Clues from dyadic interactions in early childhood predict psychotic-like experiences at the transition to adolescence in a sample enriched for psychopathology risk

Abstract

Early childhood is a critical period for socioemotional development with long-term implications for the emergence of psychopathology. However, alterations in social interactions during early childhood have not been examined as vulnerability markers for psychosis. Raters naïve to clinical outcomes coded behaviors during socially-engaged and socially-disengaged contexts of the Disruptive Behavior Diagnostic Observation Schedule (DB-DOS) video-recorded in early childhood (ages 3–5 years old; M=4.23) for a subset of child-caregiver dyads from the MAPS Study (n=93), a sample enriched for psychopathology risk by oversampling for irritability and exposure to violence. Linear regression models examined the predictive value of mutual responsive behavior (e.g., appropriate responses to the dyadic partner’s bids for engagement), positive emotion, negative emotion, and eye gaze for psychotic-like experiences (PLEs) at the “tween” wave (ages 10–13). Parallel models with internalizing/externalizing symptoms measured at pre-adolescence (n=87; ages 8–11) examined specificity. Lower responsive behavior, greater negative emotion, and lower eye gaze among child-caregiver dyads in only the socially-engaged task predicted greater PLEs, all p<.042 and multiple R²>0.045. A combined model performed better than the individual models, p=.010, multiple R²=0.139, indicating these predictors contributed unique variance. For internalizing and externalizing symptoms, only eye gaze in the socially-disengaged task was a significant predictor for externalizing symptoms. The specificity of results suggests the observed dyadic behaviors may index mechanisms of neurodevelopmental vulnerability to psychosis, which has significant implications for early detection and prevention. Studies with dense observations beginning in infancy, paired to neurophysiology, and linked to PLEs will be key to understanding neurodevelopmental mechanisms.

General Scientific Summary

Psychosis-spectrum disorders such as schizophrenia are robustly associated with impairment in social functioning, which may play a role in driving the development and maintenance of symptoms. This research examines whether observable alternations in behaviors relevant to social and emotional functioning, such as expression of emotions and eye gaze, can predict psychotic-like experiences (a low-level indicator of increased vulnerability to psychosis-spectrum disorders) approximately 7 years later.

Introduction

Psychosis is a leading cause of disability worldwide and schizophrenia alone has an economic burden exceeding US$343 billion each year in the United States (Kadakia et al., 2022). Reduced social functioning following the onset of psychosis is commonly linked with increased stress and dysfunction, which is related to both external and internalized stigma in individuals with psychosis (Hooley, 2010; Lysaker et al., 2007), impaired social cognition (Lysaker et al., 2021), impaired meta-cognitive integration of social information (Lysaker et al., 2021), or disorganized symptoms (Riehle & Lincoln, 2018). At the same time, lower social functioning may lead to the onset or worsening of psychosis symptoms (Kidd, 2013). Although lower social functioning is a robust predictor of the likelihood of an individual meeting criteria for a clinical high risk syndrome converting to a psychotic disorder (Carrión et al., 2021; Dragt et al., 2011; Lyngberg et al., 2015; Osborne et al., 2020; Tarbox et al., 2013, 2014) and is a primary concern among those diagnosed with psychosis (Lysaker et al., 2021), there is limited information about the temporal course of social functioning in individuals at elevated risk for developing psychosis. Existing theories have highlighted that increasing social support and/or reducing stressors related to social dysfunction could serve as protective factors which reduce the risk of developing a psychotic disorder (Fredrickson, 2001; Fujiwara et al., 2021; Liu et al., 2015). Interventions such as social skills training, cognitive behavioral therapy, or mindfulness-based interventions targeting improvements in social functioning have demonstrated some success in improving social functioning for individuals diagnosed with a psychotic disorder or meeting criteria for a clinical high-risk syndrome (Alvarez-Jimenez, 2018; Hooley, 2010; Turner et al., 2018). Indeed, successful treatment of other symptoms or reductions in stress have shown generalization effects where positive psychosis symptoms also decrease – even if they were not directly targeted in treatment (Brand, 2021; Buck et al., 2019; van den Berg et al., 2018; Varese et al., 2021).

