Identification of participants
We collected clinical and demographic information on all people from a defined area of south London (previously the London Borough of Camberwell) who presented with psychosis during 1988-97. All psychiatric services for the area during this period were provided by the Bethlem Royal and Maudsley NHS Trust through hospital and community teams. We identified cases from hospital computer records by generating a list of all people admitted with any possible psychotic illness as defined by ICD-9 (international classification of diseases and related health problems, ninth revision) codes 295, 295.6, 296, 296.2, 296.4, 297, 298, and 292.1, and ICD-10 (tenth revision) codes F20, 25, 22, 30, 31.3, 31.2, 31.6, 28, 29, 12.5, 16.6, 19.5, 16.75, and 19.75. We also examined case notes of all patients from the area who had psychiatric hospital records to identify people who made contact with services but were not admitted to hospital.
We checked case records to ensure the individuals were true incident cases (that is, they had not previously had contact with psychiatric services) and rated them using the operational checklist for psychotic disorders.13
Two authors, JK and JvO, carried out the ratings, and inter-rater reliability for Research Diagnostic Criteria schizophrenia14
was good (κ=0.79). The checklist is based on phenomenological descriptions in the present state examination15
and enables a computer diagnosis of Research Diagnostic Criteria schizophrenia to be made with the associated computer program.16
Ethnic and sociodemographic status
We classified ethnicity on the basis of that recorded by the patients themselves, according to categories used by the Office of Population Censuses and Surveys. We also noted the patient's and his or her parents' place of birth, when available, and any description of colour (mental state examinations routinely comment on appearance). We used this information to determine ethnicity for those patients who did not have statements of self assigned ethnicity. A check on this method was carried out by Castle et al, who compared results with those from previous direct interviews and found no errors in 34 patients.17
Because the population projections at ward level were not accurate enough to calculate population data separately for each ethnic minority, we were able to split the population into only two groups: a white group (self assigned ethnicity white) and non-white group (all other self assigned ethnicities). The non-white population was about 40% Caribbean, 30% African, and 10% other. Incident cases were assigned to either white or non-white groups. Thus the effect we measured was that of non-white ethnic minority status.
The area (about 120
000 people) was divided into electoral wards of about 10
000 people, which had different socioeconomic characteristics. We used the address at presentation to the services to identify wards for all incident cases. We estimated population data using the 1991 census18
and London Research Centre projections, which include corrections for undernumeration and information about housing, mortality, and migration.19
For the years 1992-7 our calculations were based on a linear interpolation using 1991 census data and 1997 population projection data. For the years 1988-90 we extrapolated data on the basis of the 1991-7 data. No further interpolation was attempted before 1988, so that the all analyses relate to the period 1988-97. Socioeconomic status at ward (neighbourhood) level was based on a composite deprivation score (Department of Environment index of local conditions),20
which includes unemployment, overcrowding, child poverty, lack of amenities, low earnings, no car, and low level of education (but not ethnic group).
We carried out indirect standardisation with the Research Diagnostic Criteria rates of schizophrenia for the total 10 year population as the standard and applied them to each ward, stratifying for age, sex, and ethnic minority using the ISTDIZE procedure in the Stata statistical program (StataCorp, College Station, TX). The standardisation used the stratum specific rates of the standard population to calculate the expected number of cases for each ward and the adjusted incidence rates at ward level. We calculated the standardised incidence ratio by dividing the number of cases observed by the expected number.
By using the Stata XTPOIS multilevel Poisson regression modelling procedure, we examined two types of effects: firstly, the ward random effects—that is, are the wards different with regard to the incidence of schizophrenia; and, secondly, ward and individual fixed effects—that is, does the factor being studied make neighbourhoods different with regard to the incidence of schizophrenia. We examined the fixed effects of age, sex, and non-white ethnic minority status at the individual level and deprivation and ethnic density (proportion of ethnic minorities) in thirds of distribution (highest, middle, lowest) at ward level. We carried out multilevel Poisson regression analysis to calculate incidence rate ratios for Research Diagnostic Criteria schizophrenia for individual and ward variables and to test for interaction between non-white ethnic minority status at the individual level and proportion of non-white ethnic minority at the neighbourhood level. Interaction terms were assessed by likelihood ratio tests. We adjusted associations between schizophrenia and individual level non-white ethnic minority status and ward level proportion of non-white ethnic minority for age, sex, and ward level of deprivation.