Cluster profiles of distressing psychotic-like experiences among children and associations with genetic risk, prenatal cannabis exposure, and social-environmental characteristics

Abstract

Background:

Distressing psychotic-like experiences (PLEs) in children are associated with an increased risk for psychiatric disorders. Recent studies suggest that different domains of psychotic symptoms could be associated with distinct risk factors, but less is known about PLEs. This study clustered PLEs into subgroups and explored the genetic and environmental characteristics associated with these profiles.

Methods:

Data from children (N=11,854) recruited as part of the Adolescent Brain and Cognitive Development Study v5.1 assessed PLEs using 21-items from the Prodromal-Questionnaire-Brief Child Version. K-medoid clustering of PLEs was conducted among children with at least one distressing PLE item at baseline (n=3,155). Associations of polygenic risk scores for schizophrenia (PRS-SCZ) and social-environmental characteristics with PLE subgroups were estimated using generalized multinomial mixed models, adjusting for age, sex, race/ethnicity, parental education, income-to-needs, and family psychosis history.

Results:

We identified three distressing PLE subgroups: hallucinatory-like (n=1,110), paranoid-like (n=1,229), and multiple PLE domains (n=816). Compared to those without any distressing PLEs (n=8,699), those with hallucinatory-like PLE were more likely to have had prenatal cannabis exposure (Odds Ratio(OR)=1.578, 95% CI: 1.230–2.023); paranoid-like individuals had higher PRS-SCZ (OR=1.080, 95% CI: 1.001–1.165); those with multiple PLE domains participated in less physical activities (OR=0.879, 95% CI: 0.803–0.964). All groups experienced greater childhood adversity and worse school environments.

Conclusions:

We found that those with different aspects of distressing PLEs had distinct and similar genetic and environmental characteristics. These results suggest that it may be important to consider the heterogeneity of PLE in conceptualizing the development of psychosis.

1. Introduction

Psychotic-like experiences (PLEs) include a broad range of perceptual abnormalities and paranoid-like beliefs and are relatively common in children (Linscott and van Os, 2013). Although PLEs do not guarantee a future psychosis diagnosis, those that cause significant distress may be more strongly associated with future psychopathology, including greater functional and cognitive impairments, as well as increased reliance on mental health services (Karcher et al., 2022a; Mcgrath et al., 2016; Oh et al., 2024). Understanding the development of distressing PLEs would allow us to identify steps in the pathophysiological process leading to psychosis and other adverse mental health outcomes, shifting the focus of psychiatric intervention from symptom alleviation to prevention.

A growing body of research has identified genetic and environmental risk factors associated with PLEs in children (Karcher et al., 2021). These include elevated polygenic risk scores for schizophrenia (PRS-SCZ) (Karcher et al., 2022b; Ku et al., 2024a), prenatal cannabis exposure (Fine et al., 2019; Paul et al., 2021), and adverse social-environmental conditions such as limited participation in physical activities (Ku et al., 2024a), adverse childhood experiences (Karcher et al., 2020b), and worse school environment (Park et al., 2024).

It has been postulated that distinct psychotic symptom domains may be driven by separated cognitive processes and distinct underlying mechanisms (Fletcher and Frith, 2009; Sterzer et al., 2018; Wengler et al., 2020). For instance, first-episode psychosis studies suggest that hallucinations and delusions can co-occur but may also arise independently (Compton et al., 2012; Evensen et al., 2011; Powers et al., 2025). However, PLEs have often been studied as aggregated scores of all endorsed experiences or broad distress measures (Karcher et al., 2022b), which may overlook the unique underlying configuration of distinct PLE domains such as hallucination-like and paranoid-like experiences (Betz et al., 2023; Smeets et al., 2012).

Exploring the shared and unique risk factors for specific PLE domains could allow for a more nuanced investigation of the distinct mechanisms driving their emergence. This study further investigated PLE subtypes by (1) clustering children based on their baseline PLE distress scores and (2) comparing the genetic risk, prenatal cannabis exposure, and social-environmental characteristics of these groups.