Given that it is known that early intervention has been demonstrated to improve outcomes in psychosis (e.g., Fusar-Poli, 2017; Fusar-Poli et al., 2017; Miguel Ruiz-Veguilla et al., 2012) it is exciting that there is early promise in research identifying psychosis risk prior to the onset of subthreshold positive psychosis symptoms, as there is growing evidence that risk factors such as substance use, trauma exposure, and genetic risk (Polanczyk et al., 2010; van Os et al., 2009) influence the likelihood of having psychotic-like or psychotic experiences across the psychosis spectrum in very similar ways. Therefore, it becomes increasingly possible to identify individuals with signs of underlying neurodevelopmental vulnerabilities related to risk for psychotic disorders, such as psychotic-like experiences (PLEs) or other atypicalities to predict the likelihood of an individual having experiences on the psychosis spectrum (Liu et al., 2015; Polanczyk et al., 2010; Zarubin et al., 2023). This growing body of literature predicting PLEs from variables present in infancy and early childhood suggests that neurodevelopmental differences and associated differences in behavior are present and observable in children well before the presentation of subthreshold positive symptoms of psychosis initially appear.

Based on these findings in clinical high-risk populations, examining the potential of early social interactions to predict later psychosis proneness creates an opportunity to identify and potentially mitigate psychosis risk prior to the onset of even subthreshold positive psychosis symptoms, during an early period including heightened neuroplasticity (Mittal & Wakschlag, 2017). However, most studies of psychosis begin in older youth or later, precluding rigorous prospective examination of early social dysfunction as a mechanism.

When situating social features of vulnerability to psychosis within developmental context, it is clear that crucial questions remain unanswered. As a young child, the most important social relationship is with one’s primary caregiver (Kellerman et al., 2019; Mäntymaa et al., 2003; Osborne et al., 2020; Reck & Herpertz, 2016) where reciprocal social skills are consolidated including mutual responsiveness to the dyadic partner’s cues, and affective and gaze sharing (Hay & Cook, 2007; Osborne et al., 2020). Toddlers demonstrate intentional prosocial behavior indicative of an understanding of social reciprocity, and deficits in sharing behavior at this age has been identified as an early indicator of autism spectrum disorders (ASD; Hay & Cook, 2007). Responses become more differentiated across individuals in early childhood as effortful control and more differences in self-regulation emerge (Hay & Cook, 2007). Notably, children’s interaction quality and attachment style with their caregivers is predictive of resilience, social functioning with peers, and academic achievement (Duncan et al., 2019; Fransson et al., 2016; Yoon et al., 2023). In addition to being a vehicle for the transmission of social behavior, the caregiver relationship is also a mechanism through which psychopathology is transmitted intergenerationally and could be an effective point of intervention (Liu et al., 2015; Reck & Herpertz, 2016). Because of the pervasive influence of the caregiver-child relationship and the ability to examine it in early childhood – before individuals form strong social bonds outside of the family – observed caregiver-child interactions provide uniquely valuable context in which to study and potentially intervene on an individual’s social development as divergent paths of development begin to be more evident.

Previous research has demonstrated abnormalities in responsive behavior, positive emotion, negative emotion, and eye gaze in psychosis and individuals meeting criteria for a clinical high-risk syndrome (Gupta et al., 2021; Jalbrzikowski et al., 2013; Livingstone et al., 2009; Zou et al., 2021). Each of these domains also play an important role in social interactions and may serve as mechanisms through which individuals who experience psychosis have lower social functioning (Fujiwara et al., 2021; Liu et al., 2015). Responsive behavior matches response to bids, in a way that reads the dyadic partner’s cue and responds sensitively and appropriately including social and physical bids (e.g., touch). Affect and gaze sharing are a critical elements of social interactional behavior. Positive and negative emotion encompass a range of expressions of emotion (e.g., facial expression, voice tone, body language), which can be behaviorally coded in naturalistic interactions (Fujiwara et al., 2021; Gooding et al., 2018). Eye gaze is the amount of time that an individual spends looking at another individual, including but not limited to eye contact (Gupta et al., 2021; Osborne et al., 2020). Atypicalities in these four domains of social behavior observed in child-caregiver interactions in early childhood may be developmental precursors of later social dysfunction as it relates to experiences on the psychosis spectrum.

