For this systematic review and network meta-analysis, we followed PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 and reported the search process in accordance with PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses–Search extension). The completed PRISMA 2020, PRISMA 2020 Expanded, PRISMA-S, and PRISMA for Abstract checklists are provided in Checklist 1. The protocol was registered in PROSPERO (International Prospective Register of Systematic Reviews; CRD420251056704) [36]. All PRISMA 2020 items were reviewed against the manuscript to ensure complete and transparent reporting. We systematically searched 7 electronic databases (PubMed, CINAHL, Embase, PsycINFO, Cochrane Library, Scopus, and Web of Science) for studies published between January 1, 2000, and October 1, 2025, to identify any newly published studies meeting the eligibility criteria. Search terms combined keywords and Medical Subject Headings terms related to serious games, digital games, video games, public health education, knowledge, attitudes, and diseases. The search strategy was refined in accordance with Chapter 4.4 of the Cochrane Handbook to maximize sensitivity, including the expansion of controlled vocabulary and additional free-text synonyms. An updated search was conducted in February 2026, and no additional eligible studies were identified. Full search strategies for each database are provided in Multimedia Appendix 1. Reference lists of relevant systematic reviews and meta-analyses were also screened for potentially eligible studies (Multimedia Appendix 1). Gray literature sources and clinical trial registries were not searched separately.

All records were imported into Covidence (Veritas Health Innovation, Melbourne, Australia) for deduplication, screening, and data management. Two reviewers (DH and DW) independently screened titles, abstracts, and full texts, resolving disagreements by consensus or through consultation with a third reviewer (WM). Study authors were contacted when additional clarification was required.

Eligibility criteria followed the population, intervention, comparator, outcomes, and study design framework (Multimedia Appendix 2) [37]. We included randomized controlled trials (RCTs), cluster RCTs, and pilot RCTs of digital serious games designed to improve health-related knowledge or attitudes in nonprofessional populations, including children, adolescents, adults, informal caregivers, patients, and the general public. Interventions were required to be delivered via digital platforms, such as web-based applications, mobile apps, computer software, virtual reality, augmented reality, or robot-assisted systems. Comparators included no intervention, conventional education, or digital nongame tools. The primary outcomes were changes in knowledge or attitudes, assessed using validated instruments when reported.

We excluded studies evaluating nondigital games; interventions targeting clinical treatment, rehabilitation, or professional training; and studies that did not report at least 1 primary outcome. Nonrandomized studies, qualitative research, reviews, commentaries, protocols, and conference abstracts were also excluded.

A standardized data extraction form was initially developed (DH) and subsequently refined (DH and DW) in accordance with the Cochrane Handbook for Systematic Reviews of Interventions [38]. The form was pilot-tested on a subset of studies to ensure reliability and reproducibility before full implementation. For each eligible trial, 2 reviewers (DH and DW) independently extracted data and verified the results with a third reviewer (WM). Extracted variables included study characteristics (identifier, year, country, design, and setting); participant characteristics (population group, mean age, sex distribution, sample size, and patient status); intervention details (type of digital serious game, delivery platform, educational content, duration, and follow-up, if reported); comparator details (type and format); and outcome measures (assessment tools, baseline and postintervention scores, and key findings related to knowledge or attitudes).

The primary outcomes were changes in health-related knowledge, including understanding of diseases, prevention, and health promotion, as well as changes in health-related attitudes, beliefs, perceptions, and intentions. Both outcomes were assessed using validated instruments when reported. No secondary outcomes were prespecified.

The risk of bias for each included study was assessed using the revised Cochrane risk-of-bias tool for randomized trials tool with the appropriate version applied to individually randomized and cluster-randomized trials [39]. The tool evaluates potential bias in the following domains: the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcomes, and selection of the reported results. Each trial was independently appraised by 2 reviewers (DH and DW) and categorized as low risk, some concerns, or high risk of bias. Any disagreements were resolved by consensus, with persistent discrepancies adjudicated by a third reviewer (WM).

