A systematic review and meta‐analysis of mental health service use in people who report psychotic experiences

Background Self‐reported psychotic experiences (PEs) are associated with psychopathology of all kinds, not just psychoses. However, systematic reviews on the relevance of this for health services are unavailable. Furthermore, whether association with service use is confounded by other psychopathology is unknown, and is relevant to prevention and treatment. Objectives Literature examining associations between PEs and service use was systematically reviewed. Study quality and the direction and extent of any associations were assessed, and meta‐analysis conducted. Methods Systematic review and meta‐analysis was carried out as per PRISMA guidelines. A search of electronic databases was performed based on free‐text and structured terms. Included studies were evaluated by two raters using a structured tool and estimates extracted for reporting. Results Thirteen studies were returned. We found two prospective studies, and a minority of studies accounted for concurrent psychopathology, limiting our ability to test our main hypotheses. Five studies reported associations by different types of service use. Almost all studies assessed service use by self‐report. Meta‐analysis suggested that people who reported PEs were around twice as likely to report service use compared to those who did not (pooled OR for all included studies: 2.20,95% confidence intervals (95%CI): 1.66,2.91). Conclusions There was consistent evidence of association between PEs and mental health service use at the general population level. However, evidence for causation was poor due to a limited number of studies. Whether increased service use in this group is solely attributable to PEs, and therefore whether interventions aimed at limiting/preventing PEs might be effective, requires studies focusing on the relationships between PEs, psychopathology and service use.

Nevertheless, it is unclear whether PEs themselves cause service use, or whether this relationship is confounded by concurrent psychopathology or suicidal thinking and behaviours. This has implications for treatment -if PEs are a cause of service use, then these symptoms should be the focus for intervention, however, if the association is confounded by other psychopathology then PEs should trigger assessment and treatment of concurrent psychopathological symptoms. This is the subject of the present systematic review.

| Aims and Objectives
The aim of this systematic review was to consolidate and summarise research findings on mental health service use of individuals reporting PEs.
The objectives were to: a. to summarise qualitatively, and quantitatively if possible, the association between PEs and service use, and b. to assess alternative explanations for any associations found, by evaluating whether studies attempted to account for confounding by non-psychotic psychopathology. A systematic search strategy was employed to select relevant studies for the review. Search terms referring to PEs and to service use for mental health problems were combined with terms identifying observational epidemiological studies. We included search terms used in previous reviews in the PE literature (Kaymaz et al., 2012;Kelleher et al., 2012b;Linscott & van Os, 2013a). Electronic databases (EMBASE, MedLINE, and PsychInfo) were searched since their inception for the relevant search terms, on January 1 st 2017. Search terms referring to PEs and to service use for mental health problems were combined with terms identifying observational epidemiological studies. The full list of search terms is presented in an Appendix S1, supporting information, to this paper. We also searched reference lists for three meta-analyses of PEs published since 2010 (Kaymaz et

| Selection Criteria
Both cross-sectional studies and follow-up studies were considered acceptable. We placed no restriction on year of publication. Candidate studies from the published scientific literature were selected using a combination of inclusion and exclusion criteria (see Table 1). Searches were limited to full text articles in English. Titles and abstracts were screened for all articles initially returned from the search terms.

| Data extraction
Studies meeting systematic review criteria were subject to data extraction using a form containing fields for study centre, type of geographic location, sample size, population details, duration of study, study design type, exposure, outcome measures, and results. In particular, data on the size and precision of any associations found, effects of adjustments, and stratification of results by type of service use were extracted from included studies. Citation lists for the included articles, and for the three pre-existing meta-analyses of PEs (Kaymaz et al., 2012;Kelleher et al., 2012b;Linscott & van Os, 2013b) were searched in order to identify further potential studies.

| Quality assessment
Evaluation of study quality on all included studies was done separately by two raters (V.B. and D.O.) using a risk of bias tool, the Newcastle Ottawa Scale (Wells et al., 2000) with information collected on case definition and its validity, representativeness of cases, definition and selection of controls, comparability of cases and controls (including control for confounding), ascertainment of exposure, and description of non-response. Results on this tool were used to derive a quality score for each study that was tabulated, along with details for each field. The primary summary measure of interest was the effect size comparing mental health service use in people reporting PEs to that in those not reporting PEs.
Although there were a small number of returned studies with considerable heterogeneity, meta-analysis was nevertheless attempted in order to provide a quantitative summary of the association of interest (see discussion section).

