Jumping to Conclusions and Facial Emotion Recognition in First-Episode Psychosis: Longitudinal Insights from the Gap Follow-Up Study

Abstract

Objective

Psychotic disorders are heterogeneous in their clinical presentation and outcome. While early research focused on poor prognoses in schizophrenia, recent longitudinal studies tracking first-episode psychosis (FEP) have identified more favourable outcome trajectories. This study investigates the stability and predictive value of metacognitive and social cognitive impairments—Jumping to Conclusions (JTC) bias and Facial Emotion Recognition (FER) deficits—as intermediate phenotypes of psychosis over a 5-year follow-up period.

Method

A total of 134 FEP patients and 105 population-based controls from the GAP and EU-GEI follow-up study in London were reassessed after an average of 4.8 years. JTC was measured using the 60:40 Beads task, while FER was assessed through the Degraded Facial Affect Recognition (DFAR) task. Clinical, functional, and social outcomes—including hospital admissions, symptom severity, and employment status—were evaluated. Mixed models and regression modeling examined the stability of these cognitive traits and their association with long-term outcomes.

Results

JTC and FER impairments remain stable over time, supporting their classification as intermediate phenotypes. However, neither JTC nor FER was associated with clinical outcomes (hospitalization rates, symptom severity) or social functioning (employment, independent living, relationships). A weak correlation was found between global FER impairment and negative symptoms at follow-up, but no associations emerged with real-world functional measures. Additionally, while patients demonstrated greater impairments than controls, the differences were more quantitative than qualitative, aligning with the psychosis continuum hypothesis.

Conclusions

These findings demonstrate that JTC and FER are stable in people with psychosis and controls. Therefore, they may serve as important treatment targets for early intervention in psychosis. Future research should integrate the potential role of environmental factors as well as genetic influence to deepen our understanding of cognitive impairments in psychotic disorders.

Introduction

Psychotic disorders are highly heterogenous syndromes regarding both clinical presentation and course of illness. Although early studies tended to focus on prevalent cases with schizophrenia leading to poor outcome and disability (van Os et al., 1997), more recent studies, which followed up incidence cases from their first episode of psychosis, showed that trajectories are more encouraging in terms of outcome and shed light on early predictors of poor outcome (Friis et al., 2015; Morgan et al., 2014; Murray et al., 2020). Metacognition and social cognition impairments have been extensively studied as linked to the liability for psychotic disorder as well as outcome predictors (Green et al., 2019; Halverson et al., 2019). Metacognition refers to the ability to reflect upon and regulate one’s own cognitive processes, including the capacity to assess the accuracy of one’s beliefs and integrate information from oneself and others for problem-solving (Burton, 2021; Lysaker & Dimaggio, 2014). Social cognition, on the other hand, encompasses the mental operations underlying social interactions, such as emotion perception, attributional style, and theory of mind (Green et al., 2008).

In this context, the jumping to conclusions (JTC) reasoning bias is considered a metacognitive impairment, as it reflects a tendency to make hasty decisions based on limited evidence, often accompanied by reduced awareness of one's own fallibility and overconfidence in errors (Moritz & Woodward, 2006; Ross et al., 2015). Similarly, deficits in facial emotion recognition (FER) fall under social cognition impairments, as they affect the ability to accurately perceive and interpret emotional expressions in others, which is crucial for effective interpersonal interactions (Kettle & Allen, 2019; Pinkham et al., 2014). Previous studies on jumping to conclusions (JTC) reasoning bias and facial emotion recognition (FER) impairments investigated their role as potential intermediate phenotype for psychosis in a sample of FEP patients compared to population controls (Tripoli et al., 2021, 2022). Stability over time is a defining criterion of intermediate phenotypes (Gottesman & Gould, 2003), and their role in predicting outcomes warrants further investigation.

Jumping to conclusions

The stability of jumping to conclusions in early onset and chronic patients with psychosis has been investigated. Previous longitudinal studies of only several weeks showed more unstable results of JTC (Menon et al., 2008; Woodward et al., 2009) but these effects could be biased by practice considering the short gap between tests. Longitudinal studies of almost 1-year-follow-up indicated that JTC can remain relatively stable over this period (Catalan et al., 2015; Dudley et al., 2013; Falcone et al., 2015a). However, longer investigations of stability with larger sample are required. Only two studies to date have found that JTC predicts independently social and clinical outcome (Andreou et al., 2014; Rodriguez et al., 2019). Andreou (Andreou et al., 2014) followed up 79 patients with schizophrenia and schizoaffective disorder with only a minority of 20 patients experiencing their first admission. They found that changes in the JTC task decision threshold could explain vocational improvements at 6-month-follow-up irrespective of psychopathology or cognitive changes. In a previous report on 123 patients admitted to the hospital at FEP, we found that those who were dichotomised as ‘jumpers’ at baseline had worse clinical outcome at 4-year follow-up, including more days of admissions, greater need for compulsory admission and police intervention, even when controlling for baseline IQ (Rodriguez et al., 2019).

