eTable 1. Information on the Participating and Nonparticipating Families in the Study
eTable 2. Descriptive Statistics for Psychotic Experiences
eTable 3. Univariate Model Fits for Whole Sample
eTable 4. Phenotypic Correlations Between Psychotic Experiences
eTable 5. Bivariate Model Fits for Whole Sample
eMethods. Further Details on the SPEQ Measure
Customize your JAMA Network experience by selecting one or more topics from the list below.
Zavos HMS, Freeman D, Haworth CMA, et al. Consistent Etiology of Severe, Frequent Psychotic Experiences and Milder, Less Frequent Manifestations: A Twin Study of Specific Psychotic Experiences in Adolescence. JAMA Psychiatry. 2014;71(9):1049–1057. doi:10.1001/jamapsychiatry.2014.994
The onset of psychosis is usually preceded by psychotic experiences (PE). Little is known about the etiology of PE and whether the degree of genetic and environmental influences varies across different levels of severity. A recognized challenge is to identify individuals at high risk of developing psychotic disorders prior to disease onset.
To investigate the degree of genetic and environmental influences on specific PE, assessed dimensionally, in adolescents in the community and in those who have many, frequent experiences (defined using quantitative cutoffs). We also assessed the degree of overlap in etiological influences between specific PE.
Design, Setting, and Participants
Structural equation model-fitting, including univariate and bivariate twin models, liability threshold models, DeFries-Fulker extremes analysis, and the Cherny method, was used to analyze a representative community sample of 5059 adolescent twin pairs (mean [SD] age, 16.31 [0.68] years) from England and Wales.
Main Outcomes and Measures
Psychotic experiences assessed as quantitative traits (self-rated paranoia, hallucinations, cognitive disorganization, grandiosity, and anhedonia, as well as parent-rated negative symptoms).
Genetic influences were apparent for all PE (15%-59%), with modest shared environment for hallucinations and negative symptoms (17%-24%) and significant nonshared environment (49%-64%) for the self-rated scales and 17% for parent-rated negative symptoms. Three empirical approaches converged to suggest that the etiology in extreme-scoring groups (most extreme scoring: 5%, 10%, and 15%) did not differ significantly from that of the whole distribution. There was no linear change in heritability across the distribution of PE, with the exception of a modest increase in heritability for increasing severity of parent-rated negative symptoms. Of the PE that showed covariation, this appeared to be due to shared genetic influences (bivariate heritabilities, 0.54-0.71).
Conclusions and Relevance
These findings are consistent with the concept of a psychosis continuum, suggesting that the same genetic and environmental factors influence both extreme, frequent PE and milder, less frequent manifestations in adolescents. Individual PE in adolescence, assessed quantitatively, have lower heritability estimates and higher estimates of nonshared environment than those for the liability to schizophrenia. Heritability varies by type of PE, being highest for paranoia and parent-rated negative symptoms and lowest for hallucinations.
The symptoms evident in people with psychotic disorders can also be experienced by people who are at increased risk of developing a psychotic disorder and in the general population.1 Across these populations, psychotic experiences (PE) appear to be associated with similar environmental factors (such as neighborhood deprivation and stressful life events) and to be seen in the same families.2,3 Psychotic disorders typically begin in early adulthood, but PE often first occur in adolescence.4 Individuals who report PE in childhood are at greater risk of developing psychotic disorders in adulthood.5,6
The past decade has seen increasing interest in the development of clinical interventions for individuals at high risk of psychosis.7 Understanding more about the causes of PE in adolescents is one approach that might inform the development of such interventions. In adults, twin and adoption studies suggest that both genes and environment influence risk of developing psychotic disorders.8-10 However, these studies did not address the individual PE as true dimensional quantitative traits.
There is limited understanding about the causes of PE in adolescents. Three articles11-13 describing PE (hallucinations and schizotypal traits) in adolescents (aged 13-19 years) using community twin samples of fewer than 600 pairs suggest that these experiences are moderately heritable (33%-57%), with the remaining variance explained by nonshared environmental influences (environmental influences that make children growing up in the same family different). Larger studies, using measures of the full range of positive, negative, and cognitive PE, would make it possible to test whether etiological influences vary across the distribution of severity, with particular focus on the high scorers, and to test whether different PE share the same etiological influences. Owing to the multifactorial structure of PE, a symptom-specific approach to studying the etiology of PE is encouraged.14,15 This approach has also been championed by researchers using clinical samples.16-20
The aim of this study was to examine the degree of genetic and environmental influences on specific PE in a community twin sample and in subgroups defined by extreme levels of PE (most extreme scoring: 5%, 10%, and 15%). Three empirical approaches were taken: one that categorized data to identify extreme scores and assumed an underlying liability (liability threshold model); one that used a group-based regression method (DeFries-Fulker extremes analysis); and one that tested whether there were any significant linear changes in the genetic and environmental estimates across the distribution (Cherny method). When specific PE covaried, the extent to which the relationship was due to overlapping genetic and environmental influences was investigated.
