A prospective, single-blind, 2-arm parallel pilot RCT, followed by descriptive semistructured individual interviews, was conducted between April 28, 2023, and June 27, 2025. Reporting followed CONSORT (Consolidated Standards of Reporting Trials; Checklist 1) guidelines [39], with the trial protocol available in Multimedia Appendix 1.

The institutional review board of the Hong Kong Polytechnic University (HSEARS20230223006) approved the study. All participants enrolled in this study voluntarily and signed informed consent before the formal start of the assessment after being informed that they could withdraw at any time. Each participant was introduced to the aims and content of the study, potential risks, and benefits. All participants provided written informed consent in accordance with the Declaration of Helsinki. HKD $200 (~US $26; a currency exchange rate on April 28, 2023, is applicable) supermarket vouchers were provided to participants as an incentive to compensate for their time spent on the interview and completion of questionnaires. All data, audio recordings, and written records are stored on a password-protected laptop; only the research team has access to them. In addition, all data were used for research purposes only and could not be disclosed without the participants’ consent. To ensure privacy protection, the participants’ personal information was anonymized and labeled “P1-P60.”

Participants were recruited using convenience sampling through community health centers and advertisements across Hong Kong (eg, social media). Eligibility criteria included (1) aged 60 years or older and living in the community; (2) mild-to-moderate hearing loss, defined as a pure-tone average (PTA) of 20‐50 dB across octave frequencies from 0.5 to 4 kHz in both ears; (3) no prior use of hearing aids; (4) normal cognitive performance (Montreal Cognitive Assessment score ≥26); and (5) willingness and ability to provide informed consent and comply with study procedures. Exclusion criteria were (1) a history of psychosis, mania, bipolar disorder, substance use disorder, or suicidal ideation; (2) severe or unstable medical illness; (3) significant retrocochlear pathology or an organic lesion causing hearing loss; (4) a diagnosis of probable Alzheimer disease, vascular dementia, frontotemporal dementia, or Parkinson disease; and (5) use of medications, such as antidepressants, sedatives, or antiepileptics, that may affect cognition.

This pilot RCT was not designed to definitively test intervention effects, which would require a fully powered trial. Due to the absence of established effect sizes from previous literature, a target sample of 60 participants (30 per arm), allowing for 15% attrition, was determined based on recommendations for pilot RCTs [40,41] and with consideration of the study objectives, timeline, budget, and available tablets.

The ACDT training system was co-designed by the research team and older adults with hearing loss, in collaboration with service providers in Hong Kong, and is patented (HK30110757) [42]. The 12-week training intervention consisted of five 60-minute sessions per week [42] and used a sequential dual-task training approach [33]. To enhance engagement, the system incorporates sound effects and interactive gamification elements, including likes, praise, fireworks, and leaderboards (Figure 1). In Figure 1, screenshot 1 illustrates the daily training pathway (only one module is activated per day). Screenshots 2‐4 show a listening task with background noise, in which the participant listens to a sentence spoken by a male voice, repeats it, and checks whether all words were correctly heard. Screenshots 5‐8 demonstrate different cognitive training tasks, including indicating a time using hand gestures, identifying locations, counting numbers, and reading Chinese idioms forward and backward. Screenshot 9 shows the scoring system, and Screenshot 10 presents the ranking list. It also offers participants flexibility for individual training while fostering a sense of virtual group participation through the leaderboard.

