Introduction
There has been an increased effort to include cognition as a clinically relevant dimension for characterizing premorbid trajectories to psychosis (Lam, Raine, & Lee, Reference Lam, Raine and Lee2014; Reichenberg, Reference Reichenberg2005). A recent approach, cluster analysis, delineated three distinct patterns of cognitive functioning in patients diagnosed with psychosis: high, low, and intermediate (Carter, Reference Carter2018), with the latter group showing significant internal variations (Carruthers et al., Reference Carruthers, Van Rheenen, Gurvich, Sumner and Rossell2019); this statistical approach has enhanced the understanding of premorbid adjustment and cognition trajectories. Indeed, while these variables generally presented a stable course over time (Carruthers et al., Reference Carruthers, Van Rheenen, Gurvich, Sumner and Rossell2019), cluster analyses have identified a subgroup of patients at their first episode of psychosis who showed impaired premorbid adjustment and cognitive capacities (Mohn-Haugen et al., Reference Mohn-Haugen, Mohn, Larøi, Teigset, Øie and Rund2022), which deteriorated further prior to the onset of psychosis (Green et al., Reference Green, Girshkin, Kremerskothen, Watkeys and Quidé2020).
Such a premorbid decline might have been associated with exposure to environmental risk factors (Cuesta et al., Reference Cuesta, Sánchez-Torres, Cabrera, Bioque, Merchán-Naranjo, Corripio, González-Pinto, Lobo, Bombín, de la Serna, Sanjuan, Parellada, Saiz-Ruiz and Bernardo2015; Meier et al., Reference Meier, Caspi, Ambler, Harrington, Houts, Keefe, McDonald, Ward, Poulton and Moffitt2012; Mollon & Reichenberg, Reference Mollon and Reichenberg2018; Velthorst et al., Reference Velthorst, Mollon, Murray, de Haan, Germeys, Glahn, Arango, van der Ven, Di Forti, Bernardo, Guloksuz, Delespaul, Mezquida, Amoretti, Bobes, Saiz, García-Portilla, Santos, Jiménez-López and Reichenberg2021) toward suggested mechanisms affecting brain structure, synaptic function, ion channels, glutamate neurotransmission, and inflammatory and immune processes (Tandon et al., Reference Tandon, Nasrallah, Akbarian, Carpenter, DeLisi, Gaebel, Green, Gur, Heckers, Kane, Malaspina, Meyer-Lindenberg, Murray, Owen, Smoller, Yassin and Keshavan2024). It has also been suggested that an underlying genetic susceptibility before the onset of psychosis might have contributed to deterioration, at least in some patients (Mollon & Reichenberg, Reference Mollon and Reichenberg2018; Ohi et al., Reference Ohi, Nishizawa, Sugiyama, Takai, Kuramitsu, Hasegawa, Soda, Kitaichi, Hashimoto, Ikeda and Shioiri2021; Parellada, Gomez-Vallejo, Burdeus, & Arango, Reference Parellada, Gomez-Vallejo, Burdeus and Arango2017). This latter theory was still reminiscent of Kraepelin’s early notion of ‘dementia praecox’, which was grounded on a negative connotation of schizophrenia and suggested an inevitable disease progression with cognitive decline.
In this regard, in a previous study (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023), we replicated the methodology from Dickinson et al. (Reference Dickinson, Zaidman, Giangrande, Eisenberg, Gregory and Berman2020) by combining empirical stratification of premorbid adjustment measures and current cognitive trajectories to cluster patients with psychosis. These studies tested the association between the resulting clusters and polygenic risk scores. Dickinson et al. (Reference Dickinson, Zaidman, Giangrande, Eisenberg, Gregory and Berman2020) identified a cluster of schizophrenia patients with worsening cognition prior to onset, which showed a higher polygenic risk score for schizophrenia (SCZ_PRS). In contrast, our recent study found a lower SCZ_PRS in a cluster of first-episode psychosis (FEP) patients whose premorbid functioning deteriorated before onset (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023). Notably, FEP research has provided a more accurate assessment of cognitive profiles, potentially less affected by social impairment, symptom duration, and antipsychotic treatment than in long-standing psychosis. Moreover, the cognition measurement in chronic schizophrenia patients might have been skewed toward the most severe cases, the so-called Berkson’s fallacy (Maric et al., Reference Maric, Myin-Germeys, Delespaul, de Graaf, Vollebergh and Van Os2004). This bias could be overcome by using transdiagnostic approaches.
Nonetheless, these factor analyses observed the same phenomenon at different levels, that is, premorbid trajectories and cognition characteristics of patients in their FEP and polygenic profiles. A more comprehensive analysis should also consider additional co-occurring phenomena at the premorbid level, referred to as ‘environmental risk factors’. For example, our previous study (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023) reported that, among patients, daily users of high-potency cannabis were more likely to be part of the deteriorating cluster, suggesting that this environmental risk factor often co-occurred with social and cognitive deterioration.
Similar to the polygenic risk scores, Vassos et al. (Reference Vassos, Sham, Kempton, Trotta, Stilo, Gayer-Anderson and Morgan2020) proposed an approach to measure the cumulative effect of a range of environmental factors, building an Environmental Risk Score for psychosis (ERS). This score aggregated relative risks based on the largest meta-analysis of consistently replicated environmental risk factors for psychosis, such as paternal age, urbanization, obstetric complications, childhood adversities, cannabis use, and ethnic minority membership. The present study hypothesized that specific clusters of FEP could have been differentially exposed to environmental risk factors. We used the ERS to compare derived clusters based on combined environmental exposures. In our previous report, we found that cannabis was more frequently used in the deteriorating cluster. Hence, our study also explored the putative differences in single exposures of the environmental risk factors in distinguishing within patient clusters and from controls.