The Longitudinal Experiences and Perceptions (LEAP) study assessed PE in adolescents15 drawn from the Twins Early Development study (TEDS), a general population sample of monozygotic (MZ) and dizygotic (DZ) twins born in England and Wales between 1994 and 1996.21 Ethical approval for TEDS was obtained from King’s College London, Institute of Psychiatry Ethics Committee. The study originally contacted a sample of 16 302 families who had twins between 1994 and 1996, of whom 13 488 families responded with a written consent form. Families were not contacted for the LEAP study if they had withdrawn from TEDS, had never returned any data, had no known address, or were special cases, most notably medical exclusions.
Initially, 10 874 families who participated in TEDS were invited to participate in LEAP. Of those, 5076 parents (47.6%) and 5059 twin pairs (47.5%) provided data (mean [SD], 16.32 [0.68] years). Individuals were excluded (n = 876) if they did not provide consent at first contact (when TEDS started) or, for this study, if they had a severe medical disorder, had experienced severe perinatal complications, or had unknown zygosity. The twin sample after exclusions (4743 families) was 45.0% male. Participating and nonparticipating families were largely similar in terms of sex, zygosity, ethnicity, and maternal educational level. Additional details are provided in eTable 1 in the Supplement. The nonparticipating families had higher scores on childhood behavior problems than the participating families. The difference of roughly 1 raw score between the participating and nonparticipating families, however, amounts only to an average of half a point difference on the measure (each item is rated from 0 to 2, and even small differences are significant because of the large sample size).
The Specific Psychotic Experiences Questionnaire (SPEQ)15 assesses 6 types of PE in adolescents: paranoia (15 items), hallucinations (9 items), cognitive disorganization (11 items), grandiosity (8 items), and anhedonia (10 items), all via self-report, as well as negative symptoms via parent report (10 items). The SPEQ was developed by selecting and combining items from existing scales for adults and adapting wording when necessary to be age appropriate. Age-appropriateness of items was ensured by obtaining expert clinical opinions from 3 of us (D.F., A.G.C., and P.M.) and via piloting on this age group.15 Subscales show good to excellent internal consistency (Cronbach α = 0.77-0.93) and test-retest reliability across a 9-month interval (r = 0.65-0.74).
Construct validity was assessed in terms of the principal component analysis supporting the separation of the SPEQ subscale items.15 Content validity was assessed via expert clinical opinion to judge the suitability of items for measuring adolescent PE (A.G.C., D.F., and P.M.). Validity was also assessed in terms of agreement with a second known measure of adolescent psychosis-like symptoms, the Psychotic-Like Symptoms Questionnaire.22 Individuals who reported “definitely” having any psychosis-like symptoms on the Psychotic-Like Symptoms Questionnaire had significantly more PE on all SPEQ subscales than individuals who did not report any definite psychosis-like symptoms (all significant at P < .001), with the exception of anhedonia, which was not significant. The SPEQ positive and cognitive PE subscales show significant positive correlations with the Psychotic-Like Symptoms Questionnaire quantitative score for psychosis-like symptoms (hallucinations: r = 0.60; paranoia: r = 0.48; cognitive disorganization: r = 0.41; grandiosity: r = 0.27; all P < .001).15,22 Finally, for all SPEQ subscales except anhedonia, individuals who reported a family history (having a first- or second-degree relative with schizophrenia or bipolar disorder) scored higher than individuals without a family history of psychosis (all P < .05 except hallucinations, which showed a trend in this direction). Further information on how the scales were devised is provided in the eMethods in the Supplement.
The rationale of this design is to compare the degree of resemblance among MZ twins, who share 100% of their DNA sequence, with DZ twins, who share an average of 50%. Relative differences in within-pair correlations are then used to estimate the following latent factors on the measures: additive genetic (A), shared environmental (C), and nonshared environmental (E) influences. Where correlations are higher for MZ than DZ twins, genetic influence is inferred. Within-pair similarity that is not due to genetic factors is attributed to shared environmental influences (C), which is defined as aspects of the environment that contribute to resemblance between family members. Nonshared environmental influences (E) account for individual specific factors that create differences among siblings from the same family. These are estimated from within-pair differences between MZ twins. Measurement error is included in this term.