2. Method

2.1. Participants

Participants included 11,854 children (ages 9-10) recruited from the Adolescent Brain and Cognitive Development Study (ABCD) v5.1 (https://abcdstudy.org/) with available baseline psychotic-like experience ratings. The ABCD study is a nationwide study on brain and behavioral development that recruited children aged 9-to-10 from 22 U.S. research sites (Barch et al., 2018). Participating schools (private, public, and charter) were randomly selected within a 50-mile radius of each of the research sites, representative of the sociodemographic features of the population in the USA (Garavan et al., 2018). Both participants and their parents provided written informed consent, and approval was obtained from the institutional review boards at each participating site.

2.2. Sociodemographic Characteristics

Sociodemographic characteristics, including age, sex, race, ethnicity, parental education, household income, and family history of psychosis, were gathered through parent reports and interviews during the baseline assessment. Race and ethnicity were categorized into five groups: non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, non-Hispanic other races, and Hispanic (Anglin et al., 2023; Ku et al., 2024b; Saragosa-Harris et al., 2022). Parental college education was defined as having at least one parent or caregiver who had obtained a bachelor’s degree or higher. The income-to-needs ratio was calculated by dividing the median household income by the federal poverty line for the respective household size, with values greater or less than one denoting above or below the poverty threshold (Rakesh et al., 2022). Family history of psychosis was determined by the participant having a first- or second-degree relative with psychotic symptoms using the parent-rated Family History Assessment Module Screener (Van Dijk et al., 2021).

2.3. Psychotic-like experiences (PLE)

PLEs were assessed using 21 items from the Prodromal Questionnaire-Brief Child Version (PQ-BC), where participants reported psychotic-like experiences (e.g., perceptual abnormalities, unusual thought content, etc.) over the past month (Loewy et al., 2011). PQ-BC showed adequate psychometric properties, including internal reliability and convergent validity among school-age children (Karcher et al., 2018; Karcher et al., 2020a). All PQ-BC questions were read to the participants by research assistants. Children were first asked if they had ever experienced each symptom. If they answered "yes," they were then asked, "Did it bother you?" If they answered "yes" to both questions, they were instructed to choose the number corresponding to the picture that best captured their level of distress, ranging from minimal (1) to extreme (5). The level of distress was measured for each endorsed PLE with a five-point scale of distress with a visual response scale (See Supplemental Figure 2) (Karcher et al., 2018). Based on previous research, distressing PLEs were defined as ≥3 on a five-point distress scale, as PLEs accompanied by significant distress (≥3) may be more specifically associated with psychosis risk and are more clinically relevant (Karcher et al., 2022b; Ku et al., 2024a).

2.4. Polygenic Risk Scores for Schizophrenia (PRS-SCZ)

PRS-CSx-meta (https://github.com/getian107/PRScsx) was selected to calculate multi-ancestral PRS-SCZ as done in prior research (Ahern et al., 2023; Ruan et al., 2022). The summary statistics of schizophrenia from EUR, EAS, and AFR subpopulations were extracted from the latest schizophrenia genome-wide association study (Trubetskoy et al., 2022) (supplemental methods, Table m1). PRS for each individual were calculated using genomic functional annotations and ancestry-specific LD reference panels from the UK Biobank. PRS-CSx-meta was calculated using an inverse-variance-weighted meta-analysis approach, which combined the results from multiple datasets to produce a single set of posterior effect sizes for schizophrenia (Ge et al., 2022). The resulting effect sizes from the meta-analysis were used to calculate the final PRS-SCZ for each individual. Details on genetic data were summarized in supplemental methods. Based on the analysis of principal components (PCs) and assigned genetic ancestry, the first four PCs were determined to capture population stratification (supplemental methods, Figure m1). These PCs were subsequently used as covariates in PRS-SCZ analyses to adjust for population stratification effects. This method was preferred compared to other methods (e.g., PRS-CSx and Sbaysian-RC) because our dependent variable was categorical, and PRS-CSx-meta did not require splitting the dataset using binary or continuous outcomes (Ruan et al., 2022).

