Design and setting

We obtained data from the Social Epidemiology of Psychoses in East Anglia (SEPEA) study, originally designed to investigate variation in the incidence of psychotic disorders in rural England [26]. Case ascertainment took place between 1 August 2009 and 31 January 2013 in Cambridgeshire and Peterborough NHS Foundation Trust (CPFT), and from 28 September 2009 to 28 March 2013 in the Norfolk and Suffolk NHS Foundation Trust (NSFT).

Inclusion criteria and participants

Participants accepted into EIP services at first referral for suspected psychosis were included in the present study if they met the following inclusion criteria:

1. Acceptance to EIP services
2. Aged between 16–35 in NSFT, or 17–35 in CPFT
3. Resident in the defined catchment area
4. No previous contacts with mental health services for psychotic symptoms, or previous treatment involving antipsychotic medication for more than 6 months.

Procedures

We collected sociodemographic data at baseline. Clinical information was collected at two time points: 6 months after acceptance into EIP care and at discharge from EIP services (up to 3 years of care); at these time points the clinicians responsible for care provided primary and secondary International Classification of Diseases, Tenth Revision (ICD-10) clinical diagnoses.

Outcome variable

In the present paper, our main outcome was a clinical diagnosis of ICD-10 PD (F60.X) as a primary or secondary diagnosis at either time point during EIP care (6 months post-acceptance, or discharge).

Exposure variables

We considered FEP (yes/no) as a predictor of PD. We defined FEP as a clinical diagnosis of an ICD-10 psychotic disorder (F10-33) at either time point, subsequently confirmed by a standardised research-based diagnosis using the OPCRIT assessment. OPCRIT is known to produce reliable and valid ICD-10 diagnoses for major psychotic and affective disorders based on rating of 90 signs and symptoms of disorder [29,30]. A panel of trained diagnosticians conducted OPCRIT assessments, with acceptable inter-rater reliability as previously reported [26].

We obtained data on age (continuous), sex (male/female), ethnicity, marital status, individual and parental SES and country of birth at EIP acceptance. Age was defined as age-at-first-referral to EIP care. Ethnicity was self-ascribed to one of 18 categories in the 2011 census, recoded here as a binary variable (white British versus black and minority ethnic (BME) groups). Marital status was classified as married/civil partnership, divorced/dissolved/ separated or single. Occupational data on participants and their parents was classified into five (participants) or four (parents) categories according to the National Statistics Socio-Economic Classification guidelines[31], as professional/managerial, intermediate occupations, routine/manual, and people not in employment, with students in a fifth category for participant SES.

We geo-coded participants to their small area of residence using postcode information for neighbourhood-level exposures on PD risk. We used “electoral wards” as our neighbourhood-level of analysis, with 530 wards in the SEPEA catchment area, with a median population of 3,992 people (interquartile range (IQR): 2,426–5,935). Population density was estimated based on the 2011 census population divided by area size in hectares (people per hectare (ppha)), and categorized into four equal-interval groups (0–4,000; 4,001–8,000; 8,001–12,000; and over 12,000 ppha). Multiple deprivation was defined as the proportion of households in each neighbourhood categorized as deprived on two or more of four indicators from the 2011 census covering employment, education, health, and living environment (Kirkbride et al., 2017), with four equal-interval scale categories (7.7%-18%, 18.1%-28%, 28.1%-38%, and 38.1%-47.1%).

We included waiting time (in weeks) between referral and acceptance by EIP services [32] and symptom dimensionality as potential predictors of PD in our analyses. Symptom dimensions were derived from a factor analysis of OPCRIT items, as previously described [33], for which we included scores on seven dimensions: mania, depressive symptoms, other delusions, psychomotor poverty and disorganisation, first rank delusions, paranoia and hallucinations.

Statistical analysis

We used Chi-square (χ2) tests or Fisher’s exact test (FET), and independent t-tests or Wilcoxon Rank sum-tests (RST), as appropriate, to investigate univariable differences in PD prevalence by sociodemographic and clinical characteristics. We then conducted multilevel logistic regression to examine the joint effects of these factors on PD prevalence. We used Akaike’s Information Criterion (AIC) to determine entry order into a multivariable model, with lower AIC scores indicating better model fit, and employed a forward selection method to identify the best-fitting set of predictors associated with PD. We used likelihood ratio tests (LRT) to determine the best-fitting model, with statistical significance set at p<0.05. We conducted complete case analysis due to minimal missing data (N = 39, 4.9%). Analyses were conducted in Stata version 14.

