Efficacy of digital interventions for smoking cessation by type and method: a systematic review and network meta-analysis

Abstract

Smoking cessation is the only evidence-based approach to reducing tobacco-related health risks, yet traditional interventions suffer from limited coverage. Although digital interventions show promise, their comparative efficacy across methodological frameworks and technology types remains unclear. Here we assessed digital interventions versus standard care via frequentist random-effects network meta-analysis of 152 randomized controlled trials (48.8% USA, 7.5% China). Interventions were categorized by methodology and technology type, with cross-matched subgroup analyses. Results showed that personalized interventions significantly improved smoking cessation rates compared with standard care (relative risk (RR) 1.86, 95% confidence interval (CI) 1.54–2.24), while group-customized interventions were more effective (RR 1.93, 95% CI 1.30–2.86) compared with standard digital interventions (RR 1.50, 95% CI 1.31–1.72). Among the various technology types, text message-based interventions were the most effective (RR 1.63, 95% CI 1.38–1.92). Intervention effectiveness was also influenced by age, with middle-aged individuals benefitting more than younger individuals. Short- and medium-term interventions were more effective than long-term interventions. Sensitivity analyses further confirmed these low-to-moderate findings. However, this study has some limitations, including methodological heterogeneity, potential bias and inconsistent definitions of numerical interventions. In addition, long-term follow-up data remain limited. Future studies require large-scale trials to assess long-term sustainability and population-specific responses, as well as standardization of methods and integration of data at the individual level.

A network meta-analysis of 152 randomized controlled trials finds that personalized and group-customized digital interventions, especially SMS and app-based tools, improve smoking cessation versus standard care, with greater benefits among middle-aged adults in short- to medium-term programmes.

Main

According to the World Health Organization, approximately 1.245 billion individuals worldwide were smokers in 2022, with a global smoking prevalence rate of 20.9% (ref. ¹). Smoking is a major risk factor for 40% of incident cancers and 30% of global cancer mortality, while also contributing to the pathogenesis of cardiovascular and respiratory diseases^2,3.

Although smoking cessation reduces tobacco-related morbidity, only 3–5% of smokers sustain abstinence for 6–12 months without intervention⁴. Consequently, various effective cessation strategies have been developed and validated, including pharmacotherapy, behavioural counselling and psychotherapy^5–7. However, these interventions face implementation barriers including resource intensiveness and limited accessibility in low-resource settings⁸. Furthermore, smoking prevalence exhibits an inverse socio-economic gradient, with individuals of lower education and income levels experiencing 60% higher smoking-attributable mortality risk compared with their high socio-economic status counterparts⁹. This population frequently presents with heightened nicotine dependence, diminished quit motivation and reduced self-efficacy¹⁰. Meanwhile, low- and middle-income countries report 35% fewer successful quit attempts than high-income nations, largely attributable to insufficient public health infrastructure and cost-prohibitive evidence-based interventions^11–13. Thus, addressing these compounded barriers requires targeted cessation strategies.

The World Health Organization’s Global Digital Health Strategy 2020–2025 defined digital health as “the cost-effective and secure application of information and communication technologies to support health services”, including mobile health and telemedicine¹⁴. Contemporary evidence indicates that, although over 30 systematic reviews and 300 randomized controlled trials confirm superior efficacy of offline standard care strategies compared with spontaneous quitting, the evidence for digital cessation interventions, particularly with emerging technologies, remains inconclusive^15,16. Our prior research has provided preliminary evidence supporting the efficacy of digital interventions in smoking cessation management. However, there has been little consensus on the efficacy of different types of digital intervention. A network meta-analysis (NMA) published in 2012 reported only a marginal increase in smoking cessation rates with digital interventions, with no significant outcome variation across technology types¹⁷. Since then, standardized evaluation strategies for smoking cessation interventions have yet to be updated. Over the past decade, notable technological and application advancements have been made in digital health interventions^18,19. The global expansion of mobile communication, widespread internet access, emergence of adaptive digital platforms and maturation of personalized intervention algorithms have enabled technological advancements that enable more tailored and scalable smoking cessation solutions^20,21.

Given the exponential technological advancements in the domain of digital health, there is an urgent imperative to critically reassess the clinical effectiveness of smoking cessation interventions. Therefore, we conducted a NMA to capture these recent advances. Specifically, the investigation aims to quantify the impact of diverse digital modalities on abstinence outcomes, elucidate differential efficacies across device categories and generate methodological insights to substantively inform and advance evidence-based cessation frameworks.

Results

Study selection and characteristics

The systematic search included 3,927 candidate articles, with 270 publications advancing to full-text scrutiny after preliminary title/abstract screening. The final analysis comprised 152 RCTs, including 94 trials (n = 63,955) reporting 7-day point prevalence abstinence and 58 trials (n = 66,402) documenting prolonged abstinence outcomes. The sample size totalled 117,642 individuals, characterized by a median age of 40.5 years interquartile range (IQR) 35.7–45.7 and male predominance (51.4%). Baseline cigarette consumption averaged 16.7 units daily (IQR 14.2–19.3). Study timelines spanned from 2005 to the present, with multinational representation; 48.8% were conducted in the USA and 7.5% in China, reflecting a recent acceleration in research activity. Industry sponsorship was identified in seven trials (Fig. 1 and Table 1). Full characteristics of included studies are presented in Supplementary Table 1.

Fig. 1. Study selection.

PRISMA flow diagram outlining the selection process for studies included in the present meta-analysis.