| Statistical analysis/synthesis of results
The principal summary measure for this systematic review was the odds ratio (OR). Data were entered into an electronic database and analysed quantitatively using Comprehensive Meta-Analysis Software, version 3 (Inc. B, 2014). This software allowed the meta-  nian & Laird, 1986). Given high probable heterogeneity among the studies, random effects meta-analytic models were used throughout.

| Risk of bias across studies
Risk of publication bias was evaluated by visual inspection of funnel plots and application of the Egger regression intercept method (Egger et al., 1997).

| Additional analyses
Analyses were repeated with each study removed in turn, to assess sensitivity of our inferences to the exclusion of different studies. The classic fail-safe N method (Rothstein et al., 2006) was used to estimate the number of unpublished studies that would be required in order bring the meta-analytic p value for the association to a level greater than the pre-defined alpha level of 0.05.

| Retrieved studies
The flow of articles through the study, including numbers of articles screened, assessed for eligibility and included in the review, is summarised in the diagram displayed in Figure 1. For each included study, characteristics for which data were extracted, including study, follow-up period, and exposure/outcome assessment, are summarised in Table 2, together with information on risk of bias and summary results.

| Description of included studies
Most studies did not report non-response proportion, or compare responders to the target population. All studies were conducted in highincome countries on mainly urban populations. Measures of PEs were in general more carefully validated than measures of mental health service use. Most studies were large, with numbers of participants around 1000; the smallest sample size was 103 (Daalman et al., 2016), and the largest was 11,937 (Barragán et al., 2016). Many studies used a version of the CIDI (Barragán et al., 2016;DeVylder et al., 2014;Gale et al., 2011;Smeets et al., 2013;van Nierop et al., 2011) to ascertain both PEs and mental health service use. One study employed a screening instrument for psychosis (Murphy et al., 2010), and one negative study employed a tool to ascertain self-reported symptoms of prodromal psychosis (Kobayashi et al., 2011). Nearly all studies measured use of mental health services by self-report questionnaire. No studies reported costs associated with PEs or directly assessed routine clinical data on service usage, although one study employed hospitalization registers (Werbeloff et al., 2012). One study of mental health service use in people with PEs adopted a case control design, comparing general psychiatric outpatients with a control group (Kobayashi et al., 2011). Four studies (Armando et al., 2010;DeVylder et al., 2014;Murphy et al., 2010;Nishida et al., 2008) adjusted for comorbid depression and anxiety. Two follow-up studies were included, of which one was based only on a hospitalization outcome (Werbeloff et al., 2012), and another on a total sample size of 163 subjects. Five studies reported effect sizes for different types of service use (Barragán et al., 2016;Gale et al., 2011;Murphy et al., 2010;Olfson et al., 2002;Saha et al., 2013).

| Qualitative findings for service use in people with self-reported PEs
Quality assessment agreed between both raters. All studies that found negative (Kobayashi et al., 2011), borderline (Daalman et al., 2016) or mixed (Armando et al., 2012;Murphy et al., 2010) findings for the association between PEs and mental health service use adjusted or accounted for concurrent affective psychopathology.
Among the studies with higher quality ratings, DeVylder et al. (2014), (8 quality points) found associations (odds ratios) of greater than 2 between PEs and self-reported mental health service use in both the lifetime and recently, after taking account of concurrent affective symptoms and substance use. In contrast, Murphy et al. (2010) (9 quality points) reported that only paranoid experiences were associated with using mental health care after all adjustments were made -estimates for mania, thought interferences, strange experiences and hallucinations were no longer sufficiently precise. Kobayashi et al.
(2011) (9 quality points), using a clinical definition of service use (comparing clinic attenders with non-attenders), found no good statistical evidence for an association between service use and a range of different PE scales, after adjustment for depression ratings.
All studies that did not find association between PE and service use also did not make adjustments for concurrent psychopathology.
Although Gale et al. (2011) found statistical evidence for a difference  (Continues)   Based on 13 studies. Unless otherwise stated, included studies employed self-report measures of service use, and used appropriate statistical testing and presentation of results.
in the prevalence of mental health service use among people with PEs compared to those without, they did not report this in the form of an association -for the purposes of this review, we calculated odds ratios based on reported post-estimation predicted percentages for service use for subjects with one, two or three PEs, compared to subjects with no PEs (reported in Table 1). Smeets et al. (2013) found that mental heath service was associated with both hallucinations and delusions in a random sample of Dutch adults. In another Dutch study, van Nierop et al. (2011) report that people with PEs but without psychotic disorder were nearly twice as likely to report service use for mental health problems. Among the prospective studies, Daalman et al. (2016) found that nearly 40% of people with hallucinations had used services at five year follow up, compared to 12% in a control group, and Werbeloff (2012) reported increased odds of any psychiatric hospitalization in people with psychotic symptoms based on analyses of Israeli hospitalization registry data. Nishida (2008) found that a greater proportion of Japanese students (at grades 7, 8, and 9) who reported PEs were in contact with medical services compared to those who did not.

