This study used data from a questionnaire survey examining older adults’ intention to use smart eldercare services. The survey was conducted over a 2-month period in 2020 in 2 districts of a large metropolitan city in China. The survey collected information on respondents’ sociodemographic characteristics and their awareness and use of digital meal assistance services (ie, online food ordering services). Given potential systematic differences between online and offline surveys in digital skills and service access, a mixed-mode design was adopted. Descriptive and regression analyses were conducted separately for online and offline samples. The final analytic sample comprised 1019 valid responses (n=405 online and n=614 offline). Additional details on the sampling framework and survey implementation are provided in Multimedia Appendix 1.

The offline survey used a 2-stage stratified random sampling strategy. Streets within each district were first selected using simple random sampling. Communities within selected streets were then stratified by geographic accessibility to community eldercare service stations (approximately 15-min walkability, defined as about 1.0‐1.2 km based on navigation distance) and randomly sampled within each stratum. Eligible older adults were recruited through intercept surveys, yielding 614 valid questionnaires. Further procedural details are provided in Multimedia Appendix 1.

The online survey was administered via the Wenjuanxing platform and yielded 405 valid responses. Survey links were primarily distributed by community staff, with additional responses collected through the platform’s mutual-assistance service. No incentives were provided. Eligibility was ensured through an automated age-screening item (≥60 y), and duplicate submissions were prevented by restricting 1 response per IP address. Participation was voluntary, and informed consent was implied by questionnaire submission. Additional quality control details are provided in Multimedia Appendix 1, and the completed CHERRIES (Checklist for Reporting Results of Internet E-Surveys) checklist is provided in Checklist 1.

This study involved an anonymous questionnaire survey administered online (via Wenjuanxing) and offline face-to-face. No personally identifiable information (eg, names, contact details, identification numbers, or precise addresses) or sensitive personal data (eg, medical or financial information) was collected, and no intervention, clinical procedures, or biological sample collection was involved. Under Article 32 of the Measures for Ethical Review of Life Sciences and Medical Research Involving Human Subjects, research using anonymized information data that does not cause harm to participants and does not involve sensitive personal information or commercial interests may be exempt from ethics review. Accordingly, this anonymous, minimal-risk survey study met the criteria for exemption from institutional ethics review.

Participants were informed about the study purpose, procedures, voluntary participation, and data protection. For the online survey, consent was implied by submission of the completed questionnaire; for the offline survey, oral informed consent was obtained prior to participation. Participants could withdraw at any time without consequences. No compensation was provided for participation.

All survey data were collected anonymously. Online responses were exported and stored in password-protected files, and offline paper questionnaires (if applicable) were securely stored and then digitized. Only the research team had access to the data.

EFA was conducted to identify the latent structure of older adults’ behavioral attitudes toward digital meal assistance services. Prior to analysis, questionnaires with missing values on key variables or substantial nonresponse were excluded. The final analytic sample consisted of 1019 complete questionnaires; therefore, no imputation was applied.

EFA followed standard procedures of item screening, factor extraction, rotation, and interpretation. Factors were extracted using principal component analysis, with the number of factors determined by the Kaiser criterion (eigenvalues >1). Varimax orthogonal rotation was applied. Items with factor loadings below 0.50 were considered for removal; items with cross-loadings (loading differences <0.20 across factors) or communalities below 0.40 were also evaluated and excluded when appropriate.

Binary logistic regression models were used to examine associations between behavioral attitude dimensions and both usage intention and actual service use. Given differences between online and offline samples, all models were estimated separately for each subsample. The general model specification was

where pi denotes the probability that individual i reports a higher usage intention or actual service use; Xi represents behavioral attitudes (or specific attitudinal dimensions); and Zi includes control variables (gender, age, education, and income). Results were reported as odds ratios (ORs) with 95% CIs. Given the sensitivity of the Hosmer-Lemeshow test in large samples, pseudo-R² and the Akaike information criterion were also reported as supplementary fit indicators.

Table 1 summarizes the sociodemographic characteristics of the sample. Overall, the age distribution was skewed toward younger older adults, with participants aged 60‐69 years accounting for 569 of 1019 (55.8%) of the sample, which may limit generalizability. Compared with the offline sample, the online sample was younger on average, whereas the offline sample included a higher proportion of the oldest-old (aged ≥80 y). In terms of socioeconomic characteristics, the online sample had higher educational attainment than the offline sample, with 272 of 405 (67.2%) reporting a high school education or above compared with 297 of 614 (48.4%) in the offline sample. The online sample also had a higher proportion of high-income respondents (≥8000 CNY, 1 CNY=US $0.14), whereas the offline sample included a relatively larger share of middle-income respondents (4000‐7999 CNY). Overall, the offline sample more closely resembled community-dwelling older adults, whereas the online sample was more likely to represent younger and more digitally skilled individuals. Accordingly, the findings were more applicable to younger and older adults and those with greater digital capacity, and caution was warranted when generalizing to the oldest-old or to older adults with limited digital skills; these limitations are discussed in the Limitations section.