Table 1.

Definition of digital intervention in smoking management

Methodological approach	Definition
Group customization	Interventions customized to the characteristics of specific groups, such as smoking cessation programmes designed for particular social groups.
Personalized customization	Interventions that collect individual information in advance to provide personalized smoking cessation messages, videos, tasks or discussions based on personal characteristics.
Interactive	Interventions that allow participants to engage in real-time, bidirectional communication with the intervention system, such as interactive text messages or telephone counselling.
Standard	General interventions that are non-customized and non-interactive, do not collect personal information and involve one-way communication of information.
Standard care	Traditional care methods, including print-based self-help guides, national helplines with website links (without guaranteeing live conversation) or extremely minimal guidance.
Placebo	Control measures that do not contain any active intervention components.

Technology type	Definition
Website	Smoking cessation information, videos, games, task check-ins or group discussions are provided through a website.
Application	App-based smoking cessation information, videos, games, task check-ins or group discussions.
SMS	Smoking cessation information is delivered through scheduled text messages.
Email	Smoking cessation information is delivered through scheduled emails.
Telephone	Smoking cessation interventions are provided via telephone calls or teleconferences.
Multicomponent intervention	Comprehensive smoking cessation interventions that combine two or more of the above methods.
Face-to-face	In-person counselling with a healthcare professional, offline workshops or offline care.
Standard care	Includes print-based self-help guides, national helplines with website links (without guaranteeing live conversation) or extremely minimal guidance.
Placebo	Control measures that do not contain any active intervention components.

Interventions were categorized by methodological approaches, technology types and cross-matched groups.

Primary outcome—point prevalence abstinence rate

Figures 2–4 show the comparative efficacy of digital interventions versus standard care on point prevalence abstinence. As the internal comparisons between other interventions lacked statistical significance, we focused primarily on comparisons with standard care. The league table results are provided in the Supplementary Information.

Fig. 3. Comparative efficacy of digital interventions by technology type.

a, The network structure of included studies. Circles represent interventions, with area proportional to sample size or statistical weight. Lines indicate direct comparative evidence, with thickness proportional to the number of trials. b, Comparison of the relative efficacy of each intervention against standard care. The purple diamond denotes the pooled RR point estimate. The horizontal purple bar represents the 95% CI. A total of 81 independent RCTs were included (N = 52,755).

Fig. 2. Comparative efficacy of digital interventions by methodological approach grouping.

a, The network structure of included studies. Circles represent interventions, with area proportional to sample size or statistical weight. Lines indicate direct comparative evidence, with thickness proportional to the number of trials. b, Comparison of the relative efficacy of each intervention against standard care. The purple diamond denotes the pooled RR point estimate. The horizontal purple bar represents the 95% CI. A total of 90 independent RCTs were included (N = 55,094).

Fig. 4. Comparative efficacy of digital interventions by cross-matched group.

a, The network structure of included studies. Circles represent interventions, with area proportional to sample size or statistical weight. Lines indicate direct comparative evidence, with thickness proportional to the number of trials. b, Comparison of the relative efficacy of each intervention against standard care. The purple diamond denotes the pooled RR point estimate. The horizontal purple bar represents the 95% CI. A total of 94 independent RCTs were included (N = 63,134).

Methodological approach grouping

This category comprised 90 RCTs, involving 55,094 participants, graded as low-certainty evidence. Group-customized interventions exhibited superior cessation efficacy relative to standard care (relative risk (RR) 1.93, 95% confidence interval (CI) 1.30–2.86), corresponding to a 93% increased likelihood (low-quality evidence). Personalized interventions significantly elevated quit rates versus standard care (RR 1.86, 95% CI 1.54–2.24). Interactive modalities demonstrated RR of 1.50 (95% CI 1.27–1.78; low-quality evidence), paralleling standardized digital protocols (RR 1.50, 95% CI 1.31–1.72; very low-quality evidence), while placebo showed non-significant effects compared with standard care (RR 0.87, 95% CI 0.63–1.20).

Technology type grouping

The analysis included 81 RCTs involving 52,755 participants, with overall moderate-quality evidence. SMS interventions demonstrated the highest efficacy (RR 1.63, 95% CI 1.38–1.92), indicating a 63% increased smoking cessation success compared with standard care (low-quality evidence). Telephone interventions followed closely (RR 1.59, 95% CI 1.33–1.90), improving quit rates by 59% (low-quality evidence). Multicomponent and app-based interventions also significantly improved cessation rates, with RRs of 1.56 (95% CI 1.32–1.84; low-quality evidence) and 1.53 (95% CI 1.29–1.81; moderate-quality evidence), respectively. Email (RR 1.33, 95% CI 1.07–1.65) and web-based interventions (RR 1.30, 95% CI 1.13–1.49) had comparatively smaller effects but still increased quit rates by 33% and 30%, respectively (low-quality evidence). Face-to-face interventions showed no statistically significant benefit (RR 1.13, 95% CI 0.89–1.43; very low-quality evidence). Placebo groups also had lower quit rates than standard care (RR 0.88, 95% CI 0.67–1.15).

Cross-matched group analysis

Building on the previous analysis, we conducted a cross-matched group analysis involving 94 RCTs (63,134 participants). Due to the limited studies on group-customized interventions, single-study nodes weakened network robustness. To enhance statistical power, we merged similar group-customized digital interventions into one composite category. Unmerged results, presented in the Supplementary Information, did not significantly differ.

