A bayesian perspective on cognitive bias modification interventions for alcohol and smoking addiction

A Bayesian Perspective on Cognitive Bias Modification Interventions for Alcohol and Smoking Addiction.

Author: Ruben van Beek

Supervisors: Dr. Marilisa Boffo, Dr. Maarten Marsman & Dr. Kyriaki Nikolaou Student number: 10012028

Abstract

Cognitive Bias Modification (CBM) interventions target maladaptive cognitive responses that

result from prolonged substance abuse. The goal to reduce unfavorable outcomes such as a

reduced propensity to relapse, reduced substance use, and a reduced craving for the substance.

In this systematic review we investigated the efficacy of CBM interventions in reducing

cognitive biases and clinical outcome measures in clinical and non-clinical populations of

alcohol and tobacco users. In total 16 studies were included and analyzed both individually

and with Bayesian random-effects model. For the effect of CBM interventions on alcohol

related cognitive biases a small group level effect was found though individual studies show a

large amount of variability. A group level effect for CBM interventions that target tobacco

related cognitive biases could not be estimated due to the small number of studies.

Tantalizingly, there are indications that alcohol users might benefit from CBM interventions

targeting approach bias. More research is needed to further specify the characteristics of

Introduction

The use of alcohol and tobacco is widespread. On average individuals over 15 years old

drink 6.2 liters of pure alcohol each year (WHO, 2014) and tobacco is used by approximately

one in five adults (WHO, 2015). The use of these substances is not harmless. In 2012

approximately 6% of worldwide deaths was associated with alcohol and smoking causes

approximately 10% of deaths annually (WHO, 2014; WHO 2015). Though most substance

users are aware of the potentially harmful consequences of their use, quitting or reducing

consumption successfully has proven difficult for most. Only 17% to 35% of patients that are

treated for alcohol dependence remain abstinent after one year (Miller, Walters, & Bennett,

2001). Over 80% of smokers want to quit, but less than 8% of recent smokers do so

successfully (Babb, Malarcher, Schauer, Asman, & Jamal, 2017). The cost of substance use is

high both on a personal level as on a societal level (Whiteford et al., 2013; Harwood,

Fountain, Livermore, 1998). As a result, in the past 20 years, a considerable amount of

research has focused on trying to understand the brain processes that are affected by

prolonged substance abuse, with the aim of creating research-based interventions to reverse or

ameliorate substance-induced changes in these processes.

Dual-process models of addition, explain addictive behavior as a “battle” between two

qualitatively different classes of cognitive processes: an automatic and fast unconscious

process, and a slow and controlled conscious process (Bechara, 2005; Wiers, Gladwin,

Hofmann, Salemink, & Ridderinkhof, 2013). The automatic, “hot”, or impulsive process is

responsible for an immediate and fast call to action (Bechara, 2005). The impulsive process

makes us approach a beverage when thirsty or helps avoid a potentially dangerous snake. The

controlled, “cold”, or reflective process is slower and deliberate. The reflective process is

associated with long term goals (Wiers et al., 2013). An initial response geared towards

weight loss. Note that though these processes are qualitatively different, they interact to bring

about operant behavior. Ideally, these two processes strike a balance between short term goals

and long term goals. One perspective is that the immediate reward seeking impulsive process

can be corrected by the reflective system (Bechara, 2015). In substance addiction, both the

impulsive and the reflective process have been disrupted due to repeated exposure to a

substance.

It has been found that people repeatedly exposed to a substance show a strengthening of

the impulsive process. On a neurological level, the impulsive process is driven by the

activation of the mesolimbic pathway that is part of the dopaminergic pathways in the brain.

This pathway is mainly associated with reward and reinforcement learning. Stimuli can

release dopamine in the mesolimbic pathway of the brain incentivizing behavior to approach

the stimulus, i.e. a glass of water may trigger a dopamine release resulting in approach

behavior towards the glass of water. Robinson and Berridge have shown with a series of

publications that repeated exposure to a substance sensitizes the mesolimbic dopaminergic

pathway in the brain (Berridge, Venier, Robinson, 1989; Robinson & Berridge, 1993). This

means that with every subsequent substance intake, the mesolimbic system slowly gears the

system towards a hyperreactivity for substance related stimuli or cues, the brain is sensitized

for the incentives associated with substance use. The hyperreactive mesolimbic pathway

releases a larger amount of dopamine into the brain than it would for a substance naïve user.

As a result of this maladaptation stimuli related to substance use, be it a pub or a syringe, will

elicit a stronger action tendency for substance dependent users than in substance naïve users.

Because of the innate speed of the dopaminergic projections these tendencies often operate

outside consciousness. This is why they are often attributed within dual-process models as

part of the impulsive process (Wiers et al., 2013). In other words, the impulsive process is

However, the sensitized mesolimbic pathway and its effect on behavior would not

necessarily be a problem if the reflective process would be able to exert its top down influence

on the impulsive system. However, it has been found that repeated episodes of binge drinking

or heavy marihuana use have been associated with reduced executive functioning (Tapert et

al., 2004, 2007). Overall, Goldstein and Volkow (2002) showed that dopamine mediating

structures in the frontal cortex are reduced in size in substance dependent patients. This means

that the structures that are part of the reflective system that would normally be able mediate

strengthened impulsive system, have diminished in strength. Thus, not only is the impulsive

process strengthened, but the reflective process is weakened in substance users.

Naturally methods have been developed to reverse the effects of the chain of events that

is set in motion by prolonged exposure to substances with the goal of strengthening the

reflective process and weakening the impulsive process. Among the most prevalent therapies

targeting substance use disorder are cognitive behavioral therapy (CBT) and motivational

interviewing (MI) (Carroll & Onken, 2005). MI targets motivation, by increasing the

motivation to quit or stay abstinent the long term goal of abstinence becomes more salient and

strengthens the reflective process. CBT is broad term describing a number of different

interventions targeting different mechanisms associated with substance use disorder

(McHugh, Hearon, & Otto, 2010). Though CBT’s target different mechanisms, they all

involve increasing top down control of the reflective process over the impulsive process

(Bechara, 2005). Another way to strengthen the reflective process is to increase the capacity

of the working memory. It has been found that substance dependent patients have a smaller

working memory capacity and that increasing their working memory capacity was associated

with a reduced intake of alcohol (Houben, Wiers, & Jansen, 2011).

If we can strengthen the reflective system with different types of interventions, perhaps

(CBM) interventions set out to accomplish. CBM interventions are computerized training

tasks aimed at changing the maladaptive cognitive motivational biases that are caused by

repeated exposure to substances (Wiers et al., 2013). These stimuli can be directly related to

the substance, such as a glass of beer, but also indirectly such as the general setting in which a

substance is used. The hyperreactive reaction has been disentangled into a set of different

biases. For instance, substance users have been shown to attend stimuli related to a substance

faster than non-users, they have developed an attentional bias (Field & Cox, 2008). Not only

are substance users faster at responding to substance related stimuli, they are also faster to

approach it (Mogg, Bradley, Field, & De Houwer, 2003). Thirdly, substance dependent

patients suffer from an evaluative memory bias. For instance, heavy drinkers retrieve more

positive memories related to alcohol than light drinkers (Houben & Wiers, 2006). CBM

interventions aim to reduce the effects that these biases have on substance dependent patients,

a reduced awareness of operant behavior and increased salience and tendency for action

triggered by substance related stimuli. Because the research on some CBM interventions is

scarce, we will investigate the CBM interventions that have received the most attention in

addiction literature. The intervention called Alcohol Approach Avoidance Training (A-AAT)

aims to counteract the harmful effects of alcohol approach bias by inducing an avoidance bias

towards alcohol-related stimuli (Wiers et al., 2010). By adding a competing bias the relative

weight of the approach bias is diminished and A-AAT has been shown to significantly

decrease relapse rate at one-year follow-up in alcohol dependent inpatients (Wiers, Eberl,