Facial emotion recognition

Research on facial emotion recognition in psychotic disorders has investigated whether impairment remains stable over time. Findings from cross-sectional studies comparing patients at an early stage of illness to patients at later stage and controls are mixed. Kucharska-Pietura (Kucharska-Pietura et al., 2005) reported that FER impairment was greater in patients with chronic schizophrenia than at first-episode or recent onset, inferring that the deficit worsens as the illness progresses. On the other hand, Comparelli (Comparelli et al., 2013) and Green (Green et al., 2011) found no differences on emotion recognition performance at FEP, multi episode schizophrenia, and individuals at ultra-high risk. This pattern of stability was also corroborated by longitudinal studies. Addington and colleagues’ study found minimal progression of FER alterations between prodromal and post-onset of psychosis, therefore suggesting FER impairment may be a vulnerability trait (Addington et al., 2012). Longitudinal stability of social cognition impairments including emotion processing was also described in first-episode schizophrenia at 12-month-follow-up (Horan et al., 2011). Similarly, Maat (Maat et al., 2015) reported that FER impairment remains stable over a three-year follow-up but may fluctuate with symptom severity. Antipsychotic treatment on FER remains unclear. A study found that FER impairment persisted after 30-days low-dose haloperidol treatment, despite symptom remission (Bediou et al., 2007). Another recent study also found no improvement of FER scores post-treatment with clozapine and therefore proposed that FER impairment is a long-term feature of schizophrenia (Gica et al., 2019). Social cognition might be considered a stronger predictor for functional outcome than neurocognition (Fett et al., 2011; Green et al., 2019; Halverson et al., 2019). Specifically, FER impairment has been linked to poor community and social functioning (Addington et al., 2006), including reduced ability to live independently and function at work (Kee et al., 2003) as well as a decreased interpersonal skill (Pinkham & Penn, 2006).

While prior studies have explored JTC and FER in psychosis, significant limitations remain. Many existing studies are cross-sectional or have short follow-up periods, limiting conclusions about the long-term stability of these cognitive biases. Additionally, few have examined their simultaneous predictive value for a range of outcomes—social, functional, and clinical—within the same sample, especially in early psychosis. Addressing these questions is essential to evaluate whether JTC and FER function as intermediate phenotypes and to inform interventions targeting functional recovery.

This study aims to investigate the long-term role of jumping to conclusions and facial emotion recognition in individuals with first-episode psychosis, compared to population controls. By examining these constructs in the same cohort across an extended follow-up period, our goal is to better understand their potential as stable intermediate phenotypes and predictors of key clinical and functional outcomes.

Specifically, our aims are:

• - to explore whether variation in social, functional, and clinical outcome can partly be explained by baseline JTC and FER;

• - to examine JTC and FER stability over the follow-up period.

Our hypotheses are:

• - JTC and worse FER at baseline will be associated with worse social, functional, and clinical outcome at follow-up in FEP;

• - JTC and FER will be steady over the follow-up period in both groups, supporting their candidacy as intermediate phenotypes.

Methods

Study Design

This follow-up study was part of the Biological Phenotypes, Environment, Genes and Psychosis Outcome (GAP follow up), conducted from 2016 to 2019 at the Institute of Psychiatry, Psychology, and Neuroscience, King’s College London and included 295 First Episode of Psychosis (FEP) patients and 105 controls who were followed up on average 4.8 years after baseline assessment. Participants were recruited as part of two studies conducted using the same methodology in different time frames, i.e. the Genetics and Psychosis (GAP) study occurred between 2004 and 2010 (N=410 cases; N=370 population-based controls) (Murray et al., 2020); and the London subsample of the EU-GEI study (Gayer-Anderson et al., 2020) occurred between 2010 and 2014 (N=201 cases and N=230 population-based controls). At baseline, all patients aged 18 to 65 years presenting with a first episode of psychosis to adult inpatient units in Lambeth, Southwark, and Croydon within the South London and Maudsley NHS Foundation Trust were invited to participate. Controls were recruited from the local population using a combination of random and quota sampling to represent local age, sex, and ethnic composition. Recruitment methods included random sampling from postal address lists, stratified random sampling from General Practitioner (GP) registers, and outreach through online and print advertisements as well as leaflet distribution in public spaces. Cases and controls who completed the baseline consent form and agreed to be recalled were approached via phone call, email, and post mail and asked to take part to the follow up study. All participants provided informed written consent, and ethical permission was obtained from the Institute of Psychiatry Research Ethics Committee (Ethics ref 17/ NI/0011 date 17/1/2017).

Participants

Patients included at baseline were aged 18–65 years and presented for the first time with psychotic symptoms to the psychiatric services of the South London and Maudsley (SLaM) NHS Foundation Trust, London, United Kingdom. All individuals with a suspected first episode of psychosis were potentially eligible. Inclusion criteria were: (a) at least one positive psychotic symptom lasting ≥1 day or two negative symptoms for ≥6 months during the study period; (b) age 18–64 years; and (c) residence in the catchment area at first presentation. Exclusion criteria included: (a) prior contact with specialist mental health services for psychosis outside the study period; (b) psychotic symptoms due to an organic cause (ICD-10: F09); (c) symptoms due to acute intoxication (F1X.5); (d) severe intellectual disability (IQ < 50 or F70–F79). A diagnosis of psychotic disorder (ICD-10; codes F20–F29 and F30–F33) was subsequently confirmed using the OPerational CRITeria (OPCRIT) system, which is scale composed of 90 items, including 60 covering psychopathology (McGuffin et al., 1991). All subjects who were assessed for jumping to conclusions and facial emotion recognition at baseline were included in the present study. Among these, a subsample was reassessed for neuropsychology (including measures for JTC and FER) and included in the analysis for repeated measures (see figure 1).

