Measuring attentional bias to food cues in young children using a visual search task: An eye-tracking study.

Abstract

Objective:

Attentional bias to food cues may be a risk factor for childhood obesity, yet there are few paradigms to measure such biases in young children. Therefore, the present work introduces an eye-tracking visual search task to measure attentional bias in young children.

Methods:

Fifty-one 3–6-year-olds played a game to find a target cartoon character among food (experimental condition) or toy (control condition) distractors. Children completed the experimental and toy conditions on two separate visits in randomized order. Behavioral (response latencies) and eye-tracking measures (initial orientation, initial fixation, cumulative fixation) of attention to food and toy cues were computed. Regressions were used to test for attentional bias to food versus toy cues, and whether attentional bias to food cues was related to current BMI z-score.

Results:

Children spent more cumulative time looking at food versus toy distractors and took longer to locate the target when searching through food versus toy distractors. The faster children fixated on their first food versus toy distractor was associated with higher BMI z-scores.

Conclusions:

Using a game-based paradigm employing eye-tracking, we found a behavioral attentional bias to food vs. toy distractors in young children. Further, attentional bias to food cues was associated with current BMI z-score.

1. Introduction

One in three children is overweight or obese in the United States.¹ Overweight and obesity often tracks into adulthood,^2–4 and those with obesity have increased risk for heart disease, type-2 diabetes, cancer, and other serious comorbidities.⁵ There is growing recognition of the importance of considering individual differences when designing prevention and treatment strategies for complex diseases like obesity.^5,6 Therefore, understanding individual factors that lead some children to overconsume, and ultimately gain excess weight, is essential to the development of effective obesity prevention and treatment strategies.⁷

Food cues in the environment have been shown to stimulate excess consumption.⁸ Through a process of associative learning, cues associated with palatable food intake become increasingly salient,⁹ and children may develop a stronger attentional bias (i.e., a heighted saliency of food versus non-foods in the environment) to these cues.^10,11 Additionally, research supports a genetic predisposition to excess caloric consumption in response to food cues.¹² This type of cued eating is concerning given that our obesogenic environment is replete with cues to eat and cued eating may relate to subsequent adiposity gain.^12,13 Thus, an attentional bias to food cues may lead to a greater propensity to eat in response to environmental food cues and greater subsequent weight gain.

Attentional bias may be measured indirectly or directly. Indirect assessments use behavioral measures (e.g., response latencies), but are limited because they cannot differentiate the automatic attentional processes (e.g., those that are unconscious and not controllable by a person) from the later conscious attentional stages (e.g., those that are under the conscious control).¹⁴ In contrast, direct measures of attentional bias may be measured by monitoring eye movements.¹⁵ Eye-tracking has good spatial and temporal precision, allowing researchers to isolate discrete attentional bias measures that reflect both the automatic and conscious aspects of attention.

Much of the research on the relationship between attentional bias to food cues and weight has been conducted in adults and has produced mixed results.^16–18 This discrepancy may be attributed to the use of differing paradigms that measure distinct aspects of attention in isolation. In contrast, eye-tracking studies that measured both automatic and conscious attentional bias metrics in the same task have produced a more comprehensive understanding of how attentional bias to food cue relates to weight status. Those studies have reported that compared to people with lower BMIs, individuals with higher BMIs tend to display approach /avoidance behaviors that are driven by automatic and conscious aspects of attention. For example, they automatically look at food cues faster, but consciously spend less time fixating on food cues.¹⁹ Thus, measuring attentional bias to food cues using eye-tracking may show promise in identifying children at risk of obesity.

