Because we adapted an existing theoretically based AT intervention from our previous works in the United States [32-34] and in Cambodia [35,36], we used the cultural adaptation of evidence-based interventions approach [37,38]. Although several models have been proposed to guide cultural adaptation, there is considerable consensus that cultural tailoring can be organized into 5 stages: information gathering, preliminary design, preliminary testing, refinement, and final trial [38]. Figure 1 summarizes our R21 research activities.

We integrated information gathered in the first stage to modify the original intervention. For example, using literature search and qualitative findings, we expanded our treatment content library to include strategies for coping with craving or urges that are commonly used in Southeast Asian culture (eg, chewing on ginger or lemon, chatting with friends via apps, or playing e-games on phones to keep hands busy). The original English treatment content library and the NTCC’s available materials are mainly text based with some images.

We translated all English intervention materials and assessments for use in the R33 phase into Lao, mirroring the WHO’s recommended methodology [41,42]. First, a health scientist, whose native language is Lao, led the forward translation process, focusing on conceptual (vs literal) meaning and comprehensible language for the broadest audience. Second, a more senior bilingual health expert (the Lao Site PI or a Lao co-investigator) reviewed the translation, discussed disagreements with the forward translator, edited, and finalized the forward translation. Third, 2 other bilingual health professionals independently rated the quality of the translation on 5 dimensions: conceptual equivalence, clarity in meaning, comprehensibility, use of common simple language, and cultural appropriateness. Any discrepancies were discussed to reach consensus on translation or editing. Fourth, we conducted 6 panel discussions (1 with female smokers, 1 with male smokers, and 1 with mixed-sex health care professionals at each participating hospital; 5‐7 panelists per group). The aim of this step was to evaluate material comprehensibility as well as linguistic and cultural appropriateness for the target populations, particularly to identify linguistic differences by regions (north vs south) in Lao PDR and to find and use common words. Fifth, 2 other bilingual health professionals independently backward translated the materials into English. Finally, the PI and Lao Site-PI independently reviewed and rated the backward translations for conceptual equivalence with the original English versions.

Participants (n=500) are patients who smoke recruited through SH in Vientiane and CH in Champasak province in Lao PDR. Consenting participants are randomized to one of the 2 study groups: AT (n=250) or SC (n=250). SC consists of brief advice to quit smoking delivered by research staff, self-help written materials, and a 2-week supply of nicotine replacement therapy (NRT) in the form of transdermal patches. AT consists of all SC components plus the fully automated smartphone–based treatment program that involves interactive and tailored proactive messages and images for 6 months. Participants complete in-clinic assessments at baseline and at 3, 6, and 12 months following study enrollment. Participants also complete brief weekly assessments via smartphone during the AT period so that the intervention content can be tailored to their PBM phases and constructs. Participants are blinded to their group assignment. The primary health outcome of the RCT is biochemically confirmed self-reported 7-day point prevalence abstinence 12 months post study enrollment. Secondary outcomes include abstinence at 3 and 6 months post enrollment.

Several aspects of the project, as described below, were carefully considered prior to deciding on the proposed design.

All nonemergency patients coming to SH and CH first go to a reception desk to receive a queue number and a basic medical form. A flyer that introduces this study is attached to the basic medical form. Additionally, we proactively recruit patients at the respiratory disease screening units (RDSUs) and departments or clinics specializing in women’s health (eg, gynecology, and breast and gynecologic cancer). Due to the national organization and implementation of disease control programs, all patients visiting SH or CH with respiratory symptoms are first examined at the RDSUs for type of disease, severity, and whether the disease is included in a national disease control program (eg, tuberculosis or severe acute respiratory syndrome). Patients are then referred to appropriate departments and clinics or disease control units. The RDSUs at SH and CH see about 40 and 20 patients daily, respectively. Of these, 70% are men; 60% to 70% of male patients smoke cigarettes (no estimate available for women; Setthathirath Hospital, unpublished data, 2018 and Champasak Hospital, unpublished data, 2018). Research staff proactively screen as many patients visiting the RDSU as possible for eligibility and invite eligible individuals to participate. Research staff also proactively screen as many women visiting the women’s health clinics as possible to enroll female smokers.

