Smoking is a major risk factor for cancer and cardiovascular, respiratory, and many other chronic diseases worldwide [1]. In addition to the significant health burden it imposes, smoking also incurs massive economic costs. The United States alone lost a staggering US $864.5 billion in 2020 due to this issue [2], whereas limited-income countries, such as China, with higher smoking prevalence, face similar challenges [3]. Smoking cessation is crucial for minimizing mortality risk and improving quality of life [4]. Hence, finding effective ways to promote smoking cessation among smokers continues to be a vital public health goal. However, traditional smoking cessation services such as counseling can be expensive [5,6] and poorly received because of factors such as patients’ lack of time or reluctance to seek cessation services in clinical settings [7]. These challenges necessitate the development of cost-effective models for reducing tobacco consumption.

eHealth technologies, such as websites, mobile apps, and SMS text messages, have emerged as low-cost accessible interventions. Many of these technologies offer interactive experiences to users [8], which can enhance patient adherence to cessation services [9]. As such, they are ideal tools for revolutionizing health care [10] and promoting smoking cessation for diverse user groups, including ordinary daily smokers and pregnant women who wish to quit for their children’s well-being.

Numerous studies have investigated the effectiveness of eHealth cessation interventions over the past decade, but their findings have been inconsistent [11]. For example, a systematic review of 108 studies in 2018 found evidence suggesting that web-based and mobile health (mHealth) interventions could moderately increase abstinence rates, whereas computer-assisted interventions did not show the same effect [12]. Another 2019 systematic review of 26 studies indicated that automated text messaging interventions were more effective than minimal smoking cessation support, whereas the effectiveness of mobile apps on abstinence remains unclear [13]. However, these systematic reviews have some notable limitations. Recent reviews only evaluated abstinence as the outcome variable and did not include other outcomes such as cigarette consumption [14]. In addition, the reviews did not differentiate between self-reported versus biochemically verified abstinence. Another drawback is that previous reviews primarily synthesized evidence from high-income countries, which may not be generalizable to low- and middle-income countries [15]. Finally, the most recent review, which covered the entire eHealth intervention landscape, only included studies published until 2017 [12]. Since then, the use of eHealth has skyrocketed and many new studies have been published.

With the rapid development of eHealth technologies [16], smoking cessation interventions are constantly evolving and are dynamic in nature, encompassing both delivery channels and intervention materials. Therefore, this systematic review aimed to (1) summarize recent evidence (from 2017 to mid-2022) on the effectiveness of eHealth-based smoking cessation interventions, grouped by treatment characteristics, study population, and outcome verification, and (2) assess important nonabstinence outcome indicators such as cigarette consumption and user satisfaction via narrative synthesis.

A systematic search was performed across 4 electronic databases—PubMed (including MEDLINE), PsycINFO, Embase, and Cochrane Library—to gather studies published between 2017 and mid-2022. The search strategy was first created for PubMed using a combination of keywords and Medical Subject Heading terms. To make the search more precise, keywords were mapped to Medical Subject Heading terms where possible. We later applied the search strategy to other databases, namely PsycINFO, Embase, and the Cochrane Library, using their own thesaurus terms and advanced search features. The search terms were classified into four categories: (1) smoking cessation—the theme of the intervention; (2) device—the device used to carry out the intervention; (3) intervention channel—the specific approach used to engage the participants; and (4) randomized controlled trials (RCTs)—the study design. Each database was searched accordingly. The search strategy for PubMed is shown in Table 1. Multimedia Appendix 1 [18] documents search terms for all databases. The search results were limited to studies published in English from January 1, 2017, to June 30, 2022, given that previous review papers on eHealth smoking cessation interventions primarily collected studies published before 2017 [12].

Studies were excluded if they (1) included people using smokeless tobacco products or e-cigarettes, (2) only used eHealth technology during the recruitment of participants and not as part of the intervention, and (3) had a follow-up period shorter than 3 months.

Two independent reviewers (YEF and ZZ) screened titles and abstracts for potential inclusion. A relatively good interrater reliability was achieved (proportionate agreement=81%, Cohen κ=0.61). The same pair of reviewers also independently performed a full-text review after screening for final inclusion. Any conflicts between the 2 reviewers were discussed in the presence of a third author (selected from the author list, either RW or BY), who contributed to the final consensus. The study selection process was performed using the Covidence workflow platform.

