No patients or members of the public were formally involved in setting the research questions, outcome measures, intervention design, or recruitment procedures for this trial. A total of 30 community-dwelling older adults participated in the randomized controlled trial. A convenience subsample of 10 participants contributed to the concurrent validity substudy. Participants served only as study participants and were not involved in study governance or dissemination planning.

This 2-phase study was conducted at the Hongqi Community Center in Shanghai in April-May 2025. Participant enrollment for the dataset reported here took place between April 20 and May 20, 2025. It used a sequential 2-phase mixed methods design. Phase 1 used a within-subjects concurrent validity design with repeated measures. Phase 2 used a parallel-group randomized controlled trial design with 1:1 allocation, supplemented by qualitative semistructured interviews that carefully follow the CONSORT (Consolidated Standards of Reporting Trials) guideline (Checklist 1) [25]. The Phase 2 randomized controlled trial was retrospectively registered at ClinicalTrials.gov (NCT06958653) after enrollment began (registration date: July 17, 2025). The study team initially categorized the Phase 2 component as a low-risk formative evaluation focused on user experience of a digital assessment interface and did not recognize that any randomized allocation of participants requires prospective trial registration under International Committee of Medical Journal Editors (ICMJE)/JMIR policy. Once this requirement was identified during manuscript preparation, we registered the trial and are updating the registry record to transparently reflect the outcomes and their timing as collected and reported in this manuscript. This research collected written informed consent from all participants and was reported following the Journal Article Reporting Standards (JARS) for mixed methods research [26].

Community-dwelling older adults were recruited offline from the Hongqi Community in Shanghai through community announcements, poster displays at the community center, and referrals from community health care workers. Interested individuals were screened for eligibility via telephone by a trained research assistant using a standardized eligibility checklist. Eligible participants were then invited for an in-person informed consent session prior to enrollment. Inclusion criteria for both phases were (1) age ≥60 years, (2) ability to stand unassisted for 30 seconds and walk 10 m independently, and (3) no acute illness at the time of testing. Exclusion criteria included (1) diagnosed neurological conditions affecting balance, (2) uncorrected visual or auditory impairments, (3) recent fall requiring medical attention, and (4) use of medications known to affect postural stability.

For Phase 1, a sample of 10 participants was selected for reliability studies. For Phase 2, sample size was determined a priori using G*Power 3.1 software (Heinrich Heine University Düsseldorf). Assuming a 2-tailed independent t test, α=.05, power (1-β)=.80, and a medium-to-large effect size (Cohen d=0.55) based on prior gamification studies in older adults [21], the required sample size was calculated as n=28. We recruited 30 participants (15 per group) to account for potential dropouts, though none occurred.

Systematic assessment of harms was not a primary objective of this study given the low-risk nature of balance testing in a controlled environment with safety measures. However, we monitored for adverse events during all testing sessions. Potential harms were defined as (1) falls or near-falls during balance tasks and (2) musculoskeletal discomfort or pain exacerbation. All testing was conducted with safety railings present and a trained researcher monitoring participants. No adverse events occurred during the study period. Participants were instructed to stop immediately if they experienced any discomfort, and emergency protocols were established in consultation with local health care providers.

The study was conducted in 2 sequential phases to evaluate the reliability and user experience of a GDBA tailored for community-dwelling older adults.

Data collection was conducted by trained research personnel who completed a standardized 4-hour training session on Brief-BESTest administration, equipment operation, and participant interaction protocols. Training included video-based instruction, practice sessions with pilot participants, and competency assessment by the principal investigator. Interrater reliability was established by having 2 certified physiotherapists (both with >3 years’ experience in geriatric assessment) independently score 20% (2/10) of Phase 1 participants, achieving excellent agreement. The digitalized Brief-BESTest system underwent technical validation prior to data collection, including calibration testing with known reference objects to verify measurement accuracy. For qualitative interviews, the primary interviewer (JZ) received training in semistructured interview techniques and motivational interviewing principles. Interview fidelity was maintained through use of a standardized interview guide and audio recording of all sessions for quality review.