Ethics

Ethical approval was granted by the Cambridgeshire III Local Research Ethics Committee (09/H0309/39). All data have been fully anonymized before we accessed them for analysis, and we also acquired written consent from the patient for using their medical records to be used in research.

Demographic and clinical characteristics

We identified 798 participants who met inclusion criteria, of whom 687 (86.1%) received a research-based diagnosis of FEP and 76 (9.5%) received a clinical PD diagnosis (Table 1). Fifty-two of those diagnosed with FEP (7.5%) received a comorbid PD diagnosis. The median age-at-first referral to EIP of people with and without any recorded PD was 20.5 (IQR: 18.2–23.9) and 22.6 (IQR: 19.4–27.2) years old, respectively (RST: 3.4; p<0.01). There were slightly more women (57.9%, N = 44) than men (42.1%, N = 32) with PD, whereas two thirds (67.7%, N = 489) of people without PD were men (χ2 test: 19.9; p<0.01). Participants with PD were also more likely to be white British (FET; p<0.01), from low SES (individual SES: FET, p = 0.04; parental SES: χ2 test: 11.1, p = 0.01), UK-born (FET; p<0.01), and were less likely to be diagnosed with FEP (χ2 test: 21.9; p<0.01); they also had longer median waiting times between referral and acceptance to EIP (RST: -2.0; p = 0.04). Participants with PD differed on most psychopathology dimensions at initial assessment, and were rated as having more hallucinations, paranoia and depressive symptoms than those without PD (Table 1), but fewer manic, negative and first rank delusional symptoms. There were no significant differences between participants with and without PD in neighbourhood-level population density (χ2 test: 2.4; p = 0.48) or multiple deprivation (χ2 test: 3.4; p = 0.34).

Download:

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

Table 1. Demographic and clinical characteristics of people within EIP service.

https://doi.org/10.1371/journal.pone.0234047.t001

Sub-types of PD

The majority of participants with a PD record (75%) received a diagnosis of ‘emotionally unstable’ PD (75%; see Fig 1), followed by ‘dissocial PD’ (9.2%) and ‘unspecified PD (9.2%). Due to the small sample size, we did not inspect these subtypes further in the present study.

Download:

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

Fig 1. Sub-types of PD diagnosed (N = 76) among participants (ICD-10).

https://doi.org/10.1371/journal.pone.0234047.g001

Missing data

Thirty-nine participants (4.9%) had missing data on either marital status or neighbourhood-factors (S1 Table). There were no significant differences between people with and without missing data, except that participants with missing data had lower individual and parental SES compared with those without missing data (both p<0.01).

Sociodemographic and clinical features of people with PD

Univariable modelling (Table 2) confirmed the above patterns. After multivariable modelling, receiving a PD diagnosis was associated with sex, ethnicity, parental SES, FEP status, first rank delusions and hallucinations. Thus, PD was more common amongst women (odds ratio (OR): 3.36; 95% CI: 1.97–5.73), participants with lower parental SES (OR per lower category: 1.43; 95% CI: 1.12–1.84), those without comorbid FEP (OR: 2.99; 95% CI: 1.61–5.52) and those experiencing more hallucinations (OR: 1.61; 95% CI: 1.23–2.12). PD risk was weakly associated with age-at-referral (OR per year: 0.95; 95% CI: 0.89–1.00; p = 0.059), which was retained in the model. PD risk was lower in the BME group (OR: 0.27; 95% CI: 0.11–0.71), and participants with more first rank delusions symptoms (OR: 0.74; 95% CI: 0.56–0.97).

Download:

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

Table 2. Logistic regression models for predicting receiving a PD diagnosis in the EIP service.

https://doi.org/10.1371/journal.pone.0234047.t002

Although population density and multiple deprivation were not associated with PD in univariable models (Table 2), we re-checked these variables in our final multivariable model given our a priori interest in these factors; however, neither variable improved the final model (LRT_{population density} p = 0.40); (LRT_{multiple deprivation} p = 0.89).

Main findings

To our knowledge, this is the first study to have investigated the prevalence and correlates of recording of PD diagnoses in a large, prospectively-collected sample in EIP services. While the overall prevalence of PD diagnoses was lower than we expected, at about 10%, PD diagnoses were more common for women and in those of white British ethnicity, as hypothesised. We also found that people with lower parental SES and those found not to meet research-based diagnostic criteria for FEP during their EIP care were more likely to receive a PD diagnosis. Contrary to our expectations, there was no evidence of an association between prevalence of PD and neighbourhood-level deprivation or population density in this sample. People reporting more hallucinations and fewer first-rank delusions, were more likely to receive PD diagnosis, underlining overlapping phenomenological presentations at first referral to EIP services.