Methods
Participants
FEP patients and population-based controls from the multi-centric EU-GEI study signed an informed consent form after fully explaining the research procedures. The study was carried out in compliance with the Helsinki Declaration and received ethical approval; data were pseudonymized (CORDIS, 2019; EU-GEI, 2009; Gayer-Anderson et al., Reference Gayer-Anderson, Jongsma, Di Forti, Quattrone, Velthorst, de Haan, Selten, Szöke, Llorca, Tortelli, Arango, Bobes, Bernardo, Sanjuán, Santos, Arrojo, Parellada, Tarricone, Berardi and Morgan2020; Jongsma et al., Reference Jongsma, Gayer-Anderson, Lasalvia, Quattrone, Mulè, Szöke, Selten, Turner, Arango, Tarricone, Berardi, Tortelli, Llorca, de Haan, Bobes, Bernardo, Sanjuán, Santos, Arrojo and Kirkbride2018; Leeson et al., Reference Leeson, Sharma, Harrison, Ron, Barnes and Joyce2011; Roser et al., Reference Roser, Allott, Killackey, Farhall and Cotton2015; Uren et al., Reference Uren, Cotton, Killackey, Saling and Allott2017) (Supplementary Methods, recruitment).
The Maudsley environmental risk score
This ERS was a weighted sum of the relevant environmental exposures, using effect sizes extracted from meta-analyses for each risk factor (Vassos et al., Reference Vassos, Sham, Kempton, Trotta, Stilo, Gayer-Anderson and Morgan2020), namely, ethnic minority membership, paternal age, cannabis use, childhood adversities, urbanicity, and obstetric complications. Based on our data’s availability, we could not include urbanicity because it was measured at a site rather than individual level, and obstetric complications were not collected in our sample. To validate the primary analysis, we used the Exposome (Pries et al., Reference Pries, Lage-Castellanos, Delespaul, Kenis, Luykx, Lin, Richards, Akdede, Binbay, Altinyazar, Yalinçetin, Gümüş-Akay, Cihan, Soygür, Ulaş, Cankurtaran, Kaymak, Mihaljevic, Petrovic and Guloksuz2019), built by summing log-odds weighted environmental exposures (0–1) on the EUGEI training sample, such as winter birth, hearing impairment, cannabis use, and cumulative exposure to childhood adversities (6/9 points) (Supplementary Methods, The ERS, the Exposome).
Instruments
The modified version of the Medical Research Council (MRC) socio-demographic scale collected demographic information (Mallett et al., Reference Mallett, Leff, Bhugra, Pang and Zhao2002). Given the multisite design, we could not use the same psychometric test (i.e., a reading test) among countries to assess premorbid IQ. Instead, we used nine scales from the Premorbid Adjustment Scale (PAS) (Brill, Reichenberg, Weiser, & Rabinowitz, Reference Brill, Reichenberg, Weiser and Rabinowitz2008; Cannon-Spoor, Potkin, & Wyatt, Reference Cannon-Spoor, Potkin and Wyatt1982; Rabinowitz, Levine, Brill, & Bromet, Reference Rabinowitz, Levine, Brill and Bromet2007; Shapiro et al., Reference Shapiro, Marenco, Spoor, Egan, Weinberger and Gold2009) examining premorbid social (PSF) and academic functioning (PAF), from childhood to age 11 (<12 years) and in early adolescence (12–16 years). The brief version of the Wechsler Adult Intelligence Scale (WAIS), including information, block design, digit symbol, and arithmetic subtests, estimated current IQ in patients and controls according to the imputation standardized strategy in schizophrenic patients (Missar, Gold, & Goldberg, Reference Missar, Gold and Goldberg1994; Velthorst et al., Reference Velthorst, Levine, Henquet, De Haan, Van Os, Myin-Germeys and Reichenberg2013; Wechsler, Reference Wechsler1981). The EUGEI modified version of the Cannabis Experience Questionnaire (CEQEugei-mv) (Di Forti et al., Reference Di Forti, Quattrone, Freeman, Tripoli, Gayer-Anderson, Quigley, Rodriguez, Jongsma, Ferraro, La Cascia, La Barbera, Tarricone, Berardi, Szöke, Arango, Tortelli, Velthorst, Bernardo, Del-Ben and van der Ven2019), including a section from Composite International Diagnostic Interview (CIDI), collected data on cannabis use, other substances of abuse, and tobacco in the last 12 months. We assessed childhood adversities by using two instruments: the Childhood Trauma Questionnaire (CTQ) (Bernstein, Ahluvalia, Pogge, & Handelsman, Reference Bernstein, Ahluvalia, Pogge and Handelsman1997), a 28-item self-report tool that rates the presence of physical, sexual, and emotional abuse, physical and emotional neglect, and the Childhood Experience of Care and Abuse Questionnaire interview (CECA-Q) (Bifulco, Bernazzani, Moran, & Jacobs, Reference Bifulco, Bernazzani, Moran and Jacobs2005), which additionally included information about loss of parents and bullying before the age of 17, and corroborated CTQ data (Supplementary Methods, instruments). Researchers ensured that patients assessed as soon as they achieved a stable mental state were referred to the pre-onset period. The CEQEugei-mv and PAS interviews were checked by at least one supporting data source (family, clinical notes, and other clinicians).