Reliability in this sample: internal consistency was assessed for all multi-item scales in this sample (n=30, Phase 2). The IMI subscales demonstrated good internal consistency, including interest and enjoyment (Cronbach α=.87; 95% CI 0.76‐0.93), perceived competence (α=.84; 95% CI 0.71‐0.91), and pressure and tension (α=.79; 95% CI 0.63‐0.88). The Brief-BESTest total score showed excellent internal consistency in this current sample (α=.91; 95% CI 0.83‐0.95). Test-retest reliability was not assessed in this study due to the single-session design. However, the digitalized version demonstrated excellent intrasession reliability (ICC=0.89‐0.92) across multiple automated scoring iterations of the same video recordings.

Convergent validity: convergent validity of the digitalized Brief-BESTest was supported by strong correlation with the clinician-administered version (Spearman ρ=0.91; P<.01), suggesting both methods measure the same construct.

All quantitative analyses were conducted using SPSS (IBM Corp). To assess the validity of the digitalized Brief-BESTest, balance scores from the clinician-administered and algorithm-generated assessments were compared using paired t tests or Wilcoxon signed-rank tests, depending on normality. Effect sizes were calculated using Cohen d. Interrater reliability for the clinician-administered Brief-BESTest was assessed on 20% of cases scored by 2 clinicians using the intraclass correlation coefficient (ICC, 2-way mixed model, absolute agreement). ICC values >0.75 were considered excellent, 0.40‐0.75 moderate to good, and <0.40 poor. Convergent validity was examined via Spearman correlation (ρ).

In the second phase, nonparametric tests (Mann–Whitney U) were used to compare user experience outcomes—perceived exertion, intrinsic motivation, and intention to continue use—between the experimental group and the control group. IMI subscales (interest/enjoyment, perceived competence, and pressure/tension) were analyzed separately. Significant effects were reported with Cohen d. Qualitative interview data were analyzed thematically [34] by the primary interviewer (JZ, female, PhD in digital health; CC, male, MSc in health informatics). The coding team included 2 independent analysts. Neither had prior relationships with participants. Reflexivity measures included maintaining research journals and weekly debriefing sessions during the 3-week analysis period. Analysis followed six phases: (1) familiarization, both coders independently read all 30 transcripts twice; (2) generating initial codes, line-by-line coding in NVivo 14 (Lumivero; Release 1.7.1) produced 156 initial codes; (3) searching for themes, related codes were grouped into 8 candidate themes through iterative discussion; (4) reviewing themes, candidate themes were reviewed against the dataset for internal coherence and external distinction; (5) defining themes, final themes were refined with clear operational definitions; and (6) producing the report, exemplar quotes were selected to illustrate each theme. Intercoder agreement was moderate (Cohen κ=0.68); all discrepancies were resolved through consensus discussion. Data saturation was assessed iteratively; no new themes emerged after interview 25 (experimental group: interview 13; control group: interview 12), suggesting thematic saturation was achieved. Member checking was not conducted due to the brief time frame and immediate postassessment interview design; however, findings were triangulated with quantitative data to enhance credibility.

This study was approved by the Institutional Review Board of Shanghai Jiao Tong University (approval number: H20240601I, approved on January 12, 2024). The randomized controlled trial was retrospectively registered at ClinicalTrials.gov (NCT06958653) on July 17, 2025, after enrollment began on April 20, 2025. The delay occurred because the study team initially categorized the randomized component as a low-risk formative evaluation of user experience for a digital assessment interface and did not recognize that randomized allocation requires prospective registration under ICMJE/JMIR policy; once identified during manuscript preparation, the ClinicalTrials.gov record has been corrected/updated to reflect the RCT sample size (n=30) and the substudy sample size (n=10), to designate user experience as the primary outcome domain, and to list the ABC scale as a baseline characteristic rather than an outcome, with the primary outcomes being user experience measures (perceived exertion, intrinsic motivation, and intention to continue use).

All procedures were conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent after receiving detailed verbal and written explanations of the study purpose, procedures, potential risks, and benefits. All participant data were deidentified using unique numeric codes. Video recordings captured during balance assessments contained only skeletal joint position data and did not record identifiable facial features or other personal characteristics. All images included in the manuscript are representative schematics or composite visualizations created from anonymized skeletal data; no actual participant photographs are shown. The interface screenshots and environmental setup photos were taken in empty testing environments without participants present, ensuring no individual identification is possible. All participants were informed that no identifiable images would be captured or published. Participants received 70 Chinese Yuan (approximately US $10) as compensation for their time upon completion of the study session. To minimize fall risk during balance assessments, all testing was conducted in a designated area equipped with safety railings on 3 sides.

