Specific adolescent prodromal symptoms associated with onset of psychosis in the Northern Finland Birth Cohort 1986

Several psychological symptoms in adolescence associate with later development of psychosis. However, it is unclear which symptoms specifically predict psychotic disorders rather than psychiatric disorders in general. We conducted a prospective study comparing how specific adolescent psychotic‐like symptoms, predicted psychotic and non‐psychotic hospital‐treated psychiatric disorders in the population‐based Northern Finland Birth Cohort 1986 (NFBC1986).


| INTRODUCTION
Schizophrenia and other psychotic disorders are one of the most severe causes of worldwide burden of disease (GBD, 2017). At least some psychoses are considered neurodevelopmental disorders which overtly manifest in adolescence and early adulthood (Bearden & Forsyth, 2018;Kauppi et al., 2015;Murray et al., 2017). Both positive prodromal symptoms (Lindgren et al., 2021;Tso et al., 2017) and negative features (Devoe et al., 2018) have been found to precede psychoses in clinical samples , though the predictiveness of existing criteria among adolescents is questionable (Lång et al., 2021).
Prodromal symptoms of psychosis have been studied retrospectively in large population samples and in genetic high-risk cohorts, where they are found to precede the transition to a psychotic disorder by several years (Oliver et al., 2020). Also, in a prospective general population sample with 3000 adolescents from Germany, negative and disorganized features predicted the onset of first-episode psychosis (Dominguez et al., 2010). Furthermore, in the Dunedin Study birth cohort with originally about 1000 participants, self-reported psychotic symptoms at age 11, both positive and negative, were predictive of schizophreniform psychosis at age 26 (Poulton et al., 2000). In a later analysis of the same sample, however, psychotic-like symptoms in childhood did not specifically predict schizophrenia but were associated with general mental health problems in adulthood (Fisher et al., 2013).
It is thus unclear whether psychotic-like symptoms are relatively specific precursors of later psychoses, or general markers of vulnerability to mental health problems. In the present study, putative prodromal symptoms of psychosis in adolescence were therefore studied as predictors of first onset of hospital-treated psychosis in the general-population-based Northern Finland Birth Cohort 1986, using non-psychotic psychiatric patients as controls, and in this way differentiating between symptoms specifically predictive of psychotic disorders and symptoms generally predictive of psychiatric disorders (Mäki et al., 2014).

| PROD-screen
The PROD-screen is a questionnaire developed for inquiring about psychotic-like symptoms possibly indicating heightened risk for psychosis (Heinimaa et al., 2003). The self-report form has 21 items which are derived from previous instruments, such as the Interview for the Retrospective Assessment of the Onset of Schizophrenia (IRAOS; Häfner et al., 1992), the Bonn Scale for the Assessment of Basic Symptoms (BSABS; Gross et al., 1987) Yung et al., 1996). Response alternatives are Yes or No. In the present study, the PROD-screen addressed self-reported symptoms reported by the adolescent (aged 15-16 years) cohort members within the last 6 months. Every item of the PROD-screen was considered as a separate risk indicator for later first-episode psychosis (Table 3).

| Outcome variables
All hospital treated psychiatric diagnoses of the members of NFBC1986 appearing in the CRHC between the years 2003 and 2016 for any psychiatric disorder treatment as an inpatient were identified by record linkage using personal identification codes. Outcome variables of psychotic disorders and non-psychotic psychiatric disorders included the whole follow-up period from baseline 2003 to the end of follow-up 2016. The cumulative incidence of first-episode psychosis by gender is described in Figure 1. The nationwide CRHC covers all mental and general hospitals, as well as beds in local health centres and private hospitals nationwide. In this study, psychotic disorders included all non-organic psychoses; that is, schizophrenia (ICD-10 code F20), schizoaffective psychosis (F25), affective psychosis (F30.2, We grouped participants into the exclusive groups (1) any psychotic disorder, regardless of presence of non-psychotic psychiatric disorders, (2) any non-psychotic psychiatric disorder, (3) no psychiatric disorder. Lifetime CRHC data were also used to remove participants with any previous psychiatric disorders from the follow-up analyses.