Cluster derivation
We clustered 802 FEP patients having complete information on PAS and WAIS from the EU-GEI study, based on assessments of premorbid social (PSF) and academic functioning (PAF) in early childhood (PAS < 12 years) and adolescence (PAS 12–16 years), and IQ at the time of onset (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023). To reduce collinearity between PAS < 12 and PAS 12–16 (r = 0.7 in patients for both scales) and focus on potential deterioration, we used PSF < 12 years, PAF < 12 years and change scores between <12 years and 12–16 years for both social (PSFchange) and academic domain (PAFchange), and current IQ as the input variables. Controls were not included and used as a whole reference group. We performed a Two-Step Cluster Analysis procedure in SPSS_version_24. We used the stepwise decrease in log-likelihood as the distance measure for identifying clusters and change in the Bayesian Information Criterion (BIC) to determine the number of clusters to retain (best ratio change of cluster distance at least >1.15) (Liu, Li, Dong, & Wen, Reference Liu, Li, Dong and Wen2013). We ran a 50-subject’ assignment solution by pre-determining the chosen number of clusters with random reordering. Fleiss’s kappa index established the extent of agreement in cluster assignment (Fleiss, Reference Fleiss1971) between the assignment solutions. To validate clusters, we performed repeated-measures ANOVAs to determine whether there were any statistically significant differences in PSF and PAF changes between childhood and early adolescence within the formed clusters, compared to control changes. Then, we modelled an ANCOVA to compare each cluster to controls for IQ and PAS (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023) (Supplementary Methods, cluster derivation).
Statistical analyses
Firstly, we used multinomial logistic regression to estimate the odds of belonging to each patient cluster relative to community controls at baseline for exposure to the ERS, including age, sex, and site-level urbanization as covariates. Next, to see the amount of variance explained by the ERS, we estimated the difference between the model’s pseudo-R-squares (Nagelkerke R2), including the ERS and covariates only. We then repeated case-only analyses by excluding controls and using the high-cognitive-functioning cluster as the reference category. We also repeated the Nagelkerke R2 estimation for the between-cluster variance explained by the ERS. A univariable general linear model tested the ERS differences by each cluster and controls, adjusting by age, sex, and urbanization, including post hoc multiple comparisons between groups, Bonferroni adjusted. Then, we repeated them in a subsample of matched cases and controls (Supplementary Material, sensitivity analyses). Lastly, we run multivariable general linear models, comparing each cluster with controls by different exposures in the ERS, adjusted by age, sex, and urbanization. We repeated these last multivariable models, taking into account the use of tobacco and other illegal drugs and socioeconomic status (Supplementary Material, sensitivity analyses). Finally, the same analyses tested the Exposome as a cumulative score, and then single exposures were included, adjusted by sex, age, ethnicity, and urbanization. Partial eta squared (η2) measured effect sizes.
Results
Clusters of patients
The sample included 1,263 population controls and 802 FEP patients with complete data on our variables of interest (EU-GEI, 2009) (Table 1).
Table 1. Sociodemographic and clinical characteristics of the sample by cases and controls
Legend: SD = standard deviation; df = degree of freedom; edu. = education. * aggregation of categories was made through linear regression.
We identified four transdiagnostic clusters: high-cognitive-functioning, low-cognitive-functioning, and intermediate and deteriorating functioning (Ferraro et al., Reference Ferraro, La Cascia, Quattrone, Sideli, Matranga, Capuccio, Tripoli, Gayer-Anderson, Morgan, Sami, Sham, De Haan, Velthorst, Jongsma, Kirkbride, Rutten, Richards, Roldan, Arango and Di Forti2020). The high-cognitive-functioning cluster (n = 205) displayed high IQ (Mean (M) = 106.1, sd = 14.2), slightly higher than controls’ (M = 102.6, sd = 17.6), and highly stable premorbid academic functioning. However, the premorbid sociability was steadily lower than the controls in the two ages. The low-cognitive-functioning cluster (n = 223) displayed low IQ (M = 73.9, sd = 12.7) and poor childhood premorbid academic functioning, increasing in early adolescence, and a close to average and stable premorbid sociability. The intermediate cluster (n = 224) had a middle IQ (M = 80.8, sd = 11.9) compared to the other two clusters and a low premorbid sociability and academic adjustment; both scores slightly improved during early adolescence. Finally, the deteriorating cluster (n = 150) had the same IQ as the intermediate (M = 80.6, sd = 12.9) and a normal premorbid academic and social adjustment in childhood, comparable to controls. Nonetheless, during early adolescence, significantly deviated from normal functioning, both in premorbid sociability and academic functioning (Ferraro et al., Reference Ferraro, La Cascia, Quattrone, Sideli, Matranga, Capuccio, Tripoli, Gayer-Anderson, Morgan, Sami, Sham, De Haan, Velthorst, Jongsma, Kirkbride, Rutten, Richards, Roldan, Arango and Di Forti2020) (Supplementary Results, clusters of patients). Clusters differed in some sociodemographic and clinical characteristics (Supplementary Tables 1 and 2, Supplementary Figure 1).
Clusters and ERS
In comparing the four clusters of patients with controls, the ERS explained 10.8% of extra variance compared to the covariate-only model, including age, sex, and urbanization (Nagelkerke R2 = 0.209 vs 0.101, respectively) (Δχ2 (4, 16) = 236.5, p = 5.03×10−50); it explained 2.2% of the between-cluster variance (Nagelkerke R2 = 0.090 vs 0.068, respectively) (Δχ2 (3, 9) = 17.7, p = 0.0004) in case-only analysis (parameter estimated and odd ratios in Supplementary Tables 3 and 4).
