The effects of exposure to appetitive cues on inhibitory control: A meta-analytic investigation

Inhibitory control refers to the ability to stop, change or delay a response, and is often used in order to protect higher order goals. Theoretical models suggest that appetitive cues such as pictures of alcoholic drinks or food evoke strong automatic appetitive responses which lead to transient impairments in inhibitory control, and that these effects of cues may be related to individual differences (e.g. in body mass index, or alcohol consumption). In order to investigate these claims we conducted a random effects meta-analysis of 66 effect sizes (35 alcohol, 31 food) from 37 articles that tested the effect of exposure to appetitive (alcohol/food) cues on indices of inhibitory control. The overall effect of cue exposure was small, but robust (SMD = -0.12 [95% CI -0.23, -0.02]; Z = 2.34, p = .02, I2 = 84%). Exposure to alcohol-related cues significantly impaired inhibitory control (SMD = -0.21 [95% CI = -0.32, -0.11]; Z = 4.17, p < .001), however exposure to food-related cues did not lead to impairments (SMD = -0.03 [95% CI = -0.21, 0.15]; Z = 0.36, p = .720). There was no evidence that drinking or weight status significantly moderated the effects of cues on inhibitory control. Similarly, cue modality (words, pictures, or smells) did not significantly moderate the effects. Trim and Fill analysis suggested bias in the literature, which when corrected, made the overall effect of cues non-significant. Overall, these findings provide some tentative support for theoretical claims that exposure to appetitive cues prompts transient impairments in inhibitory control. Further research is required to determine the clinical significance of these observations. However, care should be taken when drawing conclusions from a potentially biased evidence base.


Introduction
Inhibitory control refers to the ability to stop, change or delay a response that is inappropriate given current environment demands (Logan, Cowan, & Davis, 1984;Logan, Schachar, & Tannock, 1997). This (in)ability is a key component of impulsivity and executive functioning (Bickel, Jarmolowicz, Mueller, Gatchalian, & McClure, 2012), and it overlaps considerably with broader constructs such as self-control and disinhibition (Baumeister, 2014;Tarter, Kirisci, Reynolds, & Mezzich, 2004). Effective inhibition of behaviour permits the suppression of automatic appetitive responses evoked by cues related to unhealthy foods or alcohol. This may allow people the opportunity to make controlled decisions (Wiers, Gladwin, Hofmann, Salemink, & Ridderinkhof, 2013) and maintain their higher order goals such as abstinence from alcohol or weight-loss, even when tempted by environmental cues such as the sight or smell of appetising foods or alcoholic drinks (Jones, Hardman, Lawrence, & Field, 2017;Verbruggen, Best, Bowditch, Stevens, & McLaren, 2014).
Findings such as these are generally interpreted as indicating that inhibitory control is a stable trait characteristic that differs between individuals but remains fairly constant within individuals, which is why it reliably predicts between-subject variability in behaviour months or years later. However, more recent theoretical models have suggested that inhibitory control functions as a transient state which can fluctuate in response to environmental or internal 'events' (De Wit, 2009), and these short term impairments in inhibitory control may increase the immediate risk of temptation and subsequent (re) lapse. In a recent narrative review we (Jones, Christiansen, Nederkoorn, Houben, & Field, 2013a) suggested that individuals have a general stable capacity for inhibitory control, however this capacity can fluctuate (both improve and worsen) in response to environmental and internal events. Exposure to appetitive cues is one potential environmental event that may negatively influence inhibitory control, because those cues evoke automatic appetitive tendencies (Brockmeyer, Hahn, Reetz, Schmidt, & Friederich, 2015;Field, Kiernan, Eastwood, & Child, 2008;Kemps & Tiggemann, 2015;Kemps, Tiggemann, Martin, & Elliott, 2013), and these responses should conflict with inhibition of behaviour. Indeed, exposure to both alcohol (Czapla et al., 2015a,b) and food-related cues (Phelan et al., 2011) results in short-lived impairments in inhibitory control. Furthermore, these 'cue-specific' deficits in inhibition may predict greater variance in individual differences in health-related behaviours than general inhibition deficits (Houben, Nederkoorn, & Jansen, 2013;Petit, Kornreich, Noël, Verbanck, & Campanella, 2012), including ad-libitum food and alcohol consumption in the laboratory Price, Lee, & Higgs, 2016). However, as with many research questions that are studied intensively, there are some null or equivocal findings in the literature, in which cue-exposure has not impaired inhibitory control (Mainz et al., 2012;Nederkoorn, Baltus, Guerrieri, & Wiers, 2009).
The aim of the present meta-analytic investigation was to quantify the extent to which exposure to appetitive-cues (alcohol and food-related pictures) causes transient impairments in the ability to inhibit behaviour, and to identify procedural variables or participant characteristics that may moderate this effect. Following initial scoping searches, we limited our investigation to alcohol and food-related cues because the vast majority of studies in the field were limited to these domains (however, the disinhibiting effects of smoking- (Luijten, Littel, & Franken, 2011) and drug-related cues (Pike, Stoops, Fillmore, & Rush, 2013), have been investigated). We sought to identify potential moderators of the effects of appetitive cues on inhibitory control, including: drinking/weight status, the modality of cue-exposure (pictorial, lexical, olfactory), type of task used to measure inhibitory control, and to determine if cue-specific inhibitory deficits are associated with unhealthy behaviours or outcomes, such as alcohol consumption or BMI. We hypothesised that inhibitory control would be worse during or after exposure to appetitive cues compared to neutral cues, or the absence of cues. For our moderator analyses we predicted that this effect would be larger in heavier drinkers, individuals with alcohol use disorder and individuals with overweight/obesity because, theoretically, automatic appetitive responses to appetitive cues should be stronger in people who consume them more frequently (Volkow et al., 2013;Wiers et al., 2007). We had no a priori predictions regarding cue-modality, but we investigated this on the basis of findings from some individual studies which demonstrated differential effects of cues presented in different modalities on both inhibitory control and subjective craving (Boswell & Kober, 2016;Monk, Sunley, Qureshi, & Heim, 2016). Similarly, we examined the effects of different inhibitory control tasks as each have differing inhibitory pressures and may measure a different type of inhibitory control, e.g. action cancellation versus action inhibition (Eagle, Bari, & Robbins, 2008). However, we made no specific hypothesis as to which may be most affected by cue-exposure.