2.5. Prenatal Cannabis Exposure

Prenatal cannabis exposure was assessed based on parent or caregiver retrospective self-report using two distinct items: cannabis use before knowledge of pregnancy and cannabis use after knowledge of pregnancy. A combined variable of any cannabis use during pregnancy was calculated based on responses to cannabis use before and after knowledge of pregnancy. Knowledge of pregnancy was used as a proxy for developmental timing, as endocannabinoid receptors are not expressed in the fetus until 5 to 6 weeks gestation, which approximately corresponds to when mothers typically know their pregnancy (Paul et al., 2021). From these responses, three binary variables were formed with no cannabis use reported during pregnancy as the reference category: any cannabis use during pregnancy (n = 696, 5.9%), cannabis use before knowing pregnancy (n = 686, 5.8%), and cannabis use after knowing pregnancy (n = 245, 2.1%).

2.6. Social-environmental Characteristics

The positive school environment was assessed using the school environment subscale from the ABCD School Risk and Protective Factors Survey (Zucker et al., 2018). The score was calculated by the sum of six four-point subscales: availability of extracurricular activities, relationship with teachers, receiving praise for good performance, opportunities to contribute to class activities, school-parent communications, and perceived safety at school.

Physical activity levels were derived from the Sports and Activities Involvement Questionnaire (SAIQ) modeled after the Vermont Health and Behavioral Questionnaire and the Dutch Health Behavior Questionnaire at baseline (Barch et al., 2018; Huppertz et al., 2016), which collected information on the time spent per session, number of days per week, and months per year spent on 23 physical activities. The mean participation hours per week for each sport were calculated by multiplying the time spent per session by days per week and months per year, then dividing by 12 and 60 to convert to hours per week (Ku et al., 2024a).

The adverse childhood experiences were defined as summations of parent-rated child experience of 18 traumatic experiences from the Kiddie-Structured Assessment for Affective Disorders and Schizophrenia (K-SADS) and a parent-rated question from the K-SADS about whether the child was bullied at school or in the neighborhood (Barch et al., 2018; Karcher et al., 2020b).This summed ACE score approach was based on previous research examining associations between ACE and PLE in the ABCD cohort (Karcher et al., 2020b). In our analyses, we tested both 1) the summed ACE score as our primary approach in the main analyses and 2) the associations between individual components of ACE and PLE groups in the supplementary analyses to account for potential differential impacts of specific traumatic experiences.

2.7. Statistical Analyses

2.7.1. K-Medoid Clustering

To account for the zero-inflated nature of PLE distress ratings, k-medoid clustering was applied to 21 PLE distress scores among children reporting at least one distressing PLE item at baseline (n=3,155). The distributions of all 21 PQ-BC items in this sample were presented in Supplemental Methods, Figure m1. The optimal number of clusters was determined based on previous research in PLEs (Compton et al., 2012; Powers et al., 2025), a visual inspection of the distribution patterns, and the Elbow Method. Details regarding these methods are presented in the Supplemental Methods and Supplemental Figure m2.

2.7.2. Classification of PLE Items to Positive Symptoms

To interpret the clusters and the meaning of the 21 PLE items, we further mapped the 21 PLE items to the five positive psychotic symptom domains, similar to the Structured Interview for Psychosis-risk Syndromes (SIPS) (Miller et al., 2003). This involved having two psychiatric researchers, each of whom assigned each of the 21 items to the five symptom domains separately and then compared the results. The mapping for each PLE item is summarized in Supplementary Table 1. For each symptom domain, we used the maximum distress score across all assigned PLE items to estimate the overall domain distress. This guided mapping allowed us to interpret our data-driven clustering results from the 21 PLE item questionnaire to a clinical categorization of psychotic symptoms.

2.7.3. Multinomial Regression Analysis

Generalized multinomial logistic mixed-effects models were used to explore the potential association between genetic and environmental exposure variables and the PLE group membership. We chose the set of exposures based on prior literature demonstrating the associations between exposure and psychosis outcomes, which include polygenic risk for schizophrenia (Ku et al., 2024a), prenatal cannabis exposure (Paul et al., 2021), adverse childhood experiences (Karcher et al., 2020b), positive school environment (Zammit et al., 2010), and physical activities (Ku et al., 2024a). The outcome variable was the PLE group membership with the non-distressing PLE group as the reference category. We included only site as a random intercept. For random intercepts, we did not include school-level identifiers because of a large number of missing data and did not include family-level identifiers because of model convergence issues using the multinomial mixed model. Covariates included age, sex, race/ethnicity, parental education, income-to-needs ratio, and family history of psychosis. For PRS-SCZ, the model also adjusted for the first four principal components of genetic ancestry.

2.7.4. Missing Data

The multinomial regression analyses included different sample sizes due to varying percentages of missing data across different exposure variables (Supplemental Figure 1). We analyzed the data using complete-case analysis, and most variables had minimal missing data (0-3%). There were 15.92% of genetic data missing due to the lack of availability and not passing quality control (Supplemental Methods).

All analyses were conducted using R 4.4.0 and IBM SPSS Statistics software version 30.0.0.

3. Results

3.1. Sample characteristics

The study included 11,854 children (mean age 9.9 years), of whom 52.1% were male, 52.2% were non-Hispanic White, 59.5% had parents with at least a bachelor’s degree, and 8.3% had a family history of psychosis. Compared to children without distressing PLE (the Non-Distressing PLE Group; n = 8,699), those with at least one distressing PLE at baseline (n = 3,155) were younger, included fewer non-Hispanic White participants, had higher PRS-SCZ, a greater proportion of maternal cannabis use during pregnancy, more adverse social-environmental characteristics, and a greater total PLE distress score (see Table 1).

Table 1.

Sociodemographic Characteristics Based on Baseline Distressing PLE Group.

		Overall	None- distressing PLE	At least one distressing PLE	p	% Missing
Domain	n	11854	8699	3155
Demographics	Age (mean (SD))	9.91 (0.62)	9.94 (0.63)	9.85 (0.62)	<0.001	0
	Sex (%)				0.19	0
	Male	6180 (52.1)	4558 (52.4)	1622 (51.4)
	Female	5671 (47.8)	4140 (47.6)	1531 (48.5)
	Inter-sex male	3 (0.0)	1 (0.0)	2 (0.1)
	Race / Ethnicity (%)				<0.001	0
	Non-Hispanic White	6167 (52.0)	4877 (56.1)	1290 (40.9)
	Non-Hispanic Black	1782 (15.0)	1126 (12.9)	656 (20.8)
	Hispanic	2406 (20.3)	1601 (18.4)	805 (25.5)
	Non-Hispanic Asian	252 (2.1)	205 (2.4)	46 (1.5)
	Non-Hispanic Other Races	1246 (10.5)	889 (10.2)	357 (11.3)
	Parental college education (%)	7048 (59.5)	5540 (63.8)	1508 (47.8)	<0.001	0.1
	Family history of psychosis (%)	985 (8.3)	669 (7.7)	316 (10.0)	<0.001	0
	Income-to-needs ratio (%)	4.16 (3.33)	4.41 (3.36)	3.46 (3.11)	<0.001	10
Genetic Risk for Schizophrenia	PRS-SCZ (mean (SD))	0.00 (1.00)	−0.03 (1.02)	0.10 (0.95)	<0.001	17.5
Prenatal Cannabis Exposure	Any Cannabis use during pregnancy (%)	696 (5.9)	448 (5.3)	248 (7.9)	<0.001	2.9
	Cannabis use before knowing pregnancy (%)	686 (5.8)	441 (5.2)	245 (8.1)	<0.001	2.9
	Cannabis use after knowing pregnancy (%)	245 (2.1)	155 (1.8)	90 (2.9)	0.001	2.3
School Environment	School Environment (mean (SD))	19.93 (2.83)	20.09 (2.68)	19.49 (3.17)	<0.001	0.2
Physical Activity	Physical activity hours per week (mean (SD))	3.09 (3.46)	3.24 (3.51)	2.68 (3.28)	<0.001	0
Adverse Childhood Experiences	Adverse Childhood Experiences Sum (mean (SD))	0.66 (1.04)	0.60 (1.00)	0.83 (1.15)	<0.001	2.7
Psychosis-Prone Positive	P1: Unusual thought content or delusional ideas (median [IQR])	0.00 [0.00, 1.00]	0.00 [0.00, 0.00]	1.00 [0.00, 3.00]	<0.001	0
Symptom Domains	P2: Suspiciousness/persecuto ry ideas (median [IQR])	0.00 [0.00,1.00]	0.00 [0.00,0.00]	2.00 [0.00,3.00]	<0.001	0
	P3: Grandiose ideas (median [IQR])	0.00 [0.00, 0.00]	0.00 [0.00, 0.00]	0.00 [0.00, 1.00]	<0.001	0
	P4: Perceptual abnormalities/hallucinati ons (median [IQR])	0.00 [0.00, 2.00]	0.00 [0.00, 1.00]	3.00 [2.00, 4.00]	<0.001	0
	P5: Disorganized communication (median [IQR])	0.00 [0.00, 1.00]	0.00 [0.00, 0.00]	1.00 [0.00, 2.00]	<0.00 1	0
Total Distress PLE Score	Total Distress PLE Score (mean (SD))	5.00 (8.39)	1.47 (2.33)	14.73 (10.98)	<0.001	0

Abbreviations: PLE, psychotic-like experiences; PRS-SCZ, polygenic risk score for schizophrenia.

Note: P-values were determined using independent t-tests for continuous variables (age, income-to-needs ratio, genetic risk scores, school environment, physical activity, and adverse childhood experiences) and chi-squared tests for categorical variables (sex, race/ethnicity, parental college education, family history of psychosis, and prenatal cannabis exposure). Median differences for psychosis-prone symptom domains were tested using the Mann-Whitney U test. The p-values are two-tailed with statistical significance defined at p < 0.05. Missing data percentages for each variable are indicated in the "% Missing" column.

3.2. Three distinct distressing PLE cluster groups

After using K-medoid clustering on distress scores from 21 psychotic-like experience (PLE) questions, three distinct clusters emerged based on symptom domain profiles (Supplementary Table 2). The Hallucinatory-like Group (n = 1,110) consisted of high distress only in perceptual abnormalities/hallucination PLEs (Table 2, Figure 1). The Paranoid-like Group (n = 1,229) was characterized by elevated distress only in suspiciousness/persecutory PLE ideations (Table 2, Figure 1). The Multiple PLE Domain Group (n = 816) showed distressing PLEs across unusual thought content, suspicious/persecutory ideas, and hallucination PLE domains (Table 2, Figure 1).

Table 2.

Sociodemographic and Social-Environmental Characteristics Compared Among Cluster Groups of Distressing PLE.

		None- distressing PLE Group	Hallucinatory- like Group	Paranoid like Group	Multiple PLE Domains Group	p
Domain	n	8699	1110	1229	816
Demographics	Age (mean (SD))	9.94 (0.63)	9.85 (0.62)	9.87 (0.61)	9.83 (0.62)	<0.001
	Sex (%)					<0.001
	Male	4558 (52.4)	620 (55.9)	566 (46.1)	436 (53.4)
	Female	4140 (47.6)	490 (44.1)	663 (53.9)	378 (46.3)
	Inter-sex male	1 (0.0)	0 (0.0)	0 (0.0)	2 (0.2)
	Race / Ethnicity (%)					<0.001
	Non-Hispanic White	4877 (56.1)	521 (46.9)	487 (39.7)	282 (34.6)
	Non-Hispanic Black	1126 (12.9)	166 (15.0)	273 (22.2)	217 (26.6)
	Hispanic	1601 (18.4)	276 (24.9)	303 (24.7)	226 (27.7)
	Non-Hispanic Asian	205 (2.4)	20 (1.8)	22 (1.8)	4 (0.5)
	Non-Hispanic Other Races	889 (10.2)	127 (11.4)	143 (11.6)	87 (10.7)
	Parental college education (%)	5547 (63.7)	575 (51.8)	597 (48.7)	336 (41.2)	<0.001
	Income-to-needs ratio (mean (SD))	4.41 (3.36)	3.65 (3.12)	3.66 (3.31)	2.90 (2.71)	<0.001
	Family history of psychosis (%)	669 (7.7)	103 (9.3)	115 (9.4)	98 (12.0)	<0.001
Genetic Risk for Schizophrenia	PRS-SCZ (mean (SD))	−0.03 (1.01)	0.05 (0.96)	0.12 (0.95)	0.13 (0.94)	<0.001
Prenatal Cannabis Exposure	Any Cannabis use during pregnancy (%)	448 (5.3)	99 (9.3)	88 (7.4)	61 (7.8)	<0.001
	Cannabis use before knowing pregnancy (%)	441 (5.2)	99 (9.3)	85 (7.1)	61 (7.8)	<0.001
	Cannabis use after knowing pregnancy (%)	155 (1.8)	37 (3.5)	34 (2.8)	19 (2.4)	0.001
School Environment	School Environment (mean (SD))	20.09 (2.68)	19.48 (3.12)	19.60 (3.08)	19.33 (3.35)	<0.001
Physical Activities	Physical activity hours per week (mean (SD))	3.24 (3.51)	2.81 (3.29)	2.76 (3.37)	2.39 (3.09)	<0.001
Adverse Childhood Experiences	Adverse Childhood Experiences Sum Score (mean (SD))	0.60 (0.99)	0.81 (1.28)	0.80 (1.11)	0.89 (0.99)	<0.001
Psychosis-Prone Positive Symptom Domains	P1: Unusual thought content or delusional ideas (median [IQR])	0.00 [0.00, 0.00]	1.00 [0.00, 2.00]	1.00 [0.00, 2.00]	3.00 [1.00, 5.00]	<0.001
	P2: Suspiciousness/persecutory ideas (median [IQR])	0.00 [0.00, 0.00]	0.00 [0.00, 1.00]	3.00 [1.00, 4.00]	3.00 [2.00, 5.00]	<0.001
	P3: Grandiose ideas (median [IQR])	0.00 [0.00, 0.00]	0.00 [0.00, 1.00]	0.00 [0.00, 1.00]	1.00 [0.00, 1.00]	<0.001
	P4: Perceptual abnormalities/hallucinations (median [IQR])	0.00 [0.00, 1.00]	3.00 [3.00, 4.00]	2.00 [1.00, 3.00]	5.00 [4.00, 5.00]	<0.001
	P5: Disorganized communication (median [IQR])	0.00 [0.00, 0.00]	0.00 [0.00, 1.00]	0.00 [0.00, 1.00]	1.00 [1.00, 3.00]	<0.001
Total PLE Distress Score	Total Distress PLE Score (mean (SD))	1.47 (2.33)	10.45 (5.34)	9.66 (5.22)	28.20 (11.88)	<0.001

Abbreviations: PLE, psychotic-like experiences; PRS-SCZ, polygenic risk score for schizophrenia.

Note: P-values were derived from ANOVA for continuous variables (age, income-to-needs ratio, genetic risk scores, school environment, physical activity, and adverse childhood experiences) and Chi-squared tests for categorical variables (sex, race/ethnicity, parental college education, family history of psychosis, and prenatal cannabis exposure). Median differences for psychosis-prone symptom domains were tested using the Mann-Whitney U test.

Figure 1.

Distribution of maximum distress score for PLE clusters across different symptom domains.

Abbreviations: P symptoms, positive symptoms; PLE, psychotic-like experiences.

Note: The colored boxplot shows the median distress score within each symptom domain for each group. The boxes extend from the 25^th (Q1) to the 75^th (Q3) percentile, with the horizontal bolded line at the median. The whiskers represent the minimum and maximum. Each color corresponds to a positive symptom domain. The dashed line at a score of 3 indicates the distressing threshold for PLE.

Among the distressing PLE subgroups, some distinct patterns emerged: the Hallucinatory-like Group had the greatest proportion of prenatal cannabis exposure (9.3%) but the lowest PRS-SCZ (mean = 0.05); the Paranoid-like Group showed the greatest positive school environment (mean = 19.60) but intermediate PRS-SCZ (mean = 0.12); and the Multiple PLE Domain Group demonstrated the most adverse profile, with the highest PRS-SCZ (mean = 0.13), lowest physical activity (2.39 hours/week), greatest adverse childhood experiences (mean = 0.89), and highest total PLE distress scores (mean = 28.20, see Table 2).