In our Cambodian pilot study at a clinic with fewer patients, 86% (51/59) of the identified eligible smokers consented to participate, and we recruited 2‐5 participants per day; we expect a similarly high consent rate for this study. Together, SH and CH see >160,000 unique patients each year. Based on the smoking prevalence in the general population (51% in men and 7% in women), we estimate that approximately 40,000 male patients and 5600 female patients at SH and CH are current smokers. Assuming, conservatively, that only half of these smokers are eligible for the study, and even if only half of eligible smokers want to participate, there will be 10,000 available male and 1400 available female smokers in a 1-year recruitment period. Thus, our goal to enroll 500 participants (<5% of eligible smokers) at SH and CH is highly feasible.

Trial inclusion criteria include patients who are (1) aged ≥18 years, (2) self-reporting current combustible cigarette smokers (smoked at least 100 cigarettes in lifetime and currently smoke ≥1 cigarette per day), (3) willing to set a quit date within 2 weeks of study enrollment, (4) able to provide written informed consent to participate, and (5) able to read Lao (score ≥4 points on the Rapid Estimate of Adult Literacy in Medicine–Short Form [45]).

Exclusion criteria include (1) history of a medical condition that precludes the use of NRT, (2) ineligibility to participate based on medical or psychiatric conditions diagnosed by a physician or clinician, and (3) enrollment in another cessation program or current use of other cessation medications.

Participants randomized to SC receive brief advice to quit smoking delivered by research staff, NTCC’s self-help materials (developed based on the WHO’s “A guide for tobacco users to quit” [48]), and a 2-week supply of NRT (patches). SC participants are asked to complete weekly 4-item smartphone-delivered assessments about their diet (see explanation in the Brief Weekly Smartphone Assessments section) for a 6-month period.

Participants in the AT group receive the SC components (except the dietary assessments) plus proactive personalized messages for smoking cessation. The AT content is adapted from the team’s previous efforts, is informed by the R21 phase outcomes, and is designed to tap the theoretical mechanisms described in the PBM. That is, as we described in a prior study [35], AT content is designed to increase motivation, self-efficacy, and use of coping skills, while reducing nicotine withdrawal symptoms and stress. AT begins immediately after enrollment and continues for a 26-week period (about 2 messages per day). The AT approach allows for several levels of personalization for each participant. First, at baseline, participants are asked several questions about their biological sex, past quit attempts, and the presence or fear of specific health conditions. Treatment content tailored to these responses is automatically delivered throughout the treatment period. Second, there are different bins of treatment content for different cessation phases to ensure that AT targets critical mechanisms of each participant’s PBM phase, which may dynamically change during treatment. Third, participants are asked to complete brief (4 items) smartphone-delivered assessments via the Insight app during each week of the AT course. These questions vary depending on each participant’s phase (eg, current level of intrinsic motivation for preparation or maintenance phases or smoking status in the past week and self-efficacy level for all phases). Treatment content (eg, types and frequencies of messages) for the following week is based on responses to these weekly assessments and participant phases (eg, level of self-efficacy or past-week smoking status).

Our Insight platform [49], which includes a web-based central management system and an app, is used to manage and deliver the AT and weekly assessments. The Insight app will automatically deliver all interventions and assessments as prescheduled, with or without an active cellular network connection, thus ensuring timely and reliable AT delivery throughout the 6-month treatment period. Through 1-click buttons in the Insight app, participants can call or message project staff if they have questions or need assistance. Insight allows our AT to function autonomously and minimizes human involvement, making the approach very affordable for large-scale implementation in LMICs. Most importantly, Insight enables complex built-in algorithms and branching logic, allowing us to create and deliver dynamically and individually tailored treatment content as described above. Our pilot results in Cambodia demonstrate that Insight works very well in Southeast Asia; it is compatible with the GSM and 4G data networks, properly delivers messages and weekly assessments, and reliably collects and transfers data to our encrypted server. Our preliminary work also shows that the Insight platform and app work well with the Lao script (Figure 2).

Three types of data are collected at each visit: self-report data (Table 1) collected by the Health Insurance Portability and Accountability Act–compliant REDCap program on a tablet, expired air carbon monoxide (CO) assessed with a CO monitor (Vitalograph BreathCO) by research staff to verify smoking status biochemically, and medical record data (to collect clinical information, such as current diagnosed health conditions for AT personalization). The REDCap assessment is administered at baseline (~45 minutes to complete) and at each follow-up in-clinic visit (~20 min). For measures that have not been validated in Lao, we adapted them as described in the R21 phase.

