INTERSTAARS: Attention training for infants with elevated likelihood of developing ADHD: A proof-of-concept randomised controlled trial.

**Intervention:** Nine weekly sessions of gaze-contingent attention training delivered via eye-tracking technology. Infants watched animated games on a screen that responded in real-time to where they were looking. The games targeted three cognitive domains:

**Sustained attention:** Maintaining gaze on a stimulus to receive an audiovisual reward (e.g., a cartoon animal doing something fun when the infant kept looking at it)

**Working memory:** Remembering where to look to receive an audiovisual reward (e.g., a character appearing in a specific location that the infant had to recall)

**Cognitive control:** Inhibiting a previous rule to learn a new one (e.g., learning that looking left now gives a reward, even though looking right worked before)

Each session included six games, each played for a maximum of 300 seconds (5 minutes), or until the infant became disengaged for 20+ seconds. The software automatically recommended game order, pseudo-randomised across visits.

**Active Control:** Infants watched infant-friendly but non-gaze-contingent television clips under the same procedure. The set-up, duration, and audio-visual style matched the intervention as closely as possible. This is important because it controls for the effects of having a researcher visit your home, setting up equipment, and giving your infant screen time — so any differences between groups can be attributed to the gaze-contingent training specifically, not just "doing something special."

**Primary Outcome:** A composite of three attention measures (sustained attention, disengagement, and cognitive control) assessed at baseline, midpoint (after session 5), and endpoint (after session 9). The composite was analysed jointly via a multivariate ANCOVA.

**Secondary Outcomes:** Behavioural attention measures (not specified in detail in the abstract, but the paper reports no significant effects on these either).

**Sample size:** 43 infants (20 intervention, 23 control)

**Age range:** 9 to 16 months old at enrolment

**Population:** Infants with a first-degree relative (parent or older sibling) with a clinical or probable diagnosis of ADHD — this gives them an "elevated likelihood" of developing ADHD (approximately 13-fold increased risk compared to the general population)

**Setting:** Two UK sites (London and Southampton), with training sessions delivered in the family's home

**Recruitment period:** November 2015 to November 2018

**Inclusion criteria:** Residence within 2-hour travel distance of London or Southampton; at least one parent/caregiver fluent in English; first-degree relative with confirmed clinical or research-probable ADHD diagnosis

**Exclusion criteria:** Serious medical/developmental conditions (epilepsy, heart conditions, cerebral palsy, intellectual disability); significant uncorrected vision/hearing problems; significant prematurity (<36 weeks); genetic conditions (e.g., Down's syndrome); inability to reliably track infant's eyes during baseline assessment after four attempts

**Key demographic detail:** Because this is a high-risk sample (familial ADHD), these infants are not representative of the general infant population. Results may not generalise to infants without a family history of ADHD.

**Primary outcome composite (home-based assessments):**

1. **Sustained attention:** Infants watched five repetitions of each of two complex pictures, each presented until the infant looked away for >1 second. The longest of the five individual looks towards each stimulus was averaged. Minimum 4 "looks" per infant required for inclusion.

2. **Disengagement:** Infants were presented with a minimum of 48 trials where a central stimulus was followed by a peripheral stimulus. In "Overlap" trials, the central stimulus remained on screen when the peripheral appeared (requiring the infant to disengage attention). In "Baseline" trials, the central stimulus disappeared before the peripheral appeared. Disengagement scores were computed as the logged difference between mean saccadic reaction times in Baseline and Overlap conditions. Minimum 5 valid trials per condition required.

3. **Cognitive control:** Infants were presented with 18 trials where after fixating a central fixation point, a peripheral stimulus appeared. The specific details of the cognitive control task are not fully described in the abstract, but it involved rule-learning and inhibition.

**Equipment:** Tobii X2-60 eye-tracker (60 Hz sampling rate) — a portable, infant-friendly eye tracker that uses infrared light to track gaze direction in real time.

**Feasibility measures:** Uptake (how many families agreed to participate), compliance (how many sessions completed), session duration, and post-session fussiness (parent-reported).

**Safety measure:** Sleep (parent-reported, to check if training disrupted sleep patterns).

**Study Design:** Proof-of-concept Phase 2 randomised controlled trial (RCT). This is an early-stage trial designed to test whether an intervention is feasible and shows promise of efficacy before moving to larger, definitive trials.

**Randomisation:** Individual randomisation using the King's Clinical Trial Unit (KCTU) web-based service, with minimisation over trial site and infant sex. Minimisation is a method that ensures balance across groups for key prognostic factors — here, site (London vs. Southampton) and sex (male vs. female). This is important because these factors could influence outcomes independently of the intervention.

