Data Source

The Health Improvement Network (THIN) database is one of the largest UK sources of continuous primary care data, containing information on illness recording and prescribing behaviour. At the time of data collection from THIN (http://csdmruk.cegedim.com) 437 participating general practices, contributing more than 10 million patients, were included. The database is broadly representative of the UK population [15]. In the UK most people with SMI are registered with primary care [16] and the validity of general practice computer diagnoses of SMI has been established previously [17]. THIN data have also been shown to be roughly representative of UK general practice in terms of consultations and prescribing statistics [18], [19]. THIN contains records of each patient's medical conditions and symptoms, recorded during routine consultations and all prescriptions issued by GPs. Symptoms and diagnoses are classified using the Read code system, a hierarchical recording system used to record clinical summary information [20]. This creates a computerised medical history for each patient from the time they register with a general practice. In addition, the database holds information on basic demographics, urbanicity and social deprivation. Based on their residential postcode, patients are classified as residing in urban areas (population >10,000); or in towns and fringes; or in villages, hamlets and isolated areas. Social deprivation is measured using the Townsend score for the postcode sector area of residence, linked to population census data from 2001 [21]. It is a combined measure of owner-occupation, car ownership, overcrowding and unemployment [22]. The scores are defined for small areas of around 150 households, and grouped into quintiles.

Ethics Statement

The scheme for THIN to obtain and provide anonymous patient data to researchers was approved by the National Health Service South-East Multicentre Research Ethics Committee (MREC) in 2002 and scientific approval for this study was obtained from CMD Medical Research's Scientific Review Committee in March 2012.

Study Population

We included data from the date at which practices had met quality assurance criteria, namely continuously acceptable computer usage (ACU) (i.e. one medical record, one additional health data record per patient per year, and at least two prescriptions, on average per patient per year [23]) and the criteria for acceptable mortality reporting (AMR) which indicate a point at which the observed death rate for a practice corresponds to that expected based on predicted numbers of deaths derived from National statistics given the practice's demographics [24], [25].

We included all individuals aged 16 to 95 years, permanently registered for at least one year during the period from 1 January 2000 to 31 December 2010. We examined two subgroups according to age criteria: those traditionally eligible for entry to Early Intervention Services in the UK (16 to 35 years old)[26], [27] those eligible for entry to General Adult psychiatric services (16 to 65 years old).

For this study we were interested in individuals who had a first recording suggestive of a new diagnosis of SMI in their primary care records. Therefore we excluded patients who had a record of SMI prior to start of follow-up (see details of follow-up in statistical analysis) or whose first SMI record during follow-up was indicative of pre-existing SMI or repeated episodes (such as chronic paranoid schizophrenia, or manic relapse). We also excluded diagnoses made within the first year of registration, as people who received a code within the first year were more likely to be prevalent rather than incident cases [28].

Measurement of main outcome

Cases of SMI included those who had new records of a Read code for SMI (schizophrenia, bipolar disorder, other non-organic psychotic illness). A list of all SMI diagnoses was constructed using established methods [29] and cross-checked with lists of codes given in national QOF guidance.

SMI Patients were classified according to the type of diagnosis (schizophrenia, bipolar disorder, other non-organic psychosis). If patients first received a code for “other psychosis”, but had subsequent codes to indicate schizophrenia or bipolar disorder, they were reclassified as schizophrenia or bipolar accordingly. However, the date of the first diagnosis was retained as the date of the first record. Similarly, if patients first received a code to indicate inclusion on an SMI register, they were reclassified as schizophrenia, bipolar or other psychosis if they received these diagnoses subsequently. If patients received both bipolar and schizophrenia diagnoses, they were coded as their latest diagnosis (as this was considered likely to be most accurate, having considered the whole longitudinal medical history). Patients with no diagnostic codes at any time but with codes to indicate inclusion on an SMI register were excluded from the analysis. Patients receiving a diagnosis code of dementia within a year of their SMI code were excluded.