The ERS (Supplementary Table 5) was higher in the high-cognitive-functioning (β=1.4, 95% CI 0.9 1.9, η2 = 0.017), the intermediate (β=1.9, 95% CI 1.5 2.4, η2 = 0.033), the deteriorating (β=2.8, 95% CI 2.3 3.4, η2 = 0.049), and the low-cognitive-functioning group (β=2.4, 95% CI 1.9 2.8, η2 = 0.049), compared to controls. Post hoc between-clusters comparison showed that the deteriorating (meandifference = 1.4, 95% CI 0.43 2.38) and the low-cognitive-functioning clusters (meandifference = 0.9, 95% CI 0.09 1.85) had a higher ERS than the high-cognitive-functioning group (Figure 1). These results were confirmed even when limiting comparison only to patients’ clusters (Supplementary Table 6). We tested the ERS in a case-control matched sample, by age and gender and the results remained consistent (Supplementary Material, sensitivity analyses).
Figure 1. ERS by clusters of FEP and controls.
Legend: The Y axis represents ERS means and 95% Cis in each cluster of patients and controls. HIGH = high-cognitive-functioning; LOW = low-cognitive-functioning. The grey braces indicate significant differences between clusters of patients. Braces mark significant differences between patient clusters *(meandifference = 0.974, 95% CI 0.09 1.85, p = 0.018); **(meandifference = 1.41, 95% CI 0.43 2.38, p = 0.0004).
Clusters by single environmental exposures
When we analyzed each environmental exposure separately, compared to the model with covariates only (Nagelkerke R2 = 0.101), cannabis use explained 5% of the total variance (Nagelkerke R2 = 0.151), childhood adversities the 5.2% (Nagelkerke R2 = 0.153) and ethnic minority status the 3.7% (Nagelkerke R2 = 0.138).
Patients as a whole were more likely to belong to an ethnic minority (F(4, 2058) = 20.5, η2 = 0.038) and had higher exposure to cannabis use (F(4, 2058) = 26.4, η2 = 0.049) and childhood adversities (F(4, 2058) = 24.5, η2 = 0.046) than controls. Paternal age did not differentiate groups (p = 0.619) (Table 2).
Table 2. Ethnic minority, paternal age, childhood adversities, and cannabis use comparisons between clusters and controls and parameter estimates
Legend: High = high-cognitive-functioning; Low = low-cognitive-functioning.
Post hoc multiple comparisons showed that the deteriorating cluster had been more exposed to cannabis use than the intermediate cluster (meandifference = 0.48, 95% CI 0.49 0.91).
The high-cognitive-functioning cluster did not show substantial differences with controls in the likelihood of belonging to an ethnic minority (meandifference = -0.24, 95% CI −0.60 0.12). It included fewer people from ethnic minorities than the deteriorating cluster (meandifference = −0.77, 95% CI −1.29–0.24). Finally, there were no appreciable differences between clusters in exposure to childhood adversities (all p > 0.05) (Figure 2, Supplementary Table 7). These results remained consistent in cluster-only comparisons and when considering other drug and tobacco use or socio-economic status (Supplementary Tables 8 and 9 and Supplementary Figure 2).
Figure 2. Ethnic minority, paternal age, childhood adversities, and cannabis use by clusters of FEP and controls.
Legend: The Y axis represents means, standardized around the controls’ mean, and 95% CIs in each cluster of patients and controls. HIGH = high-cognitive-functioning; LOW = low-cognitive-functioning. Braces mark significant differences between the deteriorating and the intermediate (green brace) *(meandifference = 0.482, 95% CI 0.49 0.91, p = 0.018) and the high-cognitive-functioning cluster (red brace) **(meandifference = 0.771, 95% CI 0.24 1.29, p = 0.0003).
Validation analyses via Exposome
Measures to calculate Exposome score were available in 1,104 population controls and 641 FEP patients. As compared with the covariates model (Nagelkerke R2 = 0.111), its introduction in the cluster-control analysis explained 9.2% of additional variance (Nagelkerke R2 = 0.203) (Δχ2 (4, 20) = 169.2, p = 1.51×10−35), and 2% when tested among patients only (Nagelkerke R2 = 0.069) as compared with the covariate-only model (Nagelkerke R2 = 0.071).
The Exposome yielded identical results to the ERS, with controls having lower scores than all clusters of patients [F(4, 1738) = 40.7, p = 1.217×10−32] and the deteriorating group reporting the highest score, significantly higher than the high-cognitive-functioning cluster (mean_difference = 0.60, 95% CI 0.35 1.17, p = 0.029) (Supplementary Table 5 and Supplementary Figure 3).
Among Exposome risk factors, childhood adversities (F(4, 1738) = 33.4, p = 6.3×10−27) and cannabis use (F(4, 1738) = 30.9, p = 5.8×10−25) distinguished all patients from controls, the latter also highlighting between-cluster differences. Winter birth (p = 0.240) and hearing impairment (p = 0.431) were insignificant (Supplementary Table 10).
Discussion
Main findings
This study showed for the first time that individuals who were highly exposed to environmental risk factors were more likely to have lower cognitive functioning at the onset of psychosis and, more importantly, to present with deteriorating premorbid functioning. This feature distinguished them from those in a cluster with high-cognitive-functioning.
Moreover, the deteriorating cluster had a higher prevalence of ethnic minority membership and cannabis use than the other clusters. This report, coupled with the previous finding of its lower polygenic predisposition to psychosis (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023), could suggest the role of the environment in leading to premorbid social and cognitive deterioration in psychosis.
