The Adolescent Brain and Cognitive Development (ABCD; https://abcdstudy.org/) 2.0.1 release study data were used in the current analysis [28, 29]. The ABCD study is an ongoing longitudinal cohort study aimed at recruiting a representative sample of US children. At the initial stage, all families of children aged 9–10 years in geographic catchment area of study sites across the USA were contacted via schools with information about the study. Volunteer families were then screened for inclusion. Participants were purposively recruited to match national US sociodemographic factors, for example utilising targets for each of five major race/ethnicity classifications (white, African-American, Hispanic, Asian, all other). The 2.0.1 release includes baseline data on 11,873 individuals, and 1-year follow-up from 4951 of those participants, with follow-up data from the remainder of the cohort to be released at a future date. Full details of the ABCD study design are available in a special issue of Developmental Cognitive Neuroscience .
Participants’ psychotic experiences were assessed using the Prodromal Questionnaire-Brief Child version (PQ-BC). This measure was validated within the current ABCD dataset showing high reliability (Cronbach’s alpha = 0.863 for total score and 0.873 for distress subscale; . The PQ-BC is a self-report questionnaire assessing presence and distress associated with psychotic-like symptoms in children. For each psychotic experience the child is asked if they experience it (yes/no), and if they do, how much it distresses/bothers them on a pictographic 1 to 5 scale showing a human cartoon figure in various levels of distress. This questionnaire yields two outcome variables—the sum of symptoms endorsed (range 0–21) and the sum of distress reported for those symptoms (range 0–126).
For the current study, both continuous and dichotomous outcomes of the sum of symptoms were used. For the dichotomous outcome, the symptom score was transformed to a categorical 0–1 variable, where 1 indicates the presence of at least one psychotic symptom and 0 indicates no psychotic symptoms present. When applied to baseline and follow-up, this was then used to derive variables indicating ‘new onset’ of psychotic symptoms (i.e. 0 at baseline, 1 at follow-up) ‘persistence’ of psychotic symptoms (i.e. 1 at both time points), and ‘remission’ (1 at baseline, 0 at follow-up).
As an addition to our analysis plan (see Alterations to the pre-registered statistical plan for further details and rationale), we derived a count of distressing psychotic experiences which was dichotomised to represent the presence/absence of at least one distressing psychotic symptom (≥ 2 score on distress rating).
Presence of sleep disorders was assessed using the Sleep Disorder Scale for Children (SDSC), which is a parent-reported questionnaire assessing the presence of a range of sleep disorder symptoms in children . It is composed of 26 Likert items assessing the frequency of various disturbances over the past 6 months on a 1 to 5 Likert scale (1 = never, 5 = always/daily experiencing a particular issue). The total score provides a measure of sleep disturbance, for which a cutoff point at 39 has sensitivity of 0.89 and specificity of 0.74, correctly identifying 73.4% of a control group and 89.1% of sleep disordered participants.
This cutoff was used to categorise participants according to absence or presence of disturbed sleep. The categorical score at baseline and follow-up was then used to derive variables indicating ‘persistence’ of sleep disturbance (i.e. present at both time points), ‘remission’ (present at baseline, absent at follow-up), and ‘onset’ (absent at baseline, present at follow-up).
Potential confounders: sociodemographic, IQ, and medication variables
Further variables were used from the ABCD dataset to index potential confounders, defined as factors that can independently influence each of the variables of interest (sleep and psychotic experiences).
Male gender and non-white ethnicity are associated with a higher likelihood of reporting both sleep problems  and psychotic experiences . Gender (Male/Female), ethnicity (white/black/Hispanic/Asian/other) were reported within the basic demographic questionnaires of the study. Ethnicity was re-coded into white/non-white for the purposes of all analyses.
Lower socioeconomic status is also associated with increased likelihood of psychiatric disorder  and shorter sleep duration . Socioeconomic status was indexed by using the sum score (range = 0–7) of seven yes/no items in the parent demographic survey questions relating to experiences of family hardship (e.g. “in the past 12 months has there been a time when you and your immediate family needed food but couldn't afford to buy it or couldn't afford to go out to get it?”), with higher scores on this sum scale indicating lower socioeconomic status. Neighbourhood deprivation was also assessed using the area deprivation index of the home address, which provides a national percentile value (range 1–100) with higher values indicating higher levels of deprivation.
Family conflict is also associated with sleep problems  and psychotic experiences . Family conflict was indexed by the nine-item family conflict subscale of the family experiences. Each item is reported by parents as true or false (e.g. “We fight a lot in our family”), with higher values indicating higher levels of conflict (range = 0–9).
Lower IQ scores and prescription of stimulant medications have been reported to have associations with psychotic experiences  and, especially for stimulant medications, with sleep problems . Child IQ was assessed using the WISC-V matrix reasoning subscale score (range = 1–19), with higher values indicating higher IQ. Medication fields were searched for any stimulant medications (e.g. “Methylphenidate”) and their trade names (e.g. “Ritalin”) with absence or presence coded as a dichotomous 0/1 variable.
Notably, depression and anxiety were not included as potential confounders as these are consistently found to act as mediators in the causal pathway between sleep and psychotic experiences (e.g. ). Therefore, if included in statistical models as confounders, this would likely result in an underestimate of the relationship between sleep and psychotic experiences which was the primary focus of the current investigation.
The pre-registration document and the analysis code used in this study are available online at the following link: https://osf.io/8ks72/. R version 3.6.2 was used for all analyses. A list of packages and version numbers used can be found in Supplementary Material 1. The pre-registration was completed before the ABCD 2.0 data release, i.e. before the 1-year follow-up data was made available.
For each research question a set of planned regression analyses were pre-specified. In each case, the regression model was first estimated with only the key explanatory (sleep) and dependent (psychosis) variables. If a significant association was found, the analysis was repeated with potential confounders added (simultaneously) to test the robustness of the hypothesised association.
The following four research questions were tested:
Do sleep disorders and psychotic experiences co-occur?
This was investigated cross-sectionally within baseline using (a) a linear regression to test continuous association between sleep symptoms and psychotic experiences (b) a logistic regression to test if sleep symptoms (continuous) predicted presence of psychotic experiences (dichotomous) and (c) a logistic regression to test if presence of sleep disorder (dichotomous) predicted presence of psychotic experiences (dichotomous).
Do sleep disorders predict later psychotic experiences?
This was investigated by logistic regression models testing if the presence of sleep disorder (a) at baseline and (b) at both baseline and follow-up predicted psychotic experiences at 1-year follow-up.
Do sleep disorders predict persistence of psychotic experiences?
This was investigated by logistic regression models testing if the presence or persistence of sleep disorder predicted persistence (i.e. presence at both baseline and follow-up) of psychotic experiences.
Does remission of sleep disorders predict remission of psychotic experiences?
This was investigated by logistic regression models testing if the remission of sleep disorders (i.e. present at baseline, absent at follow-up) also predicted remission of psychotic experiences using logistic regression.
To additionally examine potential associations with distressing psychotic experiences, we completed the planned regression analyses with an alternative outcome: the presence/absence of at least one distressing psychotic experience. This followed further published analyses of the PQ-BC responses in the ABCD dataset that advised inclusion of distress to reduce false positives and thereby increase validity of the measure [30, 40].
In addition, we altered our analysis plan across all regression analyses based on ABCD analysis guidance by (i) including clustering by site and family to create multi-level models (otherwise known as mixed-effect models) that included site and family as random intercept factors , and (ii) repeating the analyses with scaling by population weights within the analyses (reported in supplementary materials).