Participants are asked to complete weekly assessments via smartphone for 6 months, delivered by the Insight app. Participants in the AT group receive 4‐7 smoking-related questions, depending on their cessation phase (Table 1). Although these treatment-driving questions can be considered part of the AT, we attempt to balance the effects of these weekly contacts between the treatment groups; SC participants are also asked to complete a weekly 4-item assessment via the Insight app with questions about diet. This consistency across participants also minimizes the chances of inadvertent revealing of participants’ assigned interventions. Responses are temporarily stored on smartphones and sync to our secure server whenever a connection is active; thus, we have near real-time access to the data. To carefully track participants’ completion of weekly assessments and adherence, data plans are provided for 6 months for both groups.

The primary outcome is smoking status at 12 months post enrollment. Abstinence is defined as biochemically confirmed self-reported 7-day point prevalence abstinence with expired CO <6 ppm [57]. Secondary outcomes include 3- and 6-month abstinence. We will consider several other common outcomes, such as continuous and sustained abstinence, number of quit attempts, and length of abstinence.

The study’s hypothesis is, at the 12-month follow-up, 7-day point prevalence biochemically confirmed abstinence (primary outcome) will be higher in the AT (vs SC) group.

The primary abstinence analysis will be intention-to-treat 7-day point prevalence biochemically confirmed abstinence; patients not completing follow-up assessments will be considered smokers. However, we will explore other ways of dealing with missing data as described in the Missing Data and Dropouts section. To estimate the effect of AT on intention-to-treat 7-day point prevalence biochemically confirmed abstinence rates while accounting for the potential clustering of participants recruited from the 2 clinic sites, we will use generalized linear mixed models (GLMM) analysis as reported in our previous studies [35]. Intervention groups (AT vs SC) will be estimated as a fixed effect, while the clinic will be modeled as a random effect nested within treatment condition. Sex can also be modeled as a random effect nested within clinic and treatment condition. Specifically, unadjusted and adjusted log binomial mixed models will be used to estimate the relative risk and 95% CI of the primary outcome of intention-to-treat biochemically confirmed abstinence in the AT (vs SC) group. Although the groups should be similar in baseline characteristics due to randomization, we will explore models that control for any demographic or clinical variables that differ between treatment groups at baseline (eg, sex, nicotine dependence, and reading level). We will also use GLMMs to examine changes in abstinence rates over time, while accounting for relevant baseline covariates. Similar GLMM or linear mixed model methodology, as appropriate for each outcome variable, will be used to examine other smoking-related variables, such as continuous abstinence, prolonged abstinence, and quit attempts. This project focuses on the latter 3 PBM phases, and thus, we will explore phase-specific outcomes, such as abstinence attainment and the number of days smoking or abstinent.

Based on our previous work and assuming attenuation of abstinence rates across the follow-up visits, we anticipate that the 7-day point prevalence abstinence at the 12-month follow-up will be 8% in the SC group and 17% in the AT group. With 250 participants in each group and assuming a 2-sided α of .05, we will have 86% power to detect a difference of 9% in 12-month 7-day point prevalence abstinence in the overall sample. It is unknown whether factors associated with smoking cessation differ by sex; it seems likely given the variation in smoking prevalence between sexes. Thus, we chose a sample size that gives 80% power to detect this difference in a subgroup analysis of only males (n=425, 85% of the sample). The subsample of female participants (n=75) gives 80% power to detect a 25% difference in 12-month 7-day point prevalence abstinence.

Potential treatment mechanisms will be examined via mediation analyses, with the intervention group (AT vs SC) being the independent variable, abstinence at 12 months being the outcome variable, and the hypothesized mechanisms (motivation, self-efficacy, stress, and negative affect) being potential mediators. The PROCESS macro for SPSS or SAS [58,59] or an equivalent method will be used to identify variables (eg, motivation, self-efficacy) that mediate the relationship between treatment condition and smoking cessation outcomes. This method uses an ordinary least squares path analytic framework to estimate direct and indirect effects in single and multiple mediation models, and bootstrapping methods are incorporated to generate CIs.

Treating participants lost to follow-up as smoking is a widely used strategy in smoking cessation studies. However, some researchers point to problems with this approach, especially when comparing treatment arms with differential dropout rates [60]. We will conduct sensitivity analyses to test for treatment differences assuming different missing data mechanisms. For example, we will consider a multiple imputation approach based on smoking-related patient characteristics at baseline, as well as demographics to account for potential missing-at-random mechanisms. We will also explore pattern-mixture and selection models to account for potential (and likely) missing-not-at-random mechanisms [61]. Similar findings based on these analyses will strengthen our study conclusions.