Publication bias and small-study effects were assessed using the Egger regression test (P<.10) and comparison-adjusted funnel plots within the network meta-analysis framework. Interpretation of funnel plot asymmetry was undertaken cautiously, as between-study variability and model complexity may contribute to apparent asymmetry, independent of publication bias.

The certainty of the evidence was appraised using the GRADE (Grading of Recommendations, Assessment, Development, and Evaluation) framework, following the GRADE Working Group’s guidance. Certainty was evaluated across the domains of study design, risk of bias, inconsistency, indirectness, imprecision, and potential publication bias [45]. All included randomized trials, including pilot and cluster trials, were initially rated as high-certainty evidence. Downgrading was applied when serious limitations were identified, including high risk of bias, substantial unexplained heterogeneity, indirectness of the evidence in relation to the review question, imprecision of effect estimates, or potential publication bias, in accordance with GRADE guidance [46]. Potential publication bias was also assessed. Final ratings were categorized as high, moderate, low, or very low certainty.

The database search yielded 9269 records, and a further 47 records were identified through reference searches of relevant systematic reviews and meta-analyses. After the removal of 5352 duplicates, 3917 records were screened by title and abstract (Figure 1). Of these, 3816 records were excluded, and 1 report could not be retrieved because the full text was unavailable. A total of 88 full-text articles were assessed for eligibility, of which 48 were excluded for reasons summarized in Multimedia Appendix 5. In total, 40 studies were included in the systematic review and 30 in the Bayesian network meta-analysis (Figure 2).

A total of 40 RCTs published between 2000 and 2025 were included (Table 1). Research output increased markedly after 2020, with 5 (12.5%) studies published before 2010 [47-50,76], 10 (22.5%) between 2010 and 2019 [51-57,77-79], and 25 (62.5%) since 2020 [58-75,80-86]. Studies were conducted across 19 countries, most frequently in the United States (n=13, 32.5%) [47-50,54,56,60,65,74,76,78,79,83], followed by China (n=4, 10%) [64,68,70,85], Iran (n=4, 10%) [58,62,67,73], and the Netherlands (n=3, 7.5%) [53,63,77]. All studies adopted a randomized controlled design, including 10 (25%) pilot trials and 5 (12.5%) cluster trials.

The design and characteristics of digital serious games are summarized in Multimedia Appendix 6. Interventions evolved with technological development. Early studies used computer-offline serious games (n=7, 17.5%) [47,49-52,80,85] and video-based serious games (n=6, 15%) [48,61,66,76,78,79]. Later studies adopted computer or web-based online games (n=8, 20%) [55,56,65,71,73,75,77,83] and mobile-app games (n=13, 32.5%) [54,57-59,62-64,67,68,72,81,82,84]. Since 2020, immersive formats such as virtual-reality serious games (n=3, 7.5%) [69,74,86] and augmented-reality serious games (n=2, 5%) [60,71] have become more common, and robot-assisted serious games (n=1, 2.5%) [53] were included from 2017. Three games, Watch, Discover, Think, and Act [47,50], Re-Mission [76,78], and Dental Detective [51,80], were each evaluated in 2 trials due to updated versions or applications in different populations. All iterations were therefore included in the synthesis.

A total of 8764 participants were included (intervention =4374; control =4390). Twelve studies (30%) targeted children, 11 (27.5%) adolescents, 11 (27.5%) adults, and 6 (15%) mixed populations (eg, children and adolescents or adolescents and adults). Thirteen studies (32.5%) included more women than men, 9 (n=9, 22.5%) were gender-balanced, and 18 (45%) included a higher proportion of men.

The most frequent educational topics were CD education (n=8, 20%), cancer education (n=6, 15%), sexual and reproductive health education (n=6, 15%), and nutrition and healthy lifestyle education (n=6, 15%). Other topics included vaccination and infectious-disease prevention (n=4, 10%), oral health education (n=4, 10%), medication and antimicrobial-resistance education (n=3, 7.5%), and psychological and developmental health education (n=3, 7.5%).

