We conducted a parallel 2-arm RCT with a 6- to 10-week follow-up period. Participants in the intervention group had to download and complete the quest in the “SteWARdS Antibiotic Defence” app on their smartphones, while controls received no intervention.

We recruited patients and their caregivers from 2 government-funded primary care clinics (ie, polyclinics) in Singapore. Recruitment occurred from January to March 2023, and follow-up was completed in early June 2023. The 23 polyclinics in Singapore manage 20% of Singapore’s primary health care needs by providing subsidized services such as medical treatment for acute and chronic conditions, vaccinations, and health education [22]. Participants seen at the polyclinics mainly have acute conditions that are managed outside the health care setting. Hence, while our study recruitment and baseline data collection occurred in the policlinics, the study intervention and follow-ups were administered outside the health care setting.

Participants were 18-65 years of age, had access to an Android smartphone, and were English literate. We excluded those with visual or cognitive impairment, not proficient with smartphone apps, and those unable to install Android mobile apps from the study.

We initially planned to take the intention-to-treat approach in analyzing the study data but eventually took the per-protocol approach, as our app usage data showed that participants who did not complete the game quest had too few interactions with the app to derive any benefit from the intervention. We standardized the total Knowledge and Attitude-Perception scores to 100% to illustrate the pre-post changes in scores. Correct answers are coded as 1 while wrong answers are coded as 0. Similarly, the “correct” attitudes are coded as 1 while neutral and “incorrect” attitudes are coded as 0.

Chi-square tests and Mann-Whitney U tests were used for univariate baseline comparisons. Weighted least squares regressions (weighted on preintervention scores) were used to assess the effect of the intervention on knowledge and KAP scores. The best regression model was chosen based on the lowest Akaike Information Criterion. All statistical assumptions were checked to ensure the accuracy of analyses. STATA (version 15; StataCorp LLC) was used for all statistical analyses.

Participant feedback was analyzed qualitatively using thematic analysis. OJT telephoned participants to elaborate on any ambiguous feedback and categorized the feedback into themes. ZH reviewed and discussed the categorization with OJT until a consensus was reached.

This study was approved by the National Healthcare Group Domain Specific Review Board (2022/00479) in Singapore. Informed consent was obtained from participants after the data collector had fully explained the study to them. Each participant was given a unique research ID linked to their personal details (eg, name and contact details). Only National Healthcare Group Domain Specific Review Board–approved study team members have access to participant data. All identifiable information is stored in password-protected files in the institutions’ server. Data collected from the surveys and app contained only the participants’ research IDs. Participants were reimbursed SGD 5 (US $3.82) for completing the baseline and demographic survey, SGD 10 (US $7.64) for completing the postintervention survey, and SGD 25 (US $19.10) for completing the app intervention and satisfaction survey (intervention group participants only).

We approached 1985 patients and visitors of polyclinics for 3 months, and 57.6% (1144/1985) of patients consented to take part in our study. Of these consenting participants, 41.1% (470/1144) of them did not own an Android phone, 8.7% (100/1144) did not meet our age requirements, 7.0% (80/1144) of them were not proficient in the English language, 0.6% (7/1144) of them had difficulty downloading the app, and 0.6% (7/1144) of them had visual or cognitive impairment. We block-randomized 42.0% (480/1144) of participants who were screened for eligibility into 2 study arms. Of these randomly assigned participants, 85.8% (206/240) of them in the control group and 59.1% (142/240) of them in the intervention group completed the study and had their outcomes assessed. About 40.8% (98/240) of intervention group participants did not complete the game quest (16 withdrew; 82 dropped out) in the app (CONSORT [Consolidated Standards of Reporting Trials] flowchart shown in Figure 6).

Recruitment occurred between January 2023 and March 2023, and follow-up was completed by early June 2023. We ceased recruitment when we achieved 240 participants per arm.

