To inform just-in-time interventions, it is crucial to test if CBT- and mindfulness or ACT-based messages can reduce momentary smoking urges. We will conduct an MRT (repeated within-subject randomizations of messages) to accomplish this [30,37]. In line with our existing protocol, participants first collect EMA data for 14 days, allowing us to determine high-risk situations for smoking. In the following intervention phase, participants receive tailored messages triggered by geofencing of participants’ high-risk locations for a total of 30 days. Tailoring is based on established predictors of smoking relapse (stress and presence of other smokers) [38-40]. The MRT tests the efficacy of CBT versus mindfulness or ACT versus control messages for reducing smoking urge up to 15 minutes after message delivery. Secondary outcomes include smoking or other tobacco use (including e-cigarettes), affect, and stress.

It is important to investigate if repeated messages in the MRT impact smoking behavior over time, in contrast to repeated assessment without messages. Thus, this study includes a conventional RCT component. Parallel to the MRT group, a control group completes EMA surveys only without intervention messages. This allows us to test if messages reduce smoking behavior. The primary outcome is number of cigarettes per day at the end of treatment, a 3-month follow-up, and a 6-month follow-up. Secondary analyses explore biochemically verified 7-day point prevalence abstinence, switching to e-cigarettes, and other tobacco outcomes. Post hoc dose-response analyses investigate the long-term efficacy of CBT and mindfulness or ACT messages on smoking behavior.

A crucial research question to inform future mobile interventions is how well intervention messages work in different situational contexts and when people are co-using other substances. Among MRT intervention group participants, we will explore how urge reduction message efficacy may be moderated by substance co-use and exposure to specific settings.

All study procedures are reviewed and approved by the institutional review board of the Johns Hopkins Bloomberg School of Public Health (IRB00013413).

All participants will be required to meet eligibility criteria and provide online consent prior to study involvement. To assess understanding of the information provided in the informed consent, eligible participants will be asked a series of 3 multiple-choice questions regarding the informed consent before being able to proceed with the study. Answers will be collected, and any wrong answer will result in the potential participant being prompted to review the full consent document before attempting the consent questions again. Potential participants will not be enrolled in the study if they answer the questions incorrectly 4 times. After consenting to participate but before being enrolled into the study, participants will be required to send study staff a picture of a valid identification (eg, driver’s license) that has their name, picture, and birth date to validate their age and the fact that they are a real person. Only the participant picture on the ID (headshot) will be retained for biochemical verification of smoking abstinence and will be deleted at the end of data collection.

Participant privacy and data confidentiality will be maintained throughout the study. Screening, baseline, and follow-up survey data will be collected using the Qualtrics survey platform, and data are securely stored on Johns Hopkins servers. Moreover, the study utilizes a HIPAA-compliant mobile research platform (MetricWire Catalyst) to administer EMA surveys, capture GPS location data, and deliver intervention messages. Data collected through the app are encrypted during transmission and securely stored on the platform’s servers. No personally identifiable information is stored on the participant’s device. All location data are time-stamped and linked only to unique participant identifiers. Access to raw data is limited to authorized study staff. Data used for analysis are de-identified and stored on secure, access-controlled servers at Johns Hopkins. Any publications or presentations resulting from the study will use only aggregated or anonymized data. For example, location data points may be jittered or mapped to an alternative base layer to maintain participant anonymity.

The study will use an incentive scheme to reward participants for high compliance with EMA prompts. Participants will receive US $2 for each day of participation in the EMA surveys (US $2 x 44 days = US $88) plus an extra incentive of US $90 if they complete at least 75% of the prompted assessments. Participants will also receive additional incentives for completion of baseline (US $10) and follow-up surveys at 3 months and 6 months (US $20 each). The maximum total incentive amount will be US $228 for participants in the trial.

We will use a dynamic minimization randomization approach to assign participants into the MRT intervention group (n=80) and EMA-only control group (n=80; Figure 1) while ensuring balance across arms on gender and nicotine dependence status. The variable known to robustly impact smoking cessation success is level of nicotine dependence (smoking within first 30 minutes after waking: yes/no) [45]. Gender will be used as additional variable to be balanced across arms. This adaptive approach recalculates imbalance scores in real time based on the distribution of previously assigned participants, assigning participants to the group that minimizes overall imbalance. Adaptive randomization using minimization will enable real-time assignment while ensuring balanced allocation of participants by key variables that may impact intervention outcomes.