Personalized apps showed a 77% higher quit rate compared with standard care (RR 1.77, 95% CI 1.38–2.28; low-quality evidence), while personalized websites had an RR of 1.71 (95% CI 1.28–2.29; very low-quality evidence). Standard SMS interventions increased quit success by 72% (RR 1.72, 95% CI 1.38–2.15; very low-quality evidence). The combined group-customized intervention doubled the likelihood of smoking cessation (RR 2.01, 95% CI 1.41–2.87; low-quality evidence).

Interactive SMS interventions more than doubled quit rates (RR 2.14, 95% CI 1.29–3.55; moderate-quality evidence). Interactive apps had an RR of 5.70 (95% CI 1.28–25.37); despite a large effect estimate, wide CIs indicated substantial uncertainty (very low-quality evidence). Interactive websites (RR 1.02, 95% CI 0.60–1.74), email (RR 1.16, 95% CI 0.83–1.62) and face-to-face interventions (RR 1.14, 95% CI 0.86–1.51) showed no significant benefits (very low-quality evidence).

Intervention ranking

Interventions were ranked using P-scores to reflect their relative efficacy within the network, with higher scores denoting superior efficacy. In the methodological approach grouping, personalized interventions achieved the highest relative efficacy (P-score 0.88), whereas interactive interventions had moderate relative efficacy (P-score 0.53) (Fig. 5a). In the technological grouping, SMS-based interventions demonstrated the highest relative efficacy (P-score 0.87), followed by telephone interventions (P-score 0.81) and multicomponent interventions (P-score 0.78) (Fig. 5b). Within the cross-matched analysis, interactive apps exhibited the highest relative efficacy (P-score 0.95), followed by interactive SMS interventions (P-score 0.81) and group-customized interventions (P-score 0.80) (Fig. 5c). It is important to note that clinical recommendations should not rely exclusively on P-scores; they must comprehensively consider the consistency and robustness of findings, sample sizes and the methodological quality of included studies.

Fig. 5. P-score rankings indicating relative efficacy among digital smoking cessation interventions.

a, Methodological approaches. b, Technological types. c, Cross-matched intervention groups. P-scores were used to rank interventions by their relative efficacy, ranging from 0 to 1. A red-to-yellow gradient in the heatmap visually encodes the cumulative ranking probabilities, where red denotes higher probabilities.

Subgroup analysis

Age subgroup analysis

Participants were stratified into younger (<40 years) and middle-aged (≥40 years) groups to examine age-related differences in intervention effectiveness (Extended Data Figs. 1–3). Detailed comparative results are presented in the Supplementary Information.

Extended Data Fig. 1. Methodological comparison results stratified by age.

Blue circles: Point estimates of relative risk (RR) for age <40 years; Red triangles: Point estimates of RR for age ≥40 years; Error bars (vertical lines): 95% confidence intervals (CI) of RR estimates.

Extended Data Fig. 3. Cross-matched group comparison results stratified by age.

Blue circles: Point estimates of relative risk (RR) for age <40 years; Red triangles: Point estimates of RR for age ≥40 years; Error bars (vertical lines): 95% confidence intervals (CI) of RR estimates. Arrows indicate truncated CIs exceeding the axis limits.

In the younger group, personalized interventions significantly increased smoking cessation success (RR 1.70, 95% CI 1.27–2.27). Interactive (RR 1.38, 95% CI 1.09–1.74) and standard interventions (RR 1.32, 95% CI 1.08–1.62) also showed positive effects. Group-customized interventions showed uncertain results (RR 1.59, 95% CI 0.90–2.80). In the middle-aged group, effects were stronger. Group-customized interventions substantially increased quit rates (RR 2.66, 95% CI 1.49–4.75). Personalized interventions had the largest effects (RR 2.39, 95% CI 1.75–3.25), while standard (RR 1.84, 95% CI 1.48–2.29) and interactive interventions (RR 1.75, 95% CI 1.33–2.30) were also effective.

Among younger participants, app-based interventions (RR 2.10, 95% CI 1.36–3.23) significantly improved quit rates. SMS (RR 1.54, 95% CI 1.20–1.99), telephone (RR 1.56, 95% CI 1.05–2.33) and multicomponent interventions (RR 1.35, 95% CI 1.08–1.69) were effective, whereas email, face-to-face and web interventions did not reach statistical significance. In the middle-aged group, multicomponent interventions showed the highest efficacy (RR 2.13, 95% CI 1.57–2.89). SMS (RR 1.89, 95% CI 1.46–2.44), telephone (RR 1.83, 95% CI 1.44–2.31), app-based (RR 1.83, 95% CI 1.43–2.36), web-based (RR 1.67, 95% CI 1.32–2.13) and face-to-face interventions (RR 1.42, 95% CI 1.04–1.95) were all significantly effective. Email interventions had insufficient data for evaluation in this group.

In younger adults, personalized apps markedly improved quit rates (RR 3.37, 95% CI 1.76–6.47), despite wide CIs suggesting uncertainty. Interactive SMS (RR 2.38, 95% CI 1.26–4.49) and standard SMS interventions (RR 1.67, 95% CI 1.05–2.65) were effective. Email and face-to-face interventions showed no significant effects. For the middle-aged group, group-customized apps displayed substantial but imprecise effects (RR 8.04, 95% CI 1.75–37.00). Multicomponent (RR 2.34, 95% CI 1.66–3.31), personalized apps (RR 2.04, 95% CI 1.34–3.12) and SMS interventions (RR 1.91, 95% CI 1.44–2.53) significantly increased cessation success rates.