EFA was conducted to identify the latent structure of older adults’ behavioral attitudes toward digital meal assistance services. Data suitability was confirmed by a significant Bartlett test of sphericity (χ²=286.67, P<.01) and a Kaiser-Meyer-Olkin value of 0.894, indicating adequate interitem correlations. After excluding 3 items based on predefined criteria, 16 items were retained for analysis. Five factors with eigenvalues greater than 1 were extracted, with initial eigenvalues of 6.01, 2.43, 1.82, 1.39, and 1.03. Following Varimax orthogonal rotation, the 5 factors explained 14.60%, 14.13%, 11.64%, 11.09%, and 8.92% of the total variance, respectively, yielding a cumulative explained variance of 60.38%. Detailed results are presented in Table 2.

Table 3 presents the standardized factor loadings and communalities for each retained item. All items showed strong loadings on their corresponding factors, and each factor was represented by at least 3 items, indicating a clear and stable factor structure. Reliability analysis showed that the overall scale demonstrated good internal consistency, with a Cronbach α of 0.879. In addition, correlations between the mean scores of individual behavioral attitude dimensions and the overall scale mean ranged from 0.414 to 0.762, suggesting that the dimensions were moderately correlated while maintaining adequate discriminant validity (Multimedia Appendix 2).

Based on the content characteristics of items with high factor loadings, 5 factors were labeled. Alleviating burden (F1) reflects value-based judgments regarding whether digital meal assistance services were worth using, particularly in terms of their potential to reduce the caregiving burden on adult children in daily life. When such services were perceived as helping to alleviate family caregiving pressure, older adults tended to hold more positive attitudes toward their use. Enhancing convenience (F2) captures perceived evaluations of the extent to which digital meal assistance services improve convenience in older adults’ daily lives. This factor reflected the functional use of the service, consistent with perceived usefulness in the TAM. Service acceptance (F3) represented older adults’ evaluations of the social meaning of digital meal assistance services. When service use was interpreted as indicating that adult children had not adequately fulfilled their caregiving responsibilities, attitudes toward the service tended to be negative. Digital trust (F4) reflected older adults’ attitudinal judgments regarding the safety and reliability of digital meal assistance services. Funding sources (F5) captured older adults’ perceived affordability of digital meal assistance services, consistent with perceived behavioral control in the TPB. Lower perceived affordability was associated with a lower likelihood of service use.

Notably, perceived ease of use, as defined in the TAM, did not emerge as an independent attitudinal dimension in the identified factor structure.

Table 4 summarizes mean scores across behavioral attitude dimensions by participant characteristics. Overall, older adults reported relatively higher evaluations on alleviating burden (F1) and enhancing convenience (F2). Middle- and higher-income groups generally scored higher on enhancing convenience (F2), service acceptance (F3), and digital trust (F4), whereas lower-income older adults reported higher scores on funding sources (F5) and alleviating burden (F1). Higher educational attainment was associated with more positive evaluations of alleviating burden (F1), enhancing convenience (F2), and service acceptance (F3), as well as lower concern regarding funding sources (F5). By contrast, differences in digital trust (F4) were modest across education groups. By survey mode, the online sample consistently reported higher scores across most attitude dimensions—particularly enhancing convenience (F2)—whereas differences in digital trust (F4) between online and offline samples were relatively small.

Tables 5 and 6 present logistic regression results examining associations between the 5 behavioral attitude dimensions and both intention to use and actual use of digital meal assistance services, estimated separately for the online and offline samples. All 4 models were statistically significant (P<.001), with pseudo-R² values ranging from 0.156 to 0.240 and Akaike information criterion values from 800.001 to 1002.470, indicating acceptable model fit. Hosmer-Lemeshow tests did not suggest a significant lack of fit (P value range .055-.062).

In the online sample, alleviating burden (F 1) was positively associated with both usage intention and actual use. A 1-unit increase in F1 was associated with a 61.8% increase in the odds of intending to use the service (OR 1.618, 95% CI 1.235‐2.118; P<.001) and a 74.2% increase in the odds of actual use (OR 1.742, 95% CI 1.286‐2.359; P<.001). In addition, digital trust (F4) was positively associated with usage intention, with each 1-unit increase corresponding to a 35.1% increase in the odds of intention to use (OR 1.351, 95% CI 1.054‐1.732; P=.02), and showed a marginal association with actual use (OR 1.347, 95% CI 0.986‐1.840; P=.06). Service acceptance (F3) was not significantly associated with usage intention (P=.10) but demonstrated a marginal positive association with actual use (OR 1.250, 95% CI 1.000‐1.557; P=.05).

In the offline sample, alleviating burden (F1), service acceptance (F3), and digital trust (F4) were all positively associated with usage intention. Notably, a 1-unit increase in digital trust (F4) was associated with a 22.1% increase in the odds of intention to use (OR 1.221, 95% CI 1.083‐1.376; P<.001). In contrast, funding sources (F5) showed a marginal negative association with usage intention (OR 0.806, 95% CI 0.642‐1.020; P=.07). With respect to actual use, digital trust (F4) was the only attitude dimension that remained significantly associated with service use in the offline sample, with a 1-unit increase corresponding to a 6.8% increase in the odds of actual use (OR 1.068, 95% CI 1.011‐1.128; P=.02). Although the odds ratio for alleviating burden (F1) in predicting actual use appeared large in the offline sample, this association did not reach statistical significance, likely reflecting limited statistical power and greater heterogeneity in offline respondents.