Figure 1.

Follow-up flowchart

Note. IQ=Intelligence Quotient, JTC=Jumping To Conclusions, FER=Facial Emotion Recognition.

Procedure

Data were collected through face-to-face interviews and assessments, covering repeat of the measures carried out at baseline (socio-demographic and clinical characteristics, functional data, cognitive, metacognitive and social cognitive profile) by trained researchers.

Measures

Wais III Shortened Version

The short form of the WAIS III (Velthorst et al., 2013) was administered as an indicator of general cognitive ability (IQ). This clinician-administered test includes the following subtests: Information (verbal comprehension), Block Design (reasoning and problem solving), Arithmetic (working memory) and Digit symbol-coding (processing speed). The subtests generate continuous data and together provide a reliable estimate of overall cognitive functioning.

Degraded Facial Affect Recognition (DFAR) task

The Degraded Facial Affect Recognition (DFAR) task (van ’t Wout et al., 2004) is a computerized behavioral task used to measure emotional face recognition in 64 degraded photographs of four different actors (two females, and two males) representing four emotions: angry, fearful, happy, and neutral. Subjects were asked to indicate the expression of each face by pressing a button. Continuous variables generated by DFAR performance were the number of correctly recognised: total (DFAR total), neutral (DFAR neutral), happy (DFAR happy), frightened (DFAR frightened), and angry (DFAR angry) facial expressions. This task is performance-based and does not require an interviewer.

The Benton Facial Affect Recognition test (BFRT) short form

The Benton Facial Affect Recognition test (BFRT) short form (Benton & Van Allen, 1968) was used to account for general facial recognition ability in DFAR analysis. This is a standardised, clinician-administered test comprising 13 items, each requiring participants to match unfamiliar faces. It generates a total score that was employed a continuous variable.

The Beads task

The 60:40 Beads task (Garety et al., 1991; Huq et al., 1988) was used to assess the JTC, participants are shown two jars containing red and blue beads that differ based on beads ratio 60:40. The jars are than hidden and participants are shown a series of beads (up to 20) that are drawn one at time from one of the two jars to decide which one was selected by the experimenter. The draws-to-decision (DTD) requested constitute the outcome variable in a fashion that the lower the more JTC. This task is performance-based and does not require an interviewer.

Medical Research Council (MRC) Sociodemographic Schedule

Detailed information about age, gender, self-reported ethnicity, accommodation, relationship, and employment status, level of education was collected from cases and controls using the Medical Research Council (MRC) Sociodemographic Schedule (Mallett, 1997), a structured interview conducted by trained researchers to collect nominal and ordinal data. It was used at baseline and follow-up to derive social outcome based on accommodation, employment, relationship, and education status.

OPerational CRITeria (OPCRIT)

The OPerational CRITeria (OPCRIT) system was used to assess the different aspects of psychopathology in cases. This clinician-administered diagnostic tool includes a checklist of psychopathology items and algorithms for obtaining diagnoses according to different nosographic criteria developed by McGuffin et al. (McGuffin et al., 1991). Item response modelling was previously used to develop a bi-factor model composed of general and specific dimensions of psychotic symptoms computed as continuous variables (positive, negative, disorganisation, mania, and depression) (Quattrone et al., 2019). It was administered to cases at baseline and follow-up.

Community Assessment of Psychic Experience (CAPE)

The Community Assessment of Psychic Experience (CAPE) is a 42-item self-report measurement of lifetime psychotic experiences in controls. Previous factor analyses on the CAPE showed a three-factor structure of positive, negative and depressive dimensions (Stefanis et al., 2002), generating continuous variables. It was administered to controls at baseline and follow-up.

Follow-up Psychiatric and Personal History Schedule (FU-PPHS)

Information about clinical history in cases during the follow-up period was collected using the relevant items of the structured clinical interview Follow-up Psychiatric and Personal History Schedule (FU-PPHS) (Janca & Chandrashekar, 1995). Clinical outcome was evaluated based on information about number of hospital admissions and proportion of time spent in hospital, both continuous variables.

Global Assessment of Functioning (GAF)

Functional outcome was assessed through the Global Assessment of Functioning (GAF) a clinician-rated instrument providing a continuous score ranging from 0 to 100 that measures both overall symptoms severity and disability associated with the illness at follow-up (American Psychiatric Association, 1994; Endicott et al., 1976).

Statistical analysis

Analyses were conducted in STATA 15 (StataCorp, 2017). Preliminary descriptive analyses were performed using chi-square and t-tests to examine the differences in age at baseline and follow-up, years passed between baseline and follow-up, sex, and ethnicity, between cases and controls. McNemar test and paired sample t test were used to examine difference in marital status, employment, and years in education. Paired sample t-test was used to compare performance on IQ, DFAR, and DTD between the two time points within the two groups.