Lastly, among a range of socio-environmental risk factors, childhood adversities, cannabis use, and ethnic minority membership were the most significant in distinguishing patients from the control group, highlighting their importance when collecting information on environmental risk for psychosis.
The role and significance of the ERS
Our findings showed that the high-cognitive-functioning cluster had the lowest premorbid exposure to environmental risk factors, followed by the intermediate and low-cognitive-functioning clusters with increasingly higher exposure.
Overall, we observed an inverse pattern of environmental versus polygenetic predisposition to psychosis (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023) – when the effect of one increased, the other decreased – more marked in the deteriorating group. This could suggest that lower environmental exposure is required when genetic factors are higher and vice versa in the predisposition to psychosis, in accordance with the liability threshold model (Supplementary Figure 4).
We have already discussed the unique characteristics of patients in the deteriorating cluster, where environmental exposure was the highest. Interestingly, the IQ_PRS (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023) in the deteriorating cluster was slightly higher than that of the intermediate cluster, having identical IQ (Supplementary Figure 4), so their current IQ was presumably lower than expected, also based on their optimal adjustment in childhood and later decline (Stefanatou et al., Reference Stefanatou, Karatosidi, Tsompanaki, Kattoulas, Stefanis and Smyrnis2018). These findings suggested the potential role of environmental risk factors in the loss of cognitive and social functioning in the premorbid period.
Although we did not find segregation of any diagnoses into specific clusters, the deteriorating group had the highest depressive and negative symptoms among clusters and more than one AP prescription (Supplementary Figure 1 and Supplementary Table 2). While this could have lowered IQ scores at the onset, the relationship between symptomatology and the deterioration of functioning between 12 and 16 years is more likely to be in the opposite direction, losing abilities before symptom presentation (Fusar-Poli et al., Reference Fusar-Poli, Borgwardt, Bechdolf, Addington, Riecher-Rössler, Schultze-Lutter, Keshavan, Wood, Ruhrmann, Seidman, Valmaggia, Cannon, Velthorst, De Haan, Cornblatt, Bonoldi, Birchwood, McGlashan, Carpenter and Yung2013). In contrast, the other clusters, particularly the high-cognitive-functioning cluster, presented with more genetically related functioning and a more favourable environmental context in childhood, constituted by higher socio-economic status of parents and higher education as compared with the other clusters (Supplementary Table 1) (see also Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023).
The median IQ of patients with low-cognitive-functioning (less than 74) may have included some individuals with neurodevelopmental impairment (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023; Howes & Murray, Reference Howes and Murray2014; Murray, O’Callaghan, Castle, & Lewis, Reference Murray, O’Callaghan, Castle and Lewis1992). Additionally, the previous study revealed that the low-cognitive-functioning group was the most disadvantaged in terms of polygenic liability to schizophrenia, bipolar and depressive disorder, and predisposition to lower IQ (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023). Altogether, environmental risk factors could have cumulatively and circularly contributed, with this cognitively disadvantaged profile, to the onset of psychosis (Liu et al., Reference Liu, Mendonça, Cannon, Jones, Lewis, Thompson, Zammit and Wolke2021; Sideli et al., Reference Sideli, Fisher, Murray, Sallis, Russo, Stilo, Paparelli, Wiffen, O’Connor, Pintore, Ferraro, La Cascia, La Barbera, Morgan and Di Forti2015).
It could be that individuals with lower cognitive and social functioning or with a genetic predisposition to psychiatric disorders were more likely to encounter environmental risks, like a lower socio-economic status or moving to a more disadvantaged neighbourhood (Lund et al., Reference Lund, Brooke-Sumner, Baingana, Baron, Breuer, Chandra, Haushofer, Herrman, Jordans, Kieling, Medina-Mora, Morgan, Omigbodun, Tol, Patel and Saxena2018; Maxwell, Coleman, Breen, & Vassos, Reference Maxwell, Coleman, Breen and Vassos2021). However, previous findings excluded a causal relationship between lower cognitive and social functioning on cannabis use (Ferraro et al., Reference Ferraro, Russo, O’Connor, Wiffen, Falcone, Sideli, Gardner-Sood, Stilo, Trotta, Dazzan, Mondelli, Taylor, Friedman, Sallis, La Cascia, La Barbera, David, Reichenberg, Murray and Di Forti2013, Reference Ferraro, Capuccio, Mulè, La Cascia, Sideli, Tripoli, Seminerio, Sartorio, La Barbera, Murray and Di Forti2016), childhood adversities (Sideli et al., Reference Sideli, Schimmenti, La Barbera, La Cascia, Ferraro, Aas, Alameda, Velthorst, Fisher, Caretti, Trotta, Tripoli, Quattrone, Gayer-Anderson, Seminerio, Sartorio, Marrazzo, Lasalvia, Tosato and van der Ven2022), and migration (Xu et al., Reference Xu, Vorderstrasse, McConnell, Dupre, Østbye and Wu2018).
Finally, the weighted sum of risk factors could discriminate to some degree between clusters of patients and community controls, explaining 11% of the FEP clusters/control variance, which was satisfactory compared to the 7% estimated in modelling an individual’s risk of schizophrenia using the ERS (Gillett, Vassos, & Lewis, Reference Gillett, Vassos and Lewis2018). Moreover, it was higher than the 7.9% detected in studies looking at multiple polygenic risk scores and cognitive clusters (Dickinson et al., Reference Dickinson, Zaidman, Giangrande, Eisenberg, Gregory and Berman2020; Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023).