Two reviewers (YEF and ZZ) independently performed data extraction using the same data extraction template with multiple categories for detailed information input on the Covidence platform [19]. Extracted data included the following: (1) study information (country of study, trial registration, funding sources, and declarations of interest); (2) study participants (inclusion or exclusion criteria, population characteristics, and sample size); (3) intervention and control details; (4) theoretical framework; (5) outcome measurements; and (6) key study results (abstinence rate, reduction in cigarette consumption, and satisfaction rate at ≥3 month follow-up). Any conflicts between the 2 authors were discussed between the reviewers or in the presence of a third author (either RW or BY) for final consensus.

Quality assessment was based on the GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) framework for quality assessment [17]. Two reviewers (YEF and ZZ) first conducted the risk of bias assessment independently under the guidance of the Cochrane Handbook for Systematic Reviews of Interventions [20] and the Cochrane Tobacco Addiction Group. Two reviewers assessed the risk of bias for each included study via five prespecified domains using version 2 of the Cochrane risk-of-bias tool for randomized trials (RoB 2) [21]: (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in measurement of the outcome, and (5) bias in selection of the reported result. After data extraction, each reviewer judged each domain as having low, high, or some concern. Disagreements were resolved between the 2 authors in the presence of a third author. The certainty of the evidence was rated as very low, low, moderate, or high based on the risks of bias, imprecision, inconsistency, indirectness, and publication bias.

The primary outcome of this systematic review was to evaluate the impact of eHealth-based smoking cessation interventions on the abstinence rate measured at ≥3-month follow-up via self-report or biochemical verification. Short- and long-term abstinence was defined as the abstinence result measured at the 3- and 6-month follow-ups, respectively. The measurements were 7-day point prevalence abstinence (PPA), 30-day PPA, or prolonged PPA. These measurements were used interchangeably in this review because there is evidence suggesting that such data handling does not significantly affect the results [22]. Only intention-to-treat analysis data were selected. All initially randomized participants were included, and any missing data caused by withdrawal were considered smokers based on the Cochrane Tobacco Group guidelines [23]. Nonabstinence outcomes were not subjected to meta-analysis because of the limited number of studies reporting the statistics and variation in outcome measurement standards. Subsequently, the reductions in cigarette consumption and satisfaction rates were narratively synthesized.

For the primary outcome of abstinence rate, dichotomous data on quit or smoking participants’ numbers in either the treatment or control groups at follow-ups were entered into Stata 17 software [24] to calculate the log risk ratio (RR). The included studies were stratified into different subgroups based on their study participants, eHealth interventions or controls, and outcome verification for comparable results in the meta-analysis. Where 2 or more studies were deemed comparable, we performed a meta-analysis to calculate the combined effects of the interventions on the abstinence rate. For studies that were not included in the meta-analysis, we summarized the abstinence outcomes for each study. Considering the potential treatment effect heterogeneity, differences in trial size, and the limited number of included studies, this study used the Sidik-Jonkman random-effects model method to pool log RRs and 95% CIs calculated for the abstinence outcome [25]. Heterogeneity was assessed using I² statistic, given its robustness with small sample sizes. Publication bias was assessed using funnel plots.

After removing duplicates, we identified a total of 464 studies in the initial database search. After screening and full-text assessments, 39 studies were deemed eligible for this review, of which 28 were included in the meta-analysis. Figure 1 shows the PRISMA flowchart, which illustrates the process of study selection and rationales for exclusion during full-text assessments.

All included studies underwent a risk of bias assessment based on the guidelines suggested by the Cochrane Handbook for Systematic Reviews of Interventions [20]. Bias in measurement of the outcome was rated as low risk because the Cochrane Tobacco Addiction Group stated that blinding participants made cessation interventions impossible [13]. In summary, we evaluated 26 (26/39, 67%) studies to be at low risk of bias (considered low risk of bias for all domains), 8 (8/39, 21%) studies with some concerns (considered with some concerns for at least 1 domain, but with no judgments of high risk), and 5 (5/39, 13%) studies at high risk (considered high risk of bias in at least 1 domain). The risk of bias per domain is shown in Figure 2. Incomplete outcome data was the primary cause of high risk of bias (4/39, 10%). It is worth noting that 33% (13/39) studies reported a high attrition rate (>20%), but 9 were deemed to have a low risk of bias or some concerns because there was no evidence for differential missing data.

Funnel plots were generated for each pooled result and are presented in Multimedia Appendix 2. Because of the limited number of studies in each group, assessing publication bias is challenging. Therefore, we decided to focus on the subgroup with the largest number of studies (n=13), which included long-term abstinence. Figure 4 displays the funnel plot assessing publication bias among the studies that measured long-term abstinence in adult smokers. Visual inspection of the plot revealed a relatively symmetrical distribution of the included studies, indicating that our study was unlikely to be affected by publication bias.