In the first phase, 10 older adults (mean age 64.9, SD 2.76 years; 5 men, 5 women) participated; 5 were excluded due to recent falls requiring medical attention (n=4) and declined to participate (n=1). Of these, 30% (3/10) reported a prior history of falls. The mean BMI was 26.7 kg/m² (SD 3.1).

Balance performance scores from the clinician-administered Brief-BESTest and the digitalized Brief-BESTest were 14.3 (SD 3.24) and 14.53 (SD 3.03), respectively. A paired-sample t test revealed no significant difference between the 2 assessments (P=.77; Cohen d=–0.07), indicating strong similarity in 2 sets of results.

Absolute reliability, assessed via the SE of measurement, demonstrated low variability in scores. Relative reliability was reasonable across all 6 Brief-BESTest domains for both assessment methods. ICCs for the clinician-administered version ranged from 0.84 to 0.96 (ICC=0.92; 95% CI 0.84‐0.96). For the digitalized version, ICCs ranged from 0.78 to 0.94 (ICC=0.89; 95% CI 0.78‐0.94). Mean differences (MDs) between the 2 test scores ranged from 0.2 to 0.4 points, indicating high consistency.

Convergent validity analysis, as shown in Figure 3, showed a strong and significant positive correlation between the 2 assessment methods (Spearman ρ=0.91; P<.01), confirming that the digitalized Brief-BESTest provides results comparable to the clinician-administered Brief-BESTest.

A total of 30 older adults (age mean 66.7, SD 3.93) were recruited for the study and randomized into 15 in the experimental group and 15 in the control group. The recruited participants consisted of 12 men and 18 women, with 43.34% reporting a history of falls. The BMI (kg/m²) is a mean of 26.73 (SD 3.11), with no significant difference between the experimental group and the control group (P=.45). All older adults successfully and independently completed the balance tests, with none of the participants missing in the second round of testing. The baseline information is specified in Table 1. Meanwhile, the balance ability for the control group (mean 8.67, SD 4.12) and the experimental group (mean 19.4, SD 3.60; P=.61) as well as the balance confidence for the control group (mean 0.72, SD 0.18) and the experimental group (mean 0.84, SD 0.17; P=.08) were assessed. Indicating similar balance ability and balance confidence between 2 groups of participants. No harm was documented in the experiment. The primary outcomes were user experience measures (perceived exertion, intrinsic motivation, and intention to continue use). The experiment results are specified in Table 2. The CONSORT flow diagram is specified in Figure 4.

The study examined the perceived exertion level of participants using the Borg RPE scale. The control group reported a mean perceived exertion of 12.60 (SD 2.75; 95% CI 11.1‐14.1), while the experimental group reported 9.93 (SD 2.12; 95% CI 8.8‐11.1). Participants using the GDBA had significantly lower perceived exertion levels compared to those using the digitalized Brief-BESTest (MD –2.67; 95% CI –4.60 to –0.74; P=.01; Cohen d=–1.08), representing a large effect size.

The IMI subscales were used to measure participants’ enjoyment and interest, perceived competence, and pressure and tension during the testing process.

For perceived enjoyment and interest, the experimental group reported a significantly higher level (mean 4.26, SD 1.33; 95% CI 3.5‐5.0) compared to the control group (mean 2.73, SD 1.28, 95% CI 2.0‐3.5; MD 1.53; 95% CI 0.44-2.62; P=.009; Cohen d=1.17), indicating a large effect size.

In terms of perceived competence, the experimental group scored higher (mean 4.27, SD 1.33; 95% CI 3.5‐5.0) than the control group (mean 3.13, SD 1.19; 95% CI 2.5‐3.8). The use of the GDBA resulted in a significant improvement in perceived competence (MD 1.14; 95% CI 0.19-2.09; P=.02; Cohen d=0.89), representing a large effect size.