Each session begins with a 5-minute volume adjustment to establish a comfortable listening level, followed by attention or orientation training exercises where participants answer questions related to person (identifying their name), time (date, month, or year), and place (location at the time of this writing). This was followed by 25 minutes of auditory training based on culturally adapted exercises from the Listening and Communication Enhancement program [43]. The auditory training includes speech-in-noise perception, rapid speech comprehension, and competing speaker discrimination, all contextualized within real-life scenarios such as medical visits or dim sum gatherings. Difficulty hearing a speaker in noisy environments is a common and frustrating consequence of hearing loss. In the training, participants were trained to listen to a medium-length sentence presented in background noise, repeat it aloud, and then see the sentence in writing to confirm whether they understood every word. The difficulty levels of the sentence and background noise are automatically adjusted based on the participant’s performance on the previous sentence. This training enhances the brain’s auditory processing by helping participants filter out background noise and focus on the target voice. Understanding rapid speech is another common challenge for older adults with hearing loss. In this training, the system automatically adapts the speed of sentences to help participants focus on fast talkers, aiding in comprehension. Additionally, ARHL makes listening difficult when multiple speakers are talking. During the training, participants listen to one of 3 voices (male, female, or child) as the target voice, while a competing voice is presented at varying volumes to create a challenge. This training teaches the participant’s brain to ignore a competing speaker’s voice and focus on the conversation they are engaged in. The auditory training was followed by 25 minutes of cognitive training delivered in a game-based format. These activities target executive function (forward/backward serial counting), perceptual-motor ability (arm-clock positioning), memory (forward/backward repetition of numbers/words), and complex attention (forward/backward spelling of Chinese idioms/poetry) using culturally relevant materials. Sessions concluded with a 5-minute wrap-up featuring communication strategy review (“learning classroom”) and recall of visuospatial and memory tasks.

The waitlist control group received no intervention during the study period but was given access to the ACDT after T2 data collection.

Stratified randomization based on key sociodemographic factors (eg, age, gender, and education level) was used to ensure balanced group allocation. An independent statistician assigned participants to either the ACDT or the waitlist control group in a 1:1 ratio using R software (R Core Team). Group assignments were placed in sealed opaque envelopes, and participants were informed of their allocation after completing all baseline assessments.

After obtaining informed consent, a blinded researcher collected demographic data and administered outcome measures at baseline (T0), week 6 (T1), and week 12 immediately postintervention (T2). The week-6 assessment was included to identify potential early intervention effects, monitor participant attrition patterns, and verify adherence to the intervention dosage. Prior to intervention commencement, participants attended briefing sessions on system operation and received personal accounts with instructional videos. Technical support was available via a hotline.

All the ACDT participants were invited to take part in an individual interview at T2 following a semistructured interview guide (Textbox 1). The interviews were informed by the theoretical framework of acceptability (TFA) [44], which includes 7 constructs, including affective attitude, burden, perceived effectiveness, ethicality, intervention coherence, opportunity costs, and self-efficacy [44]. The TFA was used in the present study to gain a comprehensive understanding of participants’ experiences with the ACDT, as well as their perceptions of and acceptability toward the intervention and study design. Interviews were held at community centers or in a university lab and lasted around 45‐60 minutes. All the interviews were audio recorded to facilitate data analysis. Interview data were reviewed regularly by the study team until no new topics emerged, indicating that data saturation had been reached [45].

Data analyses and plotting were performed using SPSS Statistics (version 29.0.2.0; IBM Corp) and Python (version 3.11; Python Software Foundation). A 2-sided P<.05 was set as the predetermined threshold for statistical significance. Missing data (<7% across variables) were imputed using group means to support analyses requiring complete paired observations and to avoid excluding participants due to missing covariates. A natural log transformation was applied to TMT scores at all time points to address skewness. Baseline group equivalence was confirmed via independent t tests (continuous variables) and chi-square tests (categorical variables). The primary analysis focused on within-group changes from baseline to T2. Normality of the distribution of outcome change scores was evaluated using the Shapiro–Wilk test. Based on the results, within-group changes for normally distributed variables (P≥.05) were analyzed using paired t tests, with Cohen d calculated to measure effect size. For nonnormally distributed variables (P<.05), the nonparametric Wilcoxon signed-rank test was used.