Statistical analysis was conducted in Mx (Department of Psychiatry, Virginia Institute for Psychiatric and Behavioral Genetics, Virginia Commonwealth University, Richmond). Variables were age and sex regressed, as is standard practice for quantitative genetic model fitting.23 Twin correlations were estimated for each sex and zygosity group.
Univariate models examined the influences of A, C, and E on PE. Several models were tested and compared with a saturated model: (1) a full sex-limitation model allowing for quantitative and qualitative sex differences in addition to variance differences; (2) a model allowing for quantitative and variance sex differences; (3) a no–sex differences model; and (4) a variance sex difference model.24 Models were compared using χ2 difference for nested models; the Akaike information criterion, which is equal to χ2 minus twice the df,25 was used as an aid to select the best-fitting model on the grounds of parsimony and goodness of fit.
Genetic and environmental influences across the distribution of PE were compared using 3 analytic techniques. As sex effects were not estimated, DZ opposite-sex twins were excluded from these analyses.
Liability threshold models were used to estimate the etiology of categorically defined extreme scores. These models assume that the joint distribution of twin pairs follows an underlying bivariate normal distribution.24 If the estimates of heritability and environmental influences of the liability of extreme PE at various cutoffs (5%, 10%, and 15%) are consistent, it suggests that the etiology of the liability to PE does not vary across severity.
DeFries-Fulker extremes analysis investigates the genetic and environmental influences on the difference between the mean scores of extreme groups and the whole population.26 It is designed for proband-selected data in which at least 1 twin has an extreme score and is based on regression of the co-twin to the mean of a quantitative trait score.26 A genetic link between extremes and the whole sample is implicated if significant group heritability estimates are found.
The Cherny method is an extension of the DeFries-Fulker extremes model and examines whether the relative contributions of genes and environment change linearly across the full distribution. This is implemented by including interaction effects in a regression equation, which allow for the estimation of the interaction between the heritability of a trait with the score on the trait.27
Bivariate twin models were used to assess the genetic and environmental influences on associations between specific PE where within-person correlations between the different experiences were significant and greater than 0.20.15 In bivariate analysis, MZ and DZ correlations are compared across traits, ie, one twin′s score on a trait is correlated with the co-twin′s score on another trait.28
A genetic correlation (rA) is derived from the model fitting and can vary between 0 and 1, indicating the extent to which genetic influences on one variable overlap with a second phenotype. Correlations can similarly be estimated for shared and nonshared environmental factors. The extent to which genetic, shared, and nonshared environmental factors contribute to the phenotypic correlations can also be calculated. For example, genetic influences on the correlation can be calculated by multiplying the square root of the heritability of each variable by the genetic correlation. Similar calculations can be done for shared and nonshared environmental influences.
There was some evidence of skew; therefore, variables were transformed (square root: cognitive disorganization, grandiosity, hallucinations, and paranoia; and log: negative symptoms) as required to ensure skew statistics were between −1 and 1. Descriptive statistics are given in eTable 2 in the Supplement. Twin correlations are given in Table 1. All correlations between DZ twins were less than the MZ correlations, indicating additive genetic influences on all PE. Shared environmental influences were also implicated for some PE; for example, parent-rated negative symptoms, as DZ correlations were more than half the MZ correlation. As MZ correlations were less than 1, nonshared environmental influences were also implied. There was some indication of sex differences in the etiology, indicated by the different pattern of MZ and DZ correlations for male vs female and DZ same-sex and opposite-sex pairs.
Univariate analyses are presented in Table 2. All A, C, and E model fits were acceptable (ie, not significantly worse than the saturated model). No qualitative or quantitative sex differences were evident in the genetic and environmental influences on the subscales, with the exception of hallucinations, for which heritability was higher in girls than boys (full details of model fit are shown in eTable 3 in the Supplement). All subscales were moderately heritable, ranging from 32% for hallucinations in girls to 59% for parent-rated negative symptoms, with the exception of hallucinations in boys, which showed low heritability (15%). Significant shared environmental influences were evident for hallucinations (20% for boys and 17% for girls) as well as parent-rated negative symptoms (24%). Nonshared environmental influences explained a significant proportion of the variance on all subscales (49%-64% for the self-rated scales and 17% for parent-rated negative symptoms). The high genetic and shared environment estimates for negative symptoms may, in part, be explained by shared method variance, as parents were reporting on both twins within the pair, which can inflate twin correlations.