**Blinding:** This was a double-blind trial:

Researchers who administered baseline and outcome assessments were blind to treatment condition

Parents were blind to whether their infant was receiving the active intervention or the control condition (both involved home visits with eye-tracking equipment; the control group watched videos that looked similar to the games)

Data processing and extraction were done by researchers blind to group allocation

Analysis was completed using uninformative treatment labels

This is a major strength. Blinding prevents expectation effects — if parents or researchers knew which group an infant was in, they might unconsciously behave differently, affecting the results.

**Duration:** Nine weekly sessions, with an intermediate assessment after session 5 and outcome assessment after session 9. Families could reschedule a session up to two weeks late; after that, the session was skipped. Maximum 3 skipped sessions allowed. The total intervention period was approximately 9–11 weeks per family.

**Adherence criteria:** Minimum 6 training/control sessions completed, with a minimum average session duration of 5 minutes per session (>2SD below the mean of training durations in a pilot study with typically developing infants).

**Statistical Approach:** Intention-to-treat (ITT) analysis — meaning all randomised participants were analysed in their assigned groups regardless of how much training they actually received. This is the gold standard for RCTs because it preserves the benefits of randomisation and reflects real-world effectiveness (where not everyone completes treatment). They also reported Complier-Average Causal Effect (CACE) analysis, which estimates the effect among those who actually complied with the intervention.

The primary analysis used multivariate ANCOVA (analysis of covariance) on the composite of attention measures, with baseline scores as covariates. This adjusts for any pre-existing differences between groups and increases statistical power.

**What this design can prove:**

Causality: Because of randomisation and blinding, any differences between groups at endpoint can be attributed to the intervention (not to pre-existing differences or placebo effects)

Feasibility: The trial directly tested whether home-based gaze-contingent training is practical with high-risk infants

**What this design cannot prove:**

Long-term effects: The trial only measured outcomes immediately after the 9-week intervention. It cannot tell us whether any effects would persist or translate into reduced ADHD symptoms in childhood

Clinical significance: Attention measures are not the same as ADHD diagnosis. Even if the training improved attention, it might not prevent or reduce ADHD symptoms

Generalisability: The sample is small (n=43) and specific (infants with familial ADHD risk). Results may not apply to other populations

Dose-response: The trial tested one specific dose (9 sessions, ~5 minutes each). Different doses might produce different results

**Major methodological weaknesses:**

Small sample size (n=43) — the study was powered as a proof-of-concept, not a definitive efficacy trial. The confidence intervals are wide (95% CI −0.9 to 0.2), meaning the true effect could range from a moderate negative effect to a small positive effect

The primary outcome was a composite of three measures, but the paper does not report whether individual measures showed different patterns

53% of post-training lab visits occurred outside the stated 4-week window, forcing reliance on home-based measures (which may be less controlled than lab measures)

The control condition (watching videos) may not have been truly inert — passive viewing could have some effects on attention, potentially reducing the apparent benefit of the intervention

**Primary outcome (attention composite):**

No significant difference between intervention and control groups

Effect size (ITT): ES −0.4, 95% CI −0.9 to 0.2

Effect size (CACE, compliers only): ES −0.6, 95% CI −1.6 to 0.5

Both confidence intervals cross zero, meaning the results are consistent with no effect, a small negative effect, or (less likely) a small positive effect

**Secondary outcomes (behavioural attention):**

No significant differences between groups (specific statistics not reported in abstract)

**Feasibility outcomes:**

Uptake and compliance were good — families were willing to participate and complete the sessions

No adverse effects on sleep (the training did not disrupt sleep patterns)

Small increase in post-intervention session fussiness in the intervention group compared to control (specific numbers not reported in abstract)

**Interpretation of the negative effect size:**

The intervention group performed slightly *worse* on the attention composite than the control group (ES −0.4). This is unusual and concerning. Possible explanations include:

Chance finding (the confidence interval includes zero)

The training was actually counterproductive (perhaps over-stimulating or frustrating for some infants)

The control condition (video watching) had unexpected benefits for attention

Measurement issues (the composite may not have captured the right aspects of attention)

The effect size of −0.4 means the intervention group scored about 0.4 standard deviations *lower* on the attention composite than the control group. To put this in context:

In educational interventions, an effect size of 0.2 is considered small, 0.5 medium, and 0.8 large

A negative effect of 0.4 is equivalent to the intervention group performing at approximately the 35th percentile of the control group — meaning the average infant in the intervention group scored worse than about 65% of infants in the control group

The 95% confidence interval (−0.9 to 0.2) means the true effect could range from a large negative effect (worse attention) to a small positive effect (better attention)

For comparison, the original proof-of-concept study in typically developing infants (Wass et al., 2011) found positive effects of similar gaze-contingent training. The failure to replicate in this high-risk sample suggests that infants with familial ADHD risk may respond differently to attention training, or that the training protocol needs modification.