The current study addressed questions about early differences in dyadic interaction between children and their caregivers during early childhood when we theorize that neurodevelopmental vulnerability may be expressed as early decrements in social interactional skills, well before the manifestation of most psychotic-like experiences or symptoms. Guided by dyadic interaction paradigms (Feldman, 2007; Gottman & Gottman, 2017; Guo et al., 2021; Haase, 2023; Kostøl & Kovač, 2024; Levenson, 2024; Lougheed et al., 2015; E. Lunkenheimer et al., 2020; E. S. Lunkenheimer et al., 2011; Morris et al., 2018; Thompson, 2024), observational coding of four domains of dyadic interaction – responsive behavior, positive emotion, negative emotion, and eye gaze - was completed with the video recordings of two dyadic tasks (one socially engaged task and one socially disengaged comparison task completed in early childhood). These codes were entered into models predicting psychosis proneness at long-term follow-up 6–8 years later. We hypothesized that these dyadic interaction codes in the socially engaged task would be predictive of psychosis proneness as measured through PLE scores reported at the transition to adolescence. We specifically expected less responsive behavior, less positive emotion, more negative emotion, and less eye gaze would be associated with higher PLE scores. In addition, we conducted parallel analyses using these socioemotional codes to predict internalizing and externalizing symptoms. We hypothesized that dyadic interaction at baseline would be predictive only of PLEs, as we anticipated the dyadic codes to index the precursors of the type of social dysfunction which are primary in psychosis but not in other types of psychopathology.

Methods and Materials

Participants.

Participants were a sub-sample of the Multidimensional Assessment of Preschoolers Study (MAPS) sample who returned for longitudinal follow-up at the transition to adolescence (ages 10–13; “tween” wave). The MAPS study is a longitudinal study following participants from pre-school age (ages 3–5; M = 4.23, SD = 0.67) through the transition to adolescence and enriched for psychopathology risk by oversampling for irritability and exposure to violence at initial recruitment (see Briggs-Gowan et al., 2019; Wakschlag et al., 2015). The initial recruitment included 425 pre-school aged children of whom 180 (42%) who were reascertained for a follow-up at the tween wave, funded after the conclusion of the original study to assess physical health. The sub-sample in the current analyses consists of the 93 (52% of those resascertained) dyads with both observed parent-child interactions from the pre-school baseline wave and PLEs self-reported by the child from the tween wave. Children from this subsample were not different from children of the full baseline sample in terms of age, sex, or poverty status; however, they were more likely to be African American (62% of the current sample compared to 46% of the full baseline sample). Information about the age, race, sex, poverty status, and PLE scores of the included sample is in Table 1. Poverty status was established using Census-based poverty thresholds from the 2010 census based on income and the number of adults and children in the home.

Table 1.

Demographic Metrics

	Full Sample	Subsample with pre-adolescent symptom data
Total Participants	93	87
Sex (female)	49 (52.7%)	46 (52.9%)
Age in years (S.D.)
Mean baseline age	4.23 (0.67)	4.20 (0.66)
Mean pre-adolescence age		9.23 (0.48)
Mean tween age	11.59 (0.59)	11.59 (0.59)
Poverty Status
Poor	50	47
Non-poor	43	40
Race/ethnicity
African American	58	56
Hispanic	20	20
White	14	10
Other	1	1
Clinical Scores (S.D.)
Mean PLEs	10.81 (4.64)	10.98 (4.70)
Mean internalizing symptoms		1.99 (2.90)
Mean externalizing symptoms		3.48 (4.92)

Analyses examining the specificity of the results to PLEs were conducted on the smaller group of participants (n=87) who additionally had data about internalizing and externalizing symptoms from an earlier pre-adolescent wave (ages 8–11). Comparisons of contemporaneous symptom levels was not possible because internalizing and externalizing symptoms were not assessed at the same timepoint as PLEs.

Tasks.

