The ALSPAC birth cohort consists of 14 541 pregnancies that resulted in 14 062 live births: 13 988 infants were still alive at 1 year. A total of 6455 (45.9%) children completed the PLIKSi, mean age 12.9 years (range 11.4–14.3). We compared participating children with the rest of the ALSPAC cohort on a number of variables: 49.3% of those completing the PLIKSi were male compared with 53.8% for the rest of the cohort (P<0.001, 95% CI –6.2 to –2.8); 43% of those who completed had a parental social class of manual v. 55.3% (P<0.001, 95% CI –14.1 to –10.5); 21.4% had a maternal education of below O-level (basic qualification obtained after 12 years of education) v. 38.6% (P<0.001, 95% CI –18.8 to –15.6); and 95.8% were White v. 93.1% (P<0.001, 95% CI 1.9–3.5).
In the reliability studies, according to the standard benchmarks of Landis & Koch,31
the average interrater reliability was ‘very good’ (kappa=0.72) and for the majority (75%) of individual items the kappa was above 0.6. The overall kappa value slightly improved across the three time points used to measure maintenance of reliability through the study. For the test–retest study, 163 children completed a second interview producing a test–retest kappa of 0.48, suggesting ‘fair’ agreement. This finding was consistent with findings for test–retest analyses using other semi-structured interviews such as the Present State Examination – 9th edition (PSE–9).32
In this selected group the prevalence of children displaying a psychosis-like symptom at the first visit was 13.6% falling to 8.4% at the second visit.
Frequency of psychosis-like symptoms
Overall, 38.9% (95% CI 37.5–40.1) of children self-reported experiencing one or more of the 12 symptoms (responding ‘yes’ or ‘maybe’) in the previous 6 months (). A total of 18.2% self-reported experiencing two or more symptoms and 9.3% reported experiencing three or more symptoms.
Number of children reporting psychosis-like symptoms: comparisons between self-report and observer-rated assessments (N=6435–6455)
Following the PLIKSi, the observer-rated assessments scored 13.7% (n=881, 95% CI 12.8–14.5) of children as experiencing one or more of the 12 symptoms (suspected or definite: ‘broad psychosis-like symptoms’) in the previous 6 months (). After taking account of non-response, this prevalence estimate rose to 13.8% (). Of these, 9.3% were rated with only one symptom, 2.6% rated with two or more symptoms and 1.8% had three or more symptoms. For ‘definite only’ symptoms (‘narrow psychosis-like symptoms’), 5.6% (n=364, 95% CI 5.1–6.2) of the cohort were rated positive. Finally, 2.6% (n=165, 95% CI 2.2–3.0) were rated positive for ‘definite symptoms without attributions occurring monthly or more frequently’ (‘frequent psychosis-like symptoms’).
The value of the 12 self-reported questions as screening questions was further investigated by comparing replies with the final observer rating (suspected or definite present). The positive predictive values for the questionnaire responses compared with the ratings from the clinical interviews were poor for all items except auditory hallucinations (). Of those giving positive replies to the self-report question ‘since your 12th birthday have you ever heard voices that other people can’t hear?’, after clinical cross-examination 70% were judged by interviewers to be truly experiencing this symptom. It was not possible to calculate sensitivity/specificity values because we could not, within the limits of the allocated interview time, assess further those children that replied negatively in order to investigate potential false negatives.
There were no significant differences in symptom-positive children in terms of gender or ethnicity but the prevalence of symptoms was higher for those of lower social class, with 15.2% (n=389, 95% CI 13.8–16.6) for manual compared with 12.1% (n=410, 95% CI 11.0–13.2) for non-manual (P=0.001), and in those with lower maternal educational attainment, with 15.3% (n=203, 95% CI 13.4–17.2) below O-level, 14.3% (n=314, 95% CI 12.8–15.8) at O-level and 12.2% (n=330, 95% CI 11.0–13.5) above O-level (P trend=0.005).
shows the main attributions ascribed for definite ratings on PLIKSi. Being in a hypnogogic or hypnopompic state was the most common self-attribution, with the highest attribution rate (22.2%) recorded for visual hallucinations.