On the other hand, the ERS explained only 2.2% of the between-patients variance, somewhat similar to the 2.7% explained by the PRSs in the cited study (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023). This finding is unsurprising, given that both scores were designed to differentiate between patients diagnosed with psychosis and population controls (Gillett et al., Reference Gillett, Vassos and Lewis2018).
The putative role of single exposures
Looking at single exposures, we confirmed the role of childhood adversities (Varese et al., Reference Varese, Smeets, Drukker, Lieverse, Lataster, Viechtbauer, Read, van Os and Bentall2012) and cannabis use (Di Forti et al., Reference Di Forti, Quattrone, Freeman, Tripoli, Gayer-Anderson, Quigley, Rodriguez, Jongsma, Ferraro, La Cascia, La Barbera, Tarricone, Berardi, Szöke, Arango, Tortelli, Velthorst, Bernardo, Del-Ben and van der Ven2019) in distinguishing all clusters of patients from controls, especially the deteriorating and the low-cognitive-functioning group, both having a good-to-normal premorbid social functioning. This last finding is in line with previous observations suggesting that better premorbid social functioning in patients with a history of cannabis use may have contributed to their likelihood to begin using cannabis (Ferraro et al., Reference Ferraro, La Cascia, Quattrone, Sideli, Matranga, Capuccio, Tripoli, Gayer-Anderson, Morgan, Sami, Sham, De Haan, Velthorst, Jongsma, Kirkbride, Rutten, Richards, Roldan, Arango and Di Forti2020).
Ethnic minority membership also distinguished patients (Tarricone et al., Reference Tarricone, Boydell, Kokona, Triolo, Gamberini, Sutti, Marchetta, Menchetti, Di Forti, Murray, Morgan and Berardi2016) from controls, apart from the high-cognitive-functioning cluster. It is possible that being part of the ethnic majority may be associated with socioeconomic and linguistic benefits in the high-cognitive-functioning cluster, resulting in better IQ. In contrast, minorities and migrants often have shown lower cognitive functioning due to complex mechanisms (Xu et al., Reference Xu, Vorderstrasse, McConnell, Dupre, Østbye and Wu2018). On the other hand, we could not exclude a power limitation in this lack of difference.
Interestingly, patients whose adjustment deteriorated in early adolescence showed higher exposition to cannabis use than the intermediate cluster, and they were also more likely to belong to an ethnic minority, distinguishing this group from the high-cognitive-functioning cluster. One study reported that ethnic minorities are more likely to use cannabis and other drugs (Montgomery, Dixon, & Mantey, Reference Montgomery, Dixon and Mantey2022). However, we did not find this association in the EUGEI sample (Jongsma et al., Reference Jongsma, Gayer-Anderson, Tarricone, Velthorst, van der Ven, Quattrone, di Forti, Menezes, Del-Ben, Arango, Lasalvia, Berardi, La Cascia, Bobes, Bernardo, Sanjuán, Santos, Arrojo, de Haan and Kirkbride2021). Instead, it has been hypothesized that cannabis could contribute to biases due to abnormal cognitive aberrations in the salience attribution (Wijayendran, O’Neill, & Bhattacharyya, Reference Wijayendran, O’Neill and Bhattacharyya2018), independent of being part of an ethnic minority (Anglin, Tikhonov, Tayler, & DeVylder, Reference Anglin, Tikhonov, Tayler and DeVylder2021). Indeed, we previously observed a higher probability of belonging to this cluster for patients with a premorbid frequent use of high-potency cannabis (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023). This type of cannabis has a higher addictive, dissociative (Ricci et al., Reference Ricci, Ceci, Di Carlo, Lalli, Ciavoni, Mosca, Sepede, Salone, Quattrone, Fraticelli, Maina and Martinotti2021, Reference Ricci, Ceci, Di Carlo, Di Muzio, Ciavoni, Santangelo, Di Salvo, Pettorruso, Martinotti and Maina2023), psychotic-leading (Di Forti et al., Reference Di Forti, Quattrone, Freeman, Tripoli, Gayer-Anderson, Quigley, Rodriguez, Jongsma, Ferraro, La Cascia, La Barbera, Tarricone, Berardi, Szöke, Arango, Tortelli, Velthorst, Bernardo, Del-Ben and van der Ven2019; Quattrone et al., Reference Quattrone, Ferraro, Tripoli, La Cascia, Quigley, Quattrone, Jongsma, Del Peschio, Gatto, Gayer-Anderson, Jones, Kirkbride, La Barbera, Tarricone, Tosato, Lasalvia, Szöke, Arango and Di Forti2020, Reference Quattrone, Reininghaus, Richards, Tripoli, Ferraro, Quattrone, Marino, Rodriguez, Spinazzola, Gayer-Anderson, Jongsma, Jones, La Cascia, La Barbera, Tarricone, Bonora, Tosato, Lasalvia, Szöke and D’Andrea2021) and detrimental potential on cognition and premorbid adjustment than low-potency varieties (Meier et al., Reference Meier, Caspi, Ambler, Harrington, Houts, Keefe, McDonald, Ward, Poulton and Moffitt2012; Mokrysz et al., Reference Mokrysz, Landy, Gage, Munafò, Roiser and Curran2016). Looking at this finding, we speculated that cannabis use was likely to have played an independent and strategic role in promoting premorbid social and cognitive deterioration and possibly IQ decline in this cluster of patients (Stefanatou et al., Reference Stefanatou, Karatosidi, Tsompanaki, Kattoulas, Stefanis and Smyrnis2018).
