Cardiovascular diseases (CVDs) are a group of disorders affecting the heart and blood vessels that significantly contribute to premature mortality and escalating health care expenses [1,2]. According to the Global Burden of Disease 2021 Study, the global prevalence of CVD cases has reached 612 million [3]. Regular physical activity (PA) is widely recognized as one of the most effective lifestyle interventions for managing CVD risk factors, slowing disease progression, and reducing CVD-related mortality [4]. The World Health Organization guidelines on PA recommend that individuals with CVD engage in 150-300 minutes of moderate-intensity PA, 75-150 minutes of vigorous-intensity PA, or an equivalent combination of moderate-to-vigorous intensity PA per week [5]. Moreover, as self-reported sedentary behavior (SB) is independently associated with an increased risk of CVD, irrespective of PA levels [6,7], the guidelines emphasize minimizing sedentary time and replacing it with any intensity of PA to achieve health benefits [5].

Adherence in the health domain is defined as “the extent to which a person’s behavior—taking medication, following a diet, and/or executing lifestyle changes—corresponds with agreed recommendations from a health care provider” [8]. Despite the well-documented benefits of PA, studies indicate that many individuals with CVD demonstrate poor adherence, often failing to meet the PA levels recommended by guidelines [9,10]. Promoting health behavior change and improving PA adherence in individuals with CVD is challenging due to various factors, including lack of motivation or self-efficacy, cognitive or physical limitations, and other barriers [11]. In response, there is a growing call to shift the focus of PA from its health-promoting utility to emphasizing the personal experience [12]. Since emotion plays an essential role in driving behavior, creating opportunities for individuals with CVD to “feel good” while engaging in PA and building a connection between pleasurable feelings and the activity may provide an effective strategy to enhance PA adherence.

Gamification, defined as the use of game design elements in nongame contexts [13], has shown promise in promoting adherence to PA among adults [14], offering a novel approach to modifying health behaviors [15]. Notably, studies have demonstrated that gamification is not only enjoyable for individuals with CVD but also increases the pleasure of engaging in PA [16] and supports maintaining a high level of adherence [17]. In recent years, gamification has been increasingly integrated into mobile health (mHealth), which leverages mobile computing and communication technologies—such as mobile phones and wearable sensors—to deliver medical services and information [18]. With its powerful capabilities for sensing, processing, storing, and displaying data, mHealth enables continuous tracking and collecting of PA-related information, extending gamification into daily health behaviors [15,19]. The combination of gamification and mHealth is mutually reinforcing, providing opportunities to enhance the quality and experience of CVD care [20].

Although gamification has been studied as a strategy to improve PA- and health-related outcomes in individuals with CVD [16,17,21], no systematic review on this specific topic has been published to date. Existing systematic reviews have primarily examined the application of gamification in populations with conditions such as hypertension, excess weight, and diabetes [22,23]. However, these reviews did not exclusively focus on secondary prevention or PA outcomes in individuals with established CVD. Notably, these reviews highlighted the necessity for high-quality studies [23]. They emphasized the need for randomized controlled trials (RCTs) to identify effective and acceptable gamification interventions for the self-management of CVD [22]. Furthermore, evidence regarding the effectiveness of mHealth-based gamification interventions in improving PA among individuals with CVD remains inconclusive. Given the growing body of RCTs in this area, a systematic review and meta-analysis focused on secondary prevention of CVD appears timely and warranted to address these gaps.

This study aims to quantify the effects of gamification interventions on PA in individuals with established CVD. Effective gamification outcomes should extend beyond short-term novelty effects measured immediately after the intervention to include lasting impacts assessed at the end of a predefined follow-up period [24]. Therefore, we evaluated the effects of PA both postintervention and at the end of follow-up periods as defined by the included studies. Additionally, prediction intervals (PIs) were reported alongside CIs in the meta-analyses. PIs provide an estimate of the range within which future individual observations will likely fall, offering insights often overlooked [25]. IntHout et al [26] demonstrated that implementing PIs in over 400 published meta-analyses led to completely opposite effects in more than 20% of cases. Last, a fundamental limitation of current gamification research is the inability to determine which game design elements contribute most to its efficacy [16,22]. This highlights the need for further research to identify and isolate the most active and effective gamification elements [27].