Parent-child dyads completed a battery of developmentally appropriate tasks and assessments at the pre-school wave baseline (for details see Wakschlag et al., 2015). The current analysis examined two, three-minute tasks within the Disruptive Behavior Diagnostic Observation Schedule (DB-DOS; Wakschlag et al., 2008). The DB-DOS was designed to “press” for variability in regulation/co-regulation during activities of varying motivational salience. In this sense, it is particularly well suited for detecting social dysfunction because it was specifically designed to tax children’s social-emotional capabilities via altering level of social engagement and challenge. During the “free play/rapid clean up” press, used here as a socially engaged task, the participant and their caregiver were provided with a selection of toys and were instructed to play together briefly until the timer sounded to signal “time to clean up”. To examine specificity to the interactive context, the DB-DOS “busy parent” press was selected as a socially disengaged comparison. During this “busy parent” press, caregivers were given several questionnaires to fill out while their child was given a book and expected to entertain themselves quietly nearby.

Coding of Socioemotional Functioning.

Parent and child interactive behaviors were coded from videos of parent-child interactions during the Disruptive Behavior Diagnostic Observation Schedule (DB-DOS; Wakschlag et al., 2008). Videos were holistically coded by two independent coders, who were not informed about clinical outcomes at follow-up. The coding scheme was developed for this project, guided by dyadic interaction paradigms (Feldman, 2007; Gottman & Gottman, 2017; Guo et al., 2021; Haase, 2023; Kostøl & Kovač, 2024; Levenson, 2024; Lougheed et al., 2015; E. Lunkenheimer et al., 2020; E. S. Lunkenheimer et al., 2011; Morris et al., 2018; Thompson, 2024), in four categories of behaviors: responsive behavior (e.g., making or responding to a bid for attention), positive emotion (e.g., smiling, positive tone of voice), negative emotion (e.g., frowning, expressing frustration), and eye gaze (e.g., looking at their dyadic partner). A fifth category, active touch, was coded based on the relevance of active touch for children of this age; however, due to low prevalence and variability of this behavior type in both tasks, the variable was removed prior to running analyses. No additional behaviors beyond these five were coded. Each category was given a zero (not at all), one (some), or two (a lot) rating for the degree to which the domain was observed in the child’s behavior, the parent’s behavior, and mutually (e.g., simultaneous engagement of both parent and child in behaviors of the same category), resulting in 12 codes per dyad per task. For data reduction purposes and due to recent findings in dyadic synchrony research indicating that shared socioemotional functioning might be a more powerful predictor of outcomes than individual socioemotional functioning (Wells et al., 2022), analyses were conducted on only the mutual ratings. Interrater reliability was good, with a linear weighted κ of 0.60. The two independent ratings were averaged and this averaged score of the degree to which a dyad simultaneously exhibited behaviors of each domain was used for analyses.

Clinical Outcomes.

During the follow-up assessment at the tween wave, youth completed the Adolescent Psychotic-like Symptom Screener/Community Assessment of Psychic Experiences (APSS-CAPE; Dolphin et al., 2015; Kelleher et al., 2011; Kelleher & Cannon, 2011), which is a brief 7-item questionnaire about PLEs (e.g., thinking people can read your mind, hearing voices or sounds that no one else can hear). Participants reported the frequency at which they experience common types of PLEs as never, sometimes, often, or nearly always, which were converted to a number 1–4 and summed to generate the total score used in this analysis (M = 10.81, SD = 4.64, range = 7–28). Internal consistency for the APSS-CAPE was good, with a Cronbach’s Alpha at α = 0.86.

Internalizing and externalizing symptom scores, which were used as a comparison to examine the degree to which dyadic behaviors were predictive of particularly psychosis proneness or were predictive of psychopathology risk in general, were obtained during the pre-adolescent wave. Parents were interviewed about their child’s experiences using the Diagnostic and Statistical Manual of Mental Disorders, 5^th Edition version of the Kiddie Schedule for Affective Disorders and Schizophrenia -Present and Lifetime Version (K-SADS-PL; Kaufman et al., 2000). Total scores for internalizing and externalizing symptoms were generated by summing the number of clinically-relevant symptoms present within the internalizing (major depression, generalized anxiety, and separation anxiety; M = 1.99, SD = 2.90, range = 0–16) and externalizing (oppositional defiant, conduct, and attentional deficit hyperactivity; M = 3.48, SD = 4.92, range = 0–21) domains, regardless of whether the pre-adolescent met criteria for a specific diagnosis.