Associations of jumping to conclusions and facial emotion recognition with outcome

To test the association of jumping to conclusions and facial emotion recognition with clinical outcome, we used ordinal and linear regressions testing DTD (N=134) and DFAR (N=77) scores at baseline as predictors for number of admissions (0=no other admission after the first, 1=up to 2 admissions after the first, 2=over 2 admissions) and proportion of time in hospital from baseline to follow up as outcome variables, considering baseline IQ (for JTC) and BFRT (for DFAR), age at baseline, sex and ethnicity as covariates. Linear regression models were also used to test for functional outcome prediction considering GAF symptoms and disability scores at follow-up as response variables. To explore associations of baseline DTD and DFAR scores with symptoms and psychotic-like experiences at follow-up with symptom dimensions derived from OPCRIT in cases and CAPE dimensions in controls (DTD: N=91; DFAR: N=75), we estimated Pearson’s correlation coefficients. To explore social outcome, new variables were created for both cases and controls considering changes in marital, employment, and living status between baseline and follow-up coded as 0=steady in relationship/employed/ independent living, 1=change for better, 2=steady not in relationship/unemployed/not independent; 3=change for worse. Multinomial logistic regressions of baseline DTD (N=225) and DFAR (N=152) on marital status, employment, and living arrangement changes were then estimated adjusting for case/control status, baseline IQ (for JTC) and BFRT (for DFAR), sex, age at follow up, and ethnicity.

Stability emotion recognition of jumping over to conclusions the follow-and up facial

Finally, to test the stability of jumping to conclusions (cases=64, controls=75) and facial emotion recognition (N=54, controls=75) over the years, intra-class correlation separated for cases and controls were performed. To account for group and time effects, repeated measures mixed model analysis by case control status interaction terms was performed, and baseline IQ (for DTD), BFRT score (for DFAR scores), sex, age at follow up, and ethnicity as covariates.

Results

Sample characteristics

Sociodemographic characteristics of patients at FEP and controls who were included in this study are summarised in tables 1 and 2. Controls were older and more frequently white than patients, whereas higher percentages of males were reported in patients’ group than controls (table 1).

Table 1.

Sociodemographic characteristics between patients and controls

	Patients N=134	Controls N=91	Df	Test Statistics	p value
Age at baseline (mean; sd)	29.5 (9.1)	39.1 (13.5)	223	T=6.3	<0.001
Age at follow up (mean; sd)	33.9 (9.4)	44 (14.3)	223	T=6.4	<0.001
Years passed from baseline assessment	4.4 (1.9)	5.4 (1.2)	222	T= 4.4	<0.001
(mean; sd)
Sex (male %; N)	61.2 (82)	46.2 (42)	1	Chi2=4.9	0.026
Ethnicity (%; N)
White	36.6 (49)	64.8 (59)	3	Chi2=18	<0.001
Black	52.2 (70)	26.4 (24)
Other (Mixed, Asian)	11.2 (15)	8.8 (8)

Table 2.

Sociodemographic characteristics between baseline and follow-up

	Patients=134			Controls=91
	Baseline	Follow-up	p value	Baseline	Follow-up	p value
Marital status (%; N)
Married/steady relationship	20.2 (27)	29.1 (39)	0.0771a	62.6 (57)	71.4 (65)	0.0606a
Single/separated/widowed	76.1 (102)	68.7 (92)		34.1 (31)	25.3 (23)
Missing Living independently (%; N)	3.7 (5)	2.2 (3)		3.3 (3)	3.3 (3)
Yes	54.5 (73)	57.5 (77)	0.0184a	80.2 (73)	91.2 (83)	0.0082a
No	40.3 (54)	31.3 (42)		16.5 (15)	8.8 (8)
Missing	5.2 (7)	11.2 (15)		3.3 (3)	-
Employment (%; N)
Employee (full time/part time/self-employed)	21.7 (29)	29.1 (39)	0.0858a	64.8 (59)	75.8 (69)	0.0076a
Unemployed/Economically inactive/student	74.6 (100)	68.7 (92)		31.9 (29)	20.9 (19)
Missing	3.7 (5)	2.2 (3)		3.3 (3)	3.3 (3)
Years of Education (mean; sd)	13.7 (2.9)	14.8 (3.4)	<0.001b	15.9 (3.1)	16.8 (3.5)	<0.001b

Note. ^a McNemar test; ^b Paired sample t test

Looking at sociodemographic differences between the two time points (table 2), controls changed their relationship, employment, and living status for the better, while patients showed a steadier pattern over time being more likely to be either single or separated or widowed and not employed. Nonetheless, patients showed a slight improvement in their living arrangements, gaining more independence. Furthermore, there was a small increase in the number of years spent in education between baseline and follow-up in both patients and controls.