Rinck, Becker, & Lindenmeyer, 2011; Eberl, Wiers, Pawelczack, Rinck, Becker, &

Lindenmeyer, 2013). Attentional bias modification training (ABMT) targets the attentional

bias towards substance related stimuli by training participants to shift attention away from

substance-related cues (Schoenmakers et al., 2007; 2010). It was found that alcohol addicted

to relapse after ABMT (Schoenmakers et al., 2010). A third line of CBM interventions,

evaluative conditioning (EC), target the affective value that is associated with s substance. In

EC the substance or substance related stimulus is repeated paired with a neutral stimulus,

reducing the affective value of substance related stimuli. Ideally, after EC a glass of beer is no

longer more appealing than a glass of water for a substance dependent patient (Houben,

Havermans, Nederkoorn, & Jansen, 2012). Though different paradigms have been developed,

all CBM interventions set out to reduce substance-related cognitive biases, which should lead

in turn to a decrease in substance use, craving and prolonged abstinance.

CMB interventions have received a lot of attention over the last decade (Wiers et al.,

2013; Wiers, Eberl, Rinck, Becker, & Lindenmeyer, 2011; Eberl et al, 2013). The prospect of

a computer based training to help quitting and abstinence is naturally very appealing.

Potentially these training paradigms could be executed from the patient’s own home, cutting

cost of treatment to a fraction of other interventions. However, CBM interventions have been

criticized for a number of reasons. For instance, the effect of ABMT of changing attentional

bias has not been confirmed in all cases, some studies do not find any effect (Schoorl,

Putman, & Van Der Does, 2013). Even when cognitive bias is successfully changed, the effect

of the training does not always transfer to stimuli outside the training environment (Everaert,

Mogoase, David, & Koster). This could be due to the differential effect CBM interventions

may have depending on study population or intrinsic motivation (Clarke, Notebaert, &

MacLeod, 2014). But in the worst case, CBM interventions may not work in reducing

cognitive biases or subsequent behavioral outcomes related to substance use disorder (Cristea,

Kok, & Cuijpers, 2016). However, there is also evidence that CBM interventions targeting

approach bias are efficacious (Kakoschke, Kemps, & Tiggemann, 2017). A recent paper has

shed light on the possible reasons for these divergent results in the literature (Müllig, Paulick,

students or other non-treatment seekers, these participants are likely to have a low motivation

for behavior change to quit or stay abstinent. In this paper we will only consider CBM

interventions targeting cognitive biases related to alcohol or tobacco dependence in which the

participants have the explicit goal of behavior change. We only target alcohol and tobacco

because the amount of CBM intervention studies for other substances is too small to reliably

run a meta-analysis.

Unlike the referenced studies above, we will use a Bayesian random-effects model to

evaluate the efficacy of CBM interventions. The overall low number of CBM studies targeting

alcohol and tobacco addiction may confound any results by increasing the possibility of a

false positive because of low statistical power (Cristea et al., 2016). Within the Bayesian

framework we can quantify our uncertainty of the estimated parameters, which means we can

say something about the degree of belief about those estimations. After applying a Bayesian

parameter estimation for studies individually, we will use a Bayesian random-effects model to

describe the group level parameters. In the random-effects model the effect sizes of the

individual studies are assumed to come from a single overarching distribution (Copetti et al.,

2014). We chose a random-effects model because it explicitly models the measurement error

that is associated with the effect sizes of individual studies. There are a number of advantages

to this approach (Marsman, Schönbrodt, Morey, Yao, Gelman, & Wagenmakers, 2017).

Firstly, information about the effect in the population can be estimated from the individual

studies, which allows to describe both the heterogeneity between studies as well as the mean

effect. Secondly, the hierarchical structure of the model will shrink the individual results that

are uncertain and relatively extreme to the group mean effect (Efron & Morris, 1977; Lee &

Wagenmakers, 2014). Thirdly, it is generally possible to obtain smaller posterior credible

intervals since the uncertainty of the individual study effects are reduced by the information

of between-study heterogeneity at the group level. To this aim, we aim to include type of

addiction, type of CBM intervention, number of training trials, mode of delivery, and the type

of participants as moderators in a Bayesian hierarchical random-effects regression model. We

have two main goals (1) are CBM interventions successful in reducing cognitive bias, relapse

rate, craving or substance intake for alcohol or tobacco users and (2) do different study

characteristics moderate the effect of CBM interventions on cognitive bias and

substance-related behavioral outcomes (i.e., substance use reduction, craving, and relapse rate), both for

alcohol and tobacco.

Methods Literature Search

The meta-analysis was performed according to the PRISMA guidelines (Liberati et al.,

2009). We searched PsychINFO, Medline, Web of Science, Embase, and the Cochrane

Library bibliographic databases from the earliest starting date per database until the 18th of May 2016. We used three groups of keywords covering the constructs of interest: cognitive

bias, addiction, and study type. For the cognitive bias group the main keywords were:

“cognitive bias”, “attentional bias”, “approach bias”, “response inhibition”. The second group

of keywords related to addiction included: “alcohol”, “drinking” and “tobacco”. The third

group of keywords covering intervention types consisted of: “longitudinal”, “(re)training”,

“intervention” and “task”. The full sets of keywords and subject headings for all databases

were compiled with the support of the health librarian of the University of Amsterdam. They

were generated both from a set of relevant keywords compiled by the authors and using a

reference set of articles known to the authors as meeting the inclusion criteria. For all groups

more keywords were used based on the bibliographic categorization of relevant CBM

database can be found in Appendix A. The references of the included studies were also

searched systematically for any missed study.

Selection of studies

Studies were included if they met the following eligibility criteria: (1) were published

in English; (2) included CBM interventions directed at alcohol or tobacco use; (3) tested

participants who were aware that the goal of the study was behavior change (i.e., reduction of

substance intake); (4) contained a CBM intervention, i.e., AAT (e.g., Wiers et al., 2011),

ABMT (e.g., Schoenmakers et al., 2010; Fardardi & Cox, 2009) or EC training (e.g., Houben,

Havermans, Nederkoorn, & Jansen, 2012); (5) randomly allocated participants to intervention

conditions; (6) included a comparison between a control condition (active or inactive) and a

CBM intervention (studies that featured multiple control conditions or interventions were also

included); (6) contained outcome measures of cognitive bias, relapse rate, substance use, or

craving. Screening of search results and of eligible studies was carried out independently by

the first author and a second associated researcher.

Quality assessment

To assess the quality of the included studies we used the “Risk of Bias” tool developed

by the Cochrane Collaboration (Higgins & Green, 2011). All included studies were assessed

on: (1) random sequence generation, (2) concealed allocation to the conditions, (3) blinding of

both participants and personnel, (4) blinding of outcome measurement, (5) incomplete

outcome data, (6) selective reporting, and (7) other possible sources of bias. Assessment of

the quality of the studies was conducted by the first author of this paper.