**What the authors acknowledge:**

This was a proof-of-concept trial, not powered for definitive efficacy testing

The primary outcome was a composite measure; individual components may have different sensitivities

Longer-term outcomes remain to be assessed

The study highlights challenges for pre-emptive intervention approaches

**What a critical reader would note:**

1. **Small sample size (n=43):** With only 20 infants per group, the study has limited statistical power to detect small-to-moderate effects. The wide confidence intervals reflect this uncertainty.

2. **Negative point estimate:** The intervention group performed worse than controls. While not statistically significant, this is a red flag that warrants investigation before proceeding to larger trials.

3. **Composite outcome concerns:** Combining three different attention measures into one composite may obscure important patterns. For example, the training might improve sustained attention but impair disengagement, cancelling out in the composite.

4. **Home vs. lab assessment:** The decision to use home-based measures (because lab visits were often delayed) introduces variability. Home environments are less controlled — distractions, noise, and parent presence could affect infant attention differently across families.

5. **Control condition:** Watching videos is not a true placebo. Passive viewing might train attention in a different way (e.g., sustained visual attention to a screen), potentially reducing the contrast with the active intervention.

6. **No long-term follow-up:** The study only measured immediate post-training effects. Even if attention improved, it might not translate to reduced ADHD symptoms years later. Conversely, effects might emerge only after a delay (sleeper effects).

7. **High-risk sample only:** Results may not generalise to infants without familial ADHD risk, or to infants with other risk factors (e.g., prematurity, prenatal exposure).

8. **Fussiness increase:** The small increase in post-session fussiness in the intervention group suggests the training may have been more demanding or frustrating for infants, which could affect engagement and learning.

9. **No active control for engagement:** The gaze-contingent games are interactive and adaptive, while the videos are passive. This means the intervention group had more cognitive demands during sessions, which could explain the fussiness increase and potentially the negative attention effects (if the training was too challenging).

10. **Publication bias:** This is a null result published in a reputable journal, which is commendable. However, the broader literature on cognitive training in infancy may have publication bias toward positive results, making it difficult to assess the true evidence base.

For someone running their own n=1 experiment (e.g., a parent considering attention training for their infant):

### What to test

**Specific intervention:** Gaze-contingent attention training using eye-tracking games that adapt to where your infant looks. The games target sustained attention (maintaining gaze for reward), working memory (remembering locations), and cognitive control (switching rules).

**Dose:** 9 sessions, each ~5 minutes of active game time, delivered weekly. Total intervention time: approximately 45 minutes over 9 weeks.

**Alternative:** If you don't have access to eye-tracking equipment, you could test simpler attention games (e.g., peek-a-boo, hiding objects, matching games) that target similar cognitive skills.

### Minimum meaningful duration

The trial used 9 weekly sessions. For an n=1 experiment, you would need at least 4–6 weeks to see any pattern, and ideally 8–12 weeks to assess trends.

Include a 2-week baseline period where you measure attention without any intervention, to establish your infant's typical performance.

### What to measure

**Sustained attention:** Time your infant spends looking at a single stimulus (e.g., a picture, a toy, a video) before looking away. Measure the longest single look duration across 5+ trials.

**Disengagement:** How quickly your infant can shift attention from one stimulus to another. You can test this by presenting a central toy, then introducing a peripheral toy, and timing how long it takes your infant to look at the new toy.

**Cognitive control:** Simple rule-learning games (e.g., "look left for reward" then switching to "look right for reward"). Measure how many trials your infant needs to learn the new rule.

**Fussiness:** Record how irritable or distressed your infant is during and after training sessions (1–5 scale).

**Sleep:** Track sleep duration and quality (night wakings, nap patterns) to check for adverse effects.

### Key confounds to control for

**Developmental change:** Infants develop rapidly. Any improvements in attention could be due to maturation, not training. Control for this by measuring a control skill that the training does not target (e.g., motor development, language).

**Parent interaction:** If you are delivering the training, your behaviour may change (more attention, more