Jumping to conclusions and outcome

Ordinal and linear regressions investigating prediction for clinical and functional outcome showed that the number of beads drawn at baseline was not associated with higher hospitalisation, the proportion of time spent in hospital, or GAF scores (table 3). When analysing social outcome using multinomial logistic regression models with no covariates, DTD at baseline was modestly associated with steady patterns in living, marital, and employment status (table 3). When adjusting for covariates, the effects (RRR) did not change, but the confidence intervals were wider. The number of breads drawn at baseline was not associated with symptom dimension scores (Pearson’s correlation coefficients: general r=-0.1; positive r=0.1; negative r=0.03; disorganised r=-0.02; depressive r=-0.01; mania r=-0.01) in patients nor with CAPE scores (Pearson’s correlation coefficients: total r=0.1; positive r=-0.04; negative r=0.1; depressive r=0.2) in controls at follow-up (p>0.05).

Table 3.

Predicting effects of Draws-To-Decisions (DTD) on long-term clinical, functional, and social outcome

Clinical outcome	OR (95% CI) Unadj.	OR (95% CI) Adj.
Number of Admissions	1.1 (0 .9 - 1.1)	1.1 (0 .9 - 1.2)
	B (95% CI) Unadj.	B (95% CI) Adj.
Proportion of time in hospital	-0.1 (-0.3 to 0.2)	-0.1 (-0.6 to 0.3)
Functional outcome
GAF symptom score	-0.3 (-0.9 to 0.3)	-0.6 (-1.5 to 0.2)
GAF disability score	-0.5 (-1.2 to 0.2)	-0.8 (-1.8 to 0.1)
Social outcome	RRR (95% CI) Unadj.	RRR (95% CI) Adj.
Living independently
Became independent	0.9 (0.9 – 1.1)	1 (0.9 – 1.2)
Steady not independent	0.9 (0.8 – 0.9)*	0.9 (0.8 - 1)
Did not become independent	0.8 (0.5 – 1.2)	0.7 (0.4 – 1.1)
Marital Status Change
Steady married/Engaged	1.1 (1.1 – 1.2)***	1 (0.9 – 1.2)
Got married/Engaged	1.1 (0.9 – 1.2)	1 (0.9 – 1.2)
Got separated/Divorced	1 (0.9 – 1.1)	1 (0.9 – 1.2)
Employment Change
Steady employed	1.1 (1.1 – 1.2)**	1.1 (0.9 – 1.2)
Became employed	1 (0.9 – 1.1)	1.1 (0.9 – 1.2)
Became unemployed	1 (0.9 – 1.1)	0.9 (0.8 – 1.2)

Note. ***p<0.001, **p=0.001, *p=0.007

Facial Emotion Recognition and outcome

Ordinal, linear, and multinomial logistic regression models estimating clinical, functional, and social outcomes’ predictions by DFAR scores resulted in no associations (table 4). DFAR scores were mostly weakly correlated with symptom dimensions (Pearson’s correlation coefficients ranged from -0.02 to 0.3). A statistically significant negative correlation was found between negative symptoms and overall DFAR score (r = -0.3, p = 0.0433).

Table 4.

Predicting effects of DFAR scores on long-term clinical and social outcome

		DFAR overall	DFAR fearful	DFAR angry
Clinical outcome		OR (95% CI)	OR (95% CI)	OR (95% CI)
	Number of Admissions	0.9 (0 .9 - 1)	0.9 (0 .9 - 1)	0.9 (0.9 – 1)
		B (95% CI)	B (95% CI)	B (95% CI)
	Proportion of time in hospital	-0.1 (-0.2 to 0.1)	-0.04 (-0.1 to 0.1)	-0.02 (-0.1 to 0.1)
Functional outcome
	GAF symptom score	-0.2 (-0.4 to 0.1)	-0.1 (-0.3 to 0.1)	-0.03 (-0.2 to 0.1)
	GAF symptom score	-0.1 (-0.4 to 0.2)	-0.01 (-0.2 to 0.2)	0.01 (-0.2 to 0.2)
Social outcome		RRR (95% CI)	RRR (95% CI)	RRR (95% CI)
Living independently
	Became independent	1 (0.9 – 1.1)	1 (0.9 – 1.1)	1 (0 .9 - 1)
	Steady not independent	1 (0.9 – 1.1)	1 (0.9 – 1.1)	1 (0 .9 - 1)
	Did not become independent	1.1 (0.9 – 1.4)	1.1 (0.9 – 1.3)	1 (0 .9 – 1.2)
Marital Status Change
	Steady married/Engaged	0.9 (0.9 – 1)	0.9 (0.9 – 1)	1 (0 .9 - 1)
	Got married/Engaged	0.9 (0.9 – 1)	0.9 (0.9 – 1)	1 (0 .9 - 1)
	Got separated/Divorced	0.9 (0.9 – 1.1)	0.9 (0.9 – 1)	1 (0 .9 - 1)
Employment Change
	Became employed	1 (0.9 – 1.1)	1 (0.9 – 1)	1 (0 .9 - 1)
	Steady unemployed	0.9 (0.9 – 1)	0.9 (0.9 – 1)	1 (0 .9 - 1)
	Became unemployed	1.1 (0.9 – 1.2)	0.9 (0.9 – 1)	1 (0.9 – 1.1)

Note. DFAR=Degraded Facial Affect Recognition.