This systematic review and meta-analysis aim to address key research gaps by (1) assessing the impact of gamification interventions on PA in individuals with CVD, (2) evaluating the follow-up effects of these interventions, (3) reporting the PIs to estimate the potential intervention effects in future studies, and (4) determining the most effective game design element for influencing PA behaviors.

Eligible studies met the following criteria:

As the term “gamification” gained widespread adoption in 2010 [13], we set this year as the starting point of our search. We systematically searched 7 electronic databases—Ovid MEDLINE, PubMed, Web of Science, Embase, Scopus, Cochrane, and CINAHL—for studies published between January 1, 2010, and February 3, 2024. To ensure comprehensive coverage, we manually screened reference lists and studies included in relevant systematic reviews and meta-analyses for additional eligible studies.

The search strategy, informed by previous systematic reviews [19], targeted 6 key topics: CVD, gamification, mHealth, PA, SB, and RCTs. It was initially formulated for Ovid MEDLINE and subsequently adapted for use in the other databases.

The search results were exported into EndNote (version 20.6) for document management, and duplicates were removed both automatically and manually. In total, 2 independent reviewers (Tianzhuo Y and Tianyue Y) screened the retrieved records by title, abstract, and full text to identify potentially relevant studies. A third reviewer (FL) resolved any disagreements between the reviewers.

In total, 2 independent reviewers (Tianzhuo Y and Tianyue Y) extracted and verified the data, with FL arbitrating disagreements. Data were organized into preestablished Microsoft Excel sheets, capturing study characteristics (author, year, and country), participant details (sample size, age, sex, ethnicity, and diagnosis), and outcomes (measurement tools, methods, units, data at different time points). The intervention details were described using the Template for Intervention Description and Replication (TIDieR) checklist [29], covering aspects such as Why (theoretical framework), What (materials, procedures, and game design elements of gamification intervention), Who provided (intervention provider), How (mode of delivery), Where (location), When and How much (duration and frequency), Tailoring (eg, individualized goal setting), Modifications, and How well (adherence and attrition). Game design elements were categorized based on the gamification persuasion architecture and its 7 persuasion strategies [24]. Additionally, the checklist was used to assess the reporting completeness of each intervention, with items rated as “present,” “absent,” or “unclear.”

The risk of bias was assessed using the Revised Cochrane risk-of-bias tool for randomized trials (RoB 2), which evaluates 5 aspects: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcomes, and selection of the reported results [30]. In total, 2 reviewers (Tianzhuo Y and Tianyue Y) independently performed the assessments, and any discrepancies were resolved through discussion with a third reviewer (FL).

Studies reporting PA data as mean (SD) or mean differences with SD of the differences were eligible for inclusion in the meta-analysis. Data reported as medians and interquartile ranges were also included after transformation into means and SD [31].

Characteristics and TIDieR findings were synthesized narratively, with frequencies and percentages summarized in tables. This qualitative review included all studies meeting the eligibility criteria, even if data could not be obtained for quantitative analysis. Statistical analyses were conducted using R (version 4.3.2). An overall meta-analysis was performed to determine the summary effect, with additional meta-analyses for follow-up effects and specific outcomes (eg, daily steps) when sufficient data were available. The standardized mean difference was applied to calculate effect sizes for continuous variables measured with different instruments, while the mean difference was used for those measured with similar instruments. To address small-sample bias in the included studies, we calculated the effect size Hedges g [32], which adjusts Cohen d for small-sample bias. A Hedges g of 0.20 indicates a small effect, 0.50 a moderate effect, and 0.80 a large effect [33]. Given the expected between-study heterogeneity, a random-effects model was used to pool effect sizes. To mitigate potential biases associated with the DerSimonian-Laird method, particularly in cases of few studies and high heterogeneity [34], the Sidik-Jonkman method with Knapp-Hartung adjustments was used for more robust variance estimates [35]. According to the Cochrane Handbook [36], when a study included multiple intervention groups and aimed to compare the effects of different interventions, each intervention group was treated as a separate study. The “shared” control group was split into 2 or more groups with reduced sample sizes for comparison. Only the total number of participants in the control group was divided for continuous outcomes, while the mean and SD remained unchanged.