For studies not included in the meta-analysis (n=11), we have summarized the treatment effect on abstinence outcomes in Table 5 because of their distinct features in the study population or interventions. Overall, 2 studies that used multiplatform eHealth intervention or mHealth apps accompanied by SMS text messaging reported a significant increase in abstinence at 3- and 6-month follow-ups [27,28]. A significant increase in abstinence at 6 months was reported in 2 studies using mHealth counseling and SMS messaging accompanied by pharmacotherapy, respectively, as interventions [29,30]. In addition, one study reported a text referral program to refer smokers to cessation services, which may improve cessation outcomes [31]. Finally, acceptance and commitment therapy were suggested to be more effective than US clinical practice guidelines in the context of the mHealth app group [32]. The remaining 5 studies reported no significant differences between intervention and control on cessation outcomes [33-37].

For nonabstinence outcomes, a total of 20 studies reported a reduction in cigarette consumption and user satisfaction (Multimedia Appendix 4 [28,32-34,37-50,52,53,58,60,62,64]). Among the 14 studies that reported cigarette consumption outcomes, only 3 suggested that the intervention can reduce cigarette consumption significantly compared with controls [38-40]. Finally, 19 studies assessed user satisfaction after the intervention, and 18 reported good user satisfaction, whereas the remaining 1 study specifically investigated user adherence to the program [38]. Among the studies that reported high user satisfaction, 8 compared user satisfaction between the intervention and control groups and found significantly higher satisfaction in the intervention arm [28,32,34,41-45].

This systematic review included 39 RCTs of eHealth smoking cessation interventions, published between 2017 and 2022 [27-65]. Most of the interventions were classified as mHealth interventions involving mobile SMS or app text messaging and mobile apps. This result suggests that cessation intervention delivery channels have shifted away from internet-based interventions [67] or telephone counseling [15], which were prevalent more than 5 years ago. After pooling the 28 included studies, we found mixed results across studies using different subcategories of eHealth interventions and personalization or interactive status. In addition, the intervention effect on abstinence varied among the study populations. Finally, the meta-analysis indicated that studies using biochemical verification yielded results similar compared with studies only used self-reported abstinence. Among the studies not included in the meta-analysis, approximately half (6/11, 55%) reported a statistically significant positive effect on increasing abstinence. In addition to abstinence, a small number of studies (14/39, 36%) evaluated the effects of eHealth interventions on reduced cigarette consumption. Although almost all studies assessed user satisfaction and revealed a high degree of satisfaction postintervention, less than half of them found significant differences between the intervention and control groups.

This study examined 3 types of mHealth interventions: SMS or app text messaging, stand-alone mHealth apps, and mHealth apps used alongside psycho or pharmacological therapy. These interventions have produced different treatment effects on abstinence. Our findings support a previous review [13] that suggests that SMS or app text messaging is more effective than minimal cessation support in promoting abstinence. However, our study also revealed that increasing the frequency of texting may not have a positive impact on abstinence and may even discourage adherence to the intervention [46]. In addition, a previous review found no evidence that smartphone apps can improve the likelihood of smoking cessation and called for further research in this domain [13]. In contrast, in our study, we found that more recent RCTs testing smartphone apps found an increased chance of abstinence among adult smokers in the short term. Compared with the existing knowledge, we believe this change may be due to the improvements in the overall quality of cessation apps that allow more personalized designs, which subsequently increases acceptability among smokers [68]. Finally, our study found that the use of mHealth apps in conjunction with psycho or pharmacological therapy produces abstinence results similar to those of therapy alone. However, the certainty of the evidence is low, indicating the need for further research in this area.

Our review found a high attrition rate and poor long-term treatment effects among identified mHealth smoking cessation studies, which aligns with previous reviews [12,13]. Previous interventions summarized by Belita and Sidani [69] in their systematic review have also shown high attrition rates ranging from 30% to 50% preinclusion and 10% to 50% postinclusion, with a pooled rate of 10.8% to 77%. Such a high attrition rate can easily render many potential therapies ineffective. Although mHealth technologies can adequately address some logistical factors, such as travel, they are neither important nor significantly associated with attrition rates. Therefore, priority should be given to identifying and comprehending the factors that significantly influence the attrition rate. For example, user satisfaction is an important measure of potential adherence from the perspective of eHealth developers [70]. However, we found that most of the included studies reported high satisfaction rates but still had relatively poor user adherence, as evidenced by the high attrition bias rate (13/39, 33%). This finding suggests that the assessment of user satisfaction alone may not be a reliable factor for predicting adherence, at least in terms of smoking cessation interventions. Researchers and mHealth app developers should consider narrowing the intervention scope based on demographic factors at the design stage and improving personalization based on clinical, behavioral, and health belief factors at the development stage [69]. In fact, the eHealth cessation interventions that managed to achieve the most significant increase in abstinence in our meta-analysis targeted hospitalized smokers (log RR=1.00, 95% CI 0.22-1.78; I²=3.45%) [40,47]. The success may be due to the good program adherence evidenced by the low attrition rate, thanks to the institutional environment that encourages prohealth behaviors, as well as the intervention material dedicated to this specific population.