| Confounding factors
Gender was used as a confounding factor in all analyses. Any psychiatric disorder in either parent according to CRHC data dating back to 1971 (yes vs. no) was also considered a confounding factor, as was parental educational level (highest parental education; basic at <10 years, secondary at 10-12 years, and tertiary at >12 years) at the time of the field study in adolescence, as reported by the parents.
Family structure (living with two parents vs. other) and cannabis use (never vs. ever) in adolescence were also considered confounding factors in the analyses and were obtained from the self-report questionnaire containing the PROD-screen. All these variables chosen as confounders are known to be risk factors for psychosis (Radua et al., 2018), and their distributions by group are presented in Table 1.

| Statistical analysis
To analyse the PROD-screen responses in aggregate we conducted a factor analysis using the mirt package (v. 1.34; Chalmers, 2012) in R (v. 4.1.1; R Core Team, 2021) with EM extraction and standard settings. The latent factor model included item threshold parameters (item type '2PL'), to obviate artefacts from widely varying item endorsement rates. Items were assigned to the three factors identified in our previous study  by their greatest loading in that explorative finding as follows: positive symptoms (11 items), negative symptoms (4 items), and General symptoms (5 items). One item (feeling euphoric or especially competent and important) was excluded from the subdimensions because it was not sufficiently associated with any factor. Factor scores were estimated with the expected a posteriori method. To create dichotomous predictors corresponding to individual symptoms, factor scores were dichotomized with cut-offs set to produce indicator frequencies corresponding to the mean endorsement rate of the items on that factor. For example, the cut-off for the negative factor was set to classify 8.8% of the respondents as being at higher risk, as this was the average rate of Yes responses to the four negative items.
Our main analysis was to compare the predictiveness of each PROD-screen symptom and dichotomized factor scores for psychosis and other psychiatric disorders using logistic regression, reporting odds ratios (OR) with 95% confidence intervals (CI). Models were adjusted for gender, parental psychiatric disorder, family structure, parental educational level, and the adolescent's cannabis use. In the analysis of individual symptoms, we considered p values below .05 to be statistically significant, and corrected for multiple testing with the Bonferroni method.
In addition, we also compared individual symptoms and factor scores of the PROD-screen between the psychosis and no disorder groups with the same modelling as in the main analysis. We did not consider statistically significant findings between psychotic and non-psychotic disorder groups notable if there was no difference between psychotic and no disorder groups. Items used for screening psychosis should differentiate psychotic and normal population.

| Attrition analysis
We describe the flow chart of the study and attrition in Figure 2. In the year 1986 the NFBC1986 was launched, when the pregnant mothers of the cohort members were first contacted. The adolescent field study of the NFBC1986 was conducted at the age of 15-16 years including the PROD-screen. Of the participants, 1466 did not consent to the field study and 1126 responded to less than 18 of the PROD-screen items. Of the original NFBC 1986 cohort (N = 9479), 6638 participants (70.0%) responded to the PROD-screen at the 15-16-year follow-up. In the current study, the NFBC1986 participants who had not taken part in the follow-up (N = 1126), denied F I G U R E 1 Cumulative incidences of firstepisode psychosis between ages 17 and 32 (years 2003-2016) in the Northern Finland 1986 Birth Cohort, by gender. PROD-screen questionnaire was filled in at age 15-16 (2001-2002), and the follow up of psychotic disorders started at 2003 (at age 17 years) the disclosure of information (N = 2), answered less than 18 items on