Among the 40 studies, 35 (87.5%) reported knowledge outcomes, of which 28 (80%) showed significant improvement and 7 (20%) found no between-group difference [51,54,60,72,75,77,80]. A total of 22 (55%) studies assessed attitude outcomes, with 15 (68.2%) showing positive changes and 7 (31.8%) showing no improvement [51,54,55,57,63,75,80]. A total of 30 (75%) studies were included in the Bayesian network meta-analysis [47-75,80], including those reporting knowledge outcomes (n=27, 67.5%) [47-55,57-62,64-73,75,80] and attitude outcomes (n=16, 40%) [47,50,54-57,63-67,69,70,73-75].

Among the 40 included RCTs, methodological quality was generally moderate to high (Multimedia Appendix 7). For the 35 individually randomized trials, low risk of bias was most frequently observed in the randomization process (25/35, 71.4%), deviations from intended interventions (31/35, 88.6%), and measurement of outcomes (33/35, 94.3%). “Some concerns” were mainly identified in the selection of reported results (14/35, 40%) and overall bias judgment (24/35, 68.6%), mainly due to the absence of preregistered protocols or incomplete reporting of secondary outcomes. One trial was rated as high risk of bias in the domain of deviations from intended interventions because participants and facilitators were not blinded during gameplay [51], and another trial was judged as high risk for the same reason, with substantial researcher involvement potentially influencing participant responses [68]. No other study was rated as high risk in any domain. Among the 5 cluster-randomized trials, methodological quality was similarly high; all studies were rated as low risk for the randomization process and missing outcome data, with only minor concerns regarding the selection of the reported result. Taken together, 31.4% (11/35) of studies were judged as low risk, 68.6% (24/35) as having some concerns, and none as high risk.

Across the 40 included studies, 27 reported data on knowledge outcomes, and 16 on attitude outcomes [47-75], with 21 trials contributing to both outcome categories. Compared with controls, digital serious games significantly improved public health knowledge (SMD=0.66; 95% CI 0.32‐0.99; P<.001; I²=89.1%) and showed a moderate positive effect on health attitudes (SMD=0.50; 95% CI 0.27‐0.76; P<.001; I²=80.7%) (Figure 1). Considerable heterogeneity was observed across studies (knowledge: Q=239.23; P<.001; attitude: Q=77.75; P<.001). Funnel plots showed mild asymmetry for both outcomes (Multimedia Appendix 8). Egger’s regression test indicated potential small-study effects for knowledge (P=.006) but not for attitude (P=.05).

The overall certainty of evidence, assessed using the GRADE framework, was moderate for both knowledge and attitude outcomes. Although substantial heterogeneity was present, the direction of effects remained consistent across studies, and subgroup analyses explained much of the observed variation by intervention duration, population type, and health topic. Minor methodological concerns related to randomization, allocation concealment, and small-study effects contributed to downgrading from high to moderate certainty. Indirectness and imprecision did not materially affect the certainty ratings, as all included trials directly addressed the review question and yielded precise pooled estimates (Multimedia Appendix 9).

Subgroup analyses were undertaken to explore potential sources of heterogeneity across intervention duration, study region, patient status, health topic, publication year, population type, and sex. Across both knowledge and attitude outcomes, multisession interventions consistently yielded larger effects than single-session exposure (knowledge: χ²₁=4.04; P=.04; attitude: χ²₁=4.97; P=.03), indicating that repeated game participation reinforced learning and attitude internalization. Effect sizes were also greater among nonpatient populations than among patients (knowledge: χ²₁=7.13; P=.008; attitude: χ²₁=9.97; P=.002), suggesting that individuals without disease burden may be more receptive to health information. Considerable variation was observed across health topics (knowledge: χ²₆=120.32; P<.001; attitude: χ²₆=176.14; P<.001), with cancer- and CD-focused games achieving the highest impact, whereas effects were smaller for vaccination and oral health education. Regional differences were modest but favored studies conducted in Asia (χ²4=10.18; P=.04). Publication year, age group, and sex composition did not consistently influence effect estimates (Multimedia Appendix 10).