Data extraction

Two of the authors independently coded all included studies. For each included study

publication), study characteristics, including targeted addiction disorder (tobacco or alcohol

study), population (clinical or non-clinical), and sample size; intervention characteristics,

including type of CBM intervention (ABMT, AAT, EC), number of training sessions, total

number of training trials, and intervention setting (lab or controlled); type of control condition

(active or inactive); assessment time points (pre, post, follow-ups); and types of outcome

measures for cognitive bias, substance use and craving.

In order to run the analyses we distinguished four separate groups of outcome

measures, i.e. cognitive bias, substance use reduction, craving, and relapse rate. Table 1 lists

all outcomes measure found in the included studies.

Table 1. Coding of the four categories of outcome measures.

Outcome categories Measures

1. Cognitive bias VPT, Emotional Stroop task, AAT, IAT, AMP

2. Substance use reduction DRQ, TLFB, FTND, DHQ, smoking abstinence (CO), alcohol consumption (ml), cigarettes per day

3. Craving MPSS-C, ACQ, QSU, DAQ, VAS-Craving, Likert scale

4. Relapse rate Days to relapse, 1 year follow-up, 3 month follow up

Notes.

AAT = approach-avoidance task, ACQ = alcohol craving questionnaire, AMP =Affect Misattribution Procedure, DAQ = desire for alcohol questionnaire, DHQ = drinking habits questionnaire, DRQ = drinking record questionnaire, IAT = implicit association task, MPSS-C = mood and physical symptoms scale-craving, AAT = nicotine approach-avoidance task, QSU = questionnaire of smoking urges, TLFB = timeline follow back, VAS-Craving = visual analogue craving scale, VPT = visual probe task.

For all studies that measure cognitive bias we subtracted the pre-test score on

cognitive bias from the post-test measurement for both the CBM intervention and control

condition. We chose this approach because we wanted to accommodate the difference in

cognitive bias at baseline between participants. Neglecting the a priori variability in bias

scores between participants might confound the reliability of the results (Vickers, & Altman,

2001). In most cases the data available in the papers was not sufficient to calculate the

difference scores. Therefore, we emailed the first authors of those papers and with a request to

send us the raw data so we could calculate these difference scores ourselves. For one study the

data was not obtained because the informed consent statement participants signed explicitly

stated that their data would not be shared with third parties (Elfeddali et al., 2016). Based on

the differences scores for both the CBM intervention and control condition we were able to

reduce the results of all included studies to a simple independent t-test, irrespective of the

analysis used by the respective authors. For all studies we contrasted the CBM intervention

with the sham control condition. If there was no sham control condition we contrasted the

CBM intervention with the waitlist control or no training control condition. If a study

contained multiple CBM interventions or multiple experiments, all interventions and

experiments were included in the analysis (Begh, et al., 2015; Lindgren, et al., 2015;

Schoenmakers, et al., 2010; Sharbanee, et al., 2015; Wiers, et al., 2011). Both the number of

follow-up time points and the interval between the follow-up bias measurements differed

across studies. Some studies featured a measurement of cognitive bias only directly after the

intervention (Elfaddali, De Vries, Bolman, Pronk, & Wiers, 2016; Lindgren et al., 2015;

Machulska, Zlomuzica, Rinck, Assion, & Margraf, 2016; Wiers, et al., 2011). One study only

contained a measurement for cognitive bias after three months (Kong et al., 2015). All other

studies measured cognitive bias after one month or close to one month after the intervention.

cognitive bias measurements that were one month post training or the closest time point to

one month.

Craving was mostly measured with a variety of standardized questionnaires (ACQ,

QSU-B, DAQ, MPSS-C) that have been administered both pre and post-test (Begh, et al.,

2015; Lindgren, et al., 2015; Lopes, et al., 2014; McHugh, Murray, Hearon, Calkins, Otto,

2010; Schoenmakers, et al., 2010; Wiers, et al., 2015a; Wiers, et al., 2015b). For these we can

calculate the difference scores pre and post-test just as we could with the bias scores. Two

studies use Likert scales to measure craving (Elfeddali, et al., 2016; Wiers, et al., 2011). One

study used a Visual Analogue Scale (VAS) (Wiers, et al., 2015c).

The included studies displayed a myriad of instruments that aimed to measure

substance use reduction. From carbon monoxide measured by breathalyzers to cigarettes

smoked per day to an intention to drink (Begh, et al, 2015; Lopes, et al., 2014; Lindgren, et

al., 2015). Where possible we again calculated a difference score between pre and post-test for

both the experimental and control condition (Cox, Fadardi, Hosier, & Pothos, 2015; Lindgren,

et al;, 2015; Lopes, et al., 2014; Machulska, Zlomuzica, Rinck, Assion, & Margraf, 2016;

Magurean, Constantin, & Sava, 2015; McGeary, Meadows, Amir, & Gibb, 2014; Wiers, et al.,

2015c; Wittekind, Feist, Schneider, Moritz, & Fritsche, 2015).

As for the outcome category that measured substance use reduction the dependent

measures covering relapse rate was measured in several different ways. Three studies

measured the percentage of patients at a predetermined set of follow-up time points (Begh, et

al., 2015; Elfeddali, et al., 2016; Wiers, 2011). One study measured abstinence during

treatment (Kong et al., 2015). Schoenmakers et al. (2010) measured the amount of days or

months to relapse.

We wanted to investigate the effect of CBM interventions targeting tobacco and

alcohol on four different groups of outcome categories. We had two goals for this study. The

first was to investigate the overall effect of CBM interventions on each of the four groups of

outcome categories, i.e. do CBM interventions reduce cognitive bias, craving, substance

intake or relapse. The second goal was to investigate whether different study characteristics

influence the magnitude of the effect on the outcome categories, i.e. what is the effect of the

type of addiction targeted, the type of CBM intervention, the number of training trials, the

mode of delivery, and the type of participants on each of the four outcome categories. To

answer our research questions we needed two models for all four outcome categories. The

cognitive bias and craving outcome measures contained only nominal outcome measures, the

relapse rate measures were either ordinal or dichotomous and the outcome measures related to

substance use reduction were ordinal, nominal or dichotomous.

Broadly speaking, to answer our research questions there are two groups of models

available, fixed-effects models and random-effects models. We opted for the random-effects

model because we assume that all CBM intervention studies come from an overarching

distribution of effect sizes. The included CBM studies differ on many levels and it has been

shown that these different study characteristics influence the study results (Müllig, Paulick,

Lindenmeyer, Rinck, Cina, & Wiers, 2017).

Based on previous meta-analyses on CBM interventions we expected the effect of

CBM interventions and the study characteristics to be small. Not only did we expect small

effect sizes, the overall number of studies that would meet our inclusion criteria was also

expected to be small (Cristea et al., 2016; Kakoschke, Kemps, & Tiggemann, 2017). These

two factors combined lead to the expectation finding strong evidence for CBM interventions

was unlikely. We know that frequentist statistics can have difficulty in separating the absence

Wagenmaker, 2017). Within a frequentist framework, we cannot reliably detect whether we

reject the alternative hypothesis due to the poor quality of evidence or because there is

evidence for its ineffectiveness. If we want to make claims about the efficacy of CBM

interventions it is vital that we can make this distinction. Bayesian random-effect models do

allow us to quantify the uncertainty of the parameters in the model, which gives us the

opportunity to make this distinction.