Intervention duration subgroup analysis

Interventions were classified into short-term (<3 months), medium-term (3–9 months) and long-term (>9 months) categories, with detailed visual representations provided in Extended Data Figs. 4–6.

Extended Data Fig. 4. Intervention duration subgroup results of the methodological group comparison.

Red circles: Represent relative risk (RR) data points for long term outcomes; Light green triangles: Denote RR data points for medium term outcomes; Dark blue squares: Indicate RR data points for short term outcomes; Error bars (vertical lines): 95% confidence intervals (CI) of RR estimates.

Extended Data Fig. 6. Intervention duration subgroup results of the cross-matched group comparison.

Red circles: Represent relative risk (RR) data points for long term outcomes; Light green triangles: Denote RR data points for medium term outcomes; Dark blue squares: Indicate RR data points for short term outcomes; Error bars (vertical lines): 95% confidence intervals (CI) of RR estimates. Arrows indicate truncated CIs exceeding the axis limits.

In the methodological analysis, short-term interventions demonstrated the strongest efficacy. Personalized interventions showed the highest efficacy (RR 2.29, 95% CI 1.51–3.49), increasing quit rates by 129%. Group-customized (RR 1.98, 95% CI 1.04–3.75), interactive (RR 1.92, 95% CI 1.34–2.76) and standard interventions (RR 1.74, 95% CI 1.22–2.61) also significantly improved quit success. Medium-term interventions showed slightly reduced but significant effects, with RRs for group-customized at 2.05 (95% CI 1.17–3.58), personalized at 1.97 (95% CI 1.53–2.53), interactive at 1.37 (95% CI 1.10–1.72) and standard interventions at 1.40 (95% CI 1.17–1.67). By contrast, long-term intervention effects diminished and lost statistical significance, with RRs for interactive at 1.27 (95% CI 0.84–1.92), personalized at 1.26 (95% CI 0.75–2.12) and standard at 1.30 (95% CI 0.84–2.01).

Regarding the technology types, short-term interventions achieved the highest efficacy with multicomponent (RR 2.36, 95% CI 1.38–4.05) and telephone (RR 2.27, 95% CI 1.49–3.48) interventions leading. SMS (RR 1.86, 95% CI 1.24–2.80) and web-based methods (RR 1.68, 95% CI 1.17–2.41) also performed well. Medium-term outcomes indicated sustained efficacy primarily in app-based (RR 1.61, 95% CI 1.23–2.11) and SMS interventions (RR 1.51, 95% CI 1.24–1.84), along with multicomponent and telephone interventions maintaining RRs above 1.40. Long-term effects showed a general decrease but remained significant for multicomponent (RR 1.56, 95% CI 1.15–2.11).

Cross-matched group analysis showed varied temporal efficacy. In short-term interventions, telephone-based methods exhibited high efficacy (RR 3.24, 95% CI 1.60–6.56), although the wide CIs warrant cautious interpretation. Personalized web (RR 2.75, 95% CI 1.35–5.61) and SMS interventions (RR 2.06, 95% CI 1.23–3.46) were also effective. Medium-term analyses highlighted personalized web (RR 1.94, 95% CI 1.39–2.71), personalized app (RR 2.17, 95% CI 1.47–3.19) and SMS (RR 1.57, 95% CI 1.24–1.99) interventions with maintained efficacy. In the long-term, personalized app (RR 1.77, 95% CI 1.38–2.28), SMS (RR 1.72, 95% CI 1.38–2.15) and multicomponent methods (RR 1.61, 95% CI 1.31–1.98) remained significantly effective, although efficacy was reduced compared with short-term outcomes.

Secondary outcome—prolonged abstinence

The results for prolonged abstinence rates demonstrated substantial concordance with point prevalence abstinence findings. Personalized and interactive digital interventions also showed high quit success rates in promoting prolonged abstinence, with SMS, telephone and multicomponent interventions performing well. However, we remain cautious about the robustness of the prolonged abstinence rate results for two main reasons. First, the network connectivity was relatively sparse, and the number of studies for certain interventions was limited, which may affect result reliability. Second, the biochemical verification methods used in current smoking cessation studies are inadequate for accurately detecting prolonged abstinence beyond 1 month, which may lead to reporting bias. Detailed statistical results, forest plots, network diagrams and P-scores are provided in the Supplementary Information.

Meta-regression analysis

In the network meta-regression analysis, we evaluated the effect of various covariates on intervention outcomes, including biochemical verification, baseline daily smoking amount, gender, age, use of smoking cessation medication, financial incentives and year of publication. The results showed that none of these covariates was statistically significant for the primary outcomes (P > 0.05), suggesting that these factors may not be the primary determinants of intervention effectiveness. Detailed regression analysis results can be found in the Supplementary Information.

Sensitivity analysis

To test the robustness of the results, we conducted a sensitivity analysis and reanalysed the data using Bayesian methods. The results of the Bayesian analysis were consistent with the frequentist analysis, showing no significant changes in the effects or significance of the primary interventions, further validating the reliability of the results. In addition to the Bayesian analysis, we assessed the impact of including studies involving individuals with mental health conditions on overall results and heterogeneity. After excluding these studies, the results remained stable, supporting our conclusions. Detailed sensitivity analysis results can be found in the Supplementary Information.