Effect size r for the Wilcoxon signed-rank tests was calculated based on the standardized test statistic (Z) and the total number of observations. Effect sizes were interpreted using established criteria, with thresholds of 0.1, 0.3, and 0.5 indicating small, medium, and large effects, respectively. Between-group differences across time points (baseline, wk-6, and wk-12) were evaluated using linear mixed effects models. Each model included time (categorical, 3 levels), group, and the time×group interaction as fixed effects, while adjusting for the baseline value of the respective outcome as a covariate. Within-participant correlations across time points were modeled using a first-order autoregressive residual covariance structure. Models were fitted using restricted maximum likelihood. Pairwise comparisons of estimated marginal means for time×group interactions were conducted using the least significant difference adjustment. Given the explorative nature of the study, no adjustment for multiplicity was applied to the multiple comparisons of global cognition scores.

Qualitative data were analyzed using the hybrid thematic analysis approach described by Braun and Clarke [51], combining inductive coding with deductive theme organization guided by the TFA [44]. Two researchers (IYZ and HM) independently analyzed data in Chinese to preserve the original meaning conveyed by participants. Initial codes were generated directly from the transcripts and then mapped onto TFA domains using NVivo (version 14; Lumivero). Codes that did not align with existing TFA domains were used to inform theme refinement or adaptation. All codes and preliminary themes were documented in a spreadsheet, translated into English, and then back-translated into Chinese by HM and verified by IYZ. Any discrepancies were solved with input from the study team.

A total of 60 participants were randomized. Baseline characteristics were comparable between groups regarding age, gender, education, employment status, marital status, living arrangements, and health status (Table 1). The mean age of the participants was 67.65 years (SD 4.78). Out of 60 participants, the cohort included 45 (75%) men and 15 (25%) women. Most participants (47/60, 78%) were retired and had attained an upper-secondary education or above. Seventy percent were married, while the majority (52/60, 87%) lived with a spouse, family members, or friends. Regarding health status, 42% (25/60) self-reported their health as good to very good, while 52% (31/60) reported their health as normal.

The trial proceeded according to the protocol without early termination. There were 69 candidates screened for eligibility; 4 were excluded for not meeting the eligibility criteria, and 5 withdrew prior to randomization due to health issues or loss of contact (Figure 2). Of the 60 participants who met the eligibility criteria, 30 were randomized to the ACDT group, and 24 of them completed all 60 training modules, while 6 participants completed 55 modules. The overall intervention completion rate was nearly 98% (1770/1800). At the beginning of the study, our hotline received a small number of calls regarding login issues, primarily from participants who had forgotten their passwords or were unsure how to reset them. This issue was resolved after we improved the study briefing sessions. No adverse events were reported in either the intervention or control group.

Within-group analyses showed significant decreases in TMT completion times over 12 weeks, reflecting improved performance in focused attention (intervention: mean change=–0.15; P=.02; d=–0.46; control: mean change=–0.10; P=.027; d=–0.46), and divided attention (intervention: mean change=–0.21; P=.002; d=–0.63; control: mean change=–0.11; P=.02; d=–0.45). The intervention group demonstrated moderately greater improvements than the control group (Table 2). Both groups reported a significant decrease in overall hearing handicap. No statistically significant within-group changes were observed in HK-MoCA total scores or in social relationship and loneliness scores in either group.

For outcome variables that violated the normality assumptions, Wilcoxon signed-rank tests were conducted separately within each group (Table 3). In the intervention group, significant cognitive improvements were observed in the HK-MoCA subdomains of naming (Z=2.00; P=.05; r=0.37) and visual cognition (Z=2.39; P=.02; r=0.44), while the control group showed no significant within-group changes in these areas. Both groups reported significant decreases in emotional hearing handicap; however, the intervention group demonstrated a larger effect size (Z=2.15; P=.03; r=0.39) than the control group (Z=2.03; P=.04; r=0.37). Similarly, both groups demonstrated significant improvements in 5-min delayed recall on the AVLT, with the intervention group exhibiting greater improvement (Z=4.16; P<.001; r=0.76) than the control group (Z=4.11; P<.001; r=0.75).

Changes in visual cognition, naming, divided attention, and 5-minute delayed recall at baseline, 6 weeks, and 12 weeks are shown in Figure 3 for the intervention group (blue circles) and the control group (gray circles). Each data point represents an individual participant’s score, with a horizontal offset added for clarity. Lines connect the group means at each time point, and error bars indicate 95% CIs. The intervention group showed significantly greater improvement than the control group in these functions.

According to the linear mixed effects model analysis, a small to moderate group effect was observed for 5-minute delayed recall on the AVLT (Cohen d=0.38), with the fixed effects explaining 69% of the variance (marginal R²=0.69). Baseline scores strongly predicted follow-up performance (standardized β=0.78). A significant main effect of time was identified (F_{2, 129.5}=25.70; P<.001), along with a significant group difference (F_{1, 81.2}=3.997; P=.05), indicating that the intervention group showed better memory scores across visits. The time×group interaction was nonsignificant (F_{2, 129.5}=0.022; P=.98; Figure 4).

Cognitive performance was similar between the groups at baseline. Post hoc pairwise comparisons showed that cognitive functions on HK-MoCA total scores in the intervention group increased significantly from baseline to week 6 (P=.03) and were maintained at week 12. In contrast, the control group showed a significant decline from week 6 to week 12 (P=.02) with no net change from baseline.

For hearing thresholds (Figure 5), the standardized baseline coefficient was strong (β=0.88), but the group effect remained minimal (Cohen d=0.17), with a marginal R² of 0.67. A significant time×group interaction was observed for left-ear thresholds (F_{2, 117.0}=4.470; P=.01), indicating improved thresholds in the intervention group.

The Hearing Handicap Inventory for the Elderly total score showed a significant main effect of time (F_{2, 112.6}=14.69; P<.001), with both groups reporting reduced perceived hearing handicap at weeks 6 and 12 compared with baseline. A significant group main effect was observed for the social subscale (F_{1, 62.1}=4.81; P=.03), indicating that, on average, the intervention group reported lower perceived social hearing handicap than the control group, independent of time. There was no significant time×group interaction for the social subscale (P=.24).

A total of 30 participants in the ACDT group were interviewed, and 3 key themes were generated based on the TFA [44]. No ethical issues were raised by participants. Opportunity costs refer to the trade-offs users may have to make when engaging with the intervention. None of them explicitly expressed such concerns. Instead, most rearranged daily activities to accommodate the training.

This pilot RCT provides preliminary evidence for the feasible, acceptable, and potential efficacy of ACDT for older adults with ARHL. Participants in the ACDT group demonstrated significant improvements in focused attention, divided attention, naming, and visual cognition. However, the control group’s improvement in some cognitive measures suggests possible practice effects from repeated testing, highlighting the need for active control conditions in future trials. These findings extend evidence that combined auditory-cognitive training selectively enhances executive functions and visual processing, which are critical domains in dementia risk reduction [52]. This pattern converged with qualitative findings of “enhanced attention and concentration,” “improved memory and mental clarity,” and “increased processing speed” during daily activities. Mechanistically, results align with previous research that auditory training enhances attention in older adults with hearing loss [53], cognitive training improves domain-specific functions, and dual-task trainings strengthen executive control [34]. The gamified design likely amplified these effects through enhanced engagement and motivation, which is supported by participant feedback and studies showing gamification improves adherence and training outcomes in cognitive interventions [54]. Additionally, further research is warranted to examine the effects of ACDT on dual-task listening-cognitive abilities using an active control group, as poor dual-task performance is recognized as an early indicator of dementia [30].

Notably, the ACDT group exhibited improved 5-minute delayed recall with small-to-moderate group effects and maintained higher average scores across assessments than the control group. This finding is clinically significant given delayed recall is an established early marker for cognitive decline and dementia risk [55]. Furthermore, post hoc pairwise comparisons demonstrated that HK-MoCA total scores significantly increased from baseline to week 6 and were sustained at week 12 in the ACDT group, suggesting that ACDT may help stabilize global cognition over time.