The number of individuals with a newly recorded diagnosis was determined by age (10 year age groups and service-line groups), sex, urbanicity and quintiles of Townsend score. Individual level ethnicity was not well recorded historically in THIN, therefore it was not possible to describe recording by this covariate.

Statistical Analysis

The recording of coded SMI was estimated per 100,000 person years at risk (PYAR) as the total number of new SMI cases recorded between 2000 and 2010, divided by the total number of person years of follow-up. Person-time for the denominator was estimated as the latest of: [16^th birthday, one year's registration, ACU/AMR date, start date of period], to the earliest of: [date of first incident diagnosis, date of death, date patient leaves practice, date of last data collection from the practice, end date of period].Recorded rates of all SMI and of different forms of SMI (schizophrenia, bipolar disorder, other psychosis) were estimated, according to age, gender, Townsend score and urbanicity.

Annual rates were graphed to examine the time trends. Multivariable Poisson regression models with (log) person-time as an offset, were used to examine recording of all SMI by gender, age (in 10 year age-bands) deprivation (quintiles of Townsend scores) and urbanicity (as three categories: urban, town/fringe and village/hamlet/isolated). Multilevel random intercept models were used to account for clustering of patients in practices. All analyses were carried out using STATA 12.

Schizophrenia

Schizophrenia was the least commonly recorded diagnosis of severe mental illnesses in primary care records. Up to 2004, there were between 10 and 14 new entries per 100,000 PYAR. However, by 2007 there were only around 5 new entries per 100,000 PYAR (Figure 1). In the 16–65 age group the rate was 9.2 per 100,000 PYAR (95% CI 8.7 to 9.6) and in the 16–35 age group it was 14.3 Per 100,000 PYAR (95% CI 13.4 to 15.3). Schizophrenia was more commonly recorded in men than women (adjusted IRR 0.6, 95% CI 0.54 to 0.66) (Table 1). In men the diagnosis was most commonly recorded in the 16–24 age group and recording reduced with increasing age, whereas in women there was no difference across age groups after adjustment for other factors (Table 2). Recording of schizophrenia increased with increasing social deprivation such that individuals in the most deprived quintile of Townsend score were nearly 5 times more likely to receive a diagnosis of schizophrenia, than those in the least deprived quintile (Table 1). After accounting for age, sex and social deprivation there was no difference in recording of schizophrenia in urban versus rural areas (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Time trends in GP recorded diagnosis of schizophrenia, bipolar disorder and other psychosis.

https://doi.org/10.1371/journal.pone.0082365.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Recording rate of individual diagnoses by socio-demographic factors.

https://doi.org/10.1371/journal.pone.0082365.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Recording of rate of individual diagnosis by age and gender.

https://doi.org/10.1371/journal.pone.0082365.t002

Bipolar disorder

Recording a new diagnosis of bipolar disorder ranged between 11 and 19 per 100,000 PYAR between 2000 and 2010, with a peak around 2004. However, by 2010 nearly 3 times as many people had a new record of bipolar as of schizophrenia (Figure 1) In the 16–65 and 16–35 subgroups bipolar disorder was recorded at a rate of 15.0 per 100,000 PYAR (95% CI 14.4 to 15.5) and 14.8 per 100,000 PYAR (95% CI 13.9 to 15.8) respectively over the period of study. In contrast to schizophrenia recording of bipolar disorder was more commonly recorded in women than men (Table 1). For men the first diagnosis was commonly recorded between the ages of 35–44 years, whereas women were diagnosed earlier (most commonly between 25–34 years) (Table 2). Like schizophrenia, recorded bipolar disorder increased with increasing social deprivation and the most deprived quintile was almost twice as common as the least deprived (Table 1). After accounting for age, sex and social deprivation there was no difference in recording of bipolar disorder in urban versus rural areas (Table 1).