Publication bias assessment

By comparing the adjusted funnel plots, we found clear indications of publication bias. The funnel plots showed asymmetry, with smaller studies tending to report larger intervention effects, which may result in overestimating the overall effect. The Egger’s regression test reached statistical significance (P < 0.05), indicating a high risk of publication bias. This bias may arise because newer digital interventions remain underexplored. Positive findings are more likely to be published, while negative or non-significant results are often overlooked, contributing to a positive reporting bias in the literature.

Risk-of-bias assessment

The risk-of-bias assessment indicated a high overall risk of bias in the included studies, stemming primarily from missing outcome data, potential selection bias and challenges in implementing blinding. According to the Cochrane Smoking Cessation Group, the nature of smoking cessation interventions makes it difficult to achieve full blinding for participants and researchers, inevitably introducing some bias. However, some studies managed to ensure blinding of outcome assessors, which helped reduce the risk of detection bias. Variations in randomization procedures, allocation concealment and outcome reporting across studies may have compromised the reliability of the pooled results. Detailed bias risk assessment figures and descriptions are available in the Supplementary Information.

GRADE assessment

The Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) assessment indicated the quality of evidence for the primary outcomes. The quality of evidence in the methodological group was rated as ‘low’, the digital intervention type group as ‘moderate’ and the cross-matching group as ‘low’. The quality of evidence for specific intervention methods ranged from ‘moderate’ to ‘very low’, primarily limited by a high risk of bias, insufficient transferability of study designs and the presence of publication bias. Details for the assessment are provided in the tables in the Supplementary Information.

Discussion

Digital interventions, utilizing information and communication technologies, provide cost-effective and scalable smoking cessation support, robustly supplementing traditional cessation strategies^20,22. This is particularly crucial in resource-limited and underserved regions, where such interventions can help reduce health disparities. In this study, we conducted a NMA to systematically evaluate the effectiveness of various digital interventions for smoking cessation and assess the impact of age and intervention duration on outcomes. The results revealed that personalized and group-customized digital interventions exhibited significant superiority in promoting smoking cessation, surpassing the effects of standard care and non-customized interventions. Furthermore, SMS interventions exhibited the strongest performance among digital methods, while apps and multicomponent interventions significantly outperformed standard care, establishing themselves as powerful instruments for smoking cessation. Third, middle-aged individuals demonstrated more positive responses to digital interventions, with higher effectiveness than younger populations. Lastly, short- and medium-term interventions were more effective than long-term interventions, suggesting that the duration of the intervention plays a critical role in cessation outcomes. These findings provide additional evidence supporting the use of digital interventions in smoking cessation efforts, highlighting the potential of personalized, accessible and scalable digital strategies in reducing smoking prevalence.

Research over the past decade has predominantly focused on a single type of intervention, lacking systematic comparisons across different strategies. A 2012 NMA by Chen et al. reported only a marginal increase in smoking cessation rates with digital interventions, with little difference between delivery methods¹⁷. At that time, this conclusion offered limited support for the future potential of digital health interventions. However, over the past decade, the widespread adoption of the internet and smart devices, coupled with the emergence of technologies such as artificial intelligence and machine learning, has driven breakthrough advancements in the development and application of digital health interventions. Our study’s findings demonstrate that digital health interventions not only outperform standard care but also provide evidence suggesting that digital methods could surpass face-to-face interventions. Our study was limited to RCTs on digital interventions, excluding some face-to-face intervention studies. However, this does not mean that digital methods cannot surpass face-to-face interventions. Further comprehensive analysis is needed to confirm the robustness of these results.

This study expands the current understanding of digital smoking cessation interventions by introducing a dual classification system based on methodological approach (for example, personalized and group-based) and type of digital intervention (for example, SMS, apps and multicomponent). Through cross-matching analyses, we provided a more nuanced evaluation of intervention effectiveness. For instance, SMS interventions emerged as the most effective digital modality, probably due to their high accessibility, immediacy and cost-effectiveness. These findings align with previous research¹⁵, reinforcing SMS as a powerful smoking cessation tool. In addition, our study supports personalized interventions as an important approach, as they tailor behavioural support on the basis of individual or group characteristics. This highlights the importance of precision health algorithms in smoking cessation²³ and calls for further research into artificial-intelligence-driven, adaptive digital interventions.

Compared with Chen et al.‘s NMA conducted a decade ago, we present stronger evidence in support of the application of digital health interventions. The widespread use of smartphones and mobile applications has created a vast platform for delivering personalized interventions. Modern mobile applications can adjust intervention content in real time on the basis of user behaviour and feedback and provide more precise and personalized support through data analysis and machine learning algorithms. This highly personalized intervention strategy substantially boosts smoking cessation success rates, reflecting the amplifying impact of technological advancements on intervention outcomes. Moreover, the use of artificial intelligence and machine learning allows digital interventions to intelligently identify and respond to user needs with greater precision. For instance, intelligent chatbots can simulate human conversation, providing timely psychological support and smoking cessation advice, which enhances user engagement and adherence. This level of interactivity and immediacy is difficult to achieve with traditional interventions, improving the effectiveness of the intervention and the user experience. Moreover, the rise of social media and online communities has offered new momentum for group-customized interventions. Through online social platforms, participants can gain peer support and encouragement, building robust social support networks. This group dynamic plays a crucial role in group-customized interventions, further enhancing smoking cessation success rates.

