The study took place at a total of 16 sites, including 14 community committees, 1 nursing home, and 1 community health service center in 6 districts of Beijing, China, from March 2024 to October 2024. Convenience sampling was applied during the quantitative survey phase, as it aligns with the community-level screening and voluntary nature of IC assessment under the ICOPE framework. The authors coordinated and communicated with community workers and heads of health facilities before each questionnaire survey and recruited a consecutive sample of older adults by posting group notices and conducting onsite outreach after consent was obtained. Older adults who met the following inclusion criteria were eligible to participate in the survey: (1) aged 60 years or older; (2) with normal reading, writing, and comprehension skills; and (3) voluntarily participated in this survey after providing informed consent. The exclusion criteria were as follows: (1) participants with severe functional impairments that prevented them from safely completing an exercise test, (3) participants with medically diagnosed severe cognitive impairment, (3) participants with medically diagnosed serious psychiatric disorders, and (4) participants currently experiencing an acute or worsening illness.

The sample size was estimated based on 2 primary considerations. First, the initial target sample was calculated using the Kendall model, which recommended at least 5 to 10 times the number of variables [31]. With 17 variables in this study, and considering the possibility of 20% invalid questionnaires, the required sample size was 102 to 204 cases. Second, to ensure the stability and reliability of the subsequent LPA, a sample size exceeding the commonly recommended minimum of 300 is advised [32]. A final sample of 481 older adults therefore met both the adjusted recruitment target and the methodological requirement for robust LPA.

Based on the literature review [33,34], a structured questionnaire was designed by the research team to collect general information as predictor variables, including sex, age, education level, marital status, family structure, type of labor force before retirement, per capita monthly family income, self-rated health using a Likert scale, and the number of chronic diseases.

The functional ability of participants, which results from the interplay between an individual’s IC, supportive environment, and personal characteristics [1], was assessed using 3 established scales as key outcome measures: the Barthel Index [35], Fried Frailty Phenotype [36], and HALFT scales [37] were used to assess activities of daily living, frailty, and social frailty, respectively.

The 5 key domains of IC were assessed following the instruments recommended in the new version of the WHO ICOPE guidelines [1]. The Mini-Mental State Examination [38], Geriatric Depression Scale-15 [39], WHO vision screening chart [1], whisper voice test [1], Mini Nutritional Assessment Short-Form [40], and Short Physical Performance Battery test [41] were used to assess cognition, psychological capacity, sensory capacity (vision and hearing), vitality, and locomotor capacity, respectively. The scoring rules of these instruments are summarized in Multimedia Appendix 2.

Before the formal questionnaire, the investigative team, comprising postgraduate nursing students, underwent a uniform offline training on the purpose of the study, assessment procedures, questionnaire content, assessment site setup, and language and behavioral norms. For participants who had difficulty completing the questionnaire, such as those experiencing presbyopia or slow reading speeds, the investigators helped them complete it after explaining the meaning of each item. The survey was conducted using either a paper version of the questionnaire or an application installed on tablet computers, with the content being the same in both formats. The paper version of the questionnaire was entered into the application after validity was established so that the data could be accessed for statistical analysis. At the end of the questionnaire survey, the authors invited older adults who were interested in this study to participate in follow-up qualitative interviews and retained their contact details.

The IC of older adults is influenced by many factors and exhibits differential characteristics [42]. Therefore, we attempted to find a people-oriented clustering approach to provide an objective population classification basis for customized personas [32]. LPA is a statistical method that identifies characteristics between individuals based on the estimation of their response patterns on observed indicators, providing unique insights into the heterogeneity of the target population [43].

First, Mplus (version 8.3, Muthén & Muthén) was used to analyze the latent profiles of the IC of older adults. Starting with 1 category, categories were added progressively until the maximum likelihood method was used to determine the best solution. The main evaluation indices included Akaike information criterion (AIC), Bayesian information criterion (BIC), adjusted BIC (aBIC), and entropy. The lower the AIC, BIC, and aBIC are, the better the model fit [44]. Entropy ranges from 0 to 1; the closer the value is to 1, the more accurate the model classification is, and it is generally believed that values of 0.80 or greater indicate that the classification accuracy of the model reaches 90% [45]. The Lo-Mendell-Rubin likelihood ratio test (LMRT) and the bootstrapped likelihood ratio test (BLRT) were used to compare the fit of 2 adjacent models, where low and significant P values suggest a k-class model is superior to k-1−class model [46]. To visualize the latent profiles, the estimated class-specific means from the Mplus LPA model output were exported and plotted using line charts in Microsoft Excel version 16.1.