Table 3 presents the extremes analyses. The liability threshold model results indicated genetic influences for all 6 types of extreme PE; point estimates were not significantly different across the quantitative extreme groups (5%, 10%, and 15%) and were highly similar to the heritability estimates for the whole sample. Shared environmental influences showed the same pattern as for the whole sample, that is, being significant only for hallucinations and parent-rated negative symptoms. Estimates of nonshared environmental influences on the extreme groups were also highly consistent across extreme severity groups and closely resembled the whole-sample estimates.
Transformed co-twin means were calculated by dividing the co-twin scores by the proband mean for each zygosity group. The transformed co-twin means can be interpreted as twin group correlations because they provide an indication of within-pair similarity. They were generally higher in MZ than DZ twins, suggesting additive genetic influences at the extremes (Table 3). Overall, the relationship between twins did not seem to vary substantially across the cutoff levels compared with the whole-sample twin correlations.
Group heritability estimates were consistent across the 5%, 10%, and 15% extreme groups, as indicated by similar point estimates and overlapping 95% CIs. The significant group heritability estimates indicate a genetic link between extreme PE and variation in PE in the whole sample. Group shared environment estimates also demonstrated consistency across the extremes.
Analysis using the regression-based Cherny method are presented in Table 4. There was significant linear change in heritability for only 1 of the PE, suggesting that heritability does not differ across the distribution. The exception was parent-rated negative symptoms, which showed decreases in shared environmental influences and modest increases in genetic influences with increasing negative symptoms.
Bivariate genetic analyses were conducted in the full sample for relationships between PE where phenotypic correlations were significant and above 0.20 (eTable 4 in the Supplement). Four relationships met this criterion (paranoia–hallucinations, paranoia–cognitive disorganization, hallucinations–cognitive disorganization, and cognitive disorganization–parent-rated negative symptoms). Cross-twin cross-trait (CTCT) correlations are presented in Table 1. Most MZ CTCT correlations were greater than their equivalent DZ CTCT correlation, suggesting genetic influences on the covariation. Similarly, for most comparisons, DZ CTCT correlations were greater than half the MZ CTCT correlations, suggesting shared environmental influences on the covariation. Finally, MZ CTCT correlations tended to be lower than the relevant phenotypic correlation, indicating that nonshared environmental influences also contributed to the covariation.
Bivariate twin modeling (Table 5) confirmed these observations (full model fits shown in eTable 5 in the Supplement). High genetic correlations were evident between paranoia and hallucinations, paranoia and cognitive disorganization, and hallucinations and cognitive disorganization (rA = 0.61-0.63). A moderate genetic correlation (rA = 0.27) was found between cognitive disorganization and parent-rated negative symptoms. The proportion of covariation between each pair of variables was accounted for primarily by genetic influences; bivariate heritabilities ranged from 54% (cognitive disorganization and negative symptoms) to 71% (paranoia and cognitive disorganization). Shared environmental influences were important for the relationship between cognitive disorganization and parent-rated negative symptoms only.
Moderate nonshared environmental correlations were evident between paranoia and hallucinations, paranoia and cognitive disorganization, and hallucinations and cognitive disorganization (rE = 0.24-0.33), indicating that some nonshared environmental influences are shared between different PE. A lower nonshared environmental correlation (rE = 0.10) was evident between cognitive disorganization and parent-rated negative symptoms. A significant proportion of the covariance between PE was explained by nonshared environmental influences (12%-36%).
This was the first time, to our knowledge, that individual PE assessed dimensionally in adolescence have been examined for genetic and environmental contributions. More than 5000 twins were assessed on 6 spectra of PE. We found that PE in adolescence were moderately heritable, with paranoia and parent-rated negative symptoms showing the highest heritability and hallucinations showing the lowest. Nonshared environmental influences played an important role in their etiology. Shared environmental influences were only significant for hallucinations and negative symptoms. This is consistent with previous research, which has shown a number of environmental risk factors for psychosis that may be specific to the individual, such as stressful life events, cannabis use, and childhood trauma.29-31 The low heritability estimate for hallucinations is consistent with emerging research indicating the significance of early trauma in their occurrence.32 Indeed, the heritability estimates argue for a renewed interest in the contribution of the environment to risk of PE.