| Meta-analysis
The overall meta-analytic estimate, based on adjusted estimates where available, for the association between PEs and service use from random effects analysis was 2.2 (95% confidence interval 1.66,2.91), and based on studies on a total of 73065 subjects.

| Summary of evidence
To our knowledge, this is the first systematic review addressing the question of whether people reporting PEs use more health services.
We found a small number of studies with relatively similar designs, assessing the same underlying construct in different groups of people  with PEs, with limited adjustment for alternative explanations for service use, and an emphasis on self-reported measures of service use.
The overall meta-analytic estimate of the odds ratio, comparing subjects with PEs to those without, was approximately 2, with statistical evidence indicating this was unlikely to be due to chance. Our search returned only two studies that ascertained service use prospectively, one of which only measured hospitalization, and the other had the smallest sample size of all included studies (n = 163). Measurement of PEs was generally better reported than the measurement of service use. Five studies accounted for co-morbid psychopathology in the association between PEs and mental health service use, finding attenuation of the association when this was done. All studies measured mental health service use through self-report, with the exception of one case control study which used psychiatric outpatient status as a case definition, and a study based on a hospitalization register.

| Limitations of systematic review
Our systematic review had some limitations. We only searched for full text peer-reviewed articles in English language journals. Health service use is a broad construct, and we may have missed studies meeting our definition because they used slightly different wordings for referring to mental health service use. We were unable to carry out meta-regression analyses with adequate power to detect the influence of moderating factors, for example the number of PEs reported, the type of PEs, or the type of service use. Included studies were limited to research from high-income countries, and the structure and availability of health services is highly culturally variant.
Although our analysis was based on a large pooled sample size, nonresponse was generally poorly described in individual studies.
Meta-analysis found consistency in the direction of association across studies and statistical evidence for an association between PEs and mental health service use. However statistical heterogeneity was estimated to be high (I 2 = 95.21), reflecting probable diversity in both the subjects, symptoms, and service use outcomes studies, and in the way studies were designed (Higgins & Green, 2011). Based on the scrutiny of meta-analytic estimates with one study removed, heterogeneity did not appear to be related to the inclusion/exclusion of any one particular study in this review. Moreover, only a minority of studies adjusted for confounders or reported sample non-response in detail, leading to a risk of bias. All studies that adjusted for concurrent psychopathology showed weaker associations and weaker evidence for an association than studies that did not adjust for concurrent psychopathology. Despite high heterogeneity, there was little evidence of publication bias based on inspection of the funnel plot ( Figure 3). Finally, there were only two prospective studies included in this review, limiting inferences about the direction of the association between PEs and use of mental health care.

| Explanations
Although people with PEs may be more likely to use, or have used, mental health services compared to the general population, the factors which result in help-seeking in this group remain to be determined. The findings of this review could be due to a direct effect of PEs on service use, or due to confounding, that is, PEs and service use could both result from other factors. For example, our results suggest that concurrent psychopathology, such as anxiety and depression, could play a key role in determining service use, although possibly with an independent role for PEs themselves. However, all studies that adjusted for concurrent psychopathology reported attenuated associations with mental health service use. Thus, our results are also consistent with a residual confounding explanation, that is, that improved measurement of concurrent psychopathology in people with PEs in these studies might have removed the association completely, and therefore that the associations with service use demonstrated in these studies might be accounted for by depression and other psychopathology rather than attributable to PEs in themselves. The weaker estimates found in studies that adjusted for concurrent affective symptoms are also consistent with mediation of the effect of PEs on service use by affective symptoms. Based on the findings of this review, more studies with longitudinal designs are needed to advance this question. Psychotic experiences are strongly related to suicidal thinking or behaviour (Kelleher et al., 2013;Kelleher et al., 2014a), and this relationship might also explain increased use of healthcare in people with PEs.
Although evaluation of service use is a complex concept (Gladstone et al., 2007), measurement of the details of use of services in these studies was generally very limited and varied between studies.
There is evidence that self-reported service use is liable to recall bias, and other forms of systematic measurement error (Rhodes & Fung, 2004;Rhodes et al., 2002).
Very few if any studies measured service use following the onset or demonstration of PEs. This is important because if PEs are to be a focus for primary prevention, then we must discern whether PEs are really an early marker for later service use, or occur concurrently with other psychopathology once diagnoses have already been made/help has been sought. Statistical adjustment is only one, albeit easily performed approach to accounting for the effects of other psychopathology on the measured effect-other approaches would be to match participants on the concurrent psychopathology, or restrict the sample to subjects without any other mental health problem.