In the dimension of perceived pressure and tension, the experimental group (mean 4.13, SD 1.19; 95% CI 3.5‐4.8) reported a lower sense of pressure and tension compared with the control group (mean 3.40, SD 1.12; 95% CI 2.8‐4.0), though the difference was not statistically significant (MD –0.73; 95% CI –1.60 to 0.14; P=.09; Cohen d=0.63).

Qualitative interview data provided convergent evidence supporting the quantitative findings. The significantly higher enjoyment scores in the GDBA group (quantitative) were corroborated by participants’ qualitative descriptions of the experience as “fun,” “engaging,” and “motivating.” Specifically, 12 of 15 (80%) GDBA participants mentioned real-time visual feedback as a key source of enjoyment, stating it made the assessment feel “like a game” rather than a clinical test. In contrast, control group participants described the digitalized Brief-BESTest as “straightforward” or “clinical,” with 10 of 15 (67%) stating that the lack of interactive elements made the experience feel “monotonous.”

The quantitative finding of lower perceived exertion in the GDBA group was explained qualitatively by participants’ reports that the visual tutorial and feedback helped them “learning faster on what they should do” and “forget about the effort,” suggesting reduced attention to physical strain. Several GDBA participants explicitly stated, “I didn’t feel tired because I was concentrating on understanding each task.”

The significantly higher intention to continue use (quantitative) was illuminated by qualitative themes emphasizing motivational rewards and perceived usability. As one participant noted, “If I could check my balance regularly and see improvements over time, I would definitely use it every few months.” Conversely, control group participants identified specific barriers to continued use, most frequently citing “hard to understand,” “lack of variety,” and “minimal feedback” as reasons they would not return. This triangulation of quantitative and qualitative data strengthens confidence in the conclusion that gamification elements meaningfully enhance engagement and motivation for digital balance assessment among older adults.

This study pursued two prespecified objectives: (1) to validate a computer vision–based, digitalized Brief-BESTest against a clinician-administered reference standard and (2) to test whether a gamified interface improves user experience during balance assessment among community-dwelling older adults without compromising assessment performance. Both hypotheses were supported. Phase 1 demonstrated that automated scoring by the digitalized system was statistically equivalent to expert clinician scoring and showed strong concurrent validity, consistent with the hypothesis of adequate reliability and validity. Phase 2 showed that participants assigned to the GDBA reported more favorable user experience across all primary outcomes—including lower perceived exertion, higher intrinsic motivation, and greater intention to continue use—while balance performance was comparable between the groups.

The validation findings are notable because much of the existing literature on camera-based or remote balance testing has concentrated on single-task paradigms, such as the TUG or sit-to-stand tests [10,16], which provide valuable but limited information about postural control. In contrast, the Brief-BESTest assesses 6 theoretically distinct domains of balance [35,36] and its construct validity and fall discrimination ability have been supported across community and institutionalized populations [37]. Transforming such a multidomain instrument into an automated pipeline provides a more clinically comprehensive approach than single-task paradigms [37]. The observed agreement between automated and clinician scoring is consistent with that OpenPose-based skeletal tracking can produce meaningful kinematic measurements in older adults [28]. While our findings should be interpreted as preliminary, they suggest that under standardized conditions, pose estimation algorithms can capture task-relevant movement features aligned with established scoring rubrics [28]. If replicated in larger samples and varied environments, this approach could help extend standardized balance screening beyond hospital, responding to calls for scalable digital fall risk assessment [8].