However, because of methodological differences and the use of varying clinical populations, there is a gap in our understanding of the relationship between attentional bias to food cues and BMI in young pediatric populations. For example, although eye-tracking studies have assessed the relationship between attentional bias and BMI by measuring automatic and conscious attentional biases simultaneously, they are limited to comparisons between a special population and healthy weight controls.^19,20 Understanding this relationship in a healthy weight population may help to identify those at high risk for excess weight gain at a critical time in development and before excessive weight gain occurs. For example, because learning theory predicts that overeating follows from learned associations between food cues and intake,²¹ early and consistent exposure to food cues may promote conditioned eating. By this mechanism, it is possible that young children may develop an attentional bias to food cues (especially highly palatable foods) that may lead to excess weight gain. Therefore, the purpose of this work was to develop an age-appropriate visual search paradigm to directly measure attentional bias to food cues in young children for use with eye-tracking.

Visual search paradigms require participants to locate a target object among a set of distractor items.²² When combined with eye-tracking, the visual search paradigm may also provide three direct measures of attentional bias: time to first fixation bias, initial gaze duration bias, and cumulative gaze duration bias.²³ Time to first fixation is thought to measure how an observer prioritizes objects in a scene, and is largely controlled by automatic attentional processes that involve uncontrollable reflexes to look at the onset of a particularly rewarding stimulus.¹⁷ Initial gaze duration and cumulative gaze duration biases are measures of sustained attention to stimuli in a certain category compared to a referent category.²⁴ Initial gaze duration represents the strength of a stimulus to hold one’s attention after it has been initially fixated upon. Initial durations are not purely automatic but are still largely driven by bottom-up reactionary processes, such as the motivational salience of the stimuli. Conversely, cumulative gaze duration represents the strength of a stimulus to hold one’s attention over time and has been associated with late attentional stages that are mostly influenced by top-down goal-oriented processes.²⁰

In addition to measuring both automatic and conscious attentional processes, visual search tasks are easily modified to make them child friendly. To make our visual search task age appropriate for young children, we modified the paradigm to be a hide-and-go-seek task in which the goal was to find a friendly monster hiding among food or control (toy) distractors. Because prior literature suggests that all children, irrespective of current weight status, demonstrate automatic and conscious attentional bias to food cues, it was hypothesized that all children would show time to first fixation, initial gaze duration and cumulative gaze duration biases to food cues. However, because prior literature suggests that overconsumption²⁵ and weight gain,²⁰ is only related to automatic, and not conscious, processes, we hypothesized that time to first fixation bias, but not initial gaze bias nor cumulative gaze duration bias, would be associated with current BMI; that is, children with increased BMI would have an increased time to first fixation bias.

2. Methods

2.1. Study design.

Participants were children and their parents who participated in a larger study that examined the effect of branded TV characters on food consumption. Children aged 3–6 years and one of their parents were recruited from the community using fliers, community listservs, social media, and community events. Eligibility criteria included being between the ages of 3–6-years-of-age, English fluency, absence of relevant food allergies and dietary restrictions, absence of health conditions or medication use that may impact appetite or attention span, and willingness to participate in two 1.5-hour study appointments. At the start of each visit, children were offered a standardized preload snack which consisted of approximately 1.5 times the serving of whole grain crackers (28 grams ± 2 grams), banana (50 grams ± 2 grams), cheese slices (58 grams ± 5 grams), and a small cup of water. They were given 5 minutes to eat ad libitum, and the experimenter offered 5 additional minutes only if the child asked for more time at the end of the first 5 minutes. This preload snack was followed a period in which children then viewed an age-appropriate 12-minute TV program, followed by a 10-minute period in which children were left alone in an observation room and could consume an additional snack (2 servings of EnviroKidz Choco Chimps Breakfast Cereal by Nature’s Path; 60 grams ± 2 grams) and water ad libitum. During this phase, children were also provided with crayons and paper and were given the option to color. However, because of a computer error, complete consumption and eye tracking data was only available for 20 participants, so this outcome was not analyzed. Immediately afterward, children completed the visual search task. Children were randomized to complete a version of the game that included food distractors or control toy distractors during their first visit and completed the game in the reverse order during their second visit. At the first visit, parents completed a demographics questionnaire. All study procedures were approved by The Committee for the Protection of Human Subjects at Dartmouth College. Children provided assent, and parents provided consent for their children and themselves.