Information sources and search strategy
We conducted scoping searches using three commonly used electronic databases (Scopus, PubMed and PsycInfo) in November and December 2015. We pre-registered our protocol and analysis strategy on Open Science Framework (https://osf.io/c9jf8/). Full searches were carried out in October 2017. Our literature search was guided by the Preferred Reporting Items for Systematic Review (PRISMA) guidelines. See supplementary materials for full search strategy and terms. Following identification of full text articles we conducted manual searches on reference lists, and identified further articles based on authors' knowledge. In total we identified 35 effect sizes for alcohol and 31 for food.

Eligibility criteria
All studies had to meet the following criteria in order to be included in the meta-analysis; (i) include human participants aged 18+, (ii) include alcohol or food-related (appetitive) cue exposure, i.e. olfactory or visual cues, prior to or during an inhibitory control task, (iii) a control comparison, for example exposure to neutral cues during, or the absence of cue exposure (baseline) prior to or during, an inhibitory control task. Cue exposure involved food/alcohol and neutral images/ words that were embedded into an inhibitory control task (Houben et al., 2013;, or the holding and sniffing of food/ alcohol prior to completing an inhibitory control task (Gauggel et al., 2010;Lattimore & Mead, 2015).

Outcome measure(s)
Studies were required to have an outcome measure of inhibitory control during/following appetitive cue exposure, and either a measure of inhibitory control at baseline (prior to cue-exposure) or during exposure to non-appetitive (neutral) cues. Proposed measures of inhibitory control were cross-checked against previous literature and review papers to ensure that they were validated measures (e.g. Diamond, 2013). All authors agreed on the tasks for inclusion.

Data extraction and coding
Three independent coders (JD, IK, NC) performed the searches and identified the relevant articles. After removal of duplicates, 4151 unique articles were identified. These articles were screened via title and abstract, which resulted in exclusion of 3819 articles with agreement from all coders. Data were extracted by the coders and cross-checked by the first author. In cases where insufficient data was available the authors were contacted to provide this data. If the authors did not respond to the data request and it was possible, we used Web Plot Digitizer (Version 3.10, Rohatgi, 2016) to estimate means and variances from figures presented in publications, as recommended (Jelicic Kadic, Vucic, Dosenovic, Sapunar, & Puljak, 2016;Vucic, Jelicic Kadic, & Puljak, 2015).
To code moderator variables such as drinking and weight status we first examined if any participants were described in the article in a specific way (e.g. alcohol dependent, overweight). If no explicit claims were made we made group level inferences on alcohol use using established cut-offs for 'heavy' or 'hazardous' drinking via scores on the AUDIT (score > 8 indicative of hazardous drinking (see Saunders, Aasland, Babor, De La Fuente, & Grant, 1993) or estimates of units of alcohol consumed per week (> 14 units per week indicative of heavy drinking 1 ). For weight status we examined if group mean BMI > 25 kg/ 1 Note that these cut offs differ by country. Fourteen units per week is the UK m 2 (for overweight/obese). Three studies compared alcohol-dependent participants to a control group (Noel et al., 2005;Noel et al., 2007;Sion, Jurado-Barba, Alonso, & Rubio-Valladolid, 2017), but provided no information as to whether the control group drank any alcohol. Therefore, the control groups in these studies were not included in any analysis.