The user-experience results of gamification align with previous research that indicates gamified elements can improve engagement, enjoyment, and adherence when appropriately tailored to seniors’ preferences and capabilities [19,20]. Systematic reviews have concluded that gamification can enhance engagement, enjoyment, and adherence when appropriately tailored to older users [38]. Our findings extend this body of work by focusing on assessment rather than intervention [39]. This distinction matters because the real-world value of screening and monitoring systems depends on repeated voluntary engagement over time, not only on momentary performance [40]. In assessment settings, older adults may also experience evaluative threat or worry about functional decline, which can reduce engagement and distort user experience [41,42]. The observed improvements in enjoyment, perceived competence, and intention to continue use align with self-determination theory, which posits that supporting competence and autonomy enhances intrinsic motivation [29]. Real-time feedback and avatar demonstrations likely reduced uncertainty and clarified task expectations [43], consistent with senior technology acceptance models emphasizing perceived ease of use and perceived usefulness as determinants of adoption [44]. Qualitative data reinforced these interpretations: participants frequently described the feedback as “rewarding” and the avatar guidance as cognitively supportive, echoing findings that interactive and feedback-rich systems can improve engagement among older adults [21]. The reduction in perceived exertion in the gamified group warrants consideration. Theories of attentional focus suggest that engaging, structured feedback may redirect attention away from bodily strain signals, thereby altering subjective effort perception [38]. Yet, we did not observe significant decreased pressure or tension associated with gamified features, addressing concerns that competitive or evaluative elements may heighten anxiety in older populations [21,45].

Taken together, these findings have practical implications for community screening. Measurement validity is necessary but not sufficient for real-world impact; if older adults experience assessments as confusing or effortful, uptake and repeat use will be limited. By integrating a clinically grounded instrument with older adult–tailored interaction design, systems like GDBA may increase willingness to engage in periodic screening, supporting earlier risk identification and more timely connection to fall prevention services.

Despite these promising outcomes, this study has several limitations that also suggest directions for future research. First, the single-session design precludes conclusions about long-term adherence, learning effects, and sustained motivation; longitudinal randomized trials are needed to determine whether early engagement with GDBA leads to habitual use and clinically meaningful outcomes such as reduced falls, improved independence, and lower health care use. Second, participants were relatively healthy, community-dwelling older adults from one urban district in Shanghai, limiting generalizability to frailer, rural, or digitally inexperienced populations. Future work should recruit more diverse samples and incorporate caregiver-assisted or adaptive onboarding to enhance accessibility. Third, validation occurred under optimal indoor conditions, which may not reflect home environments; subsequent studies should test system robustness under variable lighting, space, and camera angles and develop adaptive algorithms to improve accuracy. Fourth, the small Phase 1 sample and possible observer effects from the physiotherapist’s presence warrant replication in larger, more naturalistic settings. The use of a single primary physiotherapist may also have introduced assessor bias, despite strong interrater reliability on 20% of cases; future studies should use multiple assessors to verify scoring generalizability. Fifth, qualitative limitations include brief interviews, lack of interviewer blinding, and no member checking; future work should use longitudinal interviews, blinded interviewers, and participant validation. Finally, the study did not examine clinical end points or engagement mechanisms; mixed methods research should explore how gamification influences motivation and behavior change and assess cost-effectiveness, feasibility, and adaptive designs to sustain engagement and expand applicability. In addition, although the study protocol was approved by the institutional review board before recruitment, the trial was registered retrospectively at ClinicalTrials.gov after participant enrollment had begun. This may raise concerns about selective reporting. We mitigated this limitation by ensuring that the registered information, the institutional review board–approved protocol, and the outcomes and analyses reported in this paper are fully aligned, and we commit to prospective registration for all future trials of this system.

This study demonstrates the feasibility of integrating digital assessment with human-centered motivational design in a community-oriented geriatric context. The key contribution to the field lies in providing empirical evidence that evidence-based gamification design can significantly enhance older adults’ intrinsic motivation, reduce perceived physical effort, and increase intention for continued use without inducing performance anxiety or cognitive burden, thereby validating theoretical frameworks for practical application in aging populations. More broadly, the findings suggest that validity and engagement should be treated as co-equal design targets in digital health assessment systems, and gamification serves as an effective tool in enhancing user experience and engagement. In terms of real-world implications, the GDBA represents a scalable, cost-effective technology with the potential to extend accessible balance screening to underserved populations with limited clinical access (particularly in rural or resource-limited areas) and reduce health care use and fall-related morbidity through proactive community-based monitoring. If replicated in larger, prospectively registered, multienvironment studies with longitudinal follow-up, such systems could enable scalable community screening, earlier detection of balance decline, and more proactive fall-prevention referral pathways. Conceptually, this work proposes a translational framework: preserve the clinical assessment criteria in a digital format, validate against expert standards, and optimize experiential design to reduce burden and sustain voluntary engagement among older adults.