Main findings

This paper is one of the few to report on the cohort of individuals referred to an EIP service, but not meeting criteria for either an ARMS or FEP. Our results demonstrate that this group constitute a large proportion of the total referrals to an EIP service, substantially larger than those meeting ARMS criteria.

The ‘not meeting criteria’ cohort was highly enriched for non-psychotic Axis I disorders, with approximately 65% meeting criteria for a least one disorder, with mood, anxiety and substance use disorders predominating. This finding is consistent with a previous study in this service over a more limited timeframe.^{Reference O'Donoghue, Lyne, Renwick, Madigan, Kinsella and Clarke28} A further third (36.4%) had at least a secondary diagnosis. A population-based study exploring longitudinal risk demonstrated that prior psychopathology accounted for 85.5% of the outcome of developing clinical psychosis. Mood disorders have the highest population attributable fraction, which was the most common diagnostic group in our cohort. Although the relative risk for psychosis with an ARMS is high, because of its lower prevalence in the general population compared with more common mood, anxiety and substance use disorders, its population attributable fraction is far lower.^{Reference Guloksuz, Pries, Ten Have, de Graaf, van Dorsselaer and Klingenberg20} Non-psychotic pathology has also been reported as high in ARMS groups,^{Reference Albert, Tomassi, Maina and Tosato29,Reference Fusar-Poli, Nelson, Valmaggia, Yung and McGuire30} which may contribute substantially to the conversion risk in this group.^{Reference Hasmi, Pries, Ten Have, de Graaf, van Dorsselaer and Bak31} Our findings demonstrate that highly enriched Axis I morbidity and multimorbidity is a finding that is not specific to ARMS groups among EIP referrals.

The socio-occupational functioning of the ‘not meeting criteria’ cohort was examined across three variables. Although a greater proportion were not in a relationship, this association dissipated when controlled for demographic factors, likely reflecting their younger age. They were broadly similar to the FEP cohort across other metrics, with the exception of a greater proportion in long-term unemployment (adjusted odds ratio 2.83, 95% CI 1.65–4.85). The broad similarity to an FEP cohort, a group with established socio-occupational impairments,^{Reference Green, Kern, Braff and Mintz32} is a reflection of the functional morbidity in this group. Impaired social and role functioning have predictive capacity for psychosis, both within ARMS^{Reference Cornblatt, Carrión, Addington, Seidman, Walker and Cannon33} and general population^{Reference Cannon, Caspi, Moffitt, Harrington, Taylor and Murray23} samples. The significance of functional impairment as a risk marker is not only its capacity to define at-risk individuals, but also its presence as the earliest herald symptoms in the pre-illness phase of individuals with psychosis.^{Reference Cannon, Jones, Huttunen, Tanskanen, Huttunen and Rabe-Hesketh34} Risk stratification at this point offers the potential for intervention at the earliest possible stage.

Despite the fact that our FEP cohort had a majority of males, our ‘not meeting criteria’ cohort had a greater proportion again (odds ratio 1.12, 95% CI 1.03–1.71). Although classically, schizophrenia-like psychosis was thought to have a higher incidence in men, this relationship moderates when incidence is considered over the lifespan and psychotic syndromes considered are broadened. However, an age×gender interaction has been demonstrated, reflecting that before 45 years of age, overall risk rates are elevated among men.^{Reference Kirkbride, Errazuriz, Croudace, Morgan, Jackson and Boydell35} This renders our finding of the ‘not meeting criteria’ group as predominately both male and younger as significant. Clinical and functional outcomes in psychosis are also worse in men.^{Reference Barajas, Ochoa, Obiols and Lalucat-Jo36}

Ethnicity is a well-established risk factor for psychosis, with differential risk between groups.^{Reference Kirkbride, Errazuriz, Croudace, Morgan, Jackson and Boydell35} Our catchment area has previously been reported as having a marginally higher than average migrant population,^{Reference O'Donoghue, Lyne, Renwick, Lane, Madigan and Staines37} in line with that reported for this FEP cohort. When all ‘non-Irish’ nationalities were collapsed into a single category, the ‘not meeting criteria’ group had a lower proportion of non-Irish nationalities (odds ratio 0.85, 95% CI 0.77–0.91), indicating that this variable is unlikely to be a source of elevated risk. A family history of a psychotic disorder is a further well-established risk factor.^{Reference Chou, Kuo, Huang, Grainge, Valdes and See38} There was a similar proportion of both cohorts with a first- or second-degree relative with a psychotic disorder, indicating that this risk factor is equivalent between cohorts.

Finally, our study found that of those initially not meeting criteria, 41 were re-referred to the service, with 21 subsequently meeting criteria for psychosis (3.8% of all those ‘not meeting criteria’). This figure should be interpreted with caution and it is not synonymous with a conversion to psychosis rate in the sample. It is likely to be an underestimate of the true conversion rate of this cohort. Reasons for non-detection by our service of individuals who developed psychosis subsequent to assessment include being lost to care (e.g. disengaging with services or moving out of the catchment area) or treating secondary care teams not re-referring to the service (e.g. if an individual was re-referred by primary care, but the secondary care team deemed them to no longer be in their FEP). The variable duration from assessment to the end of the study period may also have been an inadequate interval for recently enrolled individuals to develop psychosis. One of the few studies to report on conversion to psychosis rates among an entire cohort of EIP service presenters found that over a median follow-up period of 7.3 years, the conversion to psychosis rate was comparable between ARMS (17.3%) and non-psychotic help-seekers (14.6%). Non-psychotic help-seekers with a previous admission had a higher rates of psychosis post-assessment than the ARMS group.^{Reference Conrad, Lewin, Sly, Schall, Halpin and Hunter39} The reported service differed from ours in several important ways, accepting referrals for adolescents aged 12–25 years and employing a pre-screening interview. To establish a definitive figure for psychosis risk in our cohort, a retrospective cohort study with longitudinal follow-up should be undertaken.