Analytic Strategy.

Descriptive statistics were calculated for each of the four categories of coded behaviors and for the outcome variables using the describe function of the psych package (version 2.3.3) in RStudio (version 2024.04.2+764 for Windows).

Then, the lm function of the stats package (version 4.3.0) was used to generate linear models examining 1) whether demographic variables (age, gender, poverty status) were predictive of later PLEs, 2) the predictive value of each of the four included dyadic variables individually, 3) the combined predictive value of the four variables, and 4) comparison analyses with the comparison task, internalizing symptoms, and externalizing symptoms. Odds ratios and their 95% confidence intervals were extracted from a logistic regression by binarizing PLE outcomes into “higher risk” and “lower risk” by grouping participants into approximately the highest quartile and lowest three quartiles on PLE scores within this sample. This study was not preregistered. The behavioral coding data, study analysis code, and original analysis output can be found on Open Science Framework. The videos used to create the behavioral coding data are not openly available to protect the anonymity of participants.

Results

1. Descriptive Statistics.

Descriptive statistics indicated that each interactive code included sufficient variability. In the socially engaged task, codes for responsive behavior ranged from 0.5 to 2 (M = 1.69, SD = 0.41), codes for positive emotion ranged from 0 to 2 (M = 1.10, SD = 0.52), codes for negative emotion ranged from 0 to 1 (M = 0.07, SD = 0.20), and codes for eye gaze ranged from 0 to 2 (M = 0.73, SD = 0.47). In the socially disengaged task, codes for responsive behavior ranged from 0 to 2 (M = 0.88, SD = 0.50), codes for positive emotion ranged from 0 to 1.5 (M = 0.37, SD = 0.43), codes for negative emotion ranged from 0 to 1 (M = 0.19, SD = 0.32), and codes for eye gaze ranged from 0 to 1.5 (M = 0.39, SD = 0.41).

2. Predictive Value of Demographic Variables

PLE score was regressed on participant’s gender, poverty status, and age at baseline to examine whether any of these demographic factors were predictive of later PLEs. None of the individual variables nor the overall model was significant, all p>.295. See Table 2.

Table 2.

Model estimates for associations between observed dyadic interactions at pre-school and PLEs at long-term follow-up. Asterisks indicate significant models or predictors at p < .05.

	β	SE	t-value	Degrees of freedom	Multiple R-squared	Adjusted R-squared	p-value
Demographic Variables				3,89	0.013	−0.021	.766
Gender	−0.205	0.981	−0.208				.835
Poverty Status	1.030	0.978	1.053				.295
Baseline Age	−0.026	0.741	−0.036				.972
Socially Engaged Task – Individual Models
Responsive Behavior	−2.966	1.144	−2.592	1,91	0.069	0.059	.011*
Positive Emotion	−1.399	0.927	−1.51	1,91	0.024	0.014	.135
Negative Emotion	5.996	2.312	2.592	1,91	0.069	0.059	.011*
Eye Gaze	−2.088	1.014	−2.06	1,91	0.045	0.034	.042*
Socially Engaged Task – Combined Model				4,88	0.139	0.100	.010*
Responsive Behavior	−1.789	1.287	−1.390				.168
Positive Emotion	−0.054	1.032	−0.052				.959
Negative Emotion	5.267	2.336	2.255				.027*
Eye Gaze	−1.680	1.080	−1.556				.124
Socially Disengaged Task – Individual Models
Responsive Behavior	−1.191	0.962	−1.239	1,91	0.017	0.006	.219
Positive Emotion	−0.821	1.111	−0.739	1,91	0.006	−0.005	.462
Negative Emotion	1.997	1.504	1.328	1,91	0.019	0.008	.188
Eye Gaze	−0.907	1.166	−0.777	1,91	0.007	−0.004	.439
Socially Disengaged Task – Combined Model				4,88	0.038	−0.005	.479
Responsive Behavior	−1.249	1.095	−1.141				.257
Positive Emotion	−0.140	1.375	−0.102				.919
Negative Emotion	2.132	1.555	1.371				.174
Eye Gaze	0.032	1.451	0.022				.983

3. Individual Predictive Value

PLE score was regressed on each of the early childhood vulnerability indicators (i.e., responsive behavior, positive emotion, negative emotion, and eye gaze) in independent models to examine the predictive value of each variable. Results indicated that lower responsive behavior (β=−2.966, p=.011, multiple R²=0.069), higher negative emotion (β=5.996, p=0.11, multiple R²=0.069), and lower eye gaze (β=−2.088, p=.042, multiple R²=0.045) were predictive of higher PLEs at long-term follow-up. See Table 2.