Though mathematically quite complex, Bayesian statistics are easily understandable

on a conceptual level. Basically we insert the data we have observed into a model describing

our a priori belief and based on the predictive success we arrive at a posterior model which is

a probabilistic expression of the true value and its relative uncertainty (Lee & Wagenmakers,

2014). This is formalized in Bayes rule:

𝑝𝑝(𝜃𝜃|𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑) = 𝑝𝑝(𝜃𝜃) ×𝑝𝑝(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑|𝜃𝜃)_{𝑝𝑝(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑)}

The starting point is our prior belief that is quantified by a prior distribution 𝑝𝑝(𝜃𝜃), we assign mass to this distribution to reflect our prior knowledge or expectations. We often have

some prior knowledge about a parameter, for instance the outside temperature in Europe is not

likely to exceed 60ºC at any time of the year. If want to model the prior distribution for the

outside temperature on a certain day, we would not assign much (perhaps any) mass to

temperatures above 60ºC. Though this prior knowledge is neglected in frequentist statistics,

Bayesians quantify it and use this prior knowledge as a starting point for analyses. The prior

distribution is updated by the predictive success of the data 𝑝𝑝(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑|𝜃𝜃)

𝑝𝑝(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑) given the possible values

assigned to the parameter of interest θ. This predictive success drives the resulting posterior distribution 𝑝𝑝(𝜃𝜃|𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑) (Marsman, Schönbrodt, Morey, Yao, Gelman, & Wagenmaker, 2017).

As mentioned above we reduced the results of the included studies targeting cognitive

bias and craving to independent t-tests. Marsman et al. (2017) recently showed that the t-tests

can be expressed as the study effects size (δs), study specific mean (µs) and the study specific

common variance (σs) can be used to conduct default Bayesian analyses. To investigate the

group level effects we take two approached. In the first approach, we estimated the effect

sizes of the studies individually, independent from the other included studies. In the second

approach, we used the hierarchical approach to pool the individual effects into a single

distribution by we extended the model with the group level parameters, the group-level mean

θ and group-level variance τ2

. This allowed us to improve the estimation of the uncertainty by

pooling the statistical information across studies and it allowed us to express the group-level

effect size and heterogeneity in an overarching distribution (Marsman et al. (2017). Like

Marsman et al. (2017) we specified standard non-informative prior distributions for

individual-study means μs, individual-study variances σs and the group-level variance τ2 based

on the Jeffreys–Zellner–Siow framework. Lastly a Cauchy (0, 1/√2) prior was used for the group-level mean θ. The group level distribution in this case refers to the efficacy of CBM interventions in lowering cognitive biases. The analyses were run separately for CBM

interventions targeting alcohol and tobacco addiction.

An effect size between 0.2-0.5 indicates a small effect, a medium effect is represented

by a value between 0.5-0.8 and large effects are signified by ES values over 0.8 (Cohen,

1988). We used IBM SPSS version 21, Stan and R for data cleaning, calculating the

difference scores and summary statistics and running both the Bayesian individual parameter

estimation and the random-effects model (IBM Corp, 2012; Stan Development Team, 2016;

Results Characteristics of the studies

Our search identified 4,854 potential studies. Duplicate papers were removed and the

titles and abstracts of the remaining 2,582 studies were screened. During the screening process

several conference abstracts could not be matched to a published study. The authors of these

conference abstracts were contacted, one additional study was identified through this process.

The remaining conference abstracts had not led to any publications. After screening the titles

and abstracts 38 studies were identified for a full text assessment. 22 studies were excluded

after closer examination, in 13 of the studies there was no explicit goals of behavior change.

The remaining 9 search results could not be traced to any publication. The flow of information

depicting the different phases of the selection process are shown in Figure 1.

In total 16 studies met the inclusion criteria, their characteristics are summarized in

Table 2. The sample sizes ranged from 32 to 509 with a total of 2935 participants. Four

studies targeted tobacco addiction and 12 targeted alcohol addiction. Six studies targeted

attentional bias (VPT or AACTP training), 10 studies targeted approach bias (AAT training)

and no studies were included that target response inhibition (Go/no-go training) were

included.

Description of the included studies

Four studies included more than one CBM intervention: one study included three

different versions of the AAT paradigm (Lindgren et al., 2015), one study included four

variations of the A-AAT intervention and also the AACTP intervention (Wiers et al., 2015c),

one study included two variants of the A-AAT (Wiers, Eberl, Rinck, Becker, & Lindenmeyer,

2011), one study included two variants of the A-AAT intervention (Wittekind, Feist,

Table 2. Characteristics of the studies included in the meta-analysis Study Type of addiction Type of CBM intervention Type of control condition N N sessions N training trials /session Training setting Type of participants Time points bias Outcome measures2 Begh et al. (2015)

Tobacco ABM (VPT) Sham training 118 5 192 Controlled Clinical BL, 4w, 8w, 3m, 6m

VPT, Stroop task, CO, MPSS-C, MPSS-M, time to relapse Clerkin et al.

(2016)

Alcohol ABM (VPT) Sham training 86 8 96 Controlled Clinical BL, 1w, 1m

VPT, DDQ, DTC, DrInC, AWSC, SIAS, SSAC, SPQ, BSAM Cox et al.

(2015)

Alcohol ABM (AACTP)

No training 148 4 Controlled Non-clinical BL, PT, 3m, 6m CSSRI-EU, DRQ, SIP, RCTQ, PACI, SWLS, Stroop task Eberl et al. (2013)

Alcohol AAT No training 509 12 200 Controlled Clinical BL, 6w A-AAT, Color Stroop, BDI, RSES, SCL90-R, AASE, AUDIT Elfaddali et

al. (2016)

Tobacco ABM(VPT) Sham training 434 6 240 Online Non-clinical

BL, 2w FTND, craving, intention to quit, self-efficacy, TACS, continued abstinence Kong et al.

(2015)

Tobacco AAT Sham training 60 4/51 260 Controlled Clinical BL, 3m S-AAT, PP, Abstinence, smoking reduction, Cotinine

Lindgren et al. (2015) S1

Alcohol AAT Sham training 295 2 600 Controlled Non-clinical BL, 1w A-AAT, DDQ, AUDIT, RAPI, drinking intention, ACQ Lindgren et al. (2015) S2

Alcohol AAT Sham training 288 2 600 Controlled Non-clinical BL, PT A-AAT, DDQ, AUDIT, RAPI, drinking intention, ACQ Machulska et al. (2016)

Tobacco AAT Sham training 205 4 250 Controlled Clinical BL, PT AAT, FTND, SCS, TAAS, attitude toward smoking, craving Schoenmaker Alcohol ABM (VPT) Sham training 43 5 528 Controlled Clinical BL, PT VPT, DAQ, time to

s et al. (2010) relapse, Sharbanee et

al. (2014)

Alcohol AAT Sham training 74 1 384 Online Non-clinical

BL, PT AAT, AUDIT, SOCRATES, ACQ, Wiers, C.E.

et al. (2015a)

Alcohol AAT Sham training 36 6 400 Controlled Clinical BL, PT Behavioral-AAT, fMRI-AAT,

Wiers, C.E. et al. (2015b)

Alcohol AAT Sham training 32 6 400 Controlled Clinical BL, PT A-AAT, DAQ Wiers, R.W.

et al. (2011)

Alcohol AAT Sham training 214 4 200 Controlled Clinical BL, 1w Craving, IAT, A-AAT, BDI, SCL-90-R Wiers, R.W. et al. (2015) Alcohol ABM (AACTP) AAT

Sham training 136 5 220 Online Non-clinical

BL, 1w AUDIT, TLFB, RCTQ, Self-efficacy

Wittekind, et al. (2015)

Alcohol AAT No training 257 1 100 Online Non-clinical

None CQSS, FTND, CDS-12, OCSS

Notes.