A trend toward significance was observed for negative symptoms and DFAR neutral (r = -0.3, p = 0.0669), while other correlations with negative symptoms were not significant (happy: r = -0.2, p = 0.1787; fearful: r = -0.2, p = 0.2410; angry: r = -0.2, p = 0.1403).

Regarding affective symptoms, a significant positive association emerged between DFAR neutral and depressive symptoms (r = 0.3, p = 0.0466). Other associations between DFAR and mania or depression were not statistically significant (e.g., DFAR happy with mania: r = 0.3, p = 0.1055; with depression: r = 0.2, p = 0.1932).

In controls, significant positive correlations were observed between DFAR overall and CAPE total scores (r = 0.3, p = 0.0374), and between DFAR fearful and CAPE depression score (r = 0.3, p = 0.0071).

IQ, Emotion Jumping Recognition to Conclusions, over time and Facial

Patients and controls who were assessed for IQ, jumping to conclusions, and facial emotion recognition were reassessed for the same measures at follow-up to evaluate their course over time (figure 2a, figure 2b). IQ scores improved over time in both groups, although not significantly in patients [controls: 104.2 (SD 18.3) vs. 108.1 (SD 19.6), t=-3.8, p= 0.0003; patients: 91.1 (SD 20.5) vs. 92.3 (SD 18.9), t=-0.8, p= 0.4492]. Mean number of beads requested did not differ between baseline and follow-up in both patients [3.7 (SD 3.8) vs. 3.3 (SD 3.3), t=0.6, p=0.5510] and controls [6.2 (SD 4.5) vs. 6.6 (SD 5.4), t=-0.5, p=0.5886], as well as DFAR scores in patients [overall: 76.1 (SD 15.9) vs. 71.6 (SD 17.5)], t=0.2, p=0.8695; neutral: 76.9 (SD 19.7) vs. 75.6 (SD 23.6)], t=0.3, p=0.7657; happy: 87.3 (SD 17.3) vs. 88.3 (SD 18.6), t=-0.3, p=0.7886; fearful: 50.6 (25.2) vs. 52.6 (SD 20.9), t=-0.5,p=0.6392; angry: 73.6 (SD 23.4) vs. 69.9 (SD 26.2), t=0.8, p=0.4121].

Figure 2a.

IQ and Draws-to-decision (DTD at baseline vs. follow-up within cases and controls

Note. IQ=Intelligence Quotient, DTD=Draws-to-decision

Figure 2b.

DFAR scores at baseline vs. follow-up within cases and controls

Note. DFAR=Degraded Facial Affect Recognition

Controls’ performance declined only for DFAR happy [92.9 (SD 8) vs. 90.6 (SD 8.4), t=2.2, p=0.0327], while the other scores did not differ notably over time [overall: 76.4 (SD 8.4) vs. 75.6 (SD 8.7), t=0.7, p=0.4597; neutral: 83.2 (SD 12.4) vs. 86.1 (SD 12.3), t=-1.7, p=0.0974; fearful: 60.9 (SD 16.9) vs. 58.3 (SD 19.8), t=1.1, p=0.2946; angry: 68.6 (SD 20.4) vs. 67.4 (SD 18.2), t=0.5, p=0.6298].

Stability of jumping to conclusions over time

To first examine the stability of data gathering between baseline and follow-up and to investigate whether the participants would show a consistent pattern of data gathering to make a decision (Supplementary Figure S1a, S1b), intraclass correlation coefficients (ICC) were calculated. For the whole sample, ICC = 0.4, F = 2.5, p<0.001; for patients’ group, ICC = 0.3, F = 1.9, p=0.011; for controls’ group, ICC=0.4, F=2.3, p<0.001.

Mixed model linear regressions estimated an effect of group on DTD over time (cases: B=-2.1, 95% CI -3.7 to -0.5, p= 0.009) but no effect of time (B=0.2, 95% CI -1 to 1.4, p= 0.765) nor groupXtime (B=-1, 95% CI -2.7 to 0.8, p= 0.272). When adding baseline IQ to the model, the effect of group decreased cases: B=-1.1, 95% CI -2.7 to 0.4, p= 0.141), and a small effect of IQ was detected (B=0.1, 95% CI 0.1 to 0.2, p< 0.001). Predictive margins for DTD across the two time points were calculated and illustrated in figure 3.

Figure 3.

Predictive margins with 95% CIs for DTD over time

Note. This marginplot describes, on average, a steady predicted number of Draws-to-Decision (y-axis) requested across the two time points (x-axis). The figure also shows that patients (red line) requested remarkably less beads over time than controls (blue lines). Error bars: 95% confidence intervals (CIs). DTD=Draws-to-decisions.