Number of children receiving a ‘definite’ observer rating for the four categories of symptoms and associated attributionsa
In the DSM–IV,33
only one of certain ‘core’ symptoms needs to be established to satisfy criteria A for schizophrenia. They must occur in the context of a clear sensorium and exclude those that occur while falling asleep or waking up. In our cohort, the prevalence of third-person hallucinations without attributions (that is, excluding hypnogogic and similar experiences) was 1.1% (n
=73, 95% CI 0.9–1.4). The prevalence of ‘bizarre’ symptoms (thought withdrawal, insertion, broadcast) and/or delusions of control (without attributions) was 2.83% (n
=182, 95% CI 2.4–3.2). Overall, the proportion of children with one or more of these ‘core’ symptoms of schizophrenia was n
=233 (3.62%, 95% CI 3.2–4.1) suspected or definite; n
=89 (1.38%, 95% CI 1.10–1.70); definite and n
=45 (0.7% 95% CI 0.5–0.9) with symptoms at a frequency of monthly or more.
IQ score and psychosis-like symptoms
At the age of 8 years, data on IQ was available for 6751 children. The mean total IQ for this sample was 104.2 (s.d.=16.4). The mean verbal score IQ was 107 (s.d.=16.8) and performance score IQ was 99.7 (s.d.=17.0). Lower IQ score was more frequent in children with low birth weight, whose mothers were of manual social class and had less education, and who lived in renting households in single-parent families (online Table DS1). Low IQ was also more common in children who were victims of bullying and whose mothers reported higher SDQ scores (online Table DS1).
As a primary analysis, we examined all children who reported ‘suspected’ or ‘definite present’ symptoms on the PLIKSi (or broad psychosis-like symptoms). There was a non-linear relationship between IQ score and broad psychosis-like symptoms (). In the crude or unadjusted analyses, the increase in prevalence was most marked in those with lower IQ scores. This pattern of results remained after adjustment. In , this non-linearity is illustrated first by giving results by IQ score in five categories. The first adjustment had a relatively modest influence on the results. The relationship between IQ and symptoms was still statistically significant in the fully adjusted model (χ2=7.43, d.f.=2, P=0.02). illustrates a ‘reverse’ J-shaped relationship with most risk associated with low IQ but some increase in risk for those of high IQ, though this was only apparent after adjustment.
Odds ratios for psychosis-like symptoms (suspected and definite) according to IQ total score quintiles before and after adjustmenta
Predicted probability of psychosis-like symptoms according to IQ scorea (fully adjusted model).
We also examined whether verbal or performance IQ was more strongly associated with psychosis-like symptoms. There was a correlation of 0.50 between verbal and performance IQ scores. There was an association between verbal IQ and PLIKS even after adjustment for performance IQ scores (verbal IQ score linear term=0.60, 95% CI 0.38–0.95; verbal IQ2=1.02, 95% CI 1.00–1.04) but no association between symptoms and performance IQ score after adjustment for verbal IQ (performance IQ=0.86, 95% CI 0.55–1.34; performance IQ2=1.01, 95% CI 0.98–1.03).
IQ and different symptom outcomes
The pattern of results for narrow psychosis-like symptoms (n=364) and frequent psychosis-like symptoms (n=165) was similar, but many of the results, particularly after the second adjustment, were not statistically significant. We also used an ordered proportional odds approach that enabled us to study all three outcomes simultaneously. We found results very similar to those reported in : the IQ score linear term after the first adjustment was 0.40 (95% CI 0.22–0.72) and IQ2 and the quadratic term was 1.04 (95% CI 1.01–1.07). The equivalent results after the second adjustment were 0.45 (95% CI 0.25–0.81) and 1.04 (95% CI 1.01–1.07).
Odds ratios for psychosis-like symptoms (suspected and definite) as linear and quadratic terms before and after adjustmenta
We investigated the possibility that attrition in the cohort might have contributed to the observed associations using imputation methods.30,34
The results in show that the same pattern of results was present and remained after our two adjustments. We compared unadjusted results in the 5328 children with complete data for IQ and symptoms with the 3449 children that also had data on all confounders. The association appeared to be stronger in the smaller data-set, whereas unadjusted results for the imputed data were closer to those for the 5328 children. This suggested that the attrition might have exaggerated the relationship between symptoms and IQ score.