3.3. Genetic, Prenatal cannabis use, and social-environmental characteristics’ association with three PLE Cluster Groups

Multinomial logistic mixed models analyzed patterns of genetic and social-environmental characteristics associated with each PLE cluster group membership, compared to those without any distressing PLEs as the reference group. Prenatal cannabis exposure was associated with significantly higher odds of membership in the Hallucinatory-like Group (adjusted Odds Ratio (OR) = 1.578, 95% CI: 1.230–2.023, < .001, see Figure 2 and Supplemental Table 3). This association was observed in both cannabis use before knowing pregnancy (adjusted OR=1.595, 95% CI: 1.244–2.045,< .001) and cannabis use after knowing pregnancy (adjusted OR=1.573, 95% CI: 1.053–2.351, p=0.027). PRS-SCZ was significantly associated with higher odds of membership in the Paranoid-like Group (adjusted OR=1.080, 95% CI: 1.001–1.165, p=0.049). Hours of weekly physical activity were significantly associated with lower odds of being in the Multiple PLE Domain Group (adjusted OR=0.879, 95% CI: 0.803–0.964, p=0.006). Adverse childhood experiences and poor school environments were significantly associated with being in all three distressing PLE groups (see Figure 2 and Supplemental Table 3). Differential associations with specific PLE groups were found when examining individual components of ACEs. Specifically, having experiences of bullying was associated with all three PLE groups (Hallucinatory-like: OR=1.439, 95% CI: 1.206-1.718, p<.001; Paranoid-like: OR=2.171, 95% CI: 1.851-2.547, p<.001; Multiple PLE Domains: OR=2.436, 95% CI: 2.022-2.935, p<.001). Witnessing domestic violence was significantly associated with both the hallucinatory-like (OR=1.416, 95% CI: 1.133–1.770, p=.002) and paranoid-like groups (OR=1.385, 95% CI: 1.110-1.728, p=.004), while experiences of sexual abuse by family members and peers were specifically associated with the hallucinatory-like group (OR=2.337, 95% CI: 1.085-5.035, p=.030; OR=1.712, 95% CI: 1.030-2.846, p=.038, respectively) (see Supplemental Table 4).

Figure 2.

Multinomial associations of cluster groups with genetic, prenatal cannabis exposure, and social-environmental characteristics.

Abbreviations: PRS-SCZ, polygenic risk score for schizophrenia; PCE, prenatal cannabis exposure.

Note: Generalized multinomial logistic mixed-effects models were used for each exposure variable separately to explore the potential association between the exposure variables and PLE group membership. The outcome variable was the PLE group membership with the minimal/no-distressing PLE group as the reference category. All analyses were clustered by sites. All continuous variables were standardized (z-scored). Models adjusted for sociodemographic covariates (age, sex, race/ethnicity, parental college education, income-to-needs ratio, family history of psychosis). For PRS-SCZ, the first four principal components were also adjusted. *p<0.05, **p<0.01, ***p<0.001. The exact Odds Ratio, CI, and P value were presented in Supplemental Table 3.

4. Discussion

This is the first study to demonstrate that different subgroups of distressing PLEs corresponding to hallucinatory-like, paranoid-like, and multiple PLE domains may be associated with different genetic and environmental characteristics. Positive psychotic symptoms like hallucinations and paranoia have been shown to occur independently and together among individuals with first-episode psychosis and those at clinical high-risk for psychosis (CHR-P) (Compton et al., 2012; Cowan and Mittal, 2021; Evensen et al., 2011). Our findings extend these prior findings to children with distressing PLE. This study identified three distinct clusters of distressing PLEs and examined their unique and overlapping associated genetic and environmental characteristics. Individuals experiencing primarily hallucinatory-like PLEs had the highest rates of prenatal cannabis exposure but lower PRS-SCZ; those experiencing predominantly paranoid-like PLEs showed higher PRS-SCZ but relatively better social environment; and those with distress across multiple PLE domains had the highest PRS-SCZ, lowest participation in physical activity, greatest sum of adverse childhood experiences, and poorest school environment. This study also found that worse school environments and adverse childhood experiences were associated with all PLE cluster subgroups. Perhaps considering the complexity and heterogeneous nature of the various domains of distressing PLEs may provide novel insight into varied and shared pathways leading to psychiatric risks in youth.

The identification of three PLE clusters—hallucinatory-like, paranoid-like, and multiple PLE domains—aligns with prior findings in first-episode psychosis and CHR-P populations. Research suggested that self-reported PLEs in CHR individuals were often categorized into persecutory ideation, bizarre experiences, and perceptual abnormalities or hallucinations (Cowan and Mittal, 2021). Research on first-episode psychosis has shown that delusions and hallucinations can co-occur or appear in isolation, potentially reflecting distinct underlying mechanisms (Compton et al., 2012; Evensen et al., 2011; Wengler et al., 2020). Recent computational models further differentiated these mechanisms: hallucinations may arise from an overreliance on prior beliefs due to low-level sensory noise, and delusions may involve disrupted higher-level processes, which increase conviction in beliefs through faulty prediction error signaling (Powers et al., 2025). Our findings suggest that the presence of distinct symptom patterns may even extend to children with PLE.

Interestingly, our study revealed that prenatal cannabis exposure was uniquely associated with being in the hallucinatory-like PLE subgroup, even after adjusting for socioeconomic status and family history of psychosis. A recent systematic review found that approximately 30% increased risk for psychotic symptoms from prenatal cannabis exposure, yet the results were heterogeneous and inconclusive (Bassalov et al., 2024), with only one study showing a significant association (Paul et al., 2021). Perhaps, this nuanced relationship might be more pronounced in a specific subgroup, potentially explaining the inconsistencies observed in prior studies. Additionally, the relatively lower PRS-SCZ observed in this subgroup suggests that environmental factors like prenatal cannabis exposure may play a more important role than genetic predispositions in this subgroup of children.

PRS-SCZ were notably higher in both paranoid-like and the multiple PLE domain groups. However, after adjusting for covariates, PRS-SCZ was only significantly associated with being in the paranoid-like group as opposed to the group without distressing PLE. This could be partially due to the higher proportion of Hispanic and non-Hispanic Black individuals in the multiple-domain group, where PRS-SCZ often show poor performance in non-European ancestries due to small sample sizes from summary statistics (Ahern et al., 2023; Karcher et al., 2022b; Trubetskoy et al., 2022). These findings demonstrate the challenges of using polygenic risk scores in diverse populations and highlight the need for further refinement and inclusivity in genetic research methodologies.

Additionally, we found that those with distress in multiple PLE domains tended to have the most adverse genetic and environmental profiles, lowest participation in physical activity, and greater total PLE distress scores. This aligns with previous findings from a general population study of 14-24-year-olds, which showed that individuals experiencing both hallucinations and delusions had greater environmental and familial risk factors, more severe symptoms, greater persistence, and higher risk of adverse clinical outcomes compared to those with a single type of symptom (Smeets et al., 2012). It is possible that early manifestation of experiences from multiple symptom domains could be more predictive of adverse psychiatric outcomes.

This study has several limitations. First, since we used only baseline data for clustering and all exposure variables and PLEs were collected at baseline, we cannot establish temporal relationships or infer causality. Second, the distribution of the questionnaire mapped onto five positive symptoms was uneven: 10 out of 21 questions targeted hallucinations, potentially leading to an overrepresentation of this domain due to higher variance in responses (Supplementary Table 1). Third, environmental measures such as prenatal cannabis exposure and adverse childhood experiences were derived from retrospective self-reports, and physical activity levels were reported by parents or caregivers, which may be subject to recall bias. Fourth, we used listwise deletion, which assumes that all predictors of missingness are included as explanatory variables in the models. If the missingness of variables (i.e., genetic risk scores) was related to an unmeasured variable, then listwise deletion may introduce bias into parameter estimates. Fifth, despite using multi-ancestral polygenic risk scores, its predictive power for non-European ancestries remains limited. Lastly, given the exploratory nature of this study, we chose not to adjust for multiple comparisons (i.e., multiple exposure variables), which may increase the risk of Type I error. This approach allowed us to identify potential associations that may be validated in larger cohorts. Future studies should aim to incorporate more follow-up data to analyze how different symptom domains evolve and their association with longitudinal outcomes.

In conclusion, this study identified clusters of distressing psychotic-like experiences (PLEs) that correspond to hallucinatory-like, paranoid-like, and multiple PLE domains, each associated with distinct and similar genetic and environmental characteristics. These findings suggest that it may be important to consider the heterogeneity of PLEs in understanding the development of psychosis. Future research with longitudinal data should track how these symptom domains evolve over time and their potential progression into more severe psychiatric conditions. This approach will enhance our understanding of the onset and development of psychosis, informing more targeted and effective interventions.