For technology implementation, we will work with the Canadian company MetricWire and their Catalyst system. This company has developed a technology platform to develop, deploy, and conduct MRTs using an app on participants’ own devices. MetricWire is HIPAA compliant, thus minimizing participant risk for loss of privacy. Further, we have successfully used their system in our previous EMA studies [46-52].

Intervention messages came from several previous studies [33,53,54]. Initially, these intervention messages were refined internally and combined with image content from free stock photo websites (Pexels, Unsplash). See Figures 2 and 3 for intervention message examples. Before inclusion in the trial, a total of 124 intervention messages were rated by an online Qualtrics panel of 301 diverse young adults (18-30 years old) who endorsed current cigarette smoking. Each participant rated 10 randomly selected messages (3010 total message ratings; 24.3 ratings per message) on dimensions of content, design, helpfulness for in-the-moment urge reduction, and helpfulness for quitting smoking [55]. Of these 124 messages, the 34 messages with lowest average scores on dimensions of helpfulness for in-the-moment urge reduction and helpfulness for quitting smoking were dropped, for a total of 90 messages that were retained and used in this trial. All intervention messages are available on the project Open Science Framework page [56] and in Multimedia Appendix 1.

Frequent assessments may lead to participant noncompliance and study attrition. Procedures are built into the study to prevent attrition over time: The app tracks participant compliance and provides feedback to participants. We will monitor compliance over time; proactively reach out to participants via calls, text messages, and email to keep them engaged; and increase contact with participants if their compliance drops below the target of a 75% response rate. We will explore potential participant problems and remind them of the importance of compliance for data quality and of the extra incentive for high compliance. Compensation details are provided in the Ethical Considerations section of the manuscript.

The MRT will determine if CBT and mindfulness or ACT messages are superior to control messages for reducing the primary outcome momentary smoking urges. Based on participants’ training data collected in the initial 14 days of EMA monitoring, intervention messages will be delivered during time periods and at high-risk locations for smoking. In the intervention phase, participants will be prompted to complete up to 3 geofence-triggered EMAs per day for a total of 30 days. Each EMA will be followed by an intervention message, and the type of message (CBT, mindfulness or ACT, control) will be randomly selected at each time point (within-subject randomization; see Figure 4). Each intervention message will also be tailored to situational factors from the EMA-pre data. Message tailoring will focus on 2 key situational triggers stemming from our conceptual framework: (1) stress (high/low) and (2) presence of other smokers (yes/no). For each situation, characterized by a combination of these 4 possible characteristics, several messages will be placed in separate message bins (4 bins for each CBT and mindfulness or ACT message, for a total of 8 bins). Control messages will thank participants for completing an assessment. Proximate outcomes will be assessed 5 minutes to 15 minutes after message delivery and include urge levels, smoking or other tobacco product use since EMA-pre survey, affect, stress, and an evaluation of the last message (eg, perceived usefulness of message, completion of suggested activity or intervention). In addition, participants will continue completing 1 brief retrospective EMA each morning. Just like the training phase, this retrospective EMA will assess cigarette, other tobacco product, alcohol and cannabis use.

A total of 80 participants will be randomized into an EMA-only control group, parallel to the MRT intervention group (Figure 1). This group will conduct a 14-day EMA-only training phase just like the MRT group but will not be switched over to the intervention phase after these initial 14 days. Instead, participants will continue the EMA-only data collection procedure for an additional 30 days (analogous to the 30-day intervention phase of the MRT). During these 30 days, the EMA-only control group will continue to receive 3 randomly prompted EMA surveys per day and an additional 3 EMA surveys triggered by smoking reports. By comparing this control group with the MRT intervention group, we will investigate if intervention message delivery results in changes in smoking behavior over time (see the Analytical Strategy Aim 2: To Test if Urge Reduction Messages Change Smoking Behavior Over Time section).

The primary proximal outcome will be change in participants’ ratings of smoking urge in EMA-post surveys, prompted up to 15 minutes after intervention message delivery, and controlling for the ratings in EMA-pre surveys. Urge will be assessed using a single item on a 5-point scale, ranging from 1 (very low) to 5 (very high).

The primary distal outcome will be change in self-reported number of cigarettes smoked per day in the past week from baseline to the 45-day, 3-month, and 6-month follow-ups.