The knowledge network comprised 14 interventions, including 7 types of digital serious games and 7 traditional or nongame comparators, forming 26 direct comparisons and 5 closed loops (Figure 3A). Between-study heterogeneity in the network meta-analysis was low (τ=2.75; 95% CrI 1.58‐4.69; τ²=7.57; network I²=8%). Digital serious games produced the greatest improvements in knowledge outcomes (Figure 4). Mobile app–based games showed significantly higher effects than traditional education (mean difference 5.46; 95% CrI 2.00‐9.39), treatment as usual (4.87; 95% CrI 1.06‐9.37), and no intervention (2.82; 95% CrI 0.09‐5.79). Computer-offline and web-based serious games also achieved superior gains compared with traditional education (4.87; 95% CrI 0.65‐7.95 and 4.12; 95% CrI 0.79‐6.05, respectively), whereas robot-assisted, virtual reality, and video-based games showed weaker comparative effects. Bayesian ranking analyses indicated that mobile app–based, computer-offline, and web-based serious games consistently ranked highest for improving knowledge outcomes (Figure 5A). Prediction intervals were calculated to reflect the expected range of treatment effects in future studies (Multimedia Appendix 11). No significant inconsistency between direct and indirect evidence was detected in node-splitting analyses (all P>.05; Multimedia Appendix 12), and the consistency and unrelated mean effects models showed nearly identical model fit (Deviance Information Criterion; DIC=109.899 vs 109.894; ΔDIC=0.005). Sensitivity analyses using alternative prior distributions produced nearly identical SUCRA values and treatment rankings, indicating robust results (Multimedia Appendix 13).

The attitude network comprised 11 interventions, including 6 types of digital serious games and 5 traditional or nongame comparators, forming 16 direct comparisons and 3 closed loops (Figure 3B). Between-study heterogeneity in the network meta-analysis was low (τ=3.19; 95% CrI 1.42‐6.53; τ²=10.20; network I²=3%). Digital serious games produced greater improvements in health attitudes compared with traditional or nongame education (Figure 4). Computer-offline, web-based, and virtual reality serious games showed the largest improvements in attitude outcomes compared with traditional education (13.28; 95% CrI 3.30‐22.92; 11.30; 95% CrI 1.53‐21.00; and 11.61; 95% CrI 1.33‐21.74, respectively), whereas video-based, face-to-face, and no-intervention conditions showed weaker or inconsistent effects. Bayesian ranking analyses indicated that computer-offline, web-based, and virtual reality serious games ranked highest for improving health attitudes (Figure 5B). Prediction intervals were calculated to reflect the expected range of treatment effects in future studies (Multimedia Appendix 11). No significant inconsistency between direct and indirect evidence was detected in node-splitting analyses (all P>.05; Multimedia Appendix 12), and the consistency and unrelated mean effects models showed similar model fit (DIC=58.47 vs 60.12; ΔDIC=1.65). Sensitivity analyses using alternative prior distributions produced nearly identical SUCRA values and treatment rankings, indicating that the results were robust (Multimedia Appendix 13).

In this systematic review and Bayesian network meta-analysis of 40 RCTs involving 8764 participants, digital serious games were associated with improvements in public health knowledge and attitudes compared with traditional or noninteractive education. Greater effects were observed with multisession interventions. Subgroup analyses indicated stronger responses among adolescents and nonpatient populations, particularly in studies conducted in Asia and in interventions addressing psychological or developmental health topics. Network meta-analysis further demonstrated differences across delivery formats: mobile app–based games ranked highest for knowledge outcomes, whereas computer-offline and web-based formats showed greater relative effectiveness for attitude change, while video-based and traditional education formats consistently ranked lower. By integrating pairwise and network meta-analysis within a cross-disease framework, this study enables comparative evaluation across formats and population contexts, addressing limitations of prior reviews restricted to single conditions or pairwise comparisons [87-89].