1: Due to a technical error half of the intervention group received sham training instead of S-AAT training

2: A-AAT = Alcohol Approach-Avoidance Task, AASE = Alcohol Abstinence Self-Efficacy Scale, AAT = Approach-Avoidance Task, ACQ = Alcohol Craving Questionnaire, AUDIT = Alcohol Use Disorder Identification Test, AWSC = Alcohol Withdrawal Symptom Checklist-Revised, BDI = Becks Depression Index, BSAM = Brief State Anxiety Measure, CO = Carbon Oxide, CDS-12 = Cigarette Dependence Scale 12, CSSRI-EU = Client Socio-Demographic and Service Receipt Inventory – European Version, CWSS = Commitment to Quitting Smoking Scale, DAQ = Desire for Alcohol

Questionnaire, DDQ = Daily Drinking Questionnaire, DRQ = Drinking Record Questionnaire, DrInC = Drinker Inventory to Cope, DTC = drinking to cope, FTND = Fagerstrom Test for Nicotine Dependence, MPSS-C = Mood and Physical Symptoms Scale-Craving, MPSS-M = Mood and Physical Symptoms Scale-Mood, OCSS = Obsessive Compulsive Smoking Scale, PACI = Personal Aspirations and Concerns Inventory, PP = 7 day Point Prevalence, RAPI = Rutgers Alcohol Problem Index, RCTQ = Readiness To Change Questionnaire, RSES = Rosenberg Self-Esteem Scale, S-AAT = Smoking Approach-Avoidance Task, SCS = Stages of Change Scale, SCL90-R = Symptom Checklist 90-R, SIAS = Social Interaction Anxiety Scale, SIP = Short Index of Problems, SOCRATES = Stages of Change Readiness and Treatment Eagerness Scale, SPQ = Speech Performance Questionnaire, SSAC = State Subjective Alcohol Craving, SWLS = Satisfaction With Life Scale, TAAS = Thoughts About Abstinence Scale, TASC = The Attention Control Scale, TLFB = Time Line Follow Back, VPT = Visual Probe Task

Figure 1. PRISMA flowchart of selection process

One study included a combination of a CBM intervention and motivational interviewing in a

factorial design (Cox, Fadardi, Hosier, & Pothos, 2015). For all studies we contrasted the

CBM intervention with the sham control condition. If there was no sham control condition we

contrasted the CBM intervention with the waitlist control or no training control condition. If a

study contained multiple CBM interventions, all interventions were included in the analysis as

separate entries (Begh, et al., 2015; Lindgren et al., 2015; Schoenmakers, et al., 2010;

Sharbanee, et al., 2015; Wiers, et al., 2011). Both the number of time points and the time

measurement of cognitive bias only directly after the intervention (Elfaddali, De Vries,

Bolman, Pronk, & Wiers, 2016; Lindgren, et al., 2015; Machulska, Zlomuzica, Rinck, Assion,

& Margraf, 2016; Wiers et al., 2011).

Quality assessment

The results of the risk of bias assessment are summarized in Figure 2. The quality

varied over the studies but there was generally a low risk of bias. For all studies the

assignment of participants to the condition was fully random or randomly stratified. Though

all participants were aware that the goal was behavior change, both the participants and

experimenter were blind to the allocation of condition in all but one of the studies. In all

studies several different outcome measures were recorded. It is unlikely that the participants

were aware that the CBM interventions were measurements as well, so the overall risk of

blinding the outcome measure was low. Furthermore, because the CBM interventions are

computerized tasks the conditions are blind to the assessor. Nine studies reported that the

differences between the attrition in the control condition versus the CBM intervention were

non-significant indicating a low risk of attrition bias. For five studies the attrition bias was

unclear and for one it was high. Only two studies had a published study protocol, for the other

14 studies it was unclear whether reporting bias was present.

Outcome Category Cognitive bias Alcohol addiction.

Individual effect sizes

Here we evaluate the results of the reanalysis of the effect sizes of the CBM

intervention studies that target alcohol and contain at least one measure of cognitive bias at a

follow-up of one month. It is evident that not all effect sizes are in the expected direction and

most display a large amount of uncertainty. The posterior effect sizes of the 19 CBM

interventions from the 8 studies targeting alcohol are shown in Figure 3. Four studies featured

several different measures of cognitive bias (Schoenmakers et al, 2010; Wiers et al., 2011;

Lindgren et al., 2015; Sharbanee et al., 2014). One study featured more than one experiment

(Lindgren et al., 2015). The posterior medians are indicated as dots and the 95% central

credible intervals are indicated by horizontal lines. The effect sizes have been coded such that

there predicted effect was positive, and we sorted their values based on the posterior medians.

We can see that the effect sizes vary greatly across studies, with values ranging from small to

large in both directions following interpretation guidelines of Cohen (1988). Three of the 95%

credible intervals do not contain the value zero and fall in the expected direction. Generally,

the credible intervals are wide and are not necessarily smaller as the sample size of the study

increases. The three smallest credible intervals do not contain the three studies with the largest

sample size. Thus, a higher sample size does not necessarily equate to more certainty in the

estimate posterior median effect size.

Random-effects model

The main results of the Bayesian random-effects model are shown in figure 4. As with

the individual estimated effect sizes, we are investigating the effect of CBM interventions

Figure 3. Individual Bayesian parameter estimation results for the directed effect sizes δ for the 19 CBM interventions from the 14 included studies that target alcohol. Posterior medians are shown as dots and the 95% credible as a horizontal line. Effect size’s were coded so that positive effect sizes represent the efficacy of CBM interventions.

Figure 4. Posterior distributions of median group level effect θ (left) and the between study

variance τ2

(right).

Moderator analyses

We were unable to run the intended moderator analysis to investigate the differential effect of the study characteristics on the effect of CBM interventions targeting alcohol or tobacco related cognitive biases. For the analyses in this paper we leaned heavily on the pioneering work of Marsman et al. (2017), his research group developed the software

solutions to run the analyses. Though code was available for the estimation of posterior effect sizes both individually and hierarchically, no solution was available to run the intended moderator analysis. When planning the project no time was reserved to develop the software tools to run this analysis. Developing these tools was deemed beyond the scope of this project.

Tobacco Addiction

Individual effect sizes

In figure 6 the individual posterior median estimates of the effect sizes δ are shown.

Unfortunately the number of studies targeting tobacco is smaller than the number of CBM interventions targeting alcohol. We can see that all posterior estimated median effect sizes are close to zero and that all of the 95% credible intervals include zero. Though there are only a

small number of observations, it is evident from this data that the effect is very small or non-existent.

Figure 5. Individual results from the Bayesian random-effects analysis for the directed effect

sizes δ for the 19 CBM interventions. The effect sizes have shrunk to the group mean as a

Figure 6. Individual Bayesian parameter estimation results for the directed effect sizes δ for the four studies targeting tobacco.

Random-effects model

Because of the small number of studies that include a CBM intervention targeting

smoking, we were unable to execute the random-effect model for that group of studies. In the

Bayesian framework we first define a prior distribution that is updated by the data and the

influence of the selected generative prior model get smaller with each iterative update cycle.

The small amount of observations does not allow us to overcome the relatively large

assumption that the prior distribution represents. A second problem in estimating the model is

observation is large. The result is that given the data that we have on tobacco CBM

interventions we cannot run the random-effect model. There is not enough data of good

enough quality to reliable converge the algorithms that guide our posterior hierarchical model.

