Design
This study used a historical cohort design based on Electronic Health Records (EHRs) from the South London and Maudsley (SLaM) NHS Trust Foundation electronic Patient Journey System (ePJS). Since 2006, comprehensive health records from over 280,000 patients in the ePJS have been de-identified and made accessible via the Clinical Record Interactive Search tool (CRIS). CRIS holds all information documented by professionals involved in the provision of specialist mental health care for all people in contact with SLaM mental healthcare services from 1 January 2007 to date [16]. SLaM covers the four London boroughs of Lambeth, Southwark, Lewisham and Croydon.
Sample
The sample comprised all cases age 16 and over who presented for the first time to SLaM between 1 January 2007 and 31 December 2010 with a diagnosis of any psychotic disorder (ICD-10 codes F20-29, 30.2, 31.2, 31.5, 32.3, and 33.3 [17]). Diagnostic information was drawn both from structured fields in the record and a natural language processing algorithm extracting diagnostic statements in text fields [16]. The natural language processing algorithm simply sought to extract any text strings associated with a diagnosis statement to supplement the existing structured fields. Individuals were followed from the date of their first diagnosis for up to 5 years.
We excluded participants without a postcode at first presentation within the four-borough catchment. This was carried out to exclude people who lived in areas covered by neighbouring mental health service providers, but who may have presented to SLaM emergency services, or to exclude referrals from outside of the catchment to specialist national services provided by SLaM. Participants of no fixed abode were also excluded by design.
Measures
Structured data were extracted on age at presentation, gender, ethnicity, education level, employment status and occupation and postcode linked in CRIS to two administrative geographical levels: Lower (LSOA) and Middle Super Output Areas (MSOA). These geographical units were used to link participants to several socio-environmental indices (see below). LSOAs and MSOAs are geographical areas made up of clusters of socially homogenous postcodes. LSOAs have an average of average 1500 residents (minimum 1000 residents) while MSOAs have an average of 7200 residents (minimum 5000 residents) [18]. Data on employment status, occupation, highest educational level and age at leaving education were only available for the following percentages of people: 32, 31, 51, and 20%. Due to the poor availability of these data, they were excluded from the analyses. Ethnicity was recorded in patient records according to the 16 + 1 ethnic data categories defined in the 2001 census. The outcomes, number of inpatient days over a 1- and 5-year follow-up from the first contact with mental health services, were calculated from ward stay tables, routinely recorded within CRIS, which contain the date of admissions and discharge from inpatient wards. The total time with services was calculated as the time between the first contact and last contact within the 5-year window.
To classify neighbourhood socioenvironmental exposure to deprivation, we used the English Indices of deprivation. These routinely collected national indicators were based on 2007 data, temporally the closest to the baseline date for most participants. Indices were provided at the LSOA level [19]. In our analyses, we assigned participants to their Index of Multiple Deprivation score for their residential address at first presentation.
Population density was determined using 2011 census data (temporally the closest to the baseline date) on the number of people per hectare [20], and based on LSOA11 codes. Ethnic density was defined as the proportion of people from the same ethnic group as the participant living in their LSOA [6], also estimated from the ONS [21] 2011 census. Data on social capital were determined using voter turnout at local government elections as a proxy [9]; however, these data were not available at the LSOA level, thus boroughs had to be used. LSOA for each person was mapped to borough to link these data, and data on the percentage voter turnout were taken from the 2009 European Parliament election [22] as this was temporally the closest to the period of study.
Social fragmentation was calculated for each LSOA from four measures of social composition in the 2011 census: unmarried adults [23], single-person households [24], households privately renting [25], and population turnover (2009–2010 data) [26]. Consistent with previous studies, we summed the z scores (number of standard deviations above/below the population mean when the distribution is normal) to create a social fragmentation index [27, 28]. Population turnover was calculated as the sum of in- and out-migration in the 12 months prior to the 2011 census [29] and was only available at the MSOA level so was calculated using MSOA11 codes.
Ethics
Ethical approval as an anonymised database for secondary analysis was originally granted in 2008 and renewed for a further 5 years in 2013 (Oxford C Research Ethics Committee, reference 08/H0606/71 + 5). The study presented in this paper has been approved by the CRIS Oversight Committee.
Analysis
Data were first described using mean values, standard deviations and ranges, or frequencies and percentages as appropriate. The relationships between area-level factors were investigated using correlations. The relationship between the number of inpatient days and each socio-environmental factor was investigated in univariate negative binomial regression models with time in contact with services treated as an offset variable. Negative binomial modelling was preferred over more typical Poisson regression for the type of outcome (count data with positively skewed distribution) to account for over-dispersion in the data. Regression analyses were conducted with adjustment for age, gender and BME status. STATA 14 [30] was used for the analyses.
Sensitivity analyses
We conducted sensitivity analysis regarding the follow-up period since lack of contact might occur for a range of reasons: the person is still living in the catchment area but is well and not in need of contact with services; or the person has left the catchment area but is unwell and being treated by another service. To address this, a number of sensitivity analyses were conducted, as follows:
-
Reanalysis of inpatient days over one year based on the full sample (less reliant on five year data) controlling for length of time in contact with services;
-
Reanalysis of inpatient days over 5 years based on those in contact with services for the full 5 years only (excludes those who might have left the area but probably sampling a more severely ill population);
-
Reanalysis of inpatient days over 1 year based on those in contact with services for the 1 year only (excludes those who might have left the area but probably sampling a more severely ill population).