Second, latent profiles were named and assigned values to classify groups. Descriptive statistics were analyzed using SPSS (version 26.0, IBM Corp) to understand the data distribution characteristics of the profiles. Variables that conformed to a normal distribution were described as mean (SD), and comparisons between groups were made using 1-way ANOVA. Count data were described as frequency and percentage. Differences in sociodemographic and health-related factors between groups were assessed by chi-square tests. Finally, multivariate logistic regression analyses were performed with latent profiles of IC as the dependent variable and variables that were statistically significant in the univariate analyses as the independent variables. To avoid modeling errors, dummy variables were used for categorical data. The impact of the predictor variables and their statistical significance were assessed through regression coefficients (β), P values, odds ratios (ORs), and 95% CIs. A P value <.05 indicated a statistically significant difference.

A naturalistic, descriptive qualitative design is well suited to mixed methods research [47], enabling older adults to provide a rich and direct description of the goals, needs, and concerns of digital monitoring regarding their IC level. Stratified purposive sampling based on the categorical results of the LPA ensured the representation of participants originating from different IC profiles. Given the sequential continuity of the mixed methods study, the inclusion and exclusion criteria established for the quantitative phase were likewise applied to this phase. The first author contacted the older adults who had accepted the invitation to be interviewed during the quantitative phase, while explaining the purpose, content, and methodology of the further study and setting the time and place of the interviews.

Semistructured interviews were conducted between October and December 2024 to collect data. In accordance with the objectives of the study and the literature review [48], the research team developed the following interview outline: (1) “What is your idea of IC?” (2) “How would you evaluate your IC? Can you share some stories from your daily life where you have felt IC come into play or decline?” (3) “What do you think about using digital health tools for IC monitoring?” (4) “What do you hope the use of digital health tools for IC monitoring will do for you?” (5) “Do you have any concerns and doubts about using digital health tools for IC monitoring?” and (6) “Do you have any suggestions or expectations regarding the digital monitoring of IC?”

The first author reviewed the results of the quantitative phase of the IC assessment with the participants before the commencement of the interviews. Participants were interviewed in relaxed and comfortable parks, cafes, and clinic rooms. A total of 2 additional individual interviews were conducted via voice calls on the Tencent Conference platform. Each interview, led by the first author and assisted by another author for observation and field notes, lasted 20 to 68 minutes. Audio interviews were transcribed into text within 24 hours using Lark intelligent transcription. The first 2 authors (ZX and YN) and 2 qualitative experts (QX and YW) proofread the transcript and verified its contextual integrity. Data collection and analysis were synchronized in this study. Recruitment of participants was stopped when no further significant new information emerged, implying data saturation [49].

The qualitative phase of data analysis was conducted with the main aim of obtaining the needs, intentions, and challenges of older adults regarding digital monitoring originating from the different IC profiles. Inductive combined with deductive content analysis was used to hand code the data to compare and integrate the similarities and differences between the different profiles horizontally [50]. Initially, the research team repeatedly reviewed interview transcripts to gain a holistic understanding of the data. Second, after interviewing 2 participants from each latent profile, the authors conducted open coding, assigning descriptive codes to meaningful text segments to develop a preliminary coding framework for guiding the remaining transcripts. Next, data from the same latent profiles were clustered and extracted contemporaneously by the first 2 authors (ZX and YN) following the attributes in the coding framework, which enabled the creation of customized personas. The framework was continuously refined and clarified throughout this process as new insights emerged or existing categories required redefinition. The final version of the framework constituted the thematic results. To capture the group-level commonalities within each persona, the primary author noted interview logs after each interview. These logs documented the author’s immediate impressions and holistic understanding of the participants, which later played a pivotal role in identifying the shared attributes that define each persona. Finally, the research team collaborated to define and describe the customized personas, outlining the participants’ differing attitudes, goals, needs, and available support that ultimately formed the subthemes for the 3 customized personas. Discrepancies between the first 2 authors (ZX and YN) during data analysis were addressed by an expert in qualitative nursing research to achieve harmonization of the results.