The findings of this study demonstrate the substantial effects of digital health interventions in smoking cessation, and the underlying mechanisms can be explained from multiple perspectives. First, personalized interventions gather and analyse individual behaviour data, psychological characteristics and smoking cessation motivation to deliver tailored support and feedback. According to social cognitive theory, self-efficacy plays a crucial role in behaviour change. Personalized interventions provide precise support and feedback that greatly enhances self-efficacy and strengthens their intrinsic motivation to quit smoking, thereby improving smoking cessation success rates^24,25. The effectiveness of SMS interventions may stem from their immediacy and widespread reach. As a convenient and low-cost communication method, SMS provides timely support and reminders at key moments, helping participants overcome challenges and temptations during the cessation process²⁶. This immediacy and continuity are consistent with the ‘timeliness’ principle of behavioural interventions, reinforcing smoking cessation motivation and behaviours²⁷. Older adults may respond more positively to digital interventions owing to a heightened perception of health risks and stronger health motivation. Although younger individuals may be more familiar with technology, older individuals often perceive health risks more acutely, which drives their motivation to engage with digital interventions. As they age, their focus on health increases, leading to a greater urgency to quit smoking²⁸. Moreover, they may have more time to engage in the interventions, resulting in higher adherence rates. It is important to note that, while we performed a subgroup analysis by age, the IQR for mean age across the studies was 35.7–45.7, and the maximum mean age did not exceed 60 years. Thus, the results for the over-40 age group are more relevant to middle-aged adults, which helps us better understand our conclusions. This further aids in understanding that differences in digital product usage between various population groups are not as pronounced as expected. For younger individuals, while their acceptance of digital technology is higher, their lower perception of health risks, fast-paced lifestyles or other stressors may impede their success in smoking cessation²⁹. This suggests that, when developing intervention strategies, tailored approaches should be implemented based on the characteristics of different age groups. In addition, the diminishing effects of interventions over time indicate the need for continuous support and maintaining participant motivation during long-term interventions. This may involve incorporating new incentive mechanisms, social support or interactive components to boost the appeal and sustainability of the interventions. For instance, the use of gamification strategies, social media engagement or regular feedback can help sustain participants’ interest and involvement, preventing a decline in smoking cessation motivation.

The strength of this study is rooted in its study design and analytical approaches. We adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines specific to NMA (PRISMA-NMA) guidelines, ensuring the methodological rigour and transparency of the systematic review and NMA. By incorporating a large number of RCTs, we believe this provides evidence on digital interventions for smoking cessation, thus strengthening the robustness and external validity of the findings. Moreover, we categorized digital interventions based on ‘methodological approach’ and ‘digital intervention type’ and performed cross-matching analysis, resulting in more detailed findings. This analytical method enhances our understanding of the differential effectiveness of various interventions and provides insights for clinical practice and policy development.

However, this study has certain limitations. First, the quality of the included studies varied; while some demonstrated methodologies, others had a risk of bias due to unclear randomization, blinding challenges and missing data. This may impact the reliability and interpretation of the findings. In addition, digital health interventions lack standardized definitions, and the focus of different studies varies. Moreover, many studies omitted key details when reporting interventions, making it difficult to accurately evaluate the specific characteristics of these interventions. To address this issue, five researchers achieved a more consistent coding framework through multiple rounds of discussion. Despite the discussions among researchers, the coding process may still have been influenced by personal bias, potentially leading to inconsistencies and blurring the clarity of comparative results. We lacked individual participant data, preventing us from conducting more granular subgroup analyses or adjusting for potential confounders such as socio-economic status or psychological characteristics. The limited number of studies on long-term interventions contributes to the increased uncertainty in the results. Certain interventions, such as email interventions in the middle-aged group, lacked data in some subgroups, limiting our ability to fully assess their effectiveness. Furthermore, our study concentrated on comparing digital interventions with standard care, leaving out a substantial number of studies comparing face-to-face interventions with standard care. Thus, further evidence is needed to compare the effectiveness of face-to-face interventions with digital ones, and we rated the evidence for face-to-face interventions as very low. Future research should address these limitations and implement appropriate corrective measures. First, high-quality randomized controlled trials should be conducted, especially those focusing on long-term intervention effects, ensuring methodological rigour and minimizing bias risk. Researchers should provide detailed reports on randomization methods, blinding procedures and data handling to increase the transparency of their studies. Second, acquiring and leveraging individual participant data can facilitate deeper subgroup analyses to explore the differential effects of interventions across diverse populations and contexts. Furthermore, integrating emerging technologies such as artificial intelligence and machine learning can facilitate the development of smarter, more personalized interventions, improving their precision and effectiveness. Leveraging big data analytics can enable real-time adjustments to intervention content and strategies, effectively meeting personalized needs. Lastly, although this study did not find that advancements in information and communication technology over different periods significantly impacted intervention effectiveness, as technology evolves, digital intervention methods and tools continue to be updated, which may lead to differences in intervention effectiveness across different timeframes. Therefore, future research should account for temporal effects and control for them in study design to enhance the accuracy and reliability of the findings.