| RESULTS
There were no statistical differences in sociodemographic factors between the psychiatric disorder groups (Table 1). When comparing psychotic psychiatric disorders to no disorder group, psychotic patients had used cannabis more often than no disorder group (χ 2 = 5.94, p = .015). When comparing separate PROD-screen symptoms among those having had psychiatric hospital treatments, we found two positive symptoms and one negative symptom specifically associated with psychotic disorders in comparison to non-psychotic mental disorders (Table 3). The positive symptoms were 'Difficulty in controlling one's speech, behaviour or facial expression while communicating' (adjusted OR 4.00, 95% CI 1.66-9.92) and 'Difficulties in understanding written text or heard speech' (adjusted OR 2.25, 95% CI 1.12-4.51). The negative symptom was 'Difficulty or uncertainty in making contact with other people' (adjusted OR 2.20, 95% CI 1.03-4.67). Of these three symptoms, one positive symptom ('Difficulty in controlling one's speech, behaviour or facial expression while communicating') remained statistically significant after Bonferroni correction for multiple comparisons.
One general symptom in the PROD-screen had an OR <1.00 ('Worrying, nervousness or anxiety'). One symptom ('Feeling euphoric or especially competent and important') had statistically significant association with psychosis onset, but it did not manage to differentiate psychosis prone participants from general population without hospital treated mental disorders. Levels of positive, negative, and general PROD-screen symptoms did not differ in the psychotic disorder group compared to the nonpsychotic psychiatric disorder group (Table 4).
We also compared psychosis to no disorder in an additional, exploratory analysis (Tables S1 and S3).We found five positive and three negative symptoms that predicted later hospital-treated psychotic disorder when comparing to no disorder group (Table S2)

| DISCUSSION
In this general-population-based prospective birth cohort study, we found that three symptoms of the PROD-screen when reported in adolescence specifically predicted the subsequent onset of psychosis and one of them remained statistically significant after the very conservative Bonferroni correction.
The symptoms that predicted psychosis in our study were present in the BSABS (Difficulty in controlling one's speech, behaviour or facial expression while communicating), both the SIPS and the BSABS (Difficulties in understanding written text or heard speech) and in the SIPS, the IRAOS, and the BSABS (Difficulty or uncertainty in making contact with other people), respectively (Heinimaa et al., 2003). These symptoms are all related to social interaction and communication.
These findings are in line with other prospective general population follow-up studies (Dominguez et al., 2010;Poulton et al., 2000).
In the Dunedin Study birth cohort, self-reported psychotic symptoms-both positive and negative-at age 11 predicted a very high risk of a schizophreniform psychosis diagnosed by structured interviews at age 26 years (Poulton et al., 2000), though a later analysis showed that this association was not specific. When comparing our study to the Dunedin study, we had a different setting. Outcome variables were based on an interview in the Dunedin study while ours were clinical diagnoses. The Dunedin study's control group included no disorders, and our study included specifically non-psychotic psychiatric disorders as a control group. Another longitudinal prospective study was conducted among adolescents and young adults from Munich aged 14-24 years. There, negative symptoms predicted positive features and were associated with later psychotic disorder (Dominguez et al., 2010).
As psychotic disorders are relatively rare, it has been emphasized that it may be impossible to predict development of psychosis in the general population (Lee et al., 2018). It is also uncertain whether early-stage treatment of prodromal symptoms reduces the  Note: Sensitivity to detect psychosis prospectively was 41% for any positive psychotic-like symptoms, with a specificity of 72%, and a positive predictive value (PPV) of 2%. Negative symptoms sensitivity 55%, specificity 76%, and PPV 3%. General symptoms sensitivity 51%, specificity 55%, and PPV 1%. a Adjusted for gender, any parental psychiatric disorder before 2003 (yes vs. no), family structure at age 16 (living with one vs. two parents), parental educational level, own cannabis use in adolescence (never vs. ever).
included ( Table 2), use of valid instruments to measure symptoms in adolescence, using the comprehensive nationwide CRHC, use of a comparison group of hospital treated non-psychotic cases, and including several confounding factors. We had an opportunity to test the symptoms of the PROD-screen in adolescence and their association with first-episode psychosis in a general population of young people over a long time, taking symptom specificity to psychosis into account. Previous studies have mainly concentrated on psychiatric outpatients and help-seeking individuals with a large variation in age (Daneault et al., 2013;Lee et al., 2018) and no previous study has examined the predictive capacity of conversions to psychosis using the PROD-screen.
The number of the youth returning the questionnaires could be considered sufficient to represent the general youth population of Northern Finland, as the cohort follow-up included 6634 participants who had filled in the PROD-screen.