Finally, paternal age did not distinguish between patients and controls, as observed in the EU-GEI case-control study by Jongsma and colleagues (Jongsma et al., Reference Jongsma, Gayer-Anderson, Tarricone, Velthorst, Van Der Ven, Quattrone, Di Forti, Menezes, Del-Ben, Arango, Jones and Kirkbride2021), possibly because of power limitations, as very few subjects had an old father.
The validation analysis via the Exposome was in line with findings obtained with the ERS. Cannabis use and childhood adversities accounted for a high proportion of the 9.2% of the variance explained by this score in our sample, while hearing impairment and winter births gave insignificant results. The role of winter birth has been challenged in previous studies (Demler, Reference Demler2011; Muntjewerff et al., Reference Muntjewerff, Ophoff, Buizer-Voskamp, Strengman and den Heijer2011) and the original construction of the Exposome (Pries et al., Reference Pries, Lage-Castellanos, Delespaul, Kenis, Luykx, Lin, Richards, Akdede, Binbay, Altinyazar, Yalinçetin, Gümüş-Akay, Cihan, Soygür, Ulaş, Cankurtaran, Kaymak, Mihaljevic, Petrovic and Guloksuz2019). Similarly to paternal age, hearing impairment irrelevance could have suffered from the small number of affected participants and the impossibility of selecting in our sample the severe forms of hearing impairment among those reported, which are more relevant in case/control samples (Shoham et al., Reference Shoham, Lewis, Hayes, McManus, Kiani, Brugha, Bebbington and Cooper2020).
If we conceptualize environmental risk factors along a temporal line (Howes & Murray, Reference Howes and Murray2014; Lipner et al., Reference Lipner, O’Brien, Pike, Ered and Ellman2022; Tandon et al., Reference Tandon, Nasrallah, Akbarian, Carpenter, DeLisi, Gaebel, Green, Gur, Heckers, Kane, Malaspina, Meyer-Lindenberg, Murray, Owen, Smoller, Yassin and Keshavan2024), hearing impairment, winter birth, and paternal age had an impact very early in life, emerging from an interplay between genetic and perinatal effects, but able to explain a tiny percentage of caseness in psychosis. Thus, detecting their effect would require more extensive meta-analytic studies (Blazer & Tucci, Reference Blazer and Tucci2019; Coury et al., Reference Coury, Lombroso, Avila-Quintero, Taylor, Flores, Szejko and Bloch2023; Janecka et al., Reference Janecka, Mill, Basson, Goriely, Spiers, Reichenberg, Schalkwyk and Fernandes2017). Childhood adversities, which occurred at a second developmental stage, have been considered more generally predisposing to a vulnerability profile of psychopathology and emotional dysregulation than to cognitive deterioration (Arango et al., Reference Arango, Díaz-Caneja, McGorry, Rapoport, Sommer, Vorstman, McDaid, Marín, Serrano-Drozdowskyj, Freedman and Carpenter2018; Pries et al., Reference Pries, Klingenberg, Menne‐Lothmann, Decoster, van Winkel, Collip, Delespaul, De Hert, Derom, Thiery, Jacobs, Wichers, Cinar, Lin, Luykx, Rutten, van Os and Guloksuz2020; Sideli et al., Reference Sideli, Schimmenti, La Barbera, La Cascia, Ferraro, Aas, Alameda, Velthorst, Fisher, Caretti, Trotta, Tripoli, Quattrone, Gayer-Anderson, Seminerio, Sartorio, Marrazzo, Lasalvia, Tosato and van der Ven2022; van Os et al., Reference van Os, Marsman, van Dam and Simons2017), thus producing an impact on the risk of psychosis but not in differentiating cognitive profiles. This was also true when we looked at the exposome, which considered the cumulative effect of childhood adversities (Pries et al., Reference Pries, Lage-Castellanos, Delespaul, Kenis, Luykx, Lin, Richards, Akdede, Binbay, Altinyazar, Yalinçetin, Gümüş-Akay, Cihan, Soygür, Ulaş, Cankurtaran, Kaymak, Mihaljevic, Petrovic and Guloksuz2019). At a later stage, we could pose cannabis use, a risk factor specific for psychosis when used in adolescence (Di Forti et al., Reference Di Forti, Quattrone, Freeman, Tripoli, Gayer-Anderson, Quigley, Rodriguez, Jongsma, Ferraro, La Cascia, La Barbera, Tarricone, Berardi, Szöke, Arango, Tortelli, Velthorst, Bernardo, Del-Ben and van der Ven2019), as was migration, which had its most significant impact as a risk factor for psychosis during adolescence (Andleeb et al., Reference Andleeb, Moltrecht, Gayer-Anderson, Arango, Arrojo, D’Andrea, Bernardo, Del-Ben, de Haan, Ferraro, La Barbera, La Cascia, Llorca, Menezes, Quattrone, Sanjuán, Selten, Szöke, Tarricone and Kirkbride2024). These risk factors could have favoured psychosis transition also throughout a detrimental effect on subjects’ acculturation and social functioning (Andleeb et al., Reference Andleeb, Moltrecht, Gayer-Anderson, Arango, Arrojo, D’Andrea, Bernardo, Del-Ben, de Haan, Ferraro, La Barbera, La Cascia, Llorca, Menezes, Quattrone, Sanjuán, Selten, Szöke, Tarricone and Kirkbride2024; Castellanos-Ryan et al., Reference Castellanos-Ryan, Pingault, Parent, Vitaro, Tremblay and Séguin2017).