Current digital intervention research, especially RCTs, focuses primarily on intervention effectiveness, but exploring only effectiveness is inadequate. There is a greater need to investigate the mediators that explain why these interventions are effective. The key to addressing this issue lies in ensuring that interventions not only produce short-term effects but also sustain participant motivation and adherence over the long term. Thus, in intervention design, relying solely on digital tools to deliver content is insufficient to ensure engagement. External factors such as incentives, social components and psychological support must be integrated to foster long-term engagement^30,31. Drawing from the findings of depression intervention studies, improving self-efficacy and reducing negative emotions are two critical mediators³². Personalized feedback, skills training and positive reinforcement can effectively strengthen participants’ confidence in successfully quitting smoking, thereby boosting self-efficacy. These improvements can be facilitated through specific digital tools, such as self-monitoring, progress tracking and reward systems embedded in apps. When participants see themselves gradually achieving small goals and gaining a sense of accomplishment, this positive feedback will further encourage their continued participation. In addition to improving self-efficacy, negative emotions such as anxiety, depression and stress frequently contribute to smoking cessation failure. Therefore, intervention design should include psychological support components, such as online counselling, emotional management courses or meditation guidance, to assist participants in managing and reducing negative emotions. Social support is another critical factor that should not be overlooked. Building online support groups or social networks can enhance interaction and mutual assistance among participants, fostering a stronger sense of social support, which in turn encourages them to maintain motivation and confidence throughout the cessation process. Therefore, beyond focusing solely on the direct effectiveness of interventions, mediating variables such as these should receive ample attention in digital smoking cessation interventions. A deeper investigation into the role of these mediators can reveal more nuanced intervention pathways, offering strong support for designing more precise and personalized interventions in the future.

Through a NMA, our study demonstrates the significant impact of digital interventions on smoking cessation, particularly highlighting the efficacy of personalized and group-customized approaches. The findings suggest that the implementation of digital technologies in smoking cessation practices enhances efficacy, offering further evidence to optimize smoking cessation strategies. Future research should emphasize high-quality studies, particularly focusing on the long-term effects of interventions, the response variations across different populations and exploring behavioural mediators to elucidate effective pathways.

Methods

This study uses NMA to systematically evaluate the comparative efficacy of digital smoking cessation interventions, adhering to the PRISMA-NMA. This study was retrospectively registered with PROSPERO (registration: CRD42024517874) in March 2024 after data collection began in January 2024. An interdisciplinary research team comprising epidemiologists, methodologists, clinical practitioners and nursing specialists collectively supervised the research design, analytical procedures and interpretative rigour. Systematic literature searches were conducted in PubMed, Embase and Cochrane Library (CENTRAL) to identify randomized controlled trials published from database inception through 10 September 2024, without linguistic restrictions. Search strategies incorporated combined keyword and MeSH terms (detailed in the Supplementary Information). Furthermore, manual searches of reference lists from pertinent systematic reviews and meta-analyses were performed to ensure study inclusion.

Two independent investigators (Shen Li and Y.L.) performed initial deduplication using EndNote v19 (Clarivate Analytics). Subsequently, three reviewers (Shen Li, Y.L., and H.S.) independently screened titles and abstracts, with studies meeting inclusion criteria advancing to full-text assessment. Five investigators (Shen Li, Y.L., S.T., H.S. and C.X.) independently conducted full-text reviews and extracted pertinent data. Discrepancies were resolved through iterative consensus discussions, with arbitration provided by a senior investigator (X.M.) when necessary. Inclusion criteria stipulated: (1) participants identified as smokers capable of effectively utilizing digital intervention systems; (2) smoking cessation interventions delivered via digital modalities; (3) control groups receiving standard care, placebo or alternative digital interventions; (4) explicit reporting of smoking cessation outcomes; and (5) randomized controlled trial designs. Exclusion criteria included studies in which pharmacotherapy was routinely provided alongside digital interventions, preventing isolation of the digital intervention’s independent effect.

To address the heterogeneity of interventions across included studies, we categorized them along two distinct dimensions: methodological approaches and technology types, enabling a more precise assessment of intervention effects. To analyse the impact associated with specific intervention characteristics, methodological frameworks were systematically combined with technological classifications to create cross-matched comparison groups (for example, personalized SMS versus standard app). This stratified analytical approach facilitates enhanced detection of variations in abstinence outcomes across different implementation paradigms.

The primary outcome was the 7-day point prevalence abstinence, defined as self-reported or biochemically validated smoking cessation during the 7 days preceding protocol-defined follow-up timepoints (for example, 30-day follow-up). The secondary outcome was prolonged abstinence, indicating sustained smoking cessation over 30 days. Data extraction was independently performed by three researchers (Y.L., Shen Li and H.S.), covering the following: authors, year of publication, country, sample size, study design, follow-up timelines, participant characteristics (age, gender and baseline cigarettes per day), intervention/control specifications and cessation counts per study arm. When studies reported serial follow-up measurements, longitudinal endpoint data received analytical prioritization to assess intervention durability and mitigate efficacy inflation. Missing or ambiguous data triggered correspondence attempts with the corresponding authors. Two assessors (Y.L. and S.T.) independently appraised bias risks through Cochrane RoB 2.0 criteria, examining randomization sequence generation, allocation masking, blinding protocols, outcome data completeness, selective reporting risks and additional bias sources³³.