Despite substantial heterogeneity observed in the pairwise meta-analysis, sensitivity analyses confirmed the stability of pooled estimates, indicating that the variability primarily reflects contextual and population-level differences rather than methodological bias. Differences across health topics highlight the influence of content relevance and narrative structure in shaping learning outcomes. Interventions targeting psychological or developmental health often incorporate self-management scenarios and emotionally salient components that may enhance perceived relevance and retention [90]. The advantage of repeated exposure is consistent with reinforcement and memory consolidation processes, whereas single-session interventions may offer insufficient opportunities for feedback and integration. Larger improvements among nonpatient populations may relate to lower baseline knowledge and reduced ceiling effects [91]. The heterogeneity observed, therefore, reflects meaningful contextual differentiation rather than instability of effect estimates.

These findings should also be interpreted considering methodological factors, including the risk of bias and the certainty of evidence. Risk-of-bias assessment using the revised Cochrane risk-of-bias tool for randomized trial tool indicated that several studies had methodological limitations that may have influenced effect estimates. The overall certainty of evidence assessed using the GRADE framework was rated as moderate for knowledge outcomes and low-to-moderate for attitudes. In addition, while CIs represent the average effect across studies, prediction intervals reflect the potential variation in effects across different implementation settings, indicating that intervention effects may vary depending on population characteristics and context.

Beyond these methodological considerations, the findings also suggest several interpretive mechanisms underlying the educational effects of serious games. Although the network meta-analysis suggested relatively consistent comparative effects across formats, the pooled estimates and ranking probabilities should still be interpreted with caution because the included studies varied in populations, health topics, and intervention characteristics. From a structural perspective, the network findings further suggest that the educational impact of serious games may operate along complementary cognitive and affective pathways. Interventions incorporating adaptive feedback, progressive challenge, and opportunities for repeated engagement are more likely to activate sustained cognitive processing, thereby facilitating the consolidation and integration of information [92]. In contrast, formats characterized by low interactivity and fixed content delivery may limit learner control and cognitive activation [93]. With respect to attitudinal outcomes, narrative-driven and role-playing designs appear more conducive to attitude change through mechanisms of perspective-taking and emotional engagement [94], while immersive simulations may intensify affective involvement through first-person experiential framing. Together, these findings indicate that knowledge gains are primarily supported by cognitive reinforcement processes, whereas attitudinal change is more closely linked to emotional immersion and social resonance. The integration of both pathways may, therefore, strengthen the overall educational impact of serious games in public health contexts.

Importantly, the relative balance between cognitive structure and experiential immersion is not only a theoretical distinction but also a practical one. Designs that prioritize deep affective engagement often require greater technological resources and infrastructural investment, whereas cognitively structured, feedback-oriented formats may be more feasible for large-scale dissemination [95]. This interplay between experiential intensity and implementation feasibility becomes particularly salient in population-level public health education [96].

Consistent with this structural tension, the network analysis highlights trade-offs between experiential depth and scalability. Virtual-reality and robot-assisted formats may achieve high experiential fidelity yet face barriers related to cost and accessibility, while mobile and web-based interventions enable broader reach, albeit sometimes with reduced experiential richness [97]. These findings suggest that innovation should not focus solely on increasing technical sophistication but rather on developing adaptive architectures capable of preserving feedback, learner autonomy, and emotional resonance across diverse delivery contexts [98,99]. Achieving equilibrium between structural fidelity and affective relevance may be essential for translating short-term knowledge improvements into sustained behavioral and attitudinal change in population health education.