| Limitations
Our register follow-up only included hospital-treated mental disorders, and we excluded those at-risk individuals who had previous substance abuse and other mental illnesses. Because psychiatric disorders requiring hospitalization are relatively rare, the numbers of adverse outcomes remained quite small despite the large original cohort population. The follow-up period was from 2003 to 2016. Some additional psychotic disorders will emerge over the coming years, which may alter symptom predictiveness, though predictiveness is likely to drop over time.
The PROD-screen was used as a mailed self-report questionnaire.
The participants answered these statements at home. It is therefore possible that some of the statements might have been misconstrued.
As Poulton et al. (2000) noted, even structured interviews have been criticized for detecting not only clinical psychosis but also false positive cases (Anthony et al., 1985;Helzer et al., 1985;Kendler et al., 1996), and self-report questionnaires do not allow for follow-up clarifications.
Furthermore, even though the CRHC has been found to be quite reliable in detecting psychoses (Isohanni et al., 1997;Mäkikyrö et al., 1998), for non-psychotic mental disorders it is a rough measure.
For this reason, we are unsure whether the results of our study may be generalized to non-psychotic mental disorders that do not require hospitalization. Our study did not include outpatient patients because we wanted to have more severe cases and better reliability of diagnoses (Isohanni et al., 1997;Mäkikyrö et al., 1998). Disorders treated in outpatient settings are generally less severe than hospital-treated, but only minority of psychotic disorders are missed by using hospitalization data only (Perälä et al., 2007). If outpatient diagnoses would have been included in this study, some more individuals from the no disorder group would have moved to the non-psychotic psychiatric disorder group.
One of the limitations, as with many longitudinal studies, is the possibility of selection bias caused by the substantial attrition over time (Zammit et al., 2013). In the present study, two thirds of the original birth cohort participated in the 15-16-year field study. In the Northern Finland Birth Cohort 1966 study, those with any psychiatric disorder participated less actively than those without psychiatric disorder (Haapea et al., 2008). However, there was minimal attrition in the follow-up as we used national register data.
The PROD-screen includes questions from the SIPS/SOPS, IRAOS, and BSABS structured interviews (Heinimaa et al., 2003). The included putative psychosis-risk symptoms are common in the general young population, and the Yes/No-response format is relatively noninformative, resulting in a low positive predictive value for the PRODscreen, as it has a limited number of symptoms. This hampers the clinical applicability of the finding; therefore, it is necessary to employ more specific questionnaires and interviewing methods for the early detection of psychosis, and/or combine questionnaires with other measures, such as biomarkers, in the future Therman et al., 2011).

| Conclusions and clinical implications
To our knowledge, this is the first general-population-based prospective study of adolescents exploring psychiatric symptoms predicting specifically the onset of hospital-treated first episode psychosis in comparison to non-psychotic disorders. We found three symptoms related with difficulties in social interaction which predicted onset of psychosis. Of these three symptoms, one positive symptom ('Difficulty in controlling one's speech, behaviour or facial expression while communicating') remained statistically significant after Bonferroni correction for multiple comparisons. This is a novel finding and should be replicated.
Psychosis-risk symptoms appear less common when assessed with gold-standard interviews rather than with self-reports, and PLEs are thus overreported with questionnaires (Horwood et al., 2008).
This study helps healthcare workers identify specific symptoms predicting psychosis. Our present results indicate that social interaction symptoms are associated specifically with onset of psychosis.

FUNDING INFORMATION
Johanna Palomäki was funded by Eemil Aaltonen Foundation. The funding source had no further role in study design, data collection and analysis, writing of the report, and in the decision to submit the report for publication.

DATA AVAILABILITY STATEMENT
NFBC data is available from the University of Oulu, Infrastructure for Population Studies. Permission to use the data can be applied for research purposes via electronic material request portal. In the use of data, we follow the EU general data protection regulation (679/2016) and Finnish Data Protection Act. The use of personal data is based on cohort participant's written informed consent at his/her latest followup study, which may cause limitations to its use. Please, contact NFBC project center (NFBCprojectcenter(at)oulu.fi) and visit the cohort website for more information.