2.2. Participants.

Seventy-three participants were enrolled, including four pilot participants who were used to refine study protocols and whose data was not included in the analyses. A further 10 participants were excluded because a computer error resulted in the loss of their anthropometric data. Of the remaining 59 participants, two dropped out after visit one and a further six children were excluded because poor eye-tracking calibration resulted in unusable data. Final analyses were conducted on 51 children and their parents.

2.3. Stimuli and Apparatus.

PsychoPy software²⁶ was used to display the stimuli and controlled timing and response operations. Stimuli were displayed on a 22-inch Elo 2201L touch screen (Elo Touch Solutions, Knoxville, TN, USA) at a screen resolution of 1920 X 1080 and refresh rate of 60 Hz. Stimuli were targets and distractors. Target stimuli included three friendly cartoon monsters that were either pink, blue, or yellow. Distractor stimuli were food images and toy images. The food images were hedonic foods downloaded from a standardized food images database.²⁷ The toy images were downloaded from the food image database’s set of control images and from the Bank of Standardized Stimuli.²⁸ Stimuli were displayed in one of 12 randomly-generated screen locations and positioned on the bottom half of the screen to improve eye-tracking accuracy and to ensure that children could reliably touch the screen while minimizing the amount of movement while seated.²⁹

2.4. Eye movement data acquisition and calibration.

Eye tracking was controlled using PyGaze software.³⁰ Forty-four participants had their binocular eye movements recorded at 60 hz using the Gazepoint GP3 HD eye-tracker (Vancouver, British Columbia). Seven participants had their binocular eye movements recorded at 500 Hz using the Eyelink 1000 (SR Research, Mississauga, ON, Canada). Participants were calibrated using PyGaze’s standard calibration procedure.

2.5. Procedure.

To ensure accurate eye-tracking with a touchscreen in young children, we used the procedure developed by O’Hanlon and Read.²⁹ Children were seated in a car seat with the center of the screen approximately 60 centimeters away from their forehead. The touchscreen and eye tracker were anchored to a customized mount that could move vertically and horizontally to accommodate for height differences between children.

The visual search game began with a pink, blue, and yellow cartoon monster presented in the middle of the screen. The child was instructed to select the monster they wanted to play the game with by the touching the preferred monster on the screen. We allowed children to choose their target monster to facilitate continued interest in game play and encourage task completion. The game started once the child made their selection. The child was told that the preferred monster would be hiding among food (or toy) images, and that their job was to find the monster. To help ensure optimal tracking conditions, children were instructed to sit with their hands on their knees and to only reach for the screen once they spotted the monster. To allow children to reach the screen without having to lean forward, they were given a stylus.

Each trial began with the presentation of a search screen that only contains the target cue (and no distractors) for 250 milliseconds (figure 1).The monster and distractor stimuli were then overlaid on the screen and stayed visible until the child correctly identified the monster’s location by tapping on the image of the monster using a stylus. Because children were forced to correctly identify the location of the monster to advance to the next trial, there was no possibility of an incorrect response. The display was comprised of the monster and 3, 5, or 8 food (or toy) distractors. A blank screen was presented for 1.5 secs after successful localization. The presentation of the blank search screen signaled the start of a new trial.

Figure 1.

An example of a trial sequence.

Children completed three practice trials to familiarize themselves with the task. Practice trials were not analyzed. Children then completed three blocks of 12 trials for a total of 36 trials. Children were provided with a break at the end of each block. Continuation of the game was initiated by the experimenter pressing the spacebar once the child verbally indicated s/he was ready to continue.