Our age criteria of 18–65 years mirrors the configuration of secondary mental health services in Ireland, and is aligned with adult service age criteria. This differs from some EIP services that focus on an adolescent and young adult population, often in the 14- to 35-year age range. In the context of the lower prevalence of psychotic disorders in the <18 year age group, the issues we raise previously, of both the limited predictive ability and capacity of current ARMS models, have been highlighted as particularly relevant in this group.^{Reference Lång, Yates, Leacy, Clarke, McNicholas and Cannon40} As we anticipate that those ‘not meeting criteria’ would constitute a larger proportion of referrals within adolescent EIP service referrals, it would be instructive to repeat our study methodology in such a service.

EIP services have several functions distinct from identifying psychosis risk, including delivering phase-specific, specialist interventions for at-threshold psychotic disorders. In this regard, the high volume of individuals identified by clinical teams as experiencing an FEP and appropriately referred for EIP interventions, remains an important and essential contribution of the service.

In a similar vein, where risk prediction is implemented, it should be for the purpose of delivering evidence-based, risk-modifying interventions. The development of risk prediction models is driven by broad acknowledgement that psychosis has developmental and symptomatic antecedents,^{Reference Insel41} and that intervening early may engender the best clinical outcomes. Such enquires are catalysed by outcomes for psychotic disorders, which have largely plateaued in recent decades.^{Reference Clarke, McDonough, Doyle and Waddington42} There is emerging evidence that screening in children and adolescents, when based on a combination of symptomatic and functional markers (distinct from either in isolation), has acceptable predictive power for later psychopathology, and that there are some promising risk-modifying interventions within this interval.^{Reference Healy and Cannon43} In the adult population, the evidence is less clear. Currently there are no definitive data supporting any risk-modifying intervention in adults,^{Reference Fusar-Poli, Davies, Solmi, Brondino, De Micheli and Kotlicka-Antczak44} nor do antipsychotics appear to have a role in modifying transition rates in ARMS samples.^{Reference Zhang, Xu, Tang, Wei, Hu and Hu45} Such studies may be challenged by the same issues discussed previously of high heterogenicity within high-risk samples, uncontrolled risk enrichment and uncertain representativeness of current ARMS cohorts for incident FEP cases.

At such a juncture, clinical concerns about prediction beget ethical ones. Models that induct individuals into a surveillance pathway, where a majority (as currently) will be ‘false positives’, raise important issues about stigmatisation and iatrogenic harm.^{Reference Ajnakina, David and Murray5} Risk prediction models are currently not accurate enough for clinical use, and the risk of adverse effects for those who screen false positive should not be underestimated. These questions are not peripheral to, but inextricably linked with, the performance of such models, as there are careful balances to be made between potential benefit and harm. Continuing research efforts need to address each aspect of risk prediction not in isolation, but in tandem, including discriminative ability, calibration and, ultimately, clinical utility.^{Reference Seyedsalehi and Lennox12}

Study strengths

This study includes complete referral data from a long-established EIP service, embedded in a public mental health service that provides universal access to the catchment area population. This allowed the examination of case trends over a broad timeframe, resulting in a large sample size.

Our service provides a standardised diagnostic assessment, using a recognised instrument (SCID), to all individuals referred. This allows for the reporting of those meeting standardised criteria for many psychotic and non-psychotic disorders. Standardised assessment templates with data on functional domains facilitated a transdiagnostic comparison between cohorts.

Finally, since the inception of the service, an electronic archive of assessment documents has been maintained, resulting in a high data retrieval rate.

Study limitations

As EIP services vary in case criteria and service configuration, caution should be exercised in generalising findings across health service models. Our service functions as a tertiary referral service and does not accept direct referrals from primary care or self-referrals. In services where this is the case, it might be expected to increase the proportion of those ‘not meeting criteria’, with a likely dilutional effect on risk factor enrichment.

Although our study provided a comprehensive description of DSM Axis I disorders, we do not assess for other conditions with standardised assessment instruments, including Axis II disorders (including personality disorders) or neurodevelopmental disorders (including attention-deficit hyperactivity disorder or autism spectrum disorder). It is possible that some of the individuals who did not meet criteria for an Axis I disorder would meet criteria for one of these disorders, and for those who did meet criteria for a disorder, it may further describe their comorbidities.

Finally, as above, an accurate figure for conversion to psychosis should be established for this group, with longitudinal follow-up. Although a portion of those developing psychosis will have been re-referred, several factors affecting service engagement limit the utility of this figure.

Implications and future research

In conclusion, to our knowledge, this is largest study to report on demographic, diagnostic and functional variables of individuals referred to, but not meeting criteria for, an EIP service. We demonstrated that this cohort is enriched for many risk factors for psychosis. Enhanced models of psychosis risk derived from current ARMS criteria may succeed in enhancing predictive capacity within selected groups, but will not address the ‘prevention paradox’ of only a limited portion of individuals entering psychosis via this route. This study is a response to calls to reconsider the current ARMS model as a starting point for defining at-risk groups, to build models that will be useful at the population level. Those referred to EIP service, but who do not meet current criteria, may be an important group for further study in this expanded search.