Outcome Categories Craving, Relapse Rate and Substance Use Reduction

We were unable to run the sets of analyses for craving, relapse rate and substance use

reduction. Due to several different reasons the project was very optimistic in its planning and

assumptions about the workload. In order to finish the project on time, it was decided to

reduce the initial scope of the project to the analyses investigating the effect of CBM

interventions on cognitive bias.

Discussion

This is the first meta-analysis to assess the efficacy of CBM interventions targeting

tobacco and alcohol using a Bayesian framework that allows us to quantify the uncertainty

associated with the parameter estimates. We performed a systematic review following the

Cochrane Handbook for Systematic Reviews of Interventions. We searched the five most

relevant bibliographic databases and found 16 studies that met our inclusion criteria. We first

examined the estimated individual effect sizes for the studies targeting alcohol using a

Bayesian meta-analytical approach. We followed up the individual parameter estimations with

the estimation of the group level characteristics using a Bayesian random-effects model. We

repeated this procedure for CBM interventions targeting tobacco, but the group level

characteristics could not be estimated because of the few studies that targeted tobacco.

Unfortunately, we were unable to run the planned moderator analyses. We were also unable to

Main effects

The individually estimated effect sizes of the reduction in cognitive bias at one-month

post-training of CBM interventions targeting alcohol vary greatly, from a medium negative

effect to a large positive effect. If we look at the associated credible intervals, we see that only

four of the 19 effect sizes have a credible interval that does not contain zero. We can conclude

that the efficacy of cognitive bias reduction at one-month post-intervention varies greatly

across studies. When we deploy the Bayesian random-effects model we see that the effects are

less extreme but still quite uncertain. When we look at the posterior group level parameter

estimates, we conclude that there is indeed a small effect with little difference between the

studies.

When we look at the four individually estimated effect sizes summarizing the efficacy

of CBM interventions targeting tobacco-related cognitive biases one-month post-intervention,

we conclude all the effects hover around zero and the confidence intervals are huge. Because

of the large amount of uncertainty and the small number of studies, we were unable to

estimate the group level parameters. The algorithms behind the estimation of the parameters

need a certain amount of information to reliably come up with estimates. In cases where there

is not enough data the algorithms produces improbable results, which was the case for this

analysis.

Overall we can conclude that the effect of CBM interventions on reducing substance

related cognitive biases one month after the training intervention is small. When we compare

our analysis with the only other known meta-analysis investigating CBM interventions in

substance addiction we see comparable though different results (Cristea et al., 2016). When

we compare the results for alcohol, we see that the confidence interval for the effect of CBM

interventions ranged from -0.01 to 0.22 which is very similar to our credible interval of 0.02

studies into account when calculating the effect sizes, it is relatively vulnerable to studies that

have extreme results. As mentioned earlier, Cristea et al.’s meta-analysis included both studies

in which participants were aware of the intervention and studies in which participants were

unware of the intervention. Another factor that could explain the discrepancy between the

results is that Cristea and colleagues aggregated different types of outcome measures in one

analysis. Unfortunately we cannot compare the results for studies investigating the efficacy of

CBM interventions targeting tobacco. Christea et al. (2016) included seven studies, because

we only included studies where the participants were explicitly aware that the goal was

behavior change we included four. This small sample size, as previously mentioned,

prevented our Bayesian meta-analysis to run properly.

Risk of bias

Overall the risk of bias of the included studies was low. All studies were Randomized

Controlled Trials (RCT) in which participants were assigned to a condition using some form

of randomization technique. The allocation to the condition was blind to both the participant

and experimenter in most cases. In all studies the participant were blind to their condition. In

all studies participants were aware that the goal was to change behavior, so technically the

participants were not fully blind to the behavioral outcome assessment. However, for the

cognitive bias measures it is likely that the participants were unaware that cognitive bias was

being measured with computerized CBM intervention training. For most studies, the

dropped-out participants were compared with the other participants on baseline measures, with no

systematic difference reported in any of the studies. As is often the case in intervention

studies, if a study does not have a published study protocol it is unclear if all dependent

measures were reported in the results section. Hence we cannot make claims about the

selective reporting of the outcome data. In this study we also did investigate publication bias

between the size of the effect and sample size of the study. However, it is more likely that

smaller effects tend to be investigated with studies with a higher sample size (Simonsohn,

2017). Furthermore, there is no objective way to gage something that has not been observed,

hence we are hesitant with running such an analysis as we seriously doubt inferences made

based on it.

Theoretical implications

The overall effect of CBM interventions on cognitive bias is small. However, there are

a wide range of different factors to take into account when judging the efficacy of CBM

interventions. As described before, different tasks target different cognitive biases that might

function differentially. The role of the different cognitive biases might differ for different

substances as well. A lot of other factors play a role as well, the length and frequency of

training and the study population are factors most likely affecting of the effect of CBM

interventions. In general CBM interventions do reduce cognitive biases, but only slightly. The

question remains whether this small effect results in clinically relevant changes in behavior.

We are currently unable to resolve the question of whether specific instances of CBM

interventions reduce bias and/or clinical outcomes for a certain defined subpopulation.

Resources should be allocated to the development of a hierarchical Bayesian meta-regression.

Limitations

For a number of reasons, we did not run all the analyses we set out to conduct. We did

not run the Bayesian hierarchical regression for the CBM intervention studies targeting

alcohol. We did not have the time to develop the software to run the regression analyses

within the planning of this project. For the random-effects analysis of the CBM interventions

targeting tobacco we did not have enough studies in our sample to run the analysis.

perform this analysis. The random-effect model for t-tests had been run before and code was

available from a previous project, code for any other data types such as ordinal or dichomous

data would have to be developed and tested. Developing software solutions to run a Bayesian

hierarchical regression model for different types of outcome data was beyond the scope of this

project.

Future directions

There are many factors that contribute to the efficacy of CBM interventions on reducing

cognitive bias. In this project, we were not able to disentangle these factors sufficiently. From

the results of this meta-analysis and a recent review (Kakoschke, Kemps, & Tiggemann,

2017) there are indications that CBM interventions targeting approach bias work better than

some other types of CBM interventions targeting other cognitive biases. Within the field of

CBM interventions targeting the cognitive biases related to substance use disorder most effort

should go to further develop CBM interventions that target approach bias in clinical

population of alcohol dependent patients (Wiers et al., 2011; Eberl et al., 2013). Once a more

concise image emerges regarding the efficacy of approach bias interventions for this group of

patients we may increase our understanding of the mechanisms underlying CBM

interventions. Without a more fundamental understanding of the mechanisms of CBM

References

Babb, S., Malarcher, A., Schauer, G., Asman, K., & Jamal, A. (2017). Quitting Smoking

Among Adults—United States, 2000–2015. MMWR. Morbidity and Mortality Weekly

Report, 65.

Bechara, A. (2005). Decision making, impulse control and loss of willpower to resist drugs: a

neurocognitive perspective. Nature neuroscience, 8(11), 1458-1463.

Begh, R., Munafò, M. R., Shiffman, S., Ferguson, S. G., Nichols, L., Mohammed, M. A.,

Holder, R. L., Sutton, S., & Aveyard, P. (2015). Lack of attentional retraining effects

in cigarette smokers attempting cessation: A proof of concept double-blind

randomised controlled trial. Drug and alcohol dependence, 149, 158-165.

Carroll, K. M., & Onken, L. S. (2005). Behavioral therapies for drug abuse. American Journal

of Psychiatry, 162(8), 1452-1460.