4. Combined Predictive Value

PLE score was also regressed on responsive behavior, positive emotion, negative emotion, and eye gaze combined in a single model to examine the combined predictive value of these variables. Results indicated that the combined model predicted more variance in later PLE scores than any of the individual models (p=.010, multiple R²=0.139), which demonstrates that these four variables were predicting unique variance (i.e., they are not all indexing a single underlying factor predicting later PLEs). See Table 2 and Figure 1. Odds ratios calculated from a logistic regression using “higher risk” and “lower risk” groups as the primary outcome are reported in Supplementary Table 1. As there were small numbers of participants in many of the cells after binarizing by PLE scores, variance in this analysis was limited. As such, the confidence intervals were large and these results should be interpreted only with caution as preliminary findings.

Figure 1.

A) Density plots of the relationships between responsive behavior, positive emotion, negative emotion, and eye gaze coded in caregiver-child dyadic interactions during a socially engaged task at pre-school age and PLE score at long-term follow-up at the transition to adolescence. Lines of best fit represent the coefficient for each predictor variable from the model in which all four variables were included. B) Density plots of the relationships between responsive behavior, positive emotion, negative emotion, and eye gaze coded in caregiver-child dyadic interactions during a socially disengaged comparison task at pre-school age and PLE score at long-term follow-up at the transition to adolescence. Lines of best fit represent the coefficient for each predictor variable from the model in which all four variables were included.

5. Comparison Analyses

Two types of comparison analysis were conducted to examine the specificity of the finding relative to the context of the observation and other forms of psychopathology.

Comparison Task.

Analyses parallel to those performed for the socially engaged task were repeated with the socially disengaged comparison task. Results indicated that none of the individual models nor the combined model were significant predictors of PLEs at long-term follow-up, all p>.174. See Table 2 and Figure 1.

Common Internalizing/Externalizing Psychopathologies.

The summary internalizing symptom scores were regressed on the demographic variables, the four dyadic variables in the socially engaged task, and the four dyadic variables in the socially disengaged task in three independent models. None of these models were significant, all p>.499. Summary externalizing symptom scores were regressed on the demographic variables, which revealed poverty status to be a significant predictor. As such, poverty status was included with the four dyadic variables in the two independent models examining the socially engaged and socially disengaged tasks. Poverty status remained a significant predictor in each of these models. Additionally, eye gaze was a significant predictor of externalizing symptoms in only the socially disengaged task and this overall model was significant. See Table 3 and Figure 2.

Table 3.

Model estimates for associations between observed dyadic interactions at pre-school and clinical outcomes at long-term follow-up. Asterisks indicate significant models or predictors at p < .05.