Meaning of findings

Our findings are partially consistent with previous studies. The prevalence of recording PD in our study was much lower compared with the 45% prevalence in a study conducted in a similar EIP setting [11], or studies carried out in secondary mental health care settings, where PD prevalence ranges from 31% to 92% [34–37]. A number of possible reasons for this may exist. For example, the present study is based on people referred to EIP services due to suspected psychosis. When psychotic symptoms become prominent enough to warrant clinical attention, the sensitivity of PD diagnosis could be weaker [3] or, alternatively, the likelihood that clinicians record a PD diagnosis may be reduced. We may have also underestimated the true prevalence of PD in our sample because, unlike previous studies [11,34–37], we did not use a structured instrument such as SCID-II to define PD. Given the naturalistic design of our cohort, we were reliant on PD diagnoses recorded during routine clinical practice. Further, while our sample was based on precise epidemiological criteria, we only assessed people accepted into EIP care; PD prevalence may be higher amongst those referred to, but not accepted by EIP services, and may indicate that these services are already effectively screening and triaging people who require onward referral to other specialist psychiatric services.

It is perhaps unsurprising that the absence of a FEP diagnosis increased the odds of receiving PD diagnosis in clinical practice, and is consistent with a small handful of people treated in EIP who, after extended evaluation, require onward referral to other specialist services. A similar result was reported in a recent study, indicating that participants at high-risk for psychosis were found to present to EIP services with more prominent personality traits and low transition to FEP at follow-up[38]. The associations we observed between PD prevalence and symptom dimensions associated with psychosis (including hallucinations and first rank delusions) is novel and underlines the phenomenological overlap and difficulties that diagnosticians may face in evaluating participants with symptoms inherent to both PD and FEP.

Participants receiving a PD diagnosis in our sample were more likely to be women, white British, and from lower SES groups, consistent with the previous literature [24,27,39,40]. While PD diagnoses were more likely to be recorded for people of white British ethnicity accepted into EIP care for suspected psychosis, our study does not provide information on the relative prevalence of PD by ethnicity in the general population. People from BME backgrounds are over-represented in FEP samples, including ours[26], so other study designs are required to determine whether the incidence or prevalence of PD varies by ethnicity.

Lastly, in contrast to our original hypothesis, we did not find associations between indicators of neighbourhood-level deprivation or population density and PD in people presenting to EIP services, primarily for psychosis. Nonetheless, we have previously shown that participants in this sample as a whole are more likely to come from more deprived and densely-populated area than the general population [26,41]. Our data suggested that people presenting to EIP services with and without a PD diagnosis tended to come from similar areas in terms of deprivation and population density, although we found strong evidence that participants who received a PD diagnosis in our sample were even more disadvantaged in terms of individual-level SES than those without PD diagnosis. Together, these findings suggest that the PD group in our sample represents an extremely socially disadvantaged group.

Limitations

This study has some limitations. First, the original SEPEA study was designed to investigate the social epidemiology of psychotic disorders and not PD as a primary outcome; thus, as discussed above, we were reliant on PD diagnoses made during clinical practice which did not necessarily entail a structured instrument for diagnosing PD. Nonetheless, however, this may reflect the real-world assessment of PD in clinical practice in EIP services. Further, diagnosis of PD used in this analysis was made at 6 months after acceptance to EIP service or at discharge. Thus, it is not clear whether people with PD had premorbid PD before EIP or if they developed PD symptoms after acceptance to EIP.

Conclusions

In contrast to previous studies, we did not find high levels of PD in a large, prospectively-collected cohort of people accepted into EIP services in England for suspected psychosis. This suggests that these services may be largely appropriately screening and triaging referrals to divert people with primary diagnoses of PD. Nonetheless, participants receiving PD diagnoses in our sample were less likely to receive a validated, research-based diagnosis of psychotic disorder while in EIP care, despite similarities in their symptoms. This symptomatic overlap highlights that difficulties in diagnostic assessment and categorisation, which may delay onward referral and treatment of personality-related problems. While careful assessment of PD symptoms at referral may further help to signpost people to appropriate services, these problems only become apparent during longitudinal assessment in EIP care.

In an already socioeconomically disadvantaged EIP cohort, people with PD reported more severe individual-level social disadvantage in terms of relationship status and occupational position, although we found no evidence to suggest people with PD came from even more deprived or densely populated neighbourhoods at first referral than other groups referred to EIP care for suspected psychosis.