This study will also assess several secondary outcomes. Proximal secondary outcomes will be collected up to 15 minutes after message delivery and include smoking or other tobacco product use (including e-cigarettes), change in rating of stress and negative affect in EMA-post surveys (controlling for the ratings in EMA-pre surveys), ratings of message helpfulness, and self-reported completion of intervention message recommendation. Distal secondary outcomes will be collected at the 45-day, 3-month, and 6-month follow-ups and include self-reported 7-day point prevalence abstinence from smoking, saliva cotinine–confirmed 7-day point prevalence abstinence (assessed using saliva test strips mailed to participants and photo confirmation) [44,62-64], change (reduction) in cigarettes smoked by at least 50% between baseline and each follow-up time point, tobacco quit attempts, change in frequency and intensity of smoking urges as assessed using 2 items of the Mood and Physical Symptoms Scale [65], change in psychological flexibility as assessed using the CompACT-15 [60], change in smoking-specific experiential avoidance as assessed with the Avoidance and Inflexibility Scale [58,59], and switching to other tobacco products including e-cigarettes from baseline to follow-up.

A sample size of 80 participants receiving 3 randomized messages per day for 30 days will provide 7200 observations. Across person-days, there will be an average of 1 prompt per day for each of the 3 conditions: CBT, mindfulness or ACT, and control (balance across person-days). Power estimations were conducted using an online tool [66] specifically developed for sample size calculations for MRTs [37,67]. Assuming a message randomization probability of 1/3 (33% CBT, 33% mindfulness or ACT, 33% control), a quadratic effect over time (initially smaller as participants are getting used to the intervention, increasing over time to peak at day 20, and decreasing thereafter as participants may get desensitized) [68], and an average compliance rate of 75% linearly decreasing over time, a sample size of 80 participants will allow for detecting a standardized mean difference between CBT, mindfulness or ACT, and control conditions on the primary outcome of self-reported smoking urge of d=0.1 with a power of 0.8 and an α level of .05. This measure of effect size is a generalization of Cohen d (standardized mean difference), with the difference that the standardization is by the average standard error over the entire study, with multiple treatments for each person. A Cohen d of 0.1 is considered a small effect. A previous EMA study found that playing Tetris (analogous to a CBT distraction technique) decreased substance use urges with a mean effect size of f²=0.12 (medium-sized effect) [69]. Given these effect sizes in the existing literature, we will be adequately powered to detect effects in Aim 1 analyses.

This aim will test between-group differences in number of cigarettes smoked over time among MRT (n=80) and EMA-only control group (n=80) participants. Sample size calculations were based on findings from an app-based mindfulness meditation trial for smoking reduction. This trial only delivered intervention content for 14 days (as opposed to the 30 days in this study) and found a significant reduction in 3.8 cigarettes per day in the intervention group compared with an increase in 0.8 cigarettes per day in the control group [70], which translates to an effect size of d=0.651 (f=0.326) [71]. Power estimation for between-group repeated-measures ANOVA and interactions were conducted with G-Power. Assuming an α level of .05, a sample size of 160 will allow us to detect effects in reduction of cigarettes per day of f=0.326 and larger with a power of 0.80. Multiple imputation will be used for missing data, and the full sample will be analyzed.

Aim 1 analyses to test the hypothesis of a proximal benefit of urge reduction messages will be conducted using a centered and weighted least squares method [72], which estimates treatment effects and allows inclusion of covariates. The method is similar to generalized estimating equations (GEEs) [73] and multilevel models in that it accounts for dependence of responses within individuals due to repeated measures via the use of robust standard errors. Moreover, the method takes advantage of sequential randomization to estimate causal treatment effects. The primary outcome will be participants’ ratings of smoking urge in EMA-post surveys, prompted up to 15 minutes after intervention message delivery and controlling for the ratings in EMA-pre surveys. The main independent variable will be message type: CBT versus mindfulness or ACT versus a control message (“Thank you for completing the survey”). We will also test differences in efficacy between CBT versus mindfulness or ACT. We will use multiple imputation to impute missing data in EMA-post surveys.

Secondary outcomes from EMA-pre to EMA-post surveys to be investigated include smoking in the 15-minute time window since message delivery, other tobacco product use (including e-cigarettes), as well as participant ratings of negative affect, stress, perceived message usefulness, and activity completion. The centered and weighted least squares method with robust standard errors [72] will be used to adjust for multiple observations nested within participants. The number of cigarettes per day and other tobacco product use (including e-cigarettes) as reported in the daily diary EMA will be analyzed as additional secondary outcome to investigate participants’ changes in smoking and other tobacco use over time while enrolled in the MRT. As poly- and multiple tobacco product use is increasingly common among young adults, we will be able to investigate if participants are switching from cigarettes to other tobacco products (including e-cigarettes) over time.