Variables of interest
The indices of inhibition used for each task are stated in Table 1. The most common tasks were the Stop Signal, Go/No-Go and Go/No-Go shifting tasks. The Stop Signal and Go/No-Go tasks require motor inhibition of a pre-potent response following a visual or auditory 'stop signal' or 'No-Go cue'. In the Stop Signal task this cue is presented following a variable delay after initial stimulus onset and therefore motor behaviour has to be cancelled, whereas in the No-Go task the No-Go cue is presented concurrently with the target stimulus, and therefore behaviour must be restrained rather than cancelled (Eagle et al., 2008). In the shifting version of the task the cues for 'Go' and 'No-Go' are switched on a block-by-block basis (Meule, 2017). In the anti-saccade task participants have to inhibit an involuntary oculomotor response (saccade) to a visual stimulus that appears in the periphery of a visual display (Hallett, 1978). In the Stroop task (Stroop, 1935) participants have to name the colour of target words whilst ignoring the semantic content of the word (e.g., the word 'red' printed in blue ink). Finally, in the flanker task participants have to categorise a target stimulus whilst ignoring distractor stimuli that appear alongside it (Eriksen & Eriksen, 1974). Stop Signal Reaction Time and Commission errors were the most common outcomes from these tasks. We also extracted and coded a number of variables for our main and supplementary analyses, including; type of task used, modality of cue exposure, drinking and weight status, and any correlations with BMI, typical alcohol use or AUDIT scores (see Table 1). We selected these variables as they were the most commonly measured across all studies.

Statistical analyses
Our main statistical analyses were carried out using Review Manager 5.3 (Cochrane Informatics & Knowledge Management Department, UK, 2014), with supplementary analyses conducted using JASP (JASP team, Version 0.8.4). All outcomes were continuous, therefore we computed the Standardised Mean Difference (SMD) effect size using the equation SMD = (M a -M c )/S p , where M a is the mean inhibitory control measure following exposure to appetitive cues, M c is the mean following neutral cues (or no cues) and S p is the pooled standard deviation. We also computed the Standard Error of the SMD; most studies used within-subject designs so we adjusted the SE based on the within-subject correlation (Elbourne et al., 2002), which was requested from authors when not presented in the article. When this correlation was not available it was estimated based on the mean correlation across all effect sizes (r∼.60). The SMD is interpreted as; 0.2 = small effect, 0.5 = moderate effect, and 0.8 = large effect. In this case a negative SMD is indicative of a reduction in inhibitory control following appetitive cue exposure. In order to ensure consistency, effect sizes from outcome measures in which a larger number was indicative of poorer inhibition (e.g. Stop Signal Reaction times, Number/proportion of commission errors) were reversed.
To assess between-study heterogeneity we used the I 2 statistic, calculated as I 2 = (Qdf/Q) x 100%, where Q is the chi-squared statistic and df is the accompanying degrees of freedom. We used random-effects models due to substantial heterogeneity between studies (Riley, Higgins, & Deeks, 2011). In the case of substantial heterogeneity we provided estimates of subgroup effects to aid interpretation of the data. Finally, to remove outlying effect sizes we calculated z-scores and identified any effect size which was an extreme value at .001 alpha level (i.e. Z > 3.30). As a result one effect size (Petit et al., 2012): Light drinkers: SMD > 6) was excluded from all subsequent analyses. This decision was made a-priori, as evidenced in our pre-registration.