Other psychosis

Compared to schizophrenia and bipolar disorders a larger group had a record of non-organic psychotic illnesses (Figure 1). An increasing number of individuals received such diagnosis between 2000 and 2004. Thereafter, the recording fell to around 20 per 100,000 person years with an increase in 2010. It was recorded at a rate of 22.3 per 100,000 PYAR (95% CI 21.6 to 23.0) in the 16–65 age group, and 29.1 per 100,000 PYAR (95% CI 27.8 to 30.4) in the 16–35 age group. Similar to schizophrenia, recording was highest for men in the 16–24 age group, but for women the recording increased over 75 years old (Table 2). Again there was an increase in recording with increasing deprivation; with nearly three times as many individuals in the most deprived group, and no statistically significant relationship with urbanicity (Table 1).

Diagnosis stability

For the majority of individuals with a record of SMI (90.3%) there was no discrepancy in diagnosis codes assigned over the 10 year study period (Table 3). However for those who had an initial record of non-organic psychosis 8.0% were subsequently coded as schizophrenia and 3.4% for bipolar disorder. A switch in code from bipolar to schizophrenia or vice versa occurred in 1.5% of the individuals (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Changes in diagnoses code.

https://doi.org/10.1371/journal.pone.0082365.t003

Strengths and Limitations

A key strength of this study is the large size of the population sample (over 4 million patients) enabling precise estimates of rates of recording in primary care. Furthermore, THIN covers the whole UK and is broadly demographically representative of UK primary care patients [18].

There are multiple challenges in estimating incidence from dynamic, longitudinal GP records. We defined our cases by Read code diagnosis as we were looking specifically at GP-recorded SMI, so these patients may not be regarded as “cases” in terms of standardised diagnostic criteria (ie ICD-10 or DSM-5) however previous research has found the diagnosis of psychosis (based on clinician reported Read codes) to be valid [17]. Because of the nature of the data, we can only state that these are newly recorded cases, rather than true incident cases of SMI; however we attempted to reduce the chance of prevalent cases being recorded as new by excluding those registered for less than one year and excluding Read codes suggestive of chronic illness (e.g. “chronic schizophrenia”).

People who received a prescription for an antipsychotic medication without an SMI code being entered were excluded, as it was unclear what the GP's working diagnosis was in such cases. This group would include those prescribed antipsychotic medication for another indication, such as behavioural disturbance, dementia or severe obsessive compulsive disorder.

The data are limited to GP-recorded SMI in-practice attendees, which reflect incidence, presentation and recording at a practice level only. Therefore it is possible that some individuals (especially younger men [45]) have been missed, as they are not registered with a GP which would be likely to disproportionately reduce the total population at risk in comparison to the number of cases (as individuals with SMI are highly likely to be registered with a GP). There is also a possibility that the patients detected (especially in older age groups) had a previous diagnosis of SMI, but that this was not recorded when they later moved into a THIN contributing practice. We attempted to limit this by excluding individuals coded within one year of registration. There may be SMI symptoms coded in free text, which would suggest our findings may be an underestimate of the true burden of disease in primary care. The rates we found are however, higher than that found in other studies identifying people mainly in secondary care settings. A number of patients (4,151) were coded as being included on the general practice SMI register, but were not picked up by our Read code list search, and as a result we excluded them from the analysis. We have not determined why these patients are included on the SMI register without also having an SMI diagnosis, but it may be that GPs have also coded as SMI those patients with mental health problems who require a lot of input but don’t truly meet the defined criteria of schizophrenia, bipolar disorder or other psychotic illness (such as chronic recurrent depression, anxiety or personality disorder). There are no formal checks made on who is added to the SMI register and there is evidence that there was confusion around the definition early on [1]. Peaks in the recording of each specific diagnosis in the year QOF was introduced (2004) suggest that GP's did a ‘catch-up’ of recording and that some of these were actually prevalent cases rather than newly diagnosed.

Another limitation is the lack of recording of ethnicity in primary care. Recording of ethnicity has improved since 2005, in particular for newly registered patients. However, there is still a large proportion without information [46]. Therefore, we did not make an attempt to establish whether certain ethnic groups were more likely to have a record of psychotic illnesses as shown elsewhere [8].