Stability time of Facial Emotion Recognition over

As per DTD, stability of emotion recognition between baseline and follow-up and consistency of individual DFAR scores across the two time points (Supplementary Figure S2a, S2b, S2c, S2d, S2e) was examined through calculations of ICCs. DFAR total showed moderate and significant stability in the whole sample (ICC = 0.4, F = 2.2, p<0.001) in patients (ICC = 0.3, F = 1.9, p=0.025), and in controls (ICC=0.5, F=2.7, p<0.001). DFAR neutral was significantly stable in the whole sample in the whole sample (ICC = 0.3, F = 1.9, p<0.001) and in controls (ICC=0.3, F=1.9, p=0.003), but did not reach significance in patients (ICC = 0.3, F = 1.7, p= 0.063). DFAR happy showed significant stability in the whole sample (ICC = 0.3, F = 1.8, p=0.001) and in controls (ICC=0.4, F=2.3, p<0.001) but not in patients (ICC = 0.2, F = 1.6, p= 0.072). DFAR fearful showed significant stability across all groups: whole sample (ICC = 0.4, F = 2.5, p<0.001), patients (ICC = 0.4, F = 2.3, p= 0.006), and controls (ICC=0.4, F=2.4, p<0.001). DFAR angry was also significantly stable across the whole sample (ICC = 0.4, F = 2.5, p<0.001), in patients (ICC = 0.4, F = 2.2, p=0.008), and in controls (ICC=0.5, F=2.7, p<0.001).

Mixed model linear regressions (figure 4a, 4b, 4c, 4d, 4e) estimated an effect of group on DFAR total over time (cases: B=-5.4, 95% CI -10.6 to -0.2, p= 0.042) and DFAR fearful (cases: B=-9.9, 95% CI -19.3 to -0.6, p= 0.036), but no effect of time (DFAR total: B=-0.8, 95% CI -4.2 to 2.6, p=0.631; DFAR fearful: B=-2.3, 95% CI -7.7 to 3.1, p=0.402) nor groupXtime (DFAR total: B=1, 95% CI -4.7 to 6.8, p= 0.732; DFAR fearful: B=5.5, 95% CI -3.7 to 14.7, p= 0.239). Generally, the effect of group was associated with a decline in performance on all the DFAR scores yet with wide 95% CI [DFAR neutral (cases: B=-6.1, 95% CI -13.3 to 1.2, p= 0.103; time: B=2.9, 95% CI -1.8 to 7.8, p=0.225; groupXtime: B=-4.9, 95% CI -13.1 to 3.2, p= 0.237); DFAR happy (cases: B=-4.6, 95% CI -10.1 to 0.9, p= 0.105; time: B=-2.5, 95% CI -6.2 to 1.2, p=0.183; groupXtime: B=3.9, 95% CI -2.3 to 10.2, p= 0.219); DFAR angry (cases: B=-1.1, 95% CI -10.1 to 7.8, p=0.802; time: B=-1.5, 95% CI -7.2 to 4.2, p=0.603; groupXtime: B=-0.5, 95% CI -10.1 to 9.2, p= 0.925)].

Figure 4a.

Predictive margins with 95% CIs for DFAR total over time

Figure 4b.

Predictive margins with 95% CIs for DFAR neutral over time

Figure 4c.

Predictive margins with 95% CIs for DFAR happy over time

Figure 4d.

Predictive margins with 95% CIs for DFAR fearful over time

Figure 4e.

Predictive margins with 95% CIs for DFAR angry over time

Note. These marginplots describe predicted DFAR scores (y-axis) across the two time points (x-axis). The figures also show that patients (red line) performance is lower over time than controls (blue line). Error bars: 95% confidence intervals (CIs). DFAR=Degraded Facial Affect Recognition.

Discussion

Studies on psychotic outcome have emphasised a high level of heterogeneity characterising the course of the illness and shed light on the role of factors associated with worse or better outcomes (Lally et al., 2017). This study considered not only clinical outcome in terms of number of admissions, time spent in hospital, symptoms severity, and level of disability, but also social outcome in terms of changes in living arrangement, marital status and employment as real-world indicators of social functioning (Tulloch et al., 2006). While controls typically followed a socially expected pattern of changes over time, becoming married or maintaining a steady relationship and employment at follow-up, the patients’ relational and employment status did not significantly change, and maintained the worse social adjustment, which is typically reported in psychosis.

Jumping to conclusions and outcome

Although jumping to conclusions as a metacognitive bias has been suggested to be relevant to social processing in terms of compromising the accurate processing of social information (Grossman & Bowie, 2020), in our study it was not significantly associated with social outcomes when adjusting for relevant covariates. While unadjusted analyses indicated a modest association between DTD at baseline and stable patterns in living, marital, and employment status, these associations were not statistically significant once covariates were included, and the wide confidence intervals suggest limited reliability of these effects. To our knowledge, only one previous study (Andreou et al., 2014) has investigated this bias as a predictor of later social functioning and found an association of jumping to conclusions with worse vocational status. However, the shorter follow-up assessment at 6 months compared to the present study could have led to a practice effect considering the relatively short gap between the two assessments. Moreover, the sample used in their study was part of a randomised controlled trial on the efficacy of metacognitive training, which could have targeted skills that in turn could have confounded the reported association. In our study, the baseline number of beads was not associated with a specific clinical or sub-clinical psychopathology profile in cases and in controls. In a previous study (Tripoli et al., 2021) jumping to conclusions was found to be associated with subclinical psychotic experiences at FEP, suggesting a role in their formation and maintenance; however, this hypothesis is not supported by the present data. Finally, JTC did not predict the number of admissions nor time spent in the hospital. In contrast, Rodriguez et al. (Rodriguez et al., 2019) reported an effect of JTC in clinical outcome after 4 years from FEP. The present study relied only on information collected face to face, whereas Rodriguez et al. used electronic records, which allowed them to reach a larger sample than the present one.

