Background and study aims
Cognitive bias modification techniques have been used to promote and sustain healthier food preferences, especially in individuals susceptible to weight gain and excess food consumption (see Allom et al.,
2016; Aulbach et al.,
2019; Jones et al.,
2016; Turton et al.,
2016; Yang et al.,
2022 for reviews). Go/No-Go, as a form of inhibitory control training purporting to modify food-specific cognitive biases, is an associative learning protocol that pairs a clear auditory or visual signal for a prescribed behavioural response with a corresponding salient stimulus (Verbruggen et al.,
2014). When implemented as a dietary intervention, the Go/No-Go training paradigm, and similarly designed tasks such as Stop-Signal, conventionally pair a salient signal denoting a stop action in the case of the latter, or response inhibition in the case of the former, with a proscribed food cue (e.g., energy-dense, hyperpalatable, ultra-processed, high fat-high sugar, etc.).
The extant body of evidence suggests that response inhibition training (RIT), an iteration of the Go/No-Go, has demonstrated modest efficacy in modulating hedonic liking, food choice, and even weight (Jones et al.,
2018). The most reliable effect has been shown through measurement of explicit evaluations of food cues, usually by ratings of liking along a visual analogue scale (Adams et al.,
2021; Chen et al.,
2018; Lawrence et al.,
2015a,
2015b; Najberg et al.,
2021; Veling et al.,
2013a,
2013b; Yang et al.,
2021,
2022). However, RIT has also been associated with changes in simulated food choice, self-selected portion sizes, and relative reinforcing value for palatable, energy-dense foods (Chen et al.,
2019; Houben & Geisen,
2018; Porter et al.,
2018; Stice et al.,
2017; Van Koningsbruggen et al.,
2014; Veling et al.,
2021). Additionally, efficacy has been observed in lab-based food intake (Houben,
2011; Houben & Jansen,
2015; Oomen et al.,
2018) and even weight loss (Lawrence et al.,
2015b; Veling et al.,
2014). Pooled effect sizes for food intake (Aulbach et al.,
2019) and evaluation (Yang et al.,
2022) tend to range between Hedges
g = 0.25–0.38 for food-specific Go/No-Go tasks specifically, higher relative to alternative bias modification tasks such as the stop-signal task (Hedges
g = 0.11–0.14) and approach-avoidance task (Hedges
g = 0.09).
The findings from previous RIT trials, indicative of both conditional efficacy and methodological heterogeneity, has highlighted the need to elucidate the nature of the RIT effect on food evaluation, including neurobehavioural mechanisms of action (Carbine & Larson,
2019; see Veling et al.,
2017 for a discussion). Theoretical frameworks of addiction may provide insight when interpreting or contextualising the food devaluation effect observed in previous RIT studies. Berridge and Robinson (
2016) have posited, after extensive research mainly in animal models, that the dopamine depletion characteristic of those demonstrating behavioural addictions for a reward-eliciting stimulus was reliably associated with an inverse augmentation of what they refer to as ‘incentive salience’, also known as ‘wanting’ more conventionally (also see Morales & Berridge,
2020 for a review). By contrast, affective response associated with consumption of the reward, or ‘liking’, remained relatively stable. Evidence to date has generally supported the hypothesis that neural pathways governing liking and wanting interact, albeit able to operate independently under certain conditions (de Araujo et al.,
2020; Roefs et al.,
2018; Volkow et al.,
2017). Although there is some evidence to suggest that RIT can modulate motivational salience of palatable foods (e.g., Houben & Giesen,
2018; Stice et al.,
2017), measures of liking and wanting are rarely measured concurrently in RIT trials. Thus, it is not certain to what degree both appetitive facets are sensitive to RIT effects, especially in varying conditions of food stimulus-specificity. Indeed, the meta-analysis by Yang and colleagues (
2022) demonstrated that devaluation effects are primarily observed with ‘trained’ relative to novel food stimuli, although this difference was not statistically significant.
Therefore, the present proof-of-concept study primarily aimed to establish the feasibility of applying a nuanced framework based on Incentive Sensitisation Theory when testing the efficacy of a food-specific RIT intervention purporting to modulate appetite and food preferences. Another salient aim was to investigate whether the generalisation of RIT devaluation effects may occur as implicit wanting and/or explicit liking when food stimuli share conspicuous nutritional properties (e.g., energy density, ultra-processed, high palatability). Evidence to date suggests that effect sizes associated with devaluation of novel food stimuli are relatively smaller than those included in the RIT task (Yang et al.,
2022). However, these comparisons are predominately based on measures of explicit liking, thus it may be interesting to reproduce this observation using other modalities of measuring food reward. Given the multidimensional nature of food-specific impulsivity (Van der Laan et al.,
2016), the final aim was to explore to what extent different dimensions of food-specific and general trait impulsivity were associated with food reward at baseline as a method to support the suitability of this outcome measure to evaluate RIT intervention efficacy. Ultimately, trends in liking and wanting for visual food stimuli based on energy density and palatability were investigated concurrently as a method of observing coherence after completion of app-based RIT relative to a control comparison.