Table 1 shows participants’ baseline characteristics. The randomization produced comparable groups as there were no significant between-group differences in participants’ sociodemographic characteristics at baseline. Participants had a mean age of 36.9 (SD 12.2) years. Our participant profile fits Singapore’s population profile in terms of ethnicity (252/348, 72.4% Chinese; 61/348, 17.5% Malay; and 24/348, 6.9% Indian), education level (128/348, 36.8% had tertiary education), and housing type (275/348, 79.0% residing in conventional public housing flats). There were also no significant between-group differences in the mean baseline knowledge score of 8.4/11 (SD 1.6), but the attitude-perception score was significantly higher in controls (mean 9.6, SD 2.6) than in intervention participants (mean 8.8, SD 3.0). In addition, 60.1% (209/348) of participants had no exposure to advertisements on antibiotic use at baseline.

The randomization produced comparable groups as there were no significant between-group differences in participants’ sociodemographic characteristics at baseline. Our participant profile fits Singapore’s population profile in terms of ethnicity and housing type.

The mean standardized knowledge score improved from 77.6% at baseline to 82.4% postgame and 83.5% 6-10 weeks post intervention among intervention group participants, while controls had a modest improvement of 75.4% from baseline to 78.1% at 6-10 weeks post baseline (Figure S1 in Multimedia Appendix 1).

The multivariable regression analyses (Table 2) showed that intervention group participants had greater improvement in knowledge than controls (between-group differences: 0.58, 95% CI 0.28-0.87) after accounting for age, educational level, and exposure to advertisements on antibiotics and AMR.

The mean standardized attitudes-perception score improved from 62.6% at baseline to 77.7% post intervention among intervention group participants, while controls had a modest improvement of 68.9% from baseline to 71.9% post intervention (Figure S1 in Multimedia Appendix 1). The proportion of intervention group participants with the “correct” attitudes and perceptions increased for all 14 statements post intervention (Figure S2 in Multimedia Appendix 1). However, intervention group participants continue to have misconceptions about antibiotic use after completing the game quest. The proportion of participants with correct responses for the statement “I need antibiotics to recover from serious symptoms from the common cold and flu” increased from 29.6% (42/142) pre- to 46.5% (66/142) postintervention, and for “I need antibiotics if I continue to have flu symptoms after two weeks” increased from 22.5% (32/142) pre- to 38.7% (55/142) postintervention.

The multivariable regression analyses (Table 3) showed that intervention group participants had greater attitudes-perception improvement than controls (0.98 [0.44-1.52]) after accounting for marital status and ethnicity.

The majority of participants agreed that the “SteWARdS Antibiotic Defence” app improved their awareness on antibiotic use (135/142, 95.1%) and AMR (136/142, 95.8%). About 73.9% (105/142) of the participants agreed that the app is easy to use, 70.4% (100/142) agreed that the app was enjoyable, and 85.2% (121/142) would recommend the app to others (Table 4).

Participants who would recommend “SteWARdS Antibiotic Defence” found the app user-friendly, fun, enjoyable, and interesting. Some participants also lauded the app’s unique learning modality and felt that the intervention addressed an important public health topic, raised their awareness of AMR, and improved their knowledge on appropriate antibiotic use (Table S3 in Multimedia Appendix 1).

Participants who would not recommend the app mainly had a poor user experience and were frustrated with technical difficulties in navigating the app. Some technologically astute players found the game boring, repetitive, and unfulfilling.

One participant found no value in recommending the app to friends or family member working in health care. Those who found the games difficult to play (mostly older adults) or experienced usability issues (eg, small fonts) also expressed dissatisfaction with the intervention.

Suggestions to improve the quality of “SteWARdS Antibiotic Defence” include the following (Table S4 in Multimedia Appendix 1):

We evaluated the effectiveness of an evidence-based serious game mobile app in improving the KAP of appropriate antibiotic use and AMR among Singapore residents. To our knowledge, this is the first RCT assessing the effectiveness of a serious game mobile app intervention for public education on appropriate antibiotic use and AMR. We also assessed the effectiveness of the app in short-term knowledge retention at 6-10 weeks post intervention, which has often been overlooked. Furthermore, we assessed the app in a real-world context in the community and not in the controlled environment of an experimental setting.