Clarke, P. J., Notebaert, L., & MacLeod, C. (2014). Absence of evidence or evidence of

absence: reflecting on therapeutic implementations of attentional bias modification.

BMC psychiatry, 14(1), 8.

Clerkin, E. M., Magee, J. C., Wells, T. T., Beard, C., & Barnett, N. P. (2016). Randomized

controlled trial of attention bias modification in a racially diverse, socially anxious,

alcohol dependent sample. Behaviour Research and Therapy, 87, 58-69.

Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 [updated March

2011]. Higgins, J.P., & Green, S. (Eds.). (2011). The Cochrane Collaboration

(Available from www.cochrane-handbook.org).

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale,

Copetti, M., Fontana, A., Graziano, G., Veneziani, F., Siena, F., Scardapane, M., Lucisano,

G., & Pellegrini, F. (2013). Advances in meta-analysis: Examples from internal

medicine to neurology. Neuroepidemiology, 42(1), 59-67.

Cox, W. M., Fadardi, J. S., Hosier, S. G., & Pothos, E. M. (2015). Differential effects and

temporal course of attentional and motivational training on excessive drinking.

Experimental and clinical psychopharmacology, 23(6), 445.

Cristea, I. A., Kok, R. N., & Cuijpers, P. (2016). The effectiveness of cognitive bias

modification interventions for substance addictions: a meta-analysis. PloS one, 11(9),

e0162226.

Eberl, C., Wiers, R. W., Pawelczack, S., Rinck, M., Becker, E. S., & Lindenmeyer, J. (2013).

Approach bias modification in alcohol dependence: do clinical effects replicate and for

whom does it work best? Developmental Cognitive Neuroscience, 4, 38-51.

Efron, B., & Morris, C. N. (1977). Stein's paradox in statistics. Scientific American, 236(5),

119-127.

Elfeddali, I., de Vries, H., Bolman, C., Pronk, T., & Wiers, R. W. (2016). A randomized

controlled trial of Web-based Attentional Bias Modification to help smokers quit.

Health Psychology, 35(8), 870.

Everaert, J., Mogoaşe, C., David, D., & Koster, E. H. (2015). Attention bias modification via single-session dot-probe training: Failures to replicate. Journal of behavior therapy

and experimental psychiatry, 49, 5-12.

Field, M., & Cox, W. M. (2008). Attentional bias in addictive behaviors: a review of its

Gelman, A. (2005, January 25). Why I don’t use the term “fixed and random effects”.

Retrieved from http://andrewgelman.com/2005/01/25/why_i_dont_use/

Goldstein, R. Z., & Volkow, N. D. (2002). Drug addiction and its underlying neurobiological

basis: neuroimaging evidence for the involvement of the frontal cortex. American

Journal of Psychiatry, 159(10), 1642-1652.

Harwood, H. J., Fountain, D., & Livermore, G. (1998). The economic costs of alcohol and

drug abuse in the United States, 1992. US Department of Health and Human Services, National Institutes of Health, National Institute on Drug Abuse, Office of Science

Policy and Communications.

Houben, K., Havermans, R. C., Nederkoorn, C., & Jansen, A. (2012). Beer à No‐Go: Learning to stop responding to alcohol cues reduces alcohol intake via reduced affective

associations rather than increased response inhibition. Addiction, 107, 1280-1287.

Houben, K., & Wiers, R. W. (2006). Assessing implicit alcohol associations with the Implicit

Association Test: Fact or artifact? Addictive Behaviors, 31, 1346–1362.

Houben, K., Wiers, R. W., & Jansen, A. (2011). Getting a grip on drinking behavior: Training

working memory to reduce alcohol abuse. Psychological Science, 22, 968–975.

IBM Corp. Released 2012. IBM SPSS Statistics for Windows, Version 21.0. Armonk, NY:

IBM Corp.

Kakoschke, N., Kemps, E., & Tiggemann, M. (2017). Approach bias modification training

and consumption: A review of the literature. Addictive Behaviors, 64, 21-28.

Kong, G., Larsen, H., Cavallo, D. A., Becker, D., Cousijn, J., Salemink, E., Collot

Re-training automatic action tendencies to approach cigarettes among adolescent smokers:

a pilot study. The American journal of drug and alcohol abuse, 41(5), 425-432.

Lee, M. D., & Wagenmakers, E. J. (2014). Bayesian cognitive modeling: A practical course.

Cambridge University Press.

Liberati, A., Altman, D. G., Tetzlaff, J., Mulrow, C., Gøtzsche, P. C., Ioannidis, J. P., Clarke,

M., Devereaux, P.J., Kleijen, J., & Moher, D. (2009). The PRISMA statement for

reporting systematic reviews and meta-analyses of studies that evaluate health care

interventions: explanation and elaboration. Annals of internal medicine, 151(4), W-65.

Lindgren, K. P., Wiers, R. W., Teachman, B. A., Gasser, M. L., Westgate, E. C., Cousijn, J.,

Enkema, M. C. & Neighbors, C. (2015). Attempted training of alcohol approach and

drinking identity associations in US undergraduate drinkers: null results from two

studies. PloS one, 10(8), e0134642.

Machulska, A., Zlomuzica, A., Rinck, M., Assion, H. J., & Margraf, J. (2016). Approach bias

modification in inpatient psychiatric smokers. Journal of psychiatric research, 76,

44-51.

Marsman, M., Schönbrodt, F. D., Morey, R. D., Yao, Y., Gelman, A., & Wagenmakers, E. J.

(2017). A Bayesian bird's eye view of ‘Replications of important results in social

psychology’. Royal Society Open Science, 4(1), 160426.

McHugh, R. K., Hearon, B. A., & Otto, M. W. (2010). Cognitive behavioral therapy for

substance use disorders. Psychiatric Clinics of North America, 33(3), 511-525.

McHugh, R. K., Murray, H. W., Hearon, B. A., Calkins, A. W., & Otto, M. W. (2010).

Attentional bias and craving in smokers: the impact of a single attentional training

Miller, W. R., Walters, S. T., & Bennett, M. E. (2001). How effective is alcoholism treatment

in the United States? Journal of studies on alcohol, 62(2), 211-220.

Mogg, K., Bradley, B. P., Field, M., & De Houwer, J. (2003). Eye movements to

smoking-related pictures in smokers: Relationship between attentional biases and implicit and

explicit measures of stimulus valence. Addiction, 98, 825–836.

Moher, D., Liberati, A., Tetzlaff, J., & Altman, D. G. (2009). Preferred reporting items for

systematic reviews and meta-analyses: the PRISMA statement. Annals of internal

medicine, 151(4), 264-269.

Mühlig, S., Paulick, J., Lindenmeyer, J., Rinck, M., Cina, R., & Wiers, R. W. (2017).

Applying the ‘Cognitive Bias Modification’ concept to smoking cessation–A

Systematic Review. SUCHT.

R Core Team (2016). R: A language and environment for statistical computing. R Foundation

for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

Robinson, T. E., & Berridge, K. C. (1993). The neural basis of drug craving: an

incentive-sensitization theory of addiction. Brain research reviews, 18(3), 247-291.

Berridge, K. C., Venier, I. L., & Robinson, T. E. (1989). Taste reactivity analysis of

6-hydroxydopamine-induced aphagia: Implications for arousal and anhedonia

hypotheses of dopamine function. Behavioral Neuroscience, 103, 36 – 45.