Characteristics of studies
The majority of studies employed within-subject (repeated measures) designs, in which participants either (a) completed similar inhibitory control tasks during or after exposure to appetitive cues and neutral cues, or (b) completed one inhibitory control task with embedded appetitive and neutral cues that permitted separate indices of inhibitory control to be computed for each type of cue (e.g. (Jones & Field, 2015;Nederkoorn et al., 2009;Petit et al., 2012)). We also identified some studies that employed between-subject designs in which participants were randomized to exposure to either appetitive or neutral cues (e.g. (Jones, Rose, Cole, & Field, 2013b;Lattimore & Mead, 2015;Muraven & Shmueli, 2006). Field and Jones (2017) employed a mixed design in which inhibitory control was measured at baseline in all participants, before a between-subjects cue exposure manipulation (appetitive, neutral), followed by a second measure of inhibitory control. In this case we took the difference between the two groups after cue exposure.
Some studies also contrasted the effects of appetitive cues on inhibitory control in different groups of participants using mixed designs. For example, heavy vs. light drinkers (Nederkoorn et al., 2009), people with alcohol dependence vs. controls (Czapla et al., 2015a,b), obese/ overweight vs. normal weight (Loeber et al., 2012). In these studies we computed within-subject comparisons based on these groups where possible to allow for individual comparisons to be included in different moderator analyses (see Table 1). Finally, some studies used multiple inhibitory control tasks or parameters (e.g., Adams, Ataya, Attwood, & Munafò, 2013) and in these cases we adjusted the sample sizes in the control conditions (N# Control /number of tasks) accordingly to ensure each comparison could be included in our pooled analyses, as recommended (Higgins & Green, 2011).

Results
The article selection process and flow is shown in Fig. 1. Following exclusion of irrelevant articles by title and abstract scanning we identified 37 full-text articles. See Table 1 for full details.

Table 1
Details of studies included in the meta-analysis.

Effect of inhibitory control task
We conducted exploratory analyses on the type of task used to operationalize inhibitory control (see Table 1 430, I 2 = 89%). However, the subgroup effect was not statistically significant (X 2 (4) = 6.64, p = .160).

Examination of bias
Visual inspection of the funnel plot (see Supplementary Fig. 2) for all studies suggested asymmetry, and Trim and Fill analyses suggested 17 effect sizes would need to be added to achieve symmetry (see Supplementary Fig. 3). Adding these effect sizes made the overall point estimate non-significant (SMD = 0.05 [95% CI −0.08, 0.18]). This suggests some degree of bias was evident across the effect sizes included. We also conducted Egger's test to formally examine asymmetry by regressing the effect size against the precision, however the test was not statistically significant (Z = −0.56, p = .574).