In summary, among a wide range of risk factors, childhood adversities, ethnic minority status, and cannabis use were the most important predictors and, when used separately, explained more variance (almost 14%) than the model with the risk summed in the ERS (11%), also due to the summative nature of the score (Vassos et al., Reference Vassos, Sham, Kempton, Trotta, Stilo, Gayer-Anderson and Morgan2020). This evidence indicated that it may be necessary to collect information on childhood abuse, ethnic minority membership, and cannabis use, at minimum, when no other information is available, to study environmental risk factors in psychosis.
Limitations
This study used a self-report approach for assessing premorbid characteristics and environmental exposures. Therefore, it was limited by the inability to distinguish between presumed causes and their possible effects. However, we used psychometrically robust measures, completed with input from at least one corroborative source of information, such as family members, clinical notes, and other clinicians (Ferraro et al., Reference Ferraro, La Cascia, Quattrone, Sideli, Matranga, Capuccio, Tripoli, Gayer-Anderson, Morgan, Sami, Sham, De Haan, Velthorst, Jongsma, Kirkbride, Rutten, Richards, Roldan, Arango and Di Forti2020), and this should have minimized the recall bias due to the retrospective method. A subsample validation analysis confirmed the reliability of patients’ self-reports on cannabis use, and the potency measure was supported by information provided in additional national reports (Di Forti et al., Reference Di Forti, Quattrone, Freeman, Tripoli, Gayer-Anderson, Quigley, Rodriguez, Jongsma, Ferraro, La Cascia, La Barbera, Tarricone, Berardi, Szöke, Arango, Tortelli, Velthorst, Bernardo, Del-Ben and van der Ven2019). We did not have data on obstetric complications or urbanicity at an individual level, but only on country of birth, whose non-correspondence with the country of residence matched a migration state (with a prevalence of 22.7% in our sample) (Tarricone et al., Reference Tarricone, D’Andrea, Jongsma, Tosato, Gayer-Anderson, Stilo, Suprani, Iyegbe, Van Der Ven, Quattrone, Di Forti, Velthorst, Rossi Menezes, Arango, Parellada, Lasalvia, La Cascia, Ferraro, Bobes and Morgan2021). Nonetheless, we adjusted by urbanization at a site level.
Our sample’s multicultural characteristics and the unique methodology used by introducing PAS with IQ measures could restrict the replicability of our findings in non-European and more homogeneous samples in which specific characteristics can more robustly emerge. In comparing the ERS and PRS among patients’ clusters and controls, the restriction of ancestry to solely white Europeans limited the ability to perform gene-environment interaction analyses. Despite the limited representation of non-European ancestry in the current sample, it is doubtful that this alone could account for the difference in environmental and genetic influences observed in the declining cluster. This is because the ERS was still highest in this group even when considering only subjects with European ancestry (Supplementary Figure 4). We recognize that other models could have been used to compare different class models, such as latent variable mixture modelling (Berlin, Williams, & Parra, Reference Berlin, Williams and Parra2014). Our study was a follow-up to an original study, this last having replication study characteristics (Dickinson et al., Reference Dickinson, Zaidman, Giangrande, Eisenberg, Gregory and Berman2020; Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023). Thus, we decided to maintain the same methodology to allow comparability.
Conclusions
The exposure to environmental risk factors partly reflected the cognitive performance at the onset of psychosis in discrete clusters of patients characterized by different patterns of premorbid functioning. It was highest among patients whose social and academic functioning deteriorated from average in the premorbid period. We have previously shown that polygenic predisposition to psychosis is unlikely related to this difference in functioning from childhood to adolescence (Ferraro et al., Reference Ferraro, Quattrone, La Barbera, La Cascia, Morgan, Kirkbride, Cardno, Sham, Tripoli, Sideli, Seminerio, Sartorio, Szoke, Tarricone, Bernardo, Rodriguez, Stilo, Gayer-Anderson, de Haan and Murray2023). Thus, although genetic predisposition to schizophrenia increased the risk in all the clusters, environmental risks, which are potentially modifiable, had a higher effect on the deteriorating group.
If correct, this could have potential clinical relevance because modifying them may have an essential role in the prevention of psychosis or at least in improving the premorbid function. It is known that patients with better premorbid adjustment presented a better prognosis (Amoretti et al., Reference Amoretti, Rosa, Mezquida, Cabrera, Ribeiro, Molina, Bioque, Lobo, González-Pinto, Fraguas, Corripio, Vieta, de la Serna, Morro, Garriga, Torrent, Cuesta and Bernardo2022), and discontinuing cannabis use, for example, predicted a better long-term outcome after onset (Schoeler et al., Reference Schoeler, Monk, Sami, Klamerus, Foglia, Brown, Camuri, Altamura, Murray and Bhattacharyya2016).
Additionally, at the onset of psychosis, patients in the deteriorating cluster had an intermediate IQ, similar to a group with consistently intermediate functioning in the premorbid period. Hence, despite having different premorbid characteristics, they could be mistaken for being part of the most common and undifferentiated intermediate group described in the literature (Green et al., Reference Green, Girshkin, Kremerskothen, Watkeys and Quidé2020). Therefore, accurately characterizing the trajectory of each patient and identifying associated environmental risk factors may be beneficial in tailoring treatment, psychoeducational, and preventive strategies.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291724003507.
Acknowledgments
Special acknowledgment to all the patients and the EU-GEI team. This study represents independent research part-funded by the NIHR Maudsley Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care. The authors report no financial relationships with commercial interests.
Author contribution
Diego Quattrone and Evangelos Vassos have equally contributed to the supervision of the work and share the last authorship.
Open access