Among MRT participants, we will explore changes in message efficacy on smoking urges over time by investigating both overall effects (across all time points) as well as linear and quadratic time trend of the intervention effect, assessed by estimating the interaction between treatment effect and day in the study with the weighted least squares method [72]. These models assess whether the relationship between a predictor and outcome changes over time. Separately for CBT and mindfulness or ACT messages, we will explore whether EMA-pre and EMA-post urge reduction changes as the intervention progresses. The same analyses will also be conducted for secondary outcomes. Analyses will provide evidence for increasing or diminishing returns of message delivery with time. Sex as a biological variable will be considered as a participant-level covariate in all models. We will also control for nicotine dependence, psychological distress, and trait negative affect.

Aim 2 analyses will use data from both the MRT group (n=80) and the EMA control group (n=80). The primary outcome will be the self-reported number of cigarettes per day in the past week at the 45-day, 3-month, and 6-month follow-ups. GEEs [73] will examine cigarettes per day at each follow-up by group (MRT vs EMA-only control). Independent variables are group membership, variables that differ by group at baseline, gender, and nicotine dependence. Our previous work has shown follow-up completion rates of up to 82% of participants at the 6-month follow-up in a Facebook smoking cessation trial with young adults [43]. Again, multiple imputation procedures will be used to impute missing data at follow-up.

To compare if delivery of CBT or mindfulness or ACT messages is associated with smoking outcomes over time, we will conduct dose-response post hoc analyses. A total of 90 message randomizations per participant will result in a range of empirical distributions of message type dose. We conducted a simulation of 90 randomizations over 80 participants in R, which resulted in a range of between 18 and 42 messages of CBT or ACT or mindfulness across participants for the intervention duration. Post hoc analyses will investigate if the extent of exposure to CBT or mindfulness or ACT messages predicts smoking outcomes. These analyses will control for baseline nicotine dependence, quit motivation, and total exposure to intervention messages.

For the secondary outcomes, we will run GEE models and mixed effects multinomial logistic regression analyses to analyze longitudinal ordinal response data to model the following outcomes across time (45 days, 3 months, 6 months): (1) self-reported 7-day point prevalence abstinence (yes/no—with sensitivity analyses for missing=smoking, complete cases, and multiple imputation), (2) saliva cotinine–confirmed 7-day point prevalence abstinence (yes/no—again with sensitivity analyses for different assumptions including self-reported nicotine replacement or other tobacco product use), (3) reduction of cigarettes by 50% or more (yes/no), (4) tobacco quit attempt (yes/no), (5) switching to other tobacco products including e-cigarettes, (6) frequency and intensity of smoking urges, and (7) ACT measures of experiential avoidance and mindfulness. Independent variables are group, gender, nicotine dependence, and any variables that differ by group at baseline.

Based on our data on high co-use of cigarettes with cannabis and alcohol [74-76], as well as changes in perceived reward of smoking cigarettes when under the influence of cannabis or alcohol [77-79], analyses in this aim will explore if message efficacy is moderated by substance co-use (whether a participant currently is under the influence of another substance [eg, cannabis, alcohol, other drugs]). Centered and weighted least squares models [72] similar to those for Aim 1 will be estimated with predictors of intervention message type (CBT, mindfulness or ACT, control), substance use (cannabis, alcohol, other drugs, none), and their interaction. Based on existing evidence on alcohol and cannabis co-use as barrier to cigarette smoking cessation, we hypothesize that intervention message efficacy will be reduced when participants are under the influence of another substance. However, analyses for this aim will enable us to explore if specific types of messages (CBT, mindfulness or ACT) are more helpful in co-use situations.

Moreover, we will explore aspects of intervention message-situation fit. We will test if specific locations (eg, home, work, bars) impact intervention message efficacy. Mixed models will be estimated containing the predictors of intervention type (CBT, mindfulness or ACT, control), location (home, work, bar, other locations), and their interaction. We will explore other situational characteristics and intervention message-situation fit based on affect, arousal, stress, and the presence of other smokers. We do not have specific hypotheses about moderation effects of location and other situational characteristics, but findings will inform future interventions using adaptive messages over time to improve intervention message-situation fit and intervention efficacy.