Discussion
The results of this meta-analytic investigation demonstrate that exposure to alcohol-related cues prompts robust, albeit small impairments in inhibitory control, although the evidence for comparable effects of food-related cues was not reliable. We observed substantial heterogeneity across effect sizes, which remained high despite several subgroup analyses that attempted to identify moderating variables. There was limited evidence to suggest that drinking status moderated the effect of alcohol-related cues, or that weight status moderated the effect of food-related cues on inhibitory control. Similarly, the modality of cue exposure did not significantly moderate our findings. Statistical correction for bias made the main effect of appetitive cues on inhibitory control no longer statistically significant.
Our primary hypothesisexposure to appetitive cues would prompt a deficit in inhibitory controlwas not fully supported. Overall, appetitive cues impaired inhibitory control, however this was driven by a significant effect for alcohol-related cues. These findings provide partial support for theoretical models which suggest inhibitory control is a transient process that is sensitive to environmental and internal events (Jones, Christiansen, Nederkoorn, Houben, & Field, 2013a;Verbruggen et al., 2014), and the transient nature of inhibitory control cues may be one psychological mechanism that underlies the influence of alcohol cues on drinking behaviour (De Wit, 2009;. We found no evidence that transient impairments in inhibitory control were associated with individual differences in alcohol or food intake over the longer term, such as (self-reported) quantity of alcohol consumed per week, hazardous drinking scores or BMI. Future studies should attempt to clarify the associations between the disinhibiting effects of appetitive cues and food or alcohol intake that is measured immediately afterwards, such as ad-libitum intake  or operant choice (Veling, Aarts, & Stroebe, 2013).
Consideration of publication bias suggests that even these small effects of appetitive cues may be inflated, and when we accounted for 'missing' (small and non-significant; k = 17) effect sizes using Trim and Fill analyses (Duval & Tweedie, 2000), the overall effect size was no longer statistically significant. This suggests that this literature is characterised by 'small study effects', often of poor methodological quality, and reporting biases that can substantially influence pooledestimates (Schwarzer, Carpenter, & Rücker, 2015). Future research should conduct well powered studies and aim to publish all data to mitigate these biases and improve our confidence in pooled estimates of effect.
As hypothesised, impairments in inhibitory control following exposure to alcohol-related cues were comparable heavy drinkers and people with alcohol dependence, but these effects were absent in light drinkers. However, the test for subgroup differences was not statistically significant. Therefore, these subgroup differences are merely suggestive, and they should be interpreted with caution because many subgroups were poorly defined and created post-hoc using median split techniques (Jones & Field, 2015;Nederkoorn et al., 2009). Future studies might use established criteria to define heavy drinking or alcohol dependence (Babor, Higgins-Biddle, Saunders, & Monteiro, 2001;Edwards, 1996), in order to determine if these subgroup differences are indeed robust.
The absence of a robust effect of food-related cues on inhibitory control was surprising but is difficult to interpret given methodological features of the original studies. Eleven (35%) of studies used the Go/ No-Go switching task for food-related cues. In this task, contingencies between making a motor response and inhibiting to high calorific vs. control cues are regularly switched on a block-by-block basis (Loeber et al., 2012). The repeated shifting of task contingencies between blocks mean that this task is likely to capture the effect of cues on inhibition and shifting, two distinct subcomponents of executive functions, and therefore this task provides an impure measure of inhibitory control (Miyake et al., 2010). Furthermore, variations on this task have also used low-calorific food cues (rather than non-food items) which may still be appetitive (He et al., 2014). Notably, when we removed effect sizes generated from Go-No/Go shifting tasks from our analysis the effect of food cues on inhibitory control was robust.
The modality in which appetitive cues were presented (e.g. visual vs. olfactory) did not moderate the effect of appetitive cues on inhibitory control. The absence of a moderation effect here should be interpreted cautiously until further direct comparisons across modalities are attempted (cf. Boswell & Kober, 2016;Monk et al., 2016). Overall, our moderator analyses suggest that we were unable to reduce the substantial heterogeneity by identifying variables that might moderate the influence of appetitive cues on inhibitory control. It is possible that other variables may influence this relationship, but we did not identify enough studies to examine this. For example, heterogeneity may have been caused by considerable variability in food-related and control images (see Table 1) and individual differences in reactivity to these cues, or differing levels of motivation to restrict unhealthy behaviours across the samples. Future studies should investigate these potential moderators in more detail.
Finally, the effect of alcohol-related cues on inhibitory control supports the recent development of Inhibitory Control Training (ICT) as a behavioural intervention, to mitigate against cue-specific inhibition deficits. ICT creates an associative link between appetitive cues and inhibition of behaviour, which is thought to extinguish the associative link between appetitive cues and approach behaviour (Stice, Lawrence, Kemps, & Veling, 2016;Verbruggen et al., 2014). Promising effects of ICT have been demonstrated for both ad-libitum food and alcohol consumption in the laboratory (Allom, Mullan, & Hagger, 2015;Jones et al., 2016).
Some limitations of our analyses should be taken into consideration. First, we only included studies that examined cue reactivity to alcohol or food-related cues. There is evidence in other domains for cue-specific impairments of inhibitory control, including smoking (Luijten et al., 2011) and illicit drug use (Pike et al., 2013;Verdejo-García et al., 2012) which we did not integrate into our current analysis due to the limited number of available studies. Similarly, although we limited our findings to adult participants (18 + in United Kingdom, the legal drinking age) research also demonstrates inhibitory deficits following alcohol cues and food cues in younger people Korucuoglu, Gladwin, & Wiers, 2015). Second, we were unable to directly measure associations between subjective and physiological cue-reactivity (e.g. craving and arousal) and inhibitory control as few studies measured this consistently. Finally, it is unknown whether these deficits in inhibitory control are a capacity deficit, or a motivational deficit (i.e. participants do not evoke effortful inhibition; Fujita, 2011). Future research should aim to overcome these limitations but also identify the mechanisms through which appetitive cues impair inhibitory control, for example through competition with attentional processes (Pessoa, Padmala, Kenzer, & Bauer, 2012), reductions in limited self-regulatory resources (Muraven & Shmueli, 2006) or a reduction in their motivations (Pessoa, 2009). Furthermore, the influence of appetitive cues on inhibitory control in real-world settings (outside of the laboratory) should be investigated in order to elucidate the significance of these deficits for health-related behaviour. An interesting way to do this would be to examine real-time cue-exposure and inhibitory control using Ecological Momentary Assessment techniques (Shiffman, 2009). Given that cuereactivity demonstrates substantial within-and between-subject variability (Serre, Fatseas, Swendsen, & Auriacombe, 2015) the real-world effects of repeated (LaRowe, Saladin, Carpenter, & Upadhyaya, 2007), cumulative and personalised cues (MacKillop et al., 2010) may identify more robust effects of exposure to appetitive cues on inhibitory control.
To conclude, the results from this meta-analytic investigation demonstrate that inhibitory control is sensitive to the presence of alcoholrelated cues. The effect of food-related cues was less robust and may be confounded by methodological features of the tasks used. Overall, these findings provide some tentative support for theoretical predictions that inhibitory control is sensitive to exposure to appetitive cues. However whether these effects are robust, and if they play an important role in health-related behaviour, are important questions for future research.