2.6. Response latency as an indirect measure of attention to food and toy distractors.

Response latencies to indicate the location of the monster were used as an indirect measure of the amount of attention given to food and toy distractors. Response latencies were calculated under the food and control conditions at each set size for each participant. To calculate an attentional bias to food cues, average response latencies during the food condition were first subtracted from the average response latency during the control condition for each set size. These values were then averaged to create a single attentional bias to food cues measure. A positive value was interpreted as an attentional bias to food cues.

2.7. Direct eye-tracking measures of attention to food and toy distractors.

Eye movement measures of attention to food and toy distractors were calculated for each participant for each set size. Time to first fixation was measured as the time it took participants to fixate their first food or toy distractor. Initial gaze duration was measured as the average amount of time a child spent fixating a food or toy distractor the first time they looked at it. Cumulative gaze duration was calculated as the total amount of time participants spent fixating food or toy distractors, summed over all fixations. For all attentional measures, a fixation duration was defined as any stationary period lasting at least 100 milliseconds.¹⁹ Areas of interest were created around distractor and target images, defined as the area of the image, 250 X 250 pixels.

The three eye-tracking measures of attention to food and toy distractors were used to calculate direct measures of attentional bias to food cues for each participant. A time to first fixation bias to food cues was calculated by first computing the difference between time to first fixation to food and toy distractors at each set size and then averaging the resulting values. A negative value represents that a child was faster to fixate their first food versus toy distractor. Similarly, an initial gaze duration bias to food cues was calculated by first computing the difference between initial gaze duration to food and toy distractors at each set size and then averaging the resulting values. A positive value represents that a child looked longer the first time they fixated a food versus toy distractor. Finally, cumulative gaze duration bias to food cues was calculated by first computing the difference between cumulative gaze duration to food versus toy distractors at each set size and averaging the resulting values. A positive number indicates that a child spent more total time looking at food versus toy distractors.

2.8. Child, parent, and household characteristics.

The parent who accompanied the child to the visit reported on their child’s age, sex, and race/ethnicity. They also reported on their own household income and highest educational level completed with values provided in Table S1. Sex was coded as male or female. Race and ethnicity were coded as white, non-Hispanic or other. Child age was treated as a continuous variable. Household income and parental education were coded as ordinal.

2.9. Child anthropometry.

The weight and height of children were measured using a Seca 763 Medical Scale and Seca 213 Stadiometer (Hamburg, Germany). Child measurements were used to compute age- and sex-adjusted BMI z-scores using the United States Center for Disease Control and Prevention 2000 growth charts.³¹

3. Statistical Analysis

All analyses were conducted using the R language and environment for statistical computing.³² We first assessed the visual search task for completeness. Of the 51 children, 40 completed all 36 trials of both game conditions. No child completed less than 6 trials of each condition. Trials were then screened to eliminate those on which a distractor was not fixated. Trials were further screened to eliminate trials with excessively long response latencies, defined as being 2.2 SD above the mean.³³

To identify potential covariates, distributions for each attentional bias to food cues measure were compared across the child, parent and household characteristics described above (see 2.8) using regression. These variables were examined because they have been shown to be related to childhood obesity,³⁰ and could potentially confound the association between attentional bias to food cues and weight. Any characteristic that was associated with at least one of the attentional measures at the P<0.10 level was included as a covariate in all adjusted analyses. Additionally, we included sex and age in all adjusted analyses. We also evaluated whether monster color was associated with any of the attentional bias measures because color differences could have potentially increased or decreased the saliency of the distractors relative to the target. There were no significant associations between monster color and any of the attentional measures, so this factor was not included in any of our adjusted analyses.

We evaluated whether participants showed an attentional bias to food versus toy distractors, by using unadjusted mixed effects regressions, nested within subject and set size. In these models, we fit response latencies, time to first fixation, initial gaze duration, and cumulative gaze duration from game condition and set size. We then repeated these regressions adjusting for selected covariates. We used a statistical significance threshold of P<0.05 when evaluating main effects in these models and others described below.