To investigate sources of heterogeneity, we first identified potential outliers, defined as studies with effect sizes that were extreme and significantly deviated from the overall effect [37]. An influence analysis was then performed using the leave-one-out method to determine which studies exerted the greatest impact on the pooled estimate and to assess whether this influence distorted the overall results [38]. Additionally, we used a Baujat plot, a diagnostic tool designed to identify studies that contribute disproportionately to heterogeneity in a meta-analysis [39]. Finally, a sensitivity analysis was conducted to evaluate the robustness of the findings by excluding studies with a high risk of bias, significant heterogeneity, or identified outliers.

The I² statistic reflects the proportion of variance in observed effects attributable to variance in true effects but does not indicate the extent to which effect sizes vary across studies. Consequently, categorizing heterogeneity as low, moderate, or high based solely on the I² statistic is not recommended [40]. In this meta-analysis, PIs were used to quantify heterogeneity [38]. PIs represent the range within which the effect size of a new study, randomly chosen from the same population as those included in the meta-analysis, is likely to fall [41]. Unlike CI, PIs provide valuable clinical decision-making insights by estimating the likely intervention effect in future studies [25]. PIs also use the same scale as the effect size, illustrating both the interval’s width and limits, which helps determine whether the intervention consistently produces beneficial effects or has the potential for harm. PIs can be calculated when a meta-analysis includes at least 3 studies [38] and are most suitable when the included studies have a low risk of bias [36]. Therefore, we calculated 95% PIs for each meta-analysis.

To evaluate the most effective game design elements, we performed a meta-regression using a multimodel inference approach. Game design elements were treated as predictors to identify the best combination and the most influential predictor overall [42]. We used the Sidik-Jonkman random-effect model to pool effect sizes and the Knapp-Hartung adjustment method to calculate the test statistic and CI. The small sample-correction Akaike’s information criterion (AICc) was applied as the evaluation criterion for the fitted model.

We assessed the quality of the evidence using the web-based version GRADEpro GDT (Grading of Recommendations Assessment, Development and Evaluation professional guideline development tool). Although RCT evidence typically starts at a high-quality rating, we evaluated the following domains to determine any necessary downgrades: (1) risk of bias, (2) inconsistency, (3) indirectness, (4) imprecision, and (5) publication bias [43].

We made the following protocol deviations to facilitate a more comprehensive synthesis of gamification interventions for individuals with CVD. First, the TIDieR checklist was used for narrative synthesis. Second, the Revised Cochrane risk-of-bias tool was used for bias assessment. Third, effect sizes were calculated using Hedges g, with CI adjusted via the Sidik-Jonkman method and the Knapp-Hartung adjustment. Fourth, all analyses were conducted using R (version 4.3.2). Last, outlier detection and influential analysis were performed to explore sources of heterogeneity instead of subgroup analyses, and 95% PIs were plotted to measure interstudy heterogeneity in the meta-analysis. No additional protocol deviations from the PROSPERO (International Prospective Register of Systematic Reviews) registration were made.

The initial search yielded 1060 records. The detailed search strategies for each database are provided in Table S1 in Multimedia Appendix 1. Additionally, 188 records were identified through manual searches of reference lists and studies included in relevant systematic reviews. After removing duplicates and screening titles and abstracts, 22 full-text papers were assessed for eligibility. In total, 6 studies met the inclusion criteria and were included in the narrative synthesis. Among them, 5 studies were eligible for inclusion in the meta-analysis. The study selection process is illustrated in Figure 1.