Data analysis

We applied a frequentist random-effects model for the NMA. The main reason for selecting the frequentist approach was its computational efficiency in handling large networks and complex models, making it well suited for the diverse interventions analysed in this study. In addition, the frequentist approach presents results as RRs with 95% CI, facilitating interpretation by clinicians. Compared with Bayesian methods, the frequentist model offers greater transparency and reproducibility, as it requires fewer subjective prior assumptions. The NMA was conducted using the ‘netmeta’ package in R software (version 4.3.3). This method allows the simultaneous comparison of multiple interventions, synthesizing direct and indirect evidence to estimate relative effects between interventions and establish rankings. The standard care group was used as the reference control group, and forest plots were constructed to visually display the effects of different interventions. A network plot was created to depict the relationships within the network, where each node represents a treatment, and the edges between nodes represent the comparisons between interventions. The presence of an edge indicates direct evidence supporting the comparison between the two interventions. The thickness of the edge is proportional to the number of studies supporting the comparison, indicating more studies supporting this comparison and therefore stronger evidence. A league table was used to present the pairwise comparisons of each intervention within the network. P-scores, ranging from 0 (indicating lower relative efficacy) to 1 (indicating higher relative efficacy), were calculated to evaluate and rank the relative effectiveness of interventions within the network. To evaluate the consistency of the network model, we examined the distribution of study group characteristics across interventions and tested the transitivity assumption for indirect comparisons. We used the node-splitting method to assess the consistency between direct and indirect evidence. If inconsistency was detected, further analysis was conducted to identify its sources and potential impacts. Global heterogeneity was assessed using the I² statistic, and an I² value exceeding 50% was considered indicative of moderate-to-severe heterogeneity³⁴.

To explore potential sources of heterogeneity, we conducted network meta-regression and subgroup analyses using the R ‘gemtc’ package, quantifying covariate effects with median posterior estimates and corresponding 95% credible intervals. Evaluated covariates encompassed publication year, participant age distribution, intervention duration, sex, baseline cigarette consumption, biochemical validation, financial incentives and adjunctive pharmacotherapy. Variables identified as significant through regression analyses informed subsequent subgroup stratifications. Age was inversely associated with abstinence success rates, potentially reflecting increased nicotine dependence and cumulative unsuccessful quit attempts among older smokers. Additional socio-behavioural mechanisms may include perceived futility of cessation in advanced age and diminished risk perceptions among elderly smokers^35,36. Population-level evidence indicates that the probability of achieving at least 6-month abstinence decreases from 13.7% among individuals aged 18–24 years to 5.0% among those aged 40–64 years³⁷. Therefore, we consider age to be an important factor affecting smoking cessation. Moreover, intervention duration is a key determinant of smoking cessation outcomes. Longer interventions support behavioural adjustments and habit formation, particularly in digital contexts that require sustained participant engagement and adaptation. Considering these clinically relevant factors, we performed subgroup analyses independent of regression significance results: (1) age subgroup analysis: participants were divided into two groups, <40 years and ≥40 years; (2) intervention duration subgroup analysis: interventions were classified as short-term (<3 months), medium-term (3–9 months) and long-term (>9 months) to evaluate the impact of intervention duration on outcomes.

Sensitivity analysis, risk of bias and outcome quality assessment

To ensure result robustness, we used a Bayesian approach as a complement to the sensitivity analysis, using non-informative priors, with the results presented as medians and 95% credible intervals. We implemented four Markov chain Monte Carlo chains with 50,000 iterations each, discarding the first 10,000 iterations as burn-in to ensure convergence, which was verified using the Gelman–Rubin diagnostic method. Evaluating publication bias in network meta-analyses is challenging because the methods typically used in traditional meta-analyses are not entirely applicable to a network structure. To resolve this, we used the comparison-adjusted funnel plot method to assess the risk of publication bias³⁸. We hypothesized that studies presenting statistically significant outcomes are more likely to be published, even when they have small sample sizes, potentially leading to asymmetry in the funnel plot. Moreover, unpublished negative results (that is, when new interventions are not significantly better than older ones) may create gaps in the funnel plot. By observing the symmetry of the funnel plot, we can make an initial assessment of the presence of publication bias. If notable asymmetry is detected, it suggests potential bias that may influence the NMA results, warranting caution in interpreting the findings. The quality of the evidence was independently assessed by two researchers (Y.L. and S.T.) using the GRADE system. Any discrepancies were resolved through consensus discussions or arbitration by a third reviewer (X.M.). These assessments categorized evidence quality as high, moderate, low or very low^39–41.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Extended data

Extended Data Fig. 2. Technology type comparison results stratified by age.

Blue circles: Point estimates of relative risk (RR) for age <40 years; Red triangles: Point estimates of RR for age ≥40 years; Error bars (vertical lines): 95% confidence intervals (CI) of RR estimates.

Extended Data Fig. 5. Intervention duration subgroup results of the technologogy type group comparison.

Red circles: Represent relative risk (RR) data points for long term outcomes; Light green triangles: Denote RR data points for medium term outcomes; Dark blue squares: Indicate RR data points for short term outcomes; Error bars (vertical lines): 95% confidence intervals (CI) of RR estimates.

Author contributions

X.M., Sheyu Li and Shen Li conceived and designed the study. Shen Li, Y.L. and H.S. performed data acquisition and extraction. Statistical analyses were conducted by Shen Li, C.X. and S.T. All authors contributed to data interpretation and manuscript revision and approved the final version. Sheyu Li, Y.J. and Y.W. supervised the project and provided critical revisions.

Peer review

Peer review information

Nature Human Behaviour thanks Ram Bajpai and Olga Perski for their contribution to the peer review of this work.

Data availability

To promote transparency and reproducibility, the analysis code and dataset have been made publicly available via GitHub at https://github.com/Xch20030923/digital-smoking-nma.git. Please note that data sharing is intended for academic research purposes only and not for other purposes.

Code availability

All statistical analyses were conducted using R software (version 4.3.3), primarily utilizing the ‘netmeta’ and ‘gemtc’ packages. All analysis scripts are available via GitHub at https://github.com/Xch20030923/digital-smoking-nma.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41562-025-02295-2.

Supplementary information

The online version contains supplementary material available at 10.1038/s41562-025-02295-2.