A total of 481 older adults, aged 60-97 years, with a mean age of 70.7 (SD 6.8) years, participated in the quantitative phase. The results in Table 1 show that most participants were female (320/481, 66.5%), aged 60-69 years (234/481, 48.6%), currently married (356/481, 74%), and living in a married couple’s household (224/481, 46.6%). Most participants had a secondary education (318/481, 66.1%), a per capita monthly household income of RMB 3000-RMB 5999 (260/481, 54.1%), and engaged in light labor (241/481, 50.1%) before retirement. In addition, the majority of older adults rated their health as “very good or good” (258/481, 53.6%) and had 1 to 3 chronic diseases (323/481, 67.2%).

The scores in the 5 domains of IC were used as exogenous variables to model the latent profiles. The z score was used to eliminate the effects of different magnitudes on data analysis during data preprocessing. The fitting statistics for modeling 1-5 latent profiles sequentially are shown in Table 2. After comprehensive consideration of the latent profile metrics and clinical significance, we identified the 3-profile model as the best option for the following reasons: (1) the values of AIC, BIC, and aBIC decreased gradually as the number of profile subgroups increased, but beginning with the 4-profile model, the LMRT was greater than .05 and the minimum classification probability was less than 10%, signifying that this category may be spurious; (2) the LMRT and BLRT results were statistically significant (P<.05) for the 3-profile model, indicating that it was better than the 2-profile solution; and (3) the entropy value of the 3-profile model exceeded 0.8, indicating good model accuracy. Therefore, the 3-profile model was selected as the most appropriate model. To verify the reliability of the LPA, we calculated the average posterior probabilities for the 3 profiles. The results are shown in Multimedia Appendix 3, where the posterior probabilities were greater than 0.8, indicating that the model results for the 3 profiles were well separated [52].

Figure 2 illustrates a graphical presentation of the z score for the 5 domains of IC in the 3 impairment profiles. Each profile was named according to its characteristics. Profile 1, constituting 19.1% (92/481) of the sample, was designated the “multisubdomain decline–IC imbalance group.” The most obvious characteristic of this group was the overall decline in IC, while only the sensory domain remained at a high level, resulting in an imbalance in IC. It should be noted that higher scores on the psychological domain indicated reduced psychological capacity. Profile 2, the “multisubdomain moderate–sensory deficit group,” included 29.1% (140/481) of the participants. This group showed a moderate IC level that was higher than those in profile 1 and was characterized by significant sensory deficits, reflecting a pattern of decline with single deficits. Profile 3 was the most prevalent in the sample, accounting for 51.8% (249/481) of cases. It was designated the “multisubdomain robust–whole balance group” owing to its exceptional performance in IC, with no significant loss of ability.

As shown in Table 1, the findings derived from the univariate analyses provide insights into the differences among the 3 latent profiles by using sociodemographic and health-related factors as predictor variables. Among these variables, age, marital status, education level, type of labor force prior to retirement, per capita monthly household income, self-rated health, and number of chronic diseases differed significantly (P<.05) among the 3 IC impairment patterns. No statistically significant differences were observed in sex and family structure.