In total, 4 studies focused on steps [16,21,44,45], while others measured moderate-to-vigorous PA [44], quantity in metabolic equivalent of task [27], and walking test distance [17]. Except for one study [27], which used the self-reported Global Physical Activity Questionnaire to assess total PA, all other studies used objective measurement tools, including wearable devices (n=3) [21,44,45], smartphone accelerometers [16], or the 6-minute walking test [17]. A detailed summary of PA-related outcomes is presented in Table S6 in Multimedia Appendix 1.

Among the 6 included studies, 5 had a low overall risk of bias, and one raised some concerns [17] due to not using appropriate analyses to estimate intervention effects (see Figure S1 in Multimedia Appendix 1). All studies were at low risk of bias in domains of the randomization process, missing outcome data, outcome measurement, and selection of the reported results. Detailed risk of bias assessments are provided in Table S7 in Multimedia Appendix 1.

In the short-term PA analysis, substantial statistical heterogeneity was observed. To this, we first identified outliers, with one study by Paldán et al [17] flagged as an outlier. Leave-one-out analyses revealed that sequential removal of each study did not significantly affect the overall effect size, which ranged from a Hedges g of 0.32 (95% CI 0.19-0.45) to 0.39 (95% CI 0.12-0.66; see Figure S2 in Multimedia Appendix 1). The Baujat plot also identified the study by Paldán et al [17] as contributing most to the heterogeneity (influence value is 14.61; see Figure S3 in Multimedia Appendix 1). Given its outlier status and concerns about the risk of bias, we excluded this study from the sensitivity analysis. After exclusion, we still obtained a statistically significant effect of a Hedges g of 0.32 (95% CI 0.19-0.45), representing a small effect with minor heterogeneity (95% PI 0.02-0.62; see Figure S4 in Multimedia Appendix 1).

Meta-regression using the multimodel inference method identified the top 5 models, ranked by increasing AICc values, with “Feedback plus Goals” having the lowest AICc (4.0) and thus showing the best fit. However, other predictor combinations had similar AICc values, making it difficult to determine a definitive “best” model. Notably, all top 5 models included the predictor “Feedback,” suggesting it may be particularly important. The top 5 models are summarized in Table S8 in Multimedia Appendix 1. As shown in Figure S5 in Multimedia Appendix 1, only 10 of the 11 game design elements were included in the predictor importance plot, as none of the studies in the meta-analysis used the element “Social support.” The plot illustrates the average importance of each predictor across all models, highlighting “Feedback” as the most important predictor (importance value is 0.71), followed by “Avatars” (importance value is 0.59).

After conducting the sensitivity analysis, the quality of evidence for short-term PA, predefined follow-up PA, and daily steps was rated as high (see Table S9 in Multimedia Appendix 1).

In total, 6 studies were included in this review, all published within the past 3 years, reflecting the growing interest in gamification interventions in the field of CVD. This meta-analysis of 1075 participants from 5 RCTs demonstrated that mHealth-based gamification interventions had a statistically significant effect on overall PA over a mean intervention duration of 3.8 months (Hedges g=0.32, 95% CI 0.19-0.45 after sensitivity analysis), indicating that gamification interventions effectively promote PA in individuals with CVD. This effect remained robust even after conducting different influence analyses, ie, the leave-one-out method.

These findings align with those of Mazeas et al [19], who assessed the impact of gamification interventions on PA in a broader population. Their study, which included healthy individuals and those with chronic diseases, demonstrated that a 12-week gamification intervention statistically significantly improved PA levels (Hedges g=0.42, 95% CI 0.14-0.69 after sensitivity analysis). However, the mean age of participants in Mazeas et al’s study was 35.7 years [19], raising questions about the generalizability of results to older individuals [46]. In contrast, the present meta-analysis focuses on participants with a mean age exceeding 59 years, addressing, to some extent, the research gap regarding the effects of gamification interventions on PA in older individuals. Nonetheless, future research is needed to comprehensively evaluate the impact of gamification interventions on older adults [47].