To explore the association between personalization status and treatment effect, we pooled studies targeting general smokers according to their personalization or interactive level and found improved abstinence results. Previous research has also suggested that such content can improve medication adherence [71] and eHealth application retention [72], thereby enhancing its effectiveness. It is not surprising that interventions that included some level of personalized or interactive content achieved a significant increase in abstinence rates both in the short and long term. By contrast, studies that lacked any personalized or interactive features showed null effects after pooling. However, caution should be exercised when interpreting these findings because of high within-group heterogeneity.

Finally, we found that studies using either biochemical verification or self-reporting for measuring abstinence showed similar treatment effects. Previous research has produced mixed results on whether self-reported abstinence is a reliable indicator of biochemically verified abstinence. Although some studies have suggested that self-reported quitting is mostly accurate [73], others have found a high proportion of self-reported quitters failing biochemical verification in clinical settings [74]. However, our review brings a new perspective to the debate, suggesting that studies using biochemical verification do not necessarily outperform those using self-reports. We found that the effect sizes of studies using both methods were consistently similar in both the short and long term (Table 4). Although biochemical verification is encouraged by the Society for Research on Nicotine and Tobacco for its scientific rigor, it can be expensive [75]. Our findings support the feasibility of eHealth-based cessation programs, which are scalable to large-scale interventions where biochemical verification is not possible. Nonetheless, given the possibility of false reporting, trials evaluating potential population-level interventions may need to be considered using biochemical verification of smoking populations that are most susceptible to false reports. For instance, a study has recommended using biochemical assessment, preferably with cotinine plasma, in intervention studies and with student populations [76]. Second, there are a variety of biochemical verification methods that target different biomarkers. This diversity has prompted an update of the 2002 Society for Research on Nicotine and Tobacco reports on whether and how to apply biomarker verification to tobacco use and abstinence [75]. Given this complexity, researchers in relevant fields should focus on standardizing currently accepted biochemical verification methods and their cut-off points to improve interstudy compatibility, rather than seeking the most accurate method.

This study provides a comprehensive and updated evaluation of the potential role of eHealth interventions in facilitating smoking cessation. Its robustness lies in the inclusion of recent well-funded studies that demonstrate the advancement of digital technology and its accessibility in both high- and lower-income nations (Multimedia Appendix 5 [27-65]). The present review included multiple outcome assessment criteria and treated populations to provide a more holistic evaluation. However, we acknowledge that this study has some limitations. First, our focus was limited to studies published in English in the last 5 years, which means that we may have overlooked relevant research conducted in other languages or low- to middle-income countries. Second, due to the significant heterogeneity of methodological design and outcome verification among studies, not all were suitable for inclusion in the meta-analysis. This limitation led to fewer studies being synthesized, which undermined the certainty of the evidence. In addition, the small sample sizes in some studies resulted in relatively large CIs for effect size estimation, making it difficult to determine a significant effect. Third, despite our attempt to synthesize the intervention effect on special populations and nonabstinence outcomes, such as cigarette consumption, the heterogeneity in outcome measurement among the collected evidence prevented us from drawing any conclusion.

Future studies could standardize the intervention evaluation strategy by following the World Health Organization Practical Guide to Monitoring and Evaluating Digital Health Interventions [77] for better comparability between trials. In addition, mobile app development should adopt a human-centered design approach and prioritize improving participant adherence and engagement [78,79] to reduce attrition and achieve better long-term cessation outcomes (≥6 months). Furthermore, research is necessary to understand the effectiveness of eHealth interventions on susceptible populations and intermediate outcomes such as reduction in cigarette consumption.

The use of eHealth technologies for smoking cessation has gained momentum in recent years. Our review highlighted the timeliness of eHealth interventions, particularly mHealth, in promoting abstinence, although their effectiveness may wane over time. Future studies could benefit from adopting a learning by doing approach and incorporating the concept of human-centered design to develop personalized intervention designs that address individual smoker needs and reduce attrition, ultimately leading to better long-term abstinence outcomes. In addition, owing to the dynamic nature of eHealth interventions, monitoring and evaluation can be challenging. Standardized evaluation strategies should be implemented to improve interstudy comparability.