Evidence from this review suggests that digital serious games can extend the reach of public health education in settings where conventional programs face limitations in coverage or engagement. The consistent advantages of mobile and web-based formats over resource-intensive technologies indicate that scalability depends more on accessibility and design efficiency than on technical sophistication [22,100]. In practice, prioritizing adaptive, feedback-driven mobile platforms may yield greater population impact than investing in high-fidelity but low-reach systems, such as virtual reality or robotics [21,101]. At the same time, stronger effects observed among adolescents and women highlight both the potential for targeted implementation and the need to address equity gaps among patients and older adults who demonstrate lower engagement. Inclusive design, tailored difficulty adjustment, and integration within existing community or school-based programs may help reduce digital exclusion and sustain long-term participation [102,103].

Beyond implementation considerations, the current evidence base remains fragmented, with substantial variation in populations, intervention formats, and outcome measures. Most randomized trials are small and exploratory, often limited to school-aged or student samples. Future research should prioritize adequately powered trials involving adult and older populations and incorporate medium- and long-term follow-up to determine whether gains in knowledge and attitudes translate into sustained behavioral change [30].

A structured implementation framework is also needed to define minimal effective exposure, establish evaluation benchmarks, and clarify ethical standards for educational gaming [104,105]. Standardization represents a critical next step. Although thematic diversity in public health education is expected, a unified evaluation framework should be developed to assess usability, implementation quality, and core design attributes—such as interactivity, feedback mechanisms, immersion, and accessibility—using validated instruments [97,98]. Establishing shared data infrastructures that systematically document intervention characteristics, engagement metrics, and outcome measures would further enhance comparability and enable cumulative synthesis across studies. Strengthening these methodological and infrastructural foundations will be essential for advancing serious game research from isolated trials toward a coherent and reproducible scientific field.

The interpretation of this synthesis should take into account the methodological and conceptual variability among the included trials, which likely contributed to the substantial heterogeneity observed in the pairwise meta-analysis. Health education topics, measurement instruments, and feedback structures differed widely, complicating direct comparison of effect sizes across studies [30]. Although subgroup analyses identified several sources of heterogeneity, the diversity in outcome definitions and analytical strategies inevitably limited the precision of pooled estimates.

In several recent studies, particularly those published after 2020, technological advancements have led serious games, such as Food Adventure Quest and Amoo, to evolve beyond stand-alone formats, increasingly integrating complementary educational components, such as video segments and classroom instruction [62,85]. Although this convergence complicates the identification of the game’s independent effects, it reflects the growing trend toward multimodal approaches in health education. It indicates that serious games are becoming embedded components of broader digital learning ecosystems that combine interactive games, web-based modules, and instructor-led components.

Reporting transparency also varied across studies. Protocol preregistration and complete reporting of secondary outcomes were often absent, resulting in the overall methodological quality being rated as moderate, despite generally low risks of bias in randomization, intervention delivery, and outcome measurement [106]. These limitations indicate that the main threat to internal validity stems from incomplete documentation rather than flawed trial conduct, underscoring the need for preregistered protocols and comprehensive reporting standards in future evaluations of serious games.

This systematic review and Bayesian network meta-analysis of RCTs provides a comprehensive evaluation of the educational impact of digital serious games in public health. Unlike earlier reviews that focused on individual interventions or specific health topics, this study compares multiple digital serious game formats within a unified analytical framework. Across 40 trials, mobile-, computer-, and web-based formats generally produced the greatest improvements in knowledge outcomes, while computer-, web-, and virtual reality–based formats showed stronger effects for attitude change. Multisession interventions sustained learning and attitudinal change more effectively than single-session exposure, highlighting the importance of reinforcement and continued engagement. The overall certainty of evidence was moderate, reflecting methodological heterogeneity across trials. These findings contribute comparative evidence to the field of digital health education and offer practical guidance for selecting scalable serious game interventions in real-world public health programs. Strengthening implementation strategies, standardizing outcome evaluation, and extending trials to underrepresented adult and older populations are important next steps.