2.1 Sample/Design
Participants were recruited from the ongoing Healthy Brain Network (HBN) initiative that seeks to create and share a 10,000-participant biobank of data from children and adolescents ages 5–21 from the New York City area[74]. Data collected includes psychiatric, cognitive, behavioral, genetic, and lifestyle information as well as MRI and EEG neuroimaging. The HBN has two main collection sites—one located on Staten Island and one in Midtown Manhattan. Participants were excluded from participation if there were outstanding safety concerns, insufficient verbal abilities, an IQ lower than 66, and neurological concerns that could interfere with MRI and EEG interpretation. Participants were asked to suspend stimulant medication meant for treatment of ADHD on the days of testing to minimize potential confounds, though participants who did not suspend medication use were still included in the study and medication use was noted on the day of assessment.
As part of the HBN survey battery, participants and their parents/guardians completed a variety of age-based questionnaires assessing basic demographic characteristics, dimensional assessments of domains associated with mental health, substance use, socioeconomic status and internet use. Information about the participant’s race was collected from parent verbal report during a structured clinical history interview. For participants under the age of 11, a trained research assistant read and explained individual items and collected responses from participants.
2.2. Measures
2.2.1 Schedule for Affective Disorders and Schizophrenia - Children’s Version (K-SADS)
Participants and their parents/guardians were independently administered the Schedule for Affective Disorders and Schizophrenia - Children’s Version (K-SADS)[75] by a trained member of the clinical team. For participants under the age of eleven, clinicians determined whether the K-SADS would be administered to both child and parent or just parent based on verbal function and expected ability to tolerate the interview. Lifetime consensus diagnoses were based on DSM–5 diagnostic criteria using information from both the parent and child along with historical records and other HBN assessments[74] when clarifications were needed.
2.2.2 Internet Addiction Test (IAT)
Young’s IAT is a 20-item questionnaire ranked on a five-point Likert-type scale.[76] Both parent and self-reported versions were used to assess PIU. The total score resulted from summing all 20 items on the questionnaire, resulting in possible scores ranging from 0–100. The internal consistency was adequate in the current sample with a cronbach’s alpha value of 0.91 for self-report and 0.95 for parent-report.
Based on previous work,[9, 45, 77] a total sum score of 40 or above was categorized as Problematic Internet Use (PIU). To check the validity of this threshold, the value of 40 was increased. For subsequent analyses, participants were then categorized as either non-problematic (absence of PIU) or problematic internet users (presence of PIU) using the cut-off value of 40.
2.2.3 The Barratt Simplified Measure of Social Status (BSMSS)
The BSMSS is a parent-reported measure of social status, which is used as a proxy for socioeconomic status (SES).[78] The total sum score used in the current report is calculated by adding the scores for the total occupation and total education subscales, which are answered in regards to both guardians. For participants with a single caregiver, total scores were calculated based solely on that caregiver’s responses. Total scores were then divided into tertiles to determine low, middle and high SES, respectively, for the demographics analysis and included as a continuous covariate for the rest. Since the presence or absence of a secondary caregiver explained a large amount of the partial missing data in the BSMSS, an additional binary variable representing this factor was also included as a covariate in all analyses.
2.2.4 Columbia Impairment Scale (CIS)
The CIS is a 13-item scale that assesses global functioning in domains of interpersonal relations, psychopathology, school performance, and use of leisure time.[79] The total sum score of all items was used for both self and parent-report versions. Internal consistency was adequate in the current sample (cronbach’s alpha for parent-report was 0.85 and 0.81 for self-report).
2.2.5 Physical Activity Questionnaire (PAQ)
The PAQ is a self-report questionnaire asking participants to recall their physical activity levels in various domains for the past seven days.[80] The PAQ-C is administered to children ages 8–14 and the PAQ-A, modified by removing items referring to school recess, is administered to ages 14–19. The total score is created using the average of all items after the first question, which are all rated on a scale from one to five. Cronbach’s alpha was 0.88 for the PAQ-C and 0.87 for the PAQ-A.
2.2.6 Sleep Disturbances Scale (SDSC)
The SDSC is a parent-reported 27-item questionnaire rated on a 5-point Likert type scale that assesses sleep disturbances in various domains (initiating and maintaining sleep, breathing, disorders of arousal, sleep-wake transition, excessive somnolence, and sleep hyperhidrosis).[81] The total sum score of all items was used in the current report. Cronbach’s alpha in the current sample was 0.82.
2.2.7 Body Composition Measures
Participant height and weight were also collected by a trained research assistant in order to calculate Body Mass Index (BMI). Electric impedance and reactance measures were collected using electrodes placed at four positions on the body: two on the right hand and two on the right foot. RJL Systems Bio-Impedance Analysis was used to calculate Fat Mass Index (FMI).
2.3 Analyses
Main analyses were completed using self-report measures (corresponding analyses using parent-report measures can be found in the supplementary material). Participants who did not complete one or more of the questionnaires required for the analyses were removed from the sample. Due to the small number of individual missing items in the remaining sample, complete cases analysis was used. All analyses were also run using the maximum number of complete datasets for each (thus resulting in differential sample sizes for each analysis) in order to determine whether results differed when analyses were run using the final sample (see Tables S7-S12). The final sample consisted of 564 (190 female) participants between the ages of 7–15 (mean = 10.80, SD = 2.16) out of the original sample of 2094 children aged 5–21. Regarding those excluded (n = 1,530), 1,310 were missing data due to key questionnaires included in the present work being introduced later in the protocol’s timeline, and the remaining 220 were missing a combination of demographic information, physical activity information, and/or KSADS administration. All statistical analyses were conducted using The R Project for Statistical Computing for Windows[82] (version 3.5.2).
To examine demographic correlates of PIU, unadjusted prevalence odds ratios were calculated using the R package epiR[83]. Next, logistic regression was used to estimate three odds ratios (ORs) models: unadjusted ORs for associations between PIU and mental disorders, ORs adjusted for age, sex, SES, site, and single caregiver, as covariates, and adjusted ORs that also accounted for comorbid mental disorders. Third, linear regression was used to assess the relationship between PIU and impairment. We again estimated three sets of models: unadjusted, adjusted for age, sex, SES, site, and single caregiver, and further adjusted for mental disorders.. Finally, we estimated associations of PIU with physical health and sleep disturbances using linear regression. Again, we first estimated unadjusted associations, followed by adjustment for demographics, and finally adding adjustment for mental disorders.