We observed a statistically significant effect size of 0.58 (95% CI 0.28-0.87) in knowledge improvement, which concurs with other serious game studies on chronic disease management among young people [26] and healthy lifestyle promotion [27]. However, a similar Singapore study on dengue prevention observed significant improvements of participants’ knowledge, attitudes, and practices from baseline but no between-group mean differences [28]. The difference in observations is likely attributed to the exposure of relevant information sources among controls in the dengue study, which attenuated the effects of the app intervention. Tertiary-educated adults were positively associated with higher postintervention scores. They may fare better at learning the more difficult AMR concepts as studies have shown that AMR concepts are hard to teach. Participants with no or limited understanding of the etiology of antibiotic resistance would face difficulty connecting AMR with antibiotic misuse [29]. Furthermore, misconceptions about antibiotic use are generally hard to rectify.

We also observed a statistically significant effect size of 0.98 (95% CI 0.44-1.52) in attitude-perception improvement with the “SteWARdS Antibiotic Defence” app. Other studies have found mixed results in the effectiveness of serious games in improving the adoption of healthy behaviors, such as fruit and vegetables intake among adolescents [15]. The effectiveness of serious games in changing attitudes and perceptions are hard to measure as studies are often multifaceted. Therefore, it is essential to have well-designed interventions targeted at the correct population, well-accepted by participants, and with attainable goals for interventions to be successful. While we recruited participants between 18 and 65 years of age, the app was originally developed for young adults as this group of people was found to have the lowest knowledge in antibiotic use and AMR [7].

Our “SteWARdS Antibiotic Defence” app received an overall positive rating, but the ratings should be interpreted with caution as the satisfaction survey was only completed by participants who finished the game quest in the app. It is unclear whether participants dropped out of the intervention due to dissatisfaction with the app or a lack of time to complete the quest. Other studies assessing the effectiveness of serious games in health education have observed high satisfaction among participants [30,31]. However, there is limited evidence on the satisfaction with health education delivered via serious games in the general population and even fewer studies report participant satisfaction. Hence, the feedback and suggestions we gathered from participants are valuable in identifying population-specific shortcomings of the intervention and informing app enhancements that could increase the success of intervention scale-ups.

Despite a modest improvement in KAP scores in the intervention group, improved health literacy may not translate to desired behavioral change [13,32]. Health literacy interventions have led to improved health outcomes in some studies, but the links between health literacy improvement and behavioral change have not been established with higher-quality studies [33]. Effective interventions are also integral to attaining successful behavioral change. One way to improve the learning outcomes, suggested by our participants, is to weave the intended learning concepts into the game mechanics rather than displaying informational text boxes for players to read. Such enhancements would possibly improve user engagement and lead to better learning outcomes [34].

A few limitations exist in our study. First, measurement errors in KAP scores might occur if intervention group participants had requested their friends or family members to help with study completion. Nonetheless, controls could also have had help in completing the postintervention survey. Therefore, any bias which attenuates the intervention's true effect is likely toward the null. Second, there is a ceiling effect for participants with near-perfect baseline scores, which could attenuate the positive effects of the intervention. Third, we randomized participants prior to baseline data collection because the knowledge questions were embedded in the app. Fourth, there may be a learning effect in the intervention group as these participants completed the knowledge questionnaire 3 times. Hence, the positive effects observed in the intervention group could be a combination of the serious game and learning effect from the immediate postgame survey. Including a study group with a traditional form of learning (eg, brochures) may enable us to assess the effect of the serious game app more accurately. Finally, we included only attendees at 2 government-funded primary care clinics (polyclinics). Despite the limitations in recruitment, the sociodemographic distribution of our participants is representative of the Singaporean population. Hence, our findings are generalizable to the rest of the country.

In addition, we noticed several older adults rejecting study participation due to a lack of interest in completing the mobile app quest. Their lack of interest implies an additional effort to change their mindsets, and serious game interventions may not be the best modality for educating older adults on public health issues. Reminders were also integral in driving the app completion rate. Hence, the actual uptake rate may be lower in the population. Future research should assess the effectiveness of interventions that facilitate long-term knowledge retention and long-lasting behavioral change in antibiotic use and AMR [35].

Our educational serious game app improves participants’ KAP on appropriate antibiotic use and AMR. Public education apps should be engaging, educational, easy to use, and have an attractive user interface. Interventions should also aim to achieve long-term knowledge retention and long-lasting behavioral change.