	β	SE	t-value	Degrees of freedom	Multiple R-squared	Adjusted R-squared	p-value
Demographic Variables – Internalizing Symptoms				3,83	0.027	−0.008	.520
Gender	−0.328	0.638	−0.513				.609
Poverty Status	0.840	0.632	1.330				.187
Baseline Age	−0.247	0.483	−0.511				.611
Socially Engaged Task – Internalizing Symptoms				4,82	0.040	−0.007	.499
Responsive Behavior	−0.388	0.868	−0.447				.656
Positive Emotion	0.225	0.716	0.314				.754
Negative Emotion	−1.289	1.560	−0.826				.411
Eye Gaze	1.096	0.732	1.498				.138
Socially Disengaged Task – Internalizing Symptoms				4,82	0.003	−0.046	.992
Responsive Behavior	−0.217	0.711	−0.306				.761
Positive Emotion	0.028	0.953	0.029				.940
Negative Emotion	0.076	1.009	0.075				.676
Eye Gaze	0.412	0.981	0.419				.977
Demographic Variables – Externalizing Symptoms				3,83	0.082	0.049	.068
Gender	0.952	1.051	0.906				.367
Poverty Status	2.550	1.040	2.452				.016*
Baseline Age	0.302	.795	0.380				.705
Socially Engaged Task – Externalizing Symptoms				5,81	0.090	0.033	.172
Responsive Behavior	−0.256	1.445	−0.177				.860
Positive Emotion	0.873	1.189	0.734				.465
Negative Emotion	−2.292	2.601	−0.881				.381
Eye Gaze	0.455	1.292	0.352				.726
Poverty Status	2.891	1.121	2.579				.012*
Socially Disengaged Task – Externalizing Symptoms				5,81	0.160	0.108	0.013*
Responsive Behavior	0.257	1.127	0.228				.820
Positive Emotion	−1.307	1.492	−0.876				.384
Negative Emotion	−0.801	1.585	−0.505				.615
Eye Gaze	4.072	1.545	2.636				.010*
Poverty Status	3.019	1.040	2.903				.005*

Figure 2.

A) Density plots of the relationships between responsive behavior, positive emotion, negative emotion, and eye gaze coded in caregiver-child dyadic interactions during a socially engaged task at pre-school age and internalizing symptom score at follow-up in preadolescence. Lines of best fit represent the coefficient for each predictor variable from the model in which all four variables were included. B) Density plots of the relationships between responsive behavior, positive emotion, negative emotion, and eye gaze coded in caregiver-child dyadic interactions during a socially engaged task at pre-school age and externalizing symptom score at follow-up in preadolescence. Lines of best fit represent the coefficient for each predictor variable from the model in which all four variables and poverty status were included.

Discussion

In the longitudinal prediction of psychosis proneness from observed social behavior in child-caregiver interaction during early childhood, lower responsive behavior, lower positive emotion, higher negative emotion, and lower eye gaze observed at pre-school accounted for about 14% of the variance in later psychosis proneness scores. The predictive value for later PLEs was specific to observed context (i.e., only during task with active social engagement) and these domains of interaction during a socially-engaged task were significant predictors only of later PLEs. Although increased eye gaze in the socially-disengaged task was predictive of later externalizing symptoms, this is thought to be indicative of a separate mechanism (e.g., children at elevated risk for developing externalizing symptoms may have behaved in ways which attracted their parent’s attention more). To our knowledge, this is the first prospective study to detect observed vulnerabilities in early social behavior as predictors of subsequent psychosis proneness. In combination, these findings provide compelling initial evidence that differences in social behaviors of child-caregiver dyads during socially-engaged tasks in early childhood may be markers of specific risk for psychosis-spectrum experiences.

The findings of the current study add to a growing line of research which suggests the possibility of identifying risk for psychosis-spectrum experiences well before the onset of subthreshold positive symptoms of psychosis (Liu et al., 2015; Polanczyk et al., 2010; Zarubin et al., 2023). In addition to adding support to neurodevelopmental perspectives on psychosis, these findings also highlight potential mechanisms of psychosis risk which are identifiable and separable from general risk for psychopathology even early in development. These results also identify potential targets for future primary prevention-focused interventions designed to prevent developmental cascades from progressing instead of attempting to remediate functioning after neurodevelopmental changes have already accumulated. It is important to note that the socially-engaged task was a DB-DOS “press”, which was created as a clinic-based assessment for the diagnosis of disruptive behaviors (Wakschlag et al., 2008) and, as such, has already been designed to be clinically informative and pragmatic. It may be fruitful to explore the use of this or a like task within the context of a broader screening (e.g., one which uses the full DB-DOS protocol) or separately for children who may have other indications of elevated psychosis risk (e.g., a first degree relative with a psychotic disorder). Nevertheless, examining the utility of survey-based assessments of social functioning rather than observation-based assessments could further expand the feasibility of scalable implementation of screening for psychosis risk.