Facial emotion recognition and outcome

Outcome prediction by social cognitive domains – even independently from neurocognition – was previously reported (Fett et al., 2011; Horan et al., 2011). Facial affect recognition impairments were hypothesised to affect real-world interactions, which in turns might lead to poor social outcome. Nonetheless, the present analysis did not find associations with changes in living, marital, and vocational status. Similarly, one longitudinal study over three years (Simons et al., 2016) and one case-control study (Janssens et al., 2012) did not find any associations between emotion recognition and social functioning in daily life. Those findings might suggest that, although standard laboratory tests are useful to discriminate impaired from normal performance, they might not detect more complex processes underpinning real-world social functioning (Koren et al., 2006). Interestingly, weak associations were detected between poor global ability to recognise facial emotions at baseline and high level of negative symptoms at follow-up. Specific deficits in recognising negative emotions might be more associated with psychotic disorders featured more negative symptoms during the course of the illness and this association was suggested as pathway to worse outcome (Pelletier-Baldelli & Holt, 2019).

Are facial jumping emotion to recognition conclusions stable and over deficits time? in

Previous literature suggested that metacognitive biases and social cognitive impairments are relatively stable following a first episode psychosis. This follow-up study sought to investigate the stability of the jumping to conclusions bias and facial emotion recognition deficits after ~5 years from first clinical admission for psychosis in patients compared to controls. Our results point towards an overall steady pattern of performance over time in both groups. Despite having shown an improved IQ score at follow-up, controls followed the same pattern of performances over time as patients differing only for having achieved better scores. This is in line with the hypothesis of a continuum in contrast with discrete categories of illness and health, where differences between psychotic and not-psychotic are not qualitative but about the amount of impairment (van Os et al., 2008). Irrespective of time, patients tend to show more impairments in facial emotion recognition than controls. This is more apparent for the global ability to recognise emotions and in particular fearful facial expressions. This is in line with previous studies (Barkl et al., 2014; Catalan et al., 2016; Daros et al., 2014; Kuharic et al., 2019; Tripoli et al., 2022), which suggested a specific deficit of recognising negative faces at first episode psychosis. Furthermore, patients tend to make hastier decisions over time than controls. It is noteworthy that when taking into account IQ, caseness is not anymore a significant predictor of jumping to conclusions over time as already reported (Tripoli et al., 2021) when comparing patients and controls at first assessment. Findings suggest a more prominent role of general cognitive ability in this reasoning bias and its maintenance over time than case status.

Strengths and limitations

The present study has to be considered in the context of several strengths and limitations. An important strength is that the follow up included data from controls from the same catchment area as patients, provided the opportunity of comparison for both social and neuropsychological trajectories between FEP and population controls over the same time period. The length of the follow up was long enough to avoid practice effect.

A potential limitation of the study was the limited number of patients with available cognitive and social cognitive data at baseline who were willing to return for a second assessment. This might have affected our power to detect the prediction of outcomes. To address this, we conducted a post-hoc power analysis, which indicated that we had adequate power to detect moderate effects in our primary group comparisons (e.g., group differences in DTD). However, the study was underpowered to detect smaller effects, particularly for group-by-time interactions and clinical outcome predictions. This limitation is reflected in the wide confidence intervals observed in these analyses, so we cannot exclude that the sample size partially explained these null findings. Moreover, since our analyses included only those who accepted testing at both baseline and follow-up, we might have missed the most unwell patients over time, potentially leading to an underestimation of cognitive and social cognitive decline. Nonetheless, an important strength is that this study relied on face-to-face assessments which enabled the collection of repeated measures for the neuropsychological assessment and gave more validity to outcome measures.

Conclusions

There is a large body of literature suggesting that metacognition and social cognition are important factors in predicting future illness trajectories and informing clinical intervention to improve patients’ quality of life. This study focused on the jumping to conclusions bias and facial emotion recognition impairments and found these traits to be more frequent in patients with psychosis compared to controls and to remain stable over 5 years of follow up. These results suggest that these cognitive deficits may reflect enduring traits rather than state-related phenomena. Although our findings did not reveal strong or consistent associations between JTC or FER and later clinical or social outcomes, the observed stability of these deficits over time suggest that they may serve as important treatment targets for cognitive remediation therapies and early interventions in psychosis. Interventions such as metacognitive training or social cognitive remediation could be explored further, particularly in the early stages of psychosis, to determine whether modifying these traits has downstream effects on functioning or symptom progression. Future research should address some of the limitations of the present study, notably the relatively small follow-up sample, which may have limited statistical power. Larger longitudinal cohorts are needed to clarify the predictive value of cognitive deficits and to assess whether these effects are moderated or mediated by other factors such as neurocognitive performance, environmental factors, or genetic influence.