Schoenmakers, T. M., de Bruin, M., Lux, I. F., Goertz, A. G., Van Kerkhof, D. H., & Wiers,

R. W. (2010). Clinical effectiveness of attentional bias modification training in

Schoenmakers, T. M., de Bruin, M., Lux, I. F., Goertz, A. G., Van Kerkhof, D. H., & Wiers,

R. W. (2010). Clinical effectiveness of attentional bias modification training in

abstinent alcoholic patients. Drug and alcohol dependence, 109(1), 30-36.

Schoenmakers, T. M., Wiers, R. W., Jones, B. T., Bruce, G., & Jansen, A. T. M. (2007).

Attentional re-training decreases attentional bias in heavy drinkers without

generalization. Addiction, 102, 399–405.

Schoorl, M., Putman, P., & Van Der Does, W. (2013). Attentional bias modification in

posttraumatic stress disorder: a randomized controlled trial. Psychotherapy and

Psychosomatics, 82, 99e105.

Sharbanee, J. M., Hu, L., Stritzke, W. G., Wiers, R. W., Rinck, M., & MacLeod, C. (2014).

The effect of approach/avoidance training on alcohol consumption is mediated by

change in alcohol action tendency. PloS one, 9(1), e85855.

Simonsohn, U. (2017, March 21). The Funnel Plot is Invalid Because of This Crazy

Assumption: r(n,d)=0. Retrieved from http://datacolada.org/58

Stan Development Team. 2016. RStan: the R interface to Stan. R package version 2.14.1.

http://mc-stan.org.

Tapert, S. F., Schweinsburg, A. D., Barlett, V. C., Brown, S. A., Frank, L. R., Brown, G. G.,

& Meloy, M. J. (2004). Blood oxygen level dependent response and spatial working

memory in adolescents with alcohol use disorders. Alcoholism: Clinical and

Experimental Research, 28, 1577–1586.

Tapert, S. F., Schweinsburg, A. D., Drummond, S. P., Paulus, M. P., Brown, S. A., Yang, T.

T., & Frank, L. R. (2007). Functional MRI of inhibitory processing in abstinent

Vickers, A. J., & Altman, D. G. (2001). Analysing controlled trials with baseline and follow

up measurements. British Medical Journal, 323(7321), 1123-1124.

Whiteford, H. A., Degenhardt, L., Rehm, J., Baxter, A. J., Ferrari, A. J., Erskine, H. E.,

Charlson, … & Vos, T. (2013). Global burden of disease attributable to mental and

substance use disorders: findings from the Global Burden of Disease Study 2010. The

Lancet, 382(9904), 1575-1586.

Wiers, C. E., Ludwig, V. U., Gladwin, T. E., Park, S. Q., Heinz, A., Wiers, R. W., ... &

Bermpohl, F. (2015). Effects of cognitive bias modification training on neural

signatures of alcohol approach tendencies in male alcohol‐dependent patients. Addiction biology, 20(5), 990-999.

Wiers, C. E., Stelzel, C., Gladwin, T. E., Park, S. Q., Pawelczack, S., Gawron, C. K., ... &

Lindenmeyer, J. (2015). Effects of cognitive bias modification training on neural

alcohol cue reactivity in alcohol dependence. American Journal of Psychiatry, 172(4),

335-343.

Wiers, R. W., Eberl, C., Rinck, M., Becker, E. S., & Lindenmeyer, J. (2011). Retraining

automatic action tendencies changes alcoholic patients’ approach bias for alcohol

and improves treatment outcome. Psychological science, 22(4), 490-497.

Wiers, R. W., Gladwin, T. E., Hofmann, W., Salemink, E., & Ridderinkhof, K. R. (2013).

Cognitive bias modification and cognitive control training in addiction and related

psychopathology: Mechanisms, clinical perspectives, and ways forward. Clinical

Psychological Science, 1(2), 192-212.

Wiers, R. W., Houben, K., Fadardi, J. S., van Beek, P., Rhemtulla, M., & Cox, W. M.

(2015c). Alcohol cognitive bias modification training for problem drinkers over the

Wiers, R. W., Schoenmakers, T., Rinck, M., & Lindenmeyer, J. (2010). Re-training

Attentional Bias And Approach Bias For Alcohol In Alcoholic Patients. Alcoholism:

Clinical & Experimental Research, 34(8), 69A.

Wiers, R. W., Rinck, M., Dictus, M., & Van den Wildenberg, E. (2009). Relatively strong

automatic appetitive action‐tendencies in male carriers of the OPRM1 G‐allele. Genes, Brain and Behavior, 8(1), 101-106.

Wittekind, C. E., Feist, A., Schneider, B. C., Moritz, S., & Fritzsche, A. (2015). The

approach-avoidance task as an online intervention in cigarette smoking: a pilot

study. Journal of behavior therapy and experimental psychiatry, 46, 115-120.

World Health Organization. (2014). Global status report on alcohol and health 2014. World

Health Organization.

World Health Organization. (2015). WHO report on the global tobacco epidemic, 2015:

Raising taxes on tobacco.

Wittekind, C. E., Feist, A., Schneider, B. C., Moritz, S., & Fritzsche, A. (2015). The

approach-avoidance task as an online intervention in cigarette smoking: a pilot

study. Journal of behavior therapy and experimental psychiatry, 46, 115-120.

World Health Organization. (2014). Global status report on alcohol and health 2014. World

Health Organization.

World Health Organization. (2015). WHO report on the global tobacco epidemic, 2015:

Appendix A

Search Template

Ruben van Beek Janneke Staaks Marilisa Boffo

Research question: cognitive bias modification interventions for the treatment of alcohol and tobacco addiction

Databases

PsycINFO 976 results (May 18, 2016) Medline 904 results (May 18, 2016) Web of Science 1.262 results (May 18, 2016) Embase 1.444 results (May 18, 2016) Cochrane library 269 results (May 18, 2016) Total 4.854 results

Total, deduplicated 2.583 results

NB. the author should be able to fill out the PRISMA statement flow diagram.

NB. At this stage, the PRISMA-P guidelines for review protocols also might be useful Other search techniques

• the following journals/conference proceedings/etc. were hand searched:

• Citation tracking (how; which articles, which database (GS or WoS))

• Expert consultation (how, who) Extra information

• Handleiding Refworks voor Systematic Reviews

• Handleiding literatuur zoeken

• Template zoekstrategieen

PsycINFO

#1 cognitive bias modification interventions

cognitive bias/ OR attentional bias/ OR selective attention/ OR approach avoidance/ OR approach behavior/ OR response inhibition/ OR cognitive control/ OR (((cognitive or memory or attention*) adj3 bias*) OR selective attention OR approach bias* OR avoid* bias* OR (approach* ADJ3 tendenc*) OR approach-avoidance OR CBM intervention* OR CBM training* OR implicit association* OR evaluative conditioning OR inhibitory control* OR inhibition task* OR inhibitory process* OR response inhibition* OR (attention* ADJ1 (retraining OR training OR modification))).ti,ab,id,tm.

Results: 32.128

#2 Addiction (smoking or alcohol)

alcoholism/ OR alcohol abuse/ OR alcohol drinking patterns/ OR drinking behavior/ OR alcohol intoxication/ OR smoking cessation/ OR tobacco smoking/ OR nicotine/ OR nicotine withdrawal/ OR (alcohol* OR binge drink* OR drug* OR heavy drink* OR nicotin* OR smoking OR smoker* OR tobacco OR cigaret* OR drinker* OR drinking behavio*).ti,ab,id. Results: 314.813