To test whether attentional bias to food cues differed depending on the number of items displayed, we included an interaction term between game version and set size. We used a threshold of P<0.10 when evaluating the statistical significance of the Wald test on the interaction because power for detecting statistical interactions is low, especially relative to the power to detect main effects.³⁴

We next examined whether BMI z-scores were associated with attentional bias to food versus toy distractors by fitting unadjusted and adjusted mixed effects linear regressions, nested within subject, predicting the four attentional bias to food measures from BMI z-scores.

4. Results

The majority of children were between 3–5 years of age (M=4.24, SD=0.91) and most children were white, non-Hispanic (92%). Ten children (19.6%) were overweight or obese, defined as having a BMI z-score greater than 1.04 (i.e., BMI percentile ≥ 85^th). A further two children (4.0%) were underweight defined as having a BMI z-score less than −1.88 (i.e., BMI percentile ≤ 3^rd). The remaining thirty-nine children in our sample had BMI z-scores corresponding to a healthy weight (76.4%). Thus, the majority of our sample was of predominately healthy weight (M= 0.21, SD = 0.99).³⁴

4.1. Child, parent, and household characteristics.

Table 1 shows the unadjusted associations between measures of attentional bias to food cues and sex and age. There were no statistically significant associations between age and response latency bias (t(49)=0.174, P=0.862); time to first fixation bias (t(49)=0.205, P=0.838); initial gaze duration bias (t(49)=−0.023, P=0.982), and cumulative gaze duration bias (t(49)=−0.210, P=0.835). There were also no associations between sex and response latency bias (t(49)=1.33, P=0.19); time to first fixation bias (t(49)=−0.122, P=0.904); initial gaze duration bias (t(49)=0.093, P=0.926); or cumulative gaze duration bias (t(49)=0.95, P=0.347). With respect to the other potential covariates, there were no differences between any of the attentional bias to food cue measures and race/ethnicity. Children of parents with higher levels of education had a greater time to first fixation bias than children of parents with lower levels of education (β=−47.41; 95% CI= (−94.66, −0.17); t(49)=−2.02, P=0.049) but there were no associations with the other attentional bias measures. In addition, children from higher income households had a greater time to first fixation bias than children from lower income households (β=−45.88; 95% CI=(−91.24, −0.53); t(49)=−2.03, P=0.048) but there was no association with the other attentional bias measures.

Table 1.

Unadjusted associations between attentional bias measures to food cues with sex and age.

		Response latency bias (msec)	Time to first fixation bias (msec)	Initial gaze duration bias (msec)	Cumulative gaze duration bias (msec)
	N	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Age, years
3	11	364.78 (1827.37)	12.52 (165.08)	41.68 (192.12)	370.79 (1686.61)
4	22	13.69 (1238.82)	−36.58 (166.47)	10.13 (149.24)	−7.98 (1128.17)
5	13	1033.87 (1597.69)	−27.71 (140.40)	46.24 (282.19)	772.82 (1733.08)
6	5	−403.21 (1108.03)	52.47 (135.41)	14.76 (145.13)	−538.54 (959.91)
β (95% CI)*		41.11 (−433.23, 515.45)	5.02 (−44.12, 54.17)	−0.70 (−62.22, 60.82)	−47.28 (−501.82, 407.27)
Sex+
Female	32	94.96 (1600.18)	−12.94 (159.11)	24.63 (180.04)	76.21 (1524.67)
Male	19	668.39 (1273.34)	−18.47 (153.23)	29.90 (221.83)	472.31 (1277.71)
β (95% CI)+		573.43 (−292.90, 1439.75)	−5.53 (−96.89, 85.83)	5.28 (−109.04, 119.60)	396.09 (−441.37, 1233.55)

^*Children age in years coded as continuous.

⁺Sex coded as binary, 0 = Female, 1 = Male.

Degrees of freedom for all analyses = 49.