The extremes analyses indicated that the heritability did not differ for individuals who reported the most severe and frequent PE compared with the full sample (liability threshold model and Cherny method) and that there was a genetic link between the extreme group and the rest of the distribution (DeFries Fulker analysis). These findings support the suggestion that psychosis exists on an etiological continuum with subclinical PE.3 They have implications for genetic studies of psychotic disorders, because if extreme, frequent PE are part of the same construct as clinically diagnosed psychotic disorders,33,34 these findings support the hypothesis that the same genes that influence symptoms within psychotic disorders also influence variation in PE in the general population. So far, 2 studies have been conducted that assessed whether PE in adolescence were associated with the same genetic variants as diagnosed schizophrenia as a whole.35,36
Previous research suggests that PE have a multifactorial structure14,15; for this reason, we analyzed domains of PE separately. The covariation between PE was explained by shared genetic influences across domains. However, not all domains correlated with one another, and genetic correlations did not reach unity, suggesting there may be some etiological influences that are distinct across different PE.
The twin design is based on several assumptions, including independence of the A, C, and E latent factors; ideally, findings should be replicated across different study designs.37,38 Self-reported data of PE has been shown to give higher means than interview data.39 It would have been advantageous to report the DeFries-Fulker and liability threshold models using even more extreme thresholds that more closely mirrored the prevalence of adult psychosis. The statistical power afforded with the etiological architecture of these scales (which involve modest amounts of A and, in some cases, C) was not high enough to estimate parameters accurately with more extreme (eg, 1%) cutoffs. The 5% cutoff included here is similar to the prevalence of the at-risk mental state,40 and a meta-analysis reported the median prevalence of adult PE to be within the ranges of the extreme group cutoffs, at 7.2%,41 but the 5% extreme cutoff does not mirror the prevalence of psychotic disorders. However, one of the other methods used for the extremes analysis, the Cherny method, was able to examine whether the relative contributions of genes and environment changed linearly across the full distribution of PE, which incorporated all individuals, including those at the very extreme. It is important to remember that nonshared environment estimates (E) include measurement error. However, the E estimates were larger than the estimated error in each scale (calculated as 1 minus Cronbach α or test-retest reliability statistic), suggesting E was important for specific PE beyond measurement error, with the exception of the parent-rated negative symptoms scale, in which error appeared to make up most of the E term.
The large sample size enabled etiological sex differences to be tested and gave power to analyze the etiology of extreme groups. It was also advantageous that the full range of positive, negative, and cognitive disorganization experiences were included, using a reliable and validated measure in a narrow age range.15
This study found significant heritability for all PE while also showing that environmental influences, particularly nonshared environment, play an important role and appear to have a more prominent role than suggested from twin studies of the liability of schizophrenia. Heritability varies by psychotic experience type, being highest for paranoia and parent-rated negative symptoms and lowest for hallucinations. These findings suggest that the same genetic and environmental causal factors influence extreme, frequent PE and milder, less frequent manifestations in adolescents. A recognized challenge is to identify individuals at high risk of developing psychotic disorders before disease onset.7 To the extent that severe, frequent PE are indicators of risk of psychosis, these findings reveal their etiological architecture and can be used to guide investigation of molecular genetic and environmental risk factors.
Submitted for Publication: September 30, 2013; final revision received March 10, 2014; accepted April 16, 2014.
Corresponding Author: Angelica Ronald, PhD, Centre for Brain and Cognitive Development, School of Psychology, Birkbeck College, Malet Street, London WC1E 7HX, England (firstname.lastname@example.org).
Published Online: July 30, 2014. doi:10.1001/jamapsychiatry.2014.994.
Author Contributions: Drs Zavos and Ronald had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Zavos, Freeman, McGuire, Plomin, Ronald.
Acquisition, analysis, or interpretation of data: Zavos, Freeman, Haworth, McGuire, Plomin, Cardno, Ronald.
Drafting of the manuscript: Zavos, Freeman, Ronald.
Critical revision of the manuscript for important intellectual content: Freeman, Haworth, McGuire, Plomin, Cardno, Ronald.
Administrative, technical, or material support: Haworth.
Statistical analysis: Zavos.
Obtained funding: Plomin, Ronald.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported by Medical Research Council grants G1100559 (Dr Ronald), G0901245, and G0500079 (Dr Plomin); UK Medical Research Council Senior Clinical Fellowship G0902308 (Dr Freeman); and the British Academy (Dr Haworth).
Role of the Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Additional Contributions: We thank the participants of TEDS for making this research possible. Special thanks to Andrew McMillan, BA, BSc, Francesca Lewis, BA, Louise Webster, BA, Neil Harvey, BSc, and Rachel Ogden, BA, and to Peter McGuffin, MB, PhD, FRCP, FRCPsych, FMedSc, for help planning the study.
Create a personal account or sign in to: