The present study is a randomized controlled trial (RCT) comparing self-report, behavioral, and biochemical outcomes of the app-AAT training to a control condition, in which the app training is not received (Additional file 1). Prior to randomization, all smokers included in the study receive standard help for smoking cessation (cf. precise description subsequently). This study employs a 2 (training: training/control) × 3 (time: pretest/posttest/follow up: 6 weeks after baseline) mixed design. The training intervention occurs over the course of 2 weeks in which smokers are given the opportunity to complete AAT-training sessions at home.
We focus on two primary outcome measures: (1) changes in smoking-related approach biases with regard to smoking stimuli presented during training (measured in milliseconds) and (2) reductions in self-reported daily cigarette smoking (measured in number of smoked cigarettes per day). Primary time points include post-training and follow-up assessments. Secondary outcome measures are (1) changes in smoking-related approach biases with regard to untrained pictures (close generalization), (2) changes in approach biases for positive pictures, (3) changes in other cognitive biases as measured by the visual dot-probe task and the Implicit Association Task (IAT) (broad generalization), (4) self-monitored smoking behavior using a cigarette tracking app, (5) biochemically verified abstinence rates defined as point prevalence with a 7-day window prior to the assessment point, (6) expired CO, and (7) other self-reported smoking behavior.
A priori power analyses were conducted using G*Power 3.1 (open-source software ). Previous research investigating the effectiveness of CBM training identified a small-to-moderate effect size [21, 35]. We conducted a power analysis for a 2 × 3 mixed design ANOVA to detect a small-to-moderate effect (Cohen’s d = 0.30, α = 0.05) for the interaction between experimental condition and time for each primary outcome and time point. Power of 0.80 and r = 0.5 for correlation between the repeated measures was assumed. Results indicated that 74 participants in total would be needed. Due to an expected attrition rate of approximately 25% at follow up (defined as dropout at any time point after completing the baseline assessment), we decided to include 100 participants, that is 50 per condition.
We would like to stress that we performed power calculations for primary outcomes exclusively. With regard to secondary outcome measures, we emphasize that the study is explorative, meaning that any significant findings on these outcome variables should be interpreted with caution until tested in further confirmatory RCTs.
Current smokers (n = 100) will be recruited at the University of Siegen (Germany) and from the general population. Figure 1 provides the Consolidated standards of reporting trials (CONSORT) diagram of participant recruitment. Participants will be recruited through flyer advertisements, radio broadcasts, television reports, and newspaper advertisement. Interested participants will receive information about the study via e-mail and will be invited to take part in a telephone interview to assess eligibility criteria. Smokers will be included if they have smoked at least six cigarettes per day for at least 6 months. Exclusion criteria are current alcohol or drug misuse or dependency, present psychiatric illness, insufficient German language skills, or uncorrected visual or auditory impairment. Those eligible for the study will be invited for three laboratory sessions. The first two sessions will be 2 weeks apart; the third session will take place approximately 6 weeks after the first (baseline) session. Full written informed consent will be obtained from each participant at study entry (at the beginning of the first session).
The study protocol (version 1: 11/2018) was approved by the local Ethics Committee of the University of Siegen (reference number ER_16_2018) and is conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. During the course of the study, ethical, legal and social aspects (ELSA) will be anticipated and addressed. Each participant will provide written informed consent to participate in the experimental procedure prior to inclusion in our study. Participation will be entirely voluntary and participants will have the right to withdraw their consent for participation at any time.
Randomization and blinding
Participants will be randomly assigned to the experimental or the control group with a 1:1 allocation ratio, according to an externally constructed randomization plan. To do so, a computer-generated randomization schedule will be employed by means of a computerized random number generator using IBM SPSS Statistics 24. Permuted block randomization will be used to ensure that study groups of approximately the same size are generated. Block size will be set to 50 participants. No other stratification will be used. The study coordinator will perform randomization. Due to the study design, it is not possible to blind participants. However, to avoid bias in the outcome assessment, research assistants concerned with data collection and/or preparation will be blind to the allocation of the participants.
Prior to randomization, participants will receive behavioral counseling and psychoeducation containing information on nicotine addiction and maintenance and short-term and long-term effects associated with cigarette smoking (about 90 min). Afterwards, smokers will be handed a self-help book (a German copy of The easy way to stop smoking by Allen Carr) to aid smoking cessation. Finally, participants will be instructed to record cigarettes smoked throughout the day for the entire study period of 6 weeks. Therefore, our behavioral interventions for smoking cessation include brief behavioral counseling, a self-help book, and the instruction to self-monitor smoking. Taken together, these three interventions constitute the TAU condition. Participants allocated to the experimental condition are then given access to the AAT-training app and are instructed to train at least once per day for 14 consecutive days. Participants perform practice training with the researchers to ensure that the training handling and concepts have been fully understood. To ensure compliance, participants will receive daily reminders via short message service (SMS) to complete the training. After completion of the study (after 6-week follow-up assessment), control condition participants will receive access to the training app.