4.2. Attentional bias to food cues.

Attention metrics to food and toy distractors at each distractor set size are shown in Figure 2. Children were approximately 338 milliseconds slower to locate the target monster in the food versus toy condition (β = −338.70; P = 0.022) (Panel A). Similarly, there was a trend to suggest that children spent approximately 253 milliseconds more looking at food distractors than toy distractors (β = −253.28; P = 0.078) (Panel D). Children were equally fast in the time to their first fixation on food and toy distractors (β = 13.17; P = 0.461) and did not differ in the amount of time they spent initially fixating food and toy distractors (β = −27.65; P = 0.182) (Panels B and C). There were no significant set size effects for any of the attention metrics, nor were there any significant set size by condition interactions. All of these results remained unchanged after adjusting for sex, age, annual household income, and parental education level.

Figure 2.

Mean attention metrics for food and toy conditions at each set size. A. Response latency. The average time needed on a trial to identify the target monster for food and toy conditions. B. Time to first fixation. The average amount of time needed to fixate the first food or toy distractor on a trial. C. Initial gaze duration. The average amount time participants spent looking at the first food or toy distractor they fixated. D. Cumulative gaze duration. The average amount of time children spent fixating food or toy distractors, summed over all fixations on each trial. Error bars represent SEM.

4.3. Association between attentional bias to food cues and BMI z-scores.

Unadjusted associations between attention to food cue metrics and BMI z-scores are shown in Figure 3. Higher BMI z-scores were associated with faster times to first fixation of food versus toy distractors (β = −50.44; P =0.022) (Panel B). BMI z-score was not associated with the average amount of time participants spent initially looking at their first food versus toy distractor (β = −8.17; P = 0.772) nor with the cumulative amount of time they fixated on food versus toy distractors (β = 77.05; P = 0.711) (Panels C and D). Similarly, the difference in response latencies between food and toy conditions was not associated with BMI z-score (β = 92.42; P = 0.609) (Panel A). These results remained unchanged after adjustment for sex, age, household income and parental education. (see Table S1).

Figure 3.

Unadjusted associations between BMI z-scores and A. Response latency bias B. Time to first fixation bias C. Initial gaze duration bias D. Cumulative gaze duration bias. Arrow directionality indicates attentional bias to food cues for each attentional bias measure.

5. Discussion

Overall, we found evidence of attentional bias to food cues measured indirectly through response latency. That is, children were slower to identify the target monster when searching through food versus toy distractors. This association appeared to be driven by the increased amount of cumulative attention that children gave to food versus toy distractors. Additionally, although children did not differ in the amount of time they took to fixate their first food versus toy images, faster times to fixate on the first food versus toy distractor were associated with higher BMI z-scores.

Our finding that higher BMI z-score was associated with increased time to first fixation to food versus toy cues contrasts previous studies that have shown that all children, not only children with higher¹⁹ or lower²⁴ BMI, tend to have an automatic attentional bias to food cues. However, this study differed from previous studies in that it measured time to first fixation as an automatic attentional bias to food cues compared to orientation bias used by previous work. An orientation bias to food cues represents the tendency of children to fixate a food cue before a control cue presented at the same time. Thus, an orientation bias, although considered an automatic process, may represent a different cognitive mechanism than the speed at which a child fixates a stimulus. Although we can only speculate, these findings may suggest that while all children have an orientation bias to food cues, only children with increased BMI demonstrate a time to first fixation bias. We were not able to measure an orientation bias in this study bias because food and toy images were never displayed at the same time.

