Study design and participants
This study took place between July 2021 and March 2022 at the Ajou University Hospital in South Korea. Patients with ADHD and healthy controls (aged 6–10 years) were recruited through advertisements placed on bulletin boards around the hospital. The study received institutional review board approval, and all participants and caregivers provided written informed consent (no. AJIRB-MED-SUR-21-240).
We included patients in the ADHD group after a psychiatrist confirmed the diagnosis according to the criteria of the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). Healthy controls were evaluated by psychiatrists for psychiatric symptoms and medical history, including ADHD. Those with a history of eye disease, autism spectrum disorder, intellectual disability, major depressive disorder, bipolar disorder, schizophrenia, Tourette syndrome, obsessive–compulsive disorder, post-traumatic stress disorder, neurological disease, or severe medical problems were excluded.
Patients with ADHD were also followed to compare symptoms by medication usage, including stimulants (methylphenidate) and non-stimulants (atomoxetine and clonidine). Among children who already used medication, primary testing took place after stopping the drug for at least one week, whereas all other children underwent primary testing before they started the drug. Follow-up testing took place 1 month after starting or re-starting ADHD medications.
Participants and caregivers completed a demographic/health questionnaire and Korean versions of the ADHD Rating Scale (ARS) and Child Behavior Checklist (CBCL). To minimise external distractions, participants were then moved to a separate room with a computer, where they underwent calibration for the eye tracker and comprehensive attention testing (CAT) while their eye movements were tracked. Given the potential effect of medication on ADHD symptoms, together with the possibility of fatigue, all testing took place in the morning or early afternoon. We repeated the CAT once during follow-up after patients had received medical treatment for ADHD.
Attention and psychopathology
Attention was evaluated with the ARS, an 18-item scale developed by DuPaul (1991) for use with children16. Symptoms are rated on 4-point Likert-type scales, ranging from 0 (never) to 3 (very often). The Korean version of ARS has internal consistency ranging from 0.77 to 0.89 and test–retest reliability of 0.8517.
Psychopathology was assessed with the CBCL, which contains 120 behavioural items that parents rate on 3-point Likert-type scales from zero to two (Not True to Very True/Often True). Items are summed to yield a syndrome scale score across three dimensions (internalising, externalising, and total behavioural problems) and six DSM-oriented scale scores. The syndrome and DSM-oriented scales have been validated18, and the Korean version of the CBCL was standardised in 199719.
Computerized CAT is the type of the computer-based CPT, and has been developed for ages 4–4920. The CAT is composed of six subtests: the simple selective attention (visual and auditory), continuous inhibition, interference selection, divided attention, and working memory tests. However, we excluded the auditory test for simple selective attention to allow comparison with the eye-tracking test, together with the divided attention and working memory tests that are only used from age 9 years. The CAT was performed using a computer, with participant understanding of text and voice guides presented at the start of each subtest checked by trained researchers.
Overall, the amended CAT took approximately 25 min to complete, including the assessments of selective visual attention (300 stimuli, 10 min), continuous inhibition (300 stimuli, 10 min), and interference selection (150 stimuli, 5 min). For the selective attention test, participants press the space bar button quickly when they see a circle figure at the center of a monitor. For the continuous inhibition test, they press the space bar when they see any figure except an X at the center of monitor. For the interference selection test, participants are instructed to pay attention to a central target while ignoring interference stimuli. Each subtest has five indicators: commission errors (CE), for the number of wrong responses; omission errors (OE), for the number of missed responses; mean reaction time (RT mean), for the average response time to the stimuli; standard deviation of reaction time (RT SD), for response time variability; and sensitivity coefficient (d′), for how successfully the target stimuli are differentiated from the non-target stimuli. Because only four indicators were calculated in eye-tracking, we excluded d′ from the comparison.
Eye-tracking apparatus and eye movement measures
Stimuli were presented on a Samsung Notebook (NT551XCJF-COM) with a 15.6-inch display, a screen resolution of 1920 × 1080 pixels, and an eye-to-screen viewing distance of approximately 50 cm. The eye-tracking apparatus (Happymind Inc. CAT test) included a host PC that tracked and computed the participant’s gaze position, as well as a display PC to present the stimuli. After downloading and running the eye-tracking programme (SeeSo; https://visual.camp/demo-archive/), eye movements were recorded at a 30 Hz sampling rate with an approximate accuracy of 1.7° (VisualCamp Co., Ltd, Seoul, Korea). Calibration to each participant in SeeSo used a five-point procedure. Online Supplementary Fig. S1 shows the graphical user interface and gaze coordinate of the eye-tracking programme.
To compare the extent of visual attention directed to the task and irrelevant regions, the participants’ field of view was divided into central and peripheral areas of interest (AOIs). As shown in online Supplementary Fig. S2, the central AOI represented the middle third of the width and length. Each subtest had four indicators: fixation ratio (FR), for the ratio of gaze fixation; mean fixation time (FT), for the average gaze fixation time to the screen; central gaze ratio (CR), for the central AOI gaze ratio; and standard deviation of gaze coordinates (gaze SD), for gaze variability. The equations used are presented in online Supplementary Fig. S3.
NCSS PASS (version 14) was used for the sample size calculation21. A recent study of eye-tracking among patients with ADHD showed that the ratio of center gaze duration between patients with ADHD (80.48%) and a healthy control group (88.35%) differed significantly according to Welch’s unequal variance t-test14. Therefore, allowing for a 5% probability of a type 1 error and a power of 80%, the minimum sample size was 29 participants in each group. Considering drop out, we decided a total sample size of 30 in each group.
We compared baseline characteristics, ARS, CBCL, CAT indicators, and eye-tracking indicators between the ADHD group before medications and the control group by independent-sample t-tests and chi-square analyses for parametric and non-parametric variables, respectively. Welch’s unequal variance t-test was used when data failed to meet the assumption of variance homogeneity. Group differences in gaze were visualised using the gaze coordinates in subtests.
Pearson’s correlation between CAT and eye-tracking indicators was evaluated before performing the regression analyses. Using the correlation matrix, we considered that indicators with r-values of > 0.7 had multicollinearity22, which we evaluated further based on a variance inflation factor (VIF) of < 523. Logistic regression then assessed the ability of the CAT indicators, eye-tracking indicators, or both indicators combined to identify group membership (control or ADHD). Sensitivity, specificity, and area under the curve (AUC) were compared against patients with ADHD by receiver operator characteristic (ROC) curve analysis. The method reported by DeLong et al. was used to compare AUC values24.
In the secondary analysis, we used paired t-tests to assess the change in ARS, CBCL, and indicators (CAT and eye-tracking) within the medication group from before to after taking medication. Differences in gaze from before to after taking medication were visualised by using gaze coordinates according to subtests.
Statistical significance was evaluated at the 5% significance level (p < 0.05), and all analyses were performed using R (version 4.1.0) and its open-source statistical packages.
This study was approved by the Ajou University Hospital Institutional Review Board (AJIRB-MED-SUR-21-240), and All participants and caregivers provided written informed consent. All the experiment protocol for involving human data was in accordance with the guidelines of Declaration of Helsinki.