When analyzing the predefined follow-up effects of gamification interventions on PA, we observed a statistically significant effect size (Hedges g=0.20, 95% CI 0.12-0.29) at a mean follow-up time of 2.5 months after the intervention ended. These findings suggest that the effects of gamification persist beyond the intervention period, although the maintenance effect diminishes over time. Similarly, Mazeas et al [19] reported a statistically significant effect at a longer follow-up period (mean 3.6 months) with a weaker effect size (Hedges g=0.15, 95% CI 0.07-0.23 after sensitivity analysis). This consistency between studies indicates that gamification is not merely a short-term novelty effect for increasing PA. However, additional research is warranted to explore the longer-term sustainability of gamification interventions, particularly beyond 6 months.

The impact of the gamification intervention on daily steps was similarly promising. The meta-analysis demonstrated a statistically significant effect of mHealth-based gamification interventions, with participants in the intervention group increasing their daily steps by 696.96 compared to the control group (Hedges g=696.96, 95% CI 327.80-1066.12). In contrast, Mazeas et al [19] reported a larger increase of 1609.56 steps per day among participants in their gamification group. This discrepancy could be attributed to differences in participant characteristics; this review exclusively included individuals with CVD, who are generally less physically active than those without CVD [48]. Nevertheless, individuals with CVD often derive greater health benefits from the same level of PA [49], which underscores the potential meaning of our findings.

Furthermore, while current guidelines emphasize the intensity of PA [5], Oja et al [50] concluded that even moderate walking intervention yields cardiovascular health benefits. Paluch et al [51] demonstrated that higher daily steps were associated with a progressively lower risk of CVD among older adults. Banach et al [52] found that an increase of just 500 steps per day was linked to a 7% reduction in cardiovascular mortality. Based on the PI, in a new study, the probability that gamification interventions would increase daily steps by at least 500 in individuals with CVD is estimated to be 59.8%. This highlights the potential of gamification interventions to statistically and clinically reduce CVD risk by promoting higher daily steps.

The meta-analyses conducted in this review all demonstrated statistically significant effects, with 95% CI consistently located on the same side of the null. However, due to between-study heterogeneity, the corresponding 95% PI spanned both sides of the null. While most PIs were primarily to the right of the null, indicating that gamification interventions are generally effective, the overlap with the null suggests that their effectiveness may vary in certain contexts. The uncertainty reflected by the PI is tied to how closely future studies align with the characteristics of the completed studies [26]. This indicates that optimizing gamification interventions based on existing studies may help minimize null results in future research.

This review has 4 main limitations. First, the small number of included studies, some of which were feasibility or pilot studies, resulted in small sample sizes and underpowered analyses. However, this limitation reflects the emerging momentum of gamification interventions as secondary prevention for CVD. Second, all studies incorporated multiple gamification elements, making it challenging to isolate the effects of individual elements. While we conducted meta-regression multimodel inference to explore this, the results remain exploratory. Further research focusing on the independent effects of specific gamification elements is needed. Third, although we had an interest in exploring whether gamification can reduce SB, none of the included studies evaluated that outcome. Designing gamification interventions targeting SB may be a valuable direction for future research. Finally, none of the studies conducted cost-effectiveness analyses, which are crucial for evaluating the public health impact of gamification interventions and guiding resource allocation. Incorporating cost-effectiveness analyses in future studies could provide evidence to support the adoption of more efficient interventions and replace fewer effective ones [94].

In conclusion, gamification interventions show promise in promoting PA, particularly in increasing daily steps among individuals with CVD. While the effects may diminish over time, their persistence during follow-up suggests that gamification is not merely a novelty effect. However, attributing effects to individual game design elements remains challenging, as no studies have independently tested their impacts. Future studies should include larger sample sizes, longer durations, and more rigorous designs to further explore the effectiveness and persistence of gamification interventions and the impact of individual game design elements.