Discussion
In this crossover study, a mobile app-based RIT intervention was tested for its concurrent effects on explicit and implicit facets of food reward. Associations between food-specific and trait impulsivity scales and food reward outcomes at baseline were also evaluated. Analyses indicated that empirical patterns in explicit liking and implicit wanting after RIT appeared to differ in a model where stimulus devaluation of non-specific (i.e., novel) food stimuli was measured. Specifically, trends found in implicit outcomes were discordant from those found in both explicit liking and wanting. Rather, reductions in explicit liking for both energy-dense and low calorie foods during the intervention were marginally significant relative to the control session. Explicit and implicit preferences for energy-dense foods at baseline were generally associated with responsiveness to food cues on average. Additionally, the app-based RIT task was rated favourably on key dimensions of intervention quality, suggesting at least a moderate level of acceptability. Overall, the results suggest that this study design is feasible, and the modality of food reward assessment may be important when testing any generalised effect of RIT. As a feasibility study, reliable conclusions cannot be drawn from the inferential tests and emphasis should be maintained on the descriptive statistics provided.
The utilisation of Berridge and Robinson’s (
2016) framework to measure and predict eating behaviours in humans has been scrutinised in the literature. Although it is beyond the scope of this study to elucidate in detail, the interested reader may wish to read critical reviews by Pool et al. (
2016), Polk et al. (
2017), and Bickel et al. (
2018). It has been argued that explicit liking, when measured similarly to the approach in this study, may not be capturing the same appetitive feedback as demonstrated in the animal models conducted by Berridge and Robinson (
2016), where liking was measured during food consumption. Indeed, explicit liking and implicit wanting tend to be highly correlated in samples representative of the general population as each measure likely captures an expectation of reward to some degree (Oustric et al.,
2020). However, divergences in liking and wanting have been demonstrated in human experiments under particular conditions such as obesity and other eating-related pathologies (Finlayson et al.,
2007b,
2011; Morales & Berridge,
2020). Another pertinent question in the context of the present study is whether this framework could provide a utility for assessment of dietary intervention efficacy. In their meta-analysis, Yang and colleagues (
2022) only found an effect of RIT when devaluation was measured explicitly, although the number of studies assessing implicit devaluation was far smaller. Although this cannot be equated to a comparison of explicit liking and implicit wanting, it may be of interest for future studies to include both types of evaluations so that coherent trends may be collated and examined in relation to observable eating behaviour such as food selection and intake.
Two notable discrepancies were detected in the observations of this study. First, the main difference between explicit liking and implicit wanting was observed in low-calorie food evaluations specifically, with a decrease being observed in the former in contrast to previous RIT trials (e.g., Lawrence et al.,
2015b). Second, trait motor impulsivity as assessed by the BIS was inversely associated with higher explicit and implicit food reward. Such unexpected observations highlight the challenges of appropriately executing the proposed study design. For example, associations between trait impulsivity scales and food reward, and RIT effects, may be state-dependent, and this sample did not demonstrate the same degree of fasting hunger sensations typically observed in ostensibly fasted subjects (e.g., Dalton & Finlayson,
2014). Moreover, the choice of food stimuli and how they are categorised may be conditional factors when observing potential effect generalisation and may additionally rely on pre-existing distinctions held by the individual (Serfas et al.,
2017). This likely introduces more error variance, and to the extent that such general effects actually exist, studies with more statistical power would likely be needed to detect them relative to effects on ‘trained’ food stimuli.
There are notable limitations in this study, thus conclusions should be drawn with caution. First, this study had a modest sample size, which may suggest an elevated probability of a type 2 error in analyses as well as overestimated effect sizes (Dechartres et al.,
2013). The wide confidence intervals produced are indicative of this fact. However, the contrast in effect sizes between explicit and implicit food reward measures suggests further investigation in a more adequately powered study may be warranted. Although the length of the washout period was standardised, it is uncertain how long effects from RIT are sustained, especially from a single session. Chen and colleagues (
2019) demonstrated that changes in food preference were sustained after 1 week after a single training session, albeit with a significantly reduced effect size (also see Adams et al.,
2021). Future studies that utilise a repeated-measures design ought to be mindful of these results when designating washout periods of adequate length to mitigate potential carryover effects. No training performance data were available, therefore, adequate learning of stimulus–response associations by each participant cannot be demonstrated, as is standard practice in RIT trials. Indeed, a meta-analysis of RIT interventional studies by Jones and colleagues (
2016) found that accuracy on inhibition trials (i.e., commission error rate) was a significant predictor of RIT efficacy to modify eating behaviours. It is therefore important that future studies record performance data when discerning between no effect or lack of compliance. Finally, the choice of control comparison did not have energy-dense food cues, which may have influenced differences between sessions independent of the training mechanism. Future studies ought to utilise different types of control tasks to provide more confidence in the reliability of these results.
In conclusion, this proof of concept study provided preliminary evidence for the feasibility of applying Berridge and Robinson’s (
2016) Incentive Sensitisation framework for assessing the efficacy of RIT to modulate appetite. Observations suggest that the LFPQ may be associated to the food responsiveness dimension of trait impulsivity at baseline. It is thus proposed that the LFPQ can be a suitable and valid instrument to assess efficacy of behavioural interventions to modify food hedonics. Effect generalisation appears to be feasible, but this may be more apparent when evaluations are measured as implicit wanting. Adequately powered, pre-registered trials are needed to reproduce these observations and infer any relationships with confidence and further examine how salient factors, such as trait impulsivity or food stimulus specificity, may moderate the RIT effect on explicit liking or implicit wanting for palatable, high-energy foods concurrently. Additionally, studies may measure liking and wanting for both trained and novel food stimuli in order make direct comparisons of these facets of reward based on stimulus specificity.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.