The training will be conducted using the AAT-training app, which is based on the AAT-training previously used by our research group . Participants assigned to the training condition will be told that the training is supposed to reduce automatic approach tendencies for cigarette cues. During training, smoking-related or positive pictures appear at the center of the mobile screen. The smoking stimuli consist of pictures of cigarettes or individuals smoking cigarettes. The positive pictures display positive social interactions (i.e., convivial gathering or playing sports) or nature scenes. We decided to use positive stimuli because of the absence of a natural control category for cigarettes. Additionally, a previous study conducted by members of our research group [18, 36] found evidence for a diminished approach tendency for natural reward stimuli in smokers. This reduced sensitivity for rewarding cues might explain why some smokers continue to smoke despite obvious detriments. Thus, we considered that next to reducing maladaptive approach tendencies for smoking cues, it would be valuable to simultaneously increase approach behavior for alternative, positive and non-hazardous activities. Baird et al.  kindly provided all pictures. Each of the picture categories contain 25 images presented in a random order. The images are rotated either 3° to the left or 3° to the right. Participants will be instructed to ignore image content and to respond to image orientation by swiping up or down. By swiping up, the picture decreases in size, whereas by swiping down, the picture increases in size, creating a sense of avoiding vs approaching the image. Thus, an indirect instruction is employed. Participants have to execute the correct movement to make the picture disappear. For the purpose of bias assessment, 12 test trials are provided at the beginning of each training session, in which smoking-related and positive pictures appear both in a swipe down and swipe up format. Afterwards, the training starts in which all smoking-related pictures appear in swipe up (avoid) format and all positive pictures appear in swipe down (approach) format. Each picture is shown twice, resulting in 100 training trials per training session. Participants are allowed to take a short break halfway through. Training sessions take approximately 10 min to complete.
To access the training, each participant is required to enter his/her personalized participant-ID before he/she can log on to the app. After completion of each training session, a training file including participant-ID, date of the training sessions, and reaction times (RTs) in milliseconds per trial will be uploaded to a server enabling study investigators to download and view the file. In doing so, we can track the number of training sessions completed by each participant and thereby assess fidelity.
Material and measures
Cognitive bias assessment
To investigate training effects and to test for generalization of training effects, three different cognitive biases will be assessed at baseline, post-training, and follow up.
Approach bias assessment
Automatic approach biases for smoking-related stimuli will be measured by means of the standard version of the nicotine-AAT (see ). At baseline, 25 smoking-related and 25 positive images derived from Baird et al.  appear on a computer screen. To test whether training effects can be generalized to pictures other than those used during bias retraining, 12 out of the 25 pictures contain images not shown during app-AAT training (close generalization). Similar to the training, images are tilted 3° to the left or 3° to the right. A joystick (Logitech Extreme 3D) is connected to the computer and participants are told to use the joystick to push images rotated to the right and to pull images rotated to the left. As a result, images shrink or grow in size dependent on arm movement. For the purpose of bias assessment, smoking and control pictures have to be pulled and pushed equally often. Each picture is shown once in push-away format und once in pull-closer format, resulting in 100 assessment trials.
An approach bias in inferred from faster picture pulling than picture pushing. That is, an approach bias score is calculated by subtracting median RTs for pulling a picture from median RTs for pushing the exact same picture. The RT is defined as the time a participant needed to execute the correct full joystick movement. Accordingly, a positive value indicates an approach tendency toward a picture category, whereas a negative value indicates an avoidance tendency. Approach biases are computed for each of the image categories (smoking vs positive; trained vs untrained) and for each of the laboratory sessions (baseline, post-training, and follow up), resulting in four bias scores at three different time points.
Attentional bias assessment
Automatic attentional biases for smoking cues will be measured by a visual dot-probe task adapted from Miller and Fillmore . The task will be operated using Inquisit Lab software and will be performed on a personal computer. Each trial starts with a 500-ms fixation cross in the center of the computer screen. Afterwards, two 13 × 18 cm pictures (a smoking-related and a control picture) appear side by side, 3 cm apart. The position of the pictures is randomly chosen as either left or right to the location of the fixation cross. After a short duration of 1000 ms, the two pictures disappear and a probe stimulus (here “X”) appears in the location of one of the pictures. A response pad (Cedrus Response Pad RB844) is attached to the computer and participants are asked to press a yellow key if the probe is left and to press a green key if the probe is right. Note that we decided to use response pads instead of serial keyboards since subject responses with standard PC keyboards are frequently associated with accuracy problems. As keyboards are not generally designed to be fast input devices in absolute terms, RTs can vary when using serial keyboards due to polling loops and the electronics in the keyboard. For instance, Plant and Turner  compared four different keyboards and found that on average, 18–34 ms were added to response times with different amounts of variability. Such timing errors are crucial given that cognitive bias assessment relies on rather small RT differences (on average 40–80 ms on a group level). Furthermore, accuracy problems in cognitive bias assessment might account for null findings, replication failure, or inconsistencies in the literature [39, 40].
The task stimuli consist of 10 smoking-related images that were matched with 10 neutral tooth-cleaning control images. This image set was kindly provided by Stippekohl and colleagues  and has been previously used in other studies conducted by members of our research group [18, 21, 36]. One major advantage of this picture set is that smoking and control pictures were carefully matched in terms of colors and shape (see ). In addition, contrary to the pictures used in the AAT, which display complex scenes, these pictures contain a simple smoking or tooth-cleaning related content without distractors. This fact is crucial since evidence hints to the fact that complex addiction-related scenes might be less effective at capturing participants’ attention and could therefore result in less attentional bias when used in visual probe tasks . Each image pair is presented four times, resulting in 40 test trials. In addition, 40 filler trials are included, which consisted of 10 pairs of neutral images. This is done to reduce possible habituation to smoking-related stimuli that might occur otherwise . Test and filler trials are presented randomly, resulting in 80 trials in total. Filler trials are not included in the final data analysis.
An attentional bias is inferred from faster responding to probes replacing a smoking-related image than to those replacing an image unrelated to smoking. To calculate an attentional bias score, RTs for probes replacing smoking pictures are subtracted from RTs for probes replacing tooth-cleaning pictures. Thus, a single measurement of attention bias at each of the three assessment times emerges. A positive value reflects an attentional bias toward smoking-related pictures while a negative value is indicative of an attention bias for tooth-cleaning pictures.
Association bias assessment
Implicit positive or negative associations for smoking will be measured by an implicit association task (IAT) . The task was adapted from Kahler and colleagues . Participants are asked to categorize positive and negative attributes (e.g., “welcomed” vs “disrespected”) and target items (e.g., pictures of a cigarette vs a chair) into predetermined categories via response pad button presses (Cedrus Response Pad RB844).
Following Greenwald et al. , the IAT is organized in seven blocks: (a) a 24-trial target discrimination block (e.g., press yellow for “smoking” vs press green for “furniture”); (b) a 24-trial attribute discrimination block (e.g., yellow for “I feel positive” vs green for “I feel negative”); (c) a 24-trial practice combined block (e.g., yellow for “smoking” or “I feel positive” vs green for “furniture” or “I feel negative”); (d) a 40-trial test combined block (same as practice); (e) a 24-trial target discrimination block, in which the target categories are reversed (e.g., yellow for “furniture” vs green for “smoking”); (f) a 24-practice combined block with reversed target categories (e.g., yellow for “furniture” or “I feel positive” vs green for “smoking” or “I feel negative”), and (g) a 40-trial test combined block (same as practice). Blocks c, d, f, and g are crucial blocks used in scoring the IAT.
Assuming that participants associate smoking with positive feelings, trials in which smoking and positive consequences share a response key are congruent, whereas trials in which smoking and negative consequences share a response key are incongruent. The IAT score will be calculated by subtracting RTs for incongruent blocks (i.e., smoking + negative; furniture + positive) from RTs for congruent blocks (i.e., smoking + positive; furniture + negative). Larger IAT scores suggest stronger implicit, positive, social associations with smoking. Similar to the attention bias calculation, a single IAT score results for each of the assessment times.
To prevent methodological confounds, the IAT is counterbalanced in two ways: first, the placements of the target and attribute stimuli labels are counterbalanced (yellow or green button). Second, two IAT orders are used: one with the congruent combination block first and one with the incongruent combination block first. The two IAT orders will be counterbalanced across participants.
Expired CO will be assessed at baseline, post-training, and at 6-week follow up using a Carbon Monoxide Monitor (piCO™ Smokerlyzer®; Bedfont Scientific Ltd).
Behavioral and self-report measures
To track daily smoking and possible changes in nicotine consumption, participants are asked to self-monitor their smoking behavior via a cigarette-tracking app, which was specifically designed for the purpose of this study. That is, participants are required to download a native journaling application and are instructed to log cigarettes during or directly after smoking. The data will then be charted on the phone and on the companion website to which it is automatically synchronized. In addition, participants will be asked to estimate their average daily nicotine consumption at pretest, post-training and follow up.
The questionnaire measures will include the Fagerström Test for Nicotine Dependence (FTND)  (German version: ), the Stages of Change Scale  (German version: ), the Thoughts About Abstinence Scale , attitudes toward smoking based on Swanson et al. , the Barratt Impulsiveness Scale (BIS) , and the Eysenck Personality Questionnaire - Revised (EPQ-R) . In addition, participants indicate their level of cigarette craving on a 6-point Likert-scale ranging from 0 (“not at all”) to 5 (“very high”). Questionnaire measures will be used to examine group differences at baseline and to assess any changes across the course of the study. At posttest, participants in the experimental condition will be also required to evaluate the app-AAT training and indicate their awareness about the training contingency.
A time schedule of enrollment, assessment visits, and AAT-app trainings for participants is shown in Fig. 2. During the first laboratory session (baseline; around 180 min) participants give informed consent and take part in a brief behavioral intervention for smoking cessation (TAU). Participants then complete the cognitive bias assessments, the questionnaire measures, and the carbon monoxide breath test. Smokers in the training condition (TAU + training) are given access to the AAT-training app. Participants are instructed to perform the app training on a daily basis for 14 consecutive days. During the second (post-training) and final (6-week follow up) laboratory session, participants complete cognitive bias assessment tasks, questionnaire measures, and the CO breath test. Participants in the experimental condition are asked to evaluate the training and indicate their awareness about training contingencies. Individuals randomized to the control condition will receive access to the training app upon completion of the final follow-up measure. Training data from these participants will not be analyzed.
Data preparation and planned analyses
Multiple imputation of missing data will be performed for intention-to-treat (ITT) analyses . To analyze changes in cognitive biases, error trials will be excluded from further analyses. Median RTs will be used to minimize the influence of outliers.
To investigate whether the app training leads to changes in the primary outcome measures, two separate 2 (condition: training vs control) × 3 (time: baseline, post-training, follow up) mixed design analysis of variance (ANOVA) will be performed. Changes in smoking approach biases and self-reported daily nicotine consumption will constitute the primary dependent variables. A separate univariate test will determine if there are significant differences in smoking-related approach biases or average nicotine consumption at baseline between the app training and the control group. Due to the random assignment to the experimental or to the control group, we expect a priori group differences to be small. Nevertheless, chance baseline imbalance between treatment arms will be taken into account, when evaluating training effects. Substantive hypotheses will be tested by means of follow-up simple effects analyses. Here, we compare means in the app training and control group (posttest and follow up, respectively), while controlling for baseline approach bias and nicotine consumption levels. We apply a mixed design ANOVA as one factor (a fixed effects factor) is a between-subjects variable (here, condition) and the other (a random effects factor) is a within-subjects variable (here, time).
To test whether the app training leads to changes in the secondary outcome measures, parametric and non-parametric statistical tests will be administered where appropriate. Close training generalization will be investigated via a 2 (condition: app-training vs control) × 2 (image content: smoking vs positive pictures) × 2 (generalization: trained vs untrained pictures) × 3 (time: baseline, posttest, follow up) mixed-design ANOVA. The 2 (condition) × 3 (time) mixed-design ANOVA will be performed separately for attentional and association biases. Multilevel modeling (MLM) will be used to estimate the growth curve for nicotine consumption as measured with the cigarette training app over time (throughout the entire study period from baseline to follow up). Time will be modeled as a continuous variable. The major advantage of MLM is that it increases power and generalizability by including all individuals regardless of missing data. To examine whether the app-AAT training leads to increased abstinence rates, for each following time point, a separate chi-square test will be carried out with the binary outcome variable (smoking vs nonsmoking) as the parameter of interest. To test the effect of training on other smoking-related variables (i.e., CO, craving, FTND, smoking attitude, motivation to quit), a 2 (condition) × 3 (time) repeated measures multivariate ANOVA (MANOVA) will be conducted. MANOVA with a significant result will be followed up with both univariate tests and discriminant factor analysis. That is, ANOVA will be carried out separately on each of the dependent variables. These analyses will in turn be followed up using contrasts. In addition, discriminant analyses will be applied, which take linear combinations of the dependent variables into account, thereby complementing univariate approaches.