/ethnic majority group. Our binary race/ethnicity variable may also have permitted residual confounding; risk of psychosis by race/ethnicity will be studied in greater detail in future EU-GEI publications.

27We used a consistent method in European catchment areas to estimate measures of the social environment, with comparable data taken from the Brazilian census. European data could only be obtained at the NUTS-2 regional level, which is larger than our catchment areas. Data from this level may have led to exposure misclassification, although the effect of this ecological bias is difficult to determine.

28Although we controlled for several risk factors simultaneously (age, sex, racial/ethnic minority status, and catchment area-level factors), we were unable to include other putative risk factors for psychosis, including cannabis use, urban birth, family history of psychosis, childhood trauma, or genetic risk. These factors are not routinely available in denominator estimates, but will be investigated in future case-control designs from the EU-GEI study.

29In this international, multicenter study we found that treated incidence of psychotic disorders varied 8-fold between catchment areas after standardization for age, sex, and racial/ethnic minority status. Rates were higher in younger people, men, racial/ethnic minorities, and areas with lower levels of owner-occupied housing, although substantial variation between catchment areas, and by broad diagnosis, remained. These results suggest that there is pronounced variation in the health care burden of psychosis worldwide.