Some limitations of the present study temper the ability to draw strong conclusions about the breadth and specificity of the observed relationships. The observed social behaviors in this analysis were in the context of the caregiver-child relationship, which is a critical bond and which has a significant role in establishing the expectations a child has for social relationships. Although this lends strength to the possibility of intervening in the child-caregiver relationship to mitigate risk and allows examination of primary social relationships for young children, it is as yet unknown whether the differences in social interaction observed here would also be present with peers and non-parental adults. It is possible that differences in dyadic interactions as observed here are related to a combination of the participant’s neurodevelopmental vulnerability and that of their caregiver (i.e., if both child and caregiver have higher genetic risk for psychosis, their interactions may be observably different because of this combination of risk across the dyad). This information was not available in the MAPS sample and would point to the need for future studies of young children from families at high genetic risk for psychosis. This highlights a methodologic limitation of the study because, although to our knowledge it is the first prospective study to link early observed social atypicalities to psychosis proneness, the MAPS protocol was not designed apriori to test this hypothesis. Neurodevelopmental studies designed for this purpose with dense sampling beginning even earlier in development (as reciprocal social behaviors and affect sharing are evident as early as first year of life and early brain:behavior atypicalities are detectable; Bosl et al., 2018; Hay & Cook, 2007) and with multi-method indicators of psychosis proneness would be optimal to expand on these foundational findings. Longitudinal, multi-method characterization of these early social patterns and their functional impact from infancy-childhood would also provide a more comprehensive picture of the pattern and persistence of social behavior most likely to presage psychosis. This dense longitudinal tracing would also shed light on which young children who exhibit these risk patterns exhibit subsequent psychosis proneness and which do not. The MAPS Study was also risk-enriched for psychopathology, so replication in population-based samples would be advantageous, as would replication where internalizing and externalizing symptom data were collected concurrently with PLEs and in larger samples with more statistical power to establish the replicability and robustness of the current findings. Assessing a risk-enriched sample increases the likelihood of some of the participants developing these experiences which have a relatively low base rate; however, it is possible that the relevant signal would not be detectable in a general population sample. While noting this limitation, the authors do expect the current analyses can be generalized somewhat beyond the current type of sample given that experiences such as exposure to domestic violence which were used as criteria in the risk-enrichment of the sample do occur in the general population and are likely linked with mechanisms of developing psychopathology (e.g., Read, 2014). Another important line of investigation in regard to specificity is in differentiation of this putative neurodevelopmental phenotype of social dysfunction as vulnerability to psychosis proneness relative to the social atypicalities of ASD. Study designs which enrich for familial psychosis risk and familial ASD risk would be fruitful.

While the idea that vulnerability to psychosis could be detected at a very young age is intriguing, these findings are but the first of a necessary program of research for delineating a comprehensive understanding of malleable processes that can be harnessed towards psychosis prevention in high-risk young children. Several future directions are identified by this analysis. First, it will be critical to examine whether the relationships between early responsive behavior, positive emotion, negative emotion, and eye gaze observed here are also observed in other samples. In addition, this analysis was the first of its kind (to the knowledge of the authors) and reports on a broad, high-level holistic coding as an initial step into exploring whether early differences in dyadic interactions might be predictive of psychosis proneness. As such, future analyses which delve into the specifics of the differences observed – for example, operationalizing and parsing apart component parts of responsive behavior – will be necessary to determine mechanisms driving social dysfunction and to provide more nuanced assessment of the degree to which dyadic interactions are altered from a typical trajectory. In addition, pursuing the viability of computational measures of behavior (e.g., algorithms quantifying positive or negative emotion, eye-tracking) for similar analyses would increase the feasibility of extending these findings by reducing the labor required to examine the behaviors. Finally, if the present findings demonstrate robustness in replication, the possibility that social skills training (which has been validated for strengthening functional social skills in young children with ASD; Choque Olsson et al., 2017) may have utility as a low-risk, low-intensity, primary prevention for psychosis risk which could be implemented in early childhood for children with high familial or environmental risk for psychosis should be considered in future research.

Given growing evidence of the continuity of psychosis-spectrum experiences (Polanczyk et al., 2010; van Os et al., 2009) and neurodevelopmental understanding of the unfolding of psychopathology, the present findings hold promise for a “healthier, earlier” approach (Wakschlag et al., 2019) to psychotic disorders and the development of primary prevention-based interventions to be implemented in the vulnerability period.