| Implications
There is now consensus that although the majority of people with PEs remain well, they are a general indicator of diagnosable psychopathology (McGorry et al., 2008;Yung et al., 2008). Indeed, research from a range of centres now indicates that PEs are a marker for a broad range of adverse outcomes, ranging from poor mental health (Saha et al., 2011a), physical health (Saha et al., 2011b), social and occupational functioning (Kelleher et al., 2014b), drug and alcohol use problems (Saha et al., 2011c), trauma (Morgan et al., 2014), social disadvantage (Morgan et al., 2014), and social isolation (Saha et al., 2012). The present study demonstrates the possible relevance of these impairments for the planning of mental health services, and the need for well designed studies on this topic. The work summarised here clearly indicates that PEs could be relevant for the indexing of a range of mental health problems, not only psychosis risk (Van Os et al., 2009). Moreover, PEs could be a reliable indicator of later need for health care, even if they are not causally related to that need; screening for PEs in primary care might be of value in facilitating prompt intervention in a sub-group at elevated risk of requiring secondary mental health care.
The absolute increase in psychosis risk attributable to psychotic symptoms is low. For example, in the meta-analysis by Kaymaz et al. (2012), the incidence rate of psychosis in those with ascertained PEs was 558 per 100 000 person years of follow up (0.6% per year), compared to 159 cases per 100 000 person-years in those without (0.2% per year).
Transitions to psychosis are much higher in enriched clinical samples, for example a meta-analysis of transition occurrence in ultra-high risk samples suggested an overall transition risk of 36% at three years follow up (Fusar-Poli et al., 2012). This lends support to the ultra-high risk approach, but also is consistent with a model where PEs represent a general marker for more severe psychopathology. The present review has demonstrated a consistent, but methodologically weak association with health service use in this population, and suggests that PEs may mark a considerable unmet mental health need in the general population.

| Future research
Many individuals who later develop psychosis experience a long duration of untreated symptoms, and not all who experience prodromal symptoms seek help (O'Callaghan et al., 2010;Wang et al., 2005).
Aside from personal barriers to access to care, such as concerns about privacy, and financial/time constraints (Eisenberg et al., 2007;Hunt & Eisenberg, 2010), there are likely to be important symptomatic and demographic predictors of service use in people with PEs. Investigating these associations requires accurate measures of mental health service use, across a broad range of settings. For example, recall bias in the evaluation of recent service use might be limited by shorter time periods of recall, or triangulation of self-report information with linked electronic health records from care providers (Stewart et al., 2009).
Depressive symptoms are common in people with PEs, and are associated with distress and need for care (Yung et al., 2007). If the association between PEs and service use is accounted for by concurrent psychopathology, and is not really causal, then intervening on PEs will not reduce use of services in people with PEs. Therefore, more detailed assessment of the role played by depressive and other mental health symptoms on use of mental health services is necessary in order to build good evidence for early intervention. In particular, PEs can be accompanied by a broad range of symptoms beyond only depressive symptoms (Kelleher et al., 2014a), suggesting that proper understanding of the relationship between PEs and use of healthcare will require measurement of a fuller range of the psychopathology that might be present with PEs.
Finally, assessing the role that PEs might play in prevention of severe mental disorders requires clarity on the temporal relationship between PEs, the existence of diagnosable mental disorder, and pathways into mental health service use. This is particularly important in light of the probably transitory nature of most PEs (Chapman et al., 1994;Hanssen et al., 2005).

| Conclusions
This review identified some PE research suggesting that people who report PEs also report more service use, and some studies report that this is true when other forms of psychopathology are adjusted for. However, in order to establish that PEs lead to service uses, irrespective of diagnosis, better studies are required. In particular studies are needed which do not rely on the self report of service use, and which are sensitive to the timing of service use and of the onset of PEs .

VB was supported by a Wellcome Trust Clinical Research Training
Fellowship (101681/Z/13/Z).