Only variables that were statistically different between the IC subgroups in the univariate analyses were included in the multivariate logistic regression analyses (Table 3). Using profile 3, the “multisubdomain robust–whole balanced group,” as the reference group, older adults aged 80 years and older are more frequently found in either profile 1 (OR 9.190, 95% CI 2.986-28.284; P<.001) or profile 2 (OR 7.018, 95% CI 2.559-19.247, P<.001). As detailed in Table 3, as education levels rose, older adults were less likely to be categorized in profile 1 or profile 2. Those with a per capita monthly household income of ≥RMB 6000, indicating a higher quality of life and social status, were less likely to be attributed to profile 1 (OR 0.163, 95% CI 0.040-0.670; P<.05). Those with better self-rated health were more likely to belong to profile 3. For example, the probability of older adults rating their self-health as “very good” joining profile 1 (OR 0.037, 95% CI 0.005-0.265; P<.001) and profile 2 (OR 0.115, 95% CI 0.017-0.773; P<.05) was reduced. Older adults with no chronic diseases were less likely to be enrolled in profile 1 (OR 0.315, 95% CI 0.105-0.943; P<.05). The results, when viewed through profile 2 as the reference group, indicated a lower possibility of older adults with a per capita monthly household income ≥RMB 6000 (OR 0.219, 95% CI 0.061-0.782; P<.05) and self-rated health of “good” (OR 0.236, 95% CI 0.060-0.935; P<.05) belonging to profile 1.

As demonstrated in Table 4, there are significant differences in activities of daily living, frailty, and social frailty among the identified latent profiles of IC impairment. Subsequent post hoc comparisons revealed a gradual decline in functioning for latent profiles 3, 2, and 1, using the 3 functional abilities as the outcome variables.

A total of 25 older adults volunteered to participate in the interviews, including 5 in profile 1, 10 in profile 2, and 10 in profile 3. Most participants were female (17/25, 68%). All participants had a mean age of 70.1 (SD 7.03, range 60-85) years, and had experienced between 0 and 4 IC domain declines. Detailed information is provided in Multimedia Appendix 4.

A total of 4 themes emerged in this study. In the hierarchical data analysis, 3 customized persons corresponding to different IC profiles were depicted, and subthemes under the themes that matched the persons were illustrated in Figures 3-5. Representative quotations were translated into English by the authors and confirmed by back-translation. In this section, the findings are reported as an integration of customized personas as follows.

Narrative persona: affects my mood—anxious evader are female and belong to profile 1. They are older than those in other profiles, have lower education and disposable income, rate their health negatively, and are more likely to be chronically ill. Insights from the interviews suggest that they are psychologically fragile and stay away from anything that may bring about emotional stress. Moreover, the lack of social support for persona A, as evidenced by inadequate family care, a desire for companionship, and a neglected position in social groups, may further diminish their physical and psychological capacity in the face of potential stress. In addition, they were skeptical about using digital technology to assess IC and reluctant to ask for help based on past experiences. As a result, persona A showed less interest in digital monitoring of IC, which may have been a psychological coping mechanism that prompted the adoption of an avoidance strategy.

Narrative persona: older adults in persona B, derived from profile 2, have a slight decline in IC, characterized by an impairment in sensory function. Compared to persona C, older adults in persona B were older, less educated, and more likely to rate their health as neutral or slightly negative. Participants expressed a desire to receive assistance from IC in the identification of potential illnesses and the acquisition of professional guidance, which was a passive strategy for treating illnesses. The acceptable attitude they displayed was derived from the Chinese proverb “capitalize on what comes,” which implied that since illness is unavoidable, it should be accepted calmly. Seemingly, this was not consistent with the Healthy Aging Framework, which posited the maintenance of IC to promote functional ability and quality of life. Moreover, they were willing to take advantage of digital health technology to assess IC but were marginally deficient in digital literacy, resulting in greater demands for functionality and ease of use. Regarding social support, they reported feeling cared for by family and friends. However, the activities and hobbies that they were compelled to abandon due to declining functionality also engendered helplessness and isolation, echoing the findings that they may be in the prefrailty and presocial frailty stages.

Narrative persona: participants in persona C are derived from profile 3 and represent older adults with high levels of IC. Compared to personas A and B, they are likely to be younger, more educated, have better disposable income, and a more positive assessment of their health. They showed a high level of interest in monitoring IC and were keen to maintain communication with their community health care team through age-friendly digital platforms for regular reminders, feedback on professional advice, and facilitation of functional improvement. Older adults in the technology-rich exploratory persona C also needed fun interventions and motivational approaches to help improve monitoring adherence, confidence, and sense of achievement. They suggested that free and community outreach is essential to increase access to IC monitoring. In addition, the interviews revealed that they were more likely to live in well-functioning family relationships and maintained good relationships with neighbors and peers through developing shared hobbies and were more autonomous and socially active.