Nevertheless, our results are consistent with previous eye-tracking research that found that automatic attentional biases to food cues may be a risk factor for overconsumption²⁵ and excess weight gain²⁰ in older children. This finding further supports the notion that retraining automatic attentional biases to food cues might be useful to decrease or prevent overeating and adiposity gain. For example, it is possible to retrain automatic attentional biases related to addictive behaviors (e.g., tobacco and alcohol) using attentional retraining paradigms.¹⁴ Although little research has examined the effect of attentional retraining on food cues in children, a pilot study showed directing attention away from food cues was related to decrease consumption in an eating in the absence of hunger paradigm in children with obesity. ³⁵

In contrast to our hypothesis and to previous work presented in Werthmann et al (2015),²⁰ we found no evidence of an initial gaze duration bias to food cues. This may be attributed to increased competition from equally salient distractors (i.e., children find both food and toys rewarding). For example, compared to other attentional bias paradigms that show a single critical versus control stimulus, our search arrays were constructed using multiple stimuli of the same category. This may have caused decreased initial gaze durations as children may switch attention between several equally compelling stimuli that are presented on the screen simultaneously. Nevertheless, and consistent with Werthmann’s work,²⁰ we found evidence of a cumulative gaze duration bias in which children spent more time looking at the food images overall compared to toy images.

To our knowledge, the present work is the first to document an attentional bias to food cues in a predominantly healthy-weight population of children as young as three. As such, our results may help to understand how these biases develop in early childhood. When viewed from an incentive salience framework, our results suggest that the development of an attentional bias to food cues is the result of associative learning between food cues and intake that occurs at a very young age, possibly prior to weight gain. Although research has shown that associative learning is one mechanism through which children as young as two may start to develop food preferences³⁶, there is no evidence that associative learning plays a direct role in the formation of an attentional bias to food cues. To date, research has been limited to studying the formation of food preferences through associative learning of food intake and social contexts of eating (e.g., the times of day at which foods are consumed).³⁶ Future research is needed to examine the formation of food preferences through associate learning of environmental food cues eating behavior.

Our finding of an attentional bias measured indirectly via response latencies, but not directly via cumulative duration bias, is surprising given that direct measures of attentional bias are generally considered more robust. A potential explanation for this observed discrepancy may be that the additional cognitive processes required to execute the motor response may also be differentially affected by food vs. non-food cues. For example, attending to a food cue may persist in one’s memory longer than a non-food cue³⁷ and impair reaction time during the search task. This hypothesis is completely speculative, though, and requires investigation.

Our results should be viewed with respect to the following limitations. The target monsters used in the task were not visually matched on shape and color with food and toy distractors. It is therefore possible that attention was drawn to food and toy distractors because of low-level image statistics or visual saliency differences with the target monster. Similarly, we did not balance the food and toy stimuli for attractiveness and brightness, and this may have resulted in attention differences to each type of distractor. This may have affected the finding of a response bias latency to food cues but would not have affected the association between time to first fixation bias to food cues and BMI z-score. Finally, although the number of trials in our task was purposely kept low to encourage game completion, it is possible that this resulted in imprecise measures of attentional bias. This may have reduced our power to observe associations between the attentional bias measures and weight status. Given that the majority of children were able to complete all 30 trials as planned, future research may attempt to increase the number of trials to increase the robustness of the attentional measures. In addition, because of our small sample size we were unable to test for effect modifiers, such as age.

The present work has several strengths. One of its largest is that it is the first to investigate the association between attentional bias to food cues and BMI in children as young as three. Similarly, it is innovative to measure this association in a predominately healthy-weight sample. This study is also among the first to gamify an attentional bias to food cues paradigm. Developing age-appropriate games using established paradigms may improve protocol compliance and reduce participant burden in young children, while providing robust measures of attentional bias.

In conclusion, we have introduced a novel age-appropriate game to assess indirect and direct measures of attentional bias to food cues in young children. We found evidence of a general attentional bias to food versus toy distractors and a cross-sectional relationship between time to first fixation bias to food cues and BMI z-score in young children. This game-based paradigm may represent a robust and engaging way to measure attentional bias to food cues in young children.

Highlights.

• There is a gap in our understanding of the relationship between attentional bias to food cues and weight in young pediatric populations.

• The present works introduces a novel eye-tracking visual search task to measure attentional bias to food cues in young children.

• Our findings demonstrate an association between attentional bias to food cues and BMI z-score in a predominantly healthy weight pediatric population.