Hearing threshold analysis revealed minimal overall between-group differences, which means that ACDT did not produce a significant improvement in PTA compared to the control group, and this finding is consistent with the recognized stability of peripheral hearing loss over short time intervals [56]. While the significant time×group interaction observed for left-ear thresholds may indicate training-induced neuroplasticity in central auditory pathways [57], these preliminary findings warrant confirmation in larger trials. Future studies should investigate whether specific training components can be optimized for bilateral benefits and determine if observed neuroplastic changes translate to measurable functional communication improvements.

Interestingly, the control group reported lower perceived overall hearing handicap but higher emotional hearing handicap than the ACDT group. One possible explanation is that the ACDT participants became more aware of their hearing challenges, leading to more accurate self-reporting. In contrast, the control group may have had lower insight into their hearing limitations, potentially resulting in underreporting. This interpretation aligns with qualitative findings and researcher observations during data collection, wherein ACDT participants frequently described increased awareness of hearing-related challenges in their daily lives. Previous research also supports that targeted interventions can improve auditory awareness, which may initially increase self-reported handicap but ultimately facilitate adaptive behavior change [58].

In our study, participants reported enhanced recognition of auditory and attentional limitations, which subsequently prompted adaptive modifications in communication behaviors. Many participants described integrating training into their daily routines and adopting compensatory strategies to improve attention and listening engagement, which may reflect a meaningful sign of behavior change. However, longitudinal investigation is required to verify sustainability. ACDT also fostered perceived accomplishment and strengthened self-efficacy, potentially encouraging sustained engagement in beneficial behavior changes. Collectively, these findings position ACDT as a catalyst for positive behavioral adaptations, benefiting not only participants but also their families.

In this study, no statistically significant changes were observed in social relationship or loneliness scores for either group. A possible explanation is that participants were comparatively young, most were married and living with a spouse or other family members, and they demonstrated a positive attitude toward adopting this new training to address their hearing problems. Future studies should test ACDT among individuals at higher risk of loneliness and social isolation.

While the intervention was generally well-received, participant feedback identified opportunities for enhancement through more personalized and interactive training content. Integration of artificial intelligence (AI) can dynamically tailor training modules to individual abilities and progression patterns, with real-time adjustments in difficulty and personalized feedback to optimize training efficacy while maintaining engagement. Machine learning algorithms could further enhance ecological validity by generating complex auditory environments that simulate real-world scenarios, improving training transferability to daily communication challenges. Incorporating AI-powered analytics could continuously monitor user performance trajectories, identify areas for improvement, and deliver targeted interventions to maximize auditory and cognitive gains. Future iterations of ACDT should prioritize the development of AI-enhanced features, including personalized content, interactive feedback, and real-time progress tracking. Engaging users and their families in the co-design process is essential to ensure that the platform meets their needs and preferences.

This study demonstrates several strengths, including (1) multimodal assessment of auditory and cognitive outcomes, (2) ecologically valid gamified design that could enhance participant engagement, and (3) novel documentation of self-awareness as a facilitator for adaptive behavior change. However, several limitations warrant consideration: (1) the modest sample size and absence of longitudinal follow-up limit detection of subtle neuroplasticity effects, such as the observed hearing threshold trend; (2) the waitlist control design cannot exclude attention bias and isolate the specific efficacy of dual-task training; and (3) although ARHL has been reported to be more common, more severe, and to have an earlier onset in men than in women [59], the homogeneous participant characteristics regarding education level and hearing loss severity, along with a predominance of male participants, may restrict generalizability of our findings to broader ARHL populations.

This pilot RCT suggests that ACDT is feasible and acceptable for older adults with ARHL, with preliminary evidence supporting its potential efficacy. The intervention appeared to benefit both auditory and cognitive domains, as well as self-awareness of hearing loss, which may facilitate adaptive behavioral adjustments. Future development could explore AI-enhanced personalization and interactivity to improve ecological relevance. Further trial with larger, more diverse samples, active control conditions, and longitudinal follow-up would help clarify ACDT’s sustained effects on auditory-cognitive health, dual-task listening-cognitive abilities, and real-world functioning.