This study used an explanatory sequential mixed methods design [38]. The overall study flow is presented in Figure 1. Quantitative data were first collected via a cross-sectional questionnaire survey. Participants for the qualitative phase were drawn from the quantitative sample and selected purposively based on the survey results. Semistructured interviews were then conducted. Similar mixed methods approaches have previously been used to investigate the adoption and acceptance of emerging technologies [39,40].

The reporting of this mixed methods study was guided by the GRAMMS (Good Reporting of A Mixed Methods Study) checklist [41], the quantitative cross-sectional component was reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist [42], and the qualitative component was reported in accordance with the COREQ (Consolidated Criteria for Reporting Qualitative Research) checklist [43]. The completed checklists are provided in Multimedia Appendix 1.

Quantitative data were collected using a paper-based questionnaire administered by trained researchers. After enrollment, participants completed the questionnaire independently whenever possible. For patients who were unable to complete the questionnaire independently because of poststroke upper limb dysfunction, difficulty writing, or fatigue, but who were able to understand the questionnaire content and communicate their response choices verbally, data were collected using a standardized researcher-assisted procedure. A trained researcher (YY) read each item and its response options aloud and recorded the answers selected by the patient without interpreting the items, suggesting answers, or otherwise influencing the responses. Family members or caregivers did not complete any questionnaire items on behalf of the patients. Each returned questionnaire was reviewed on-site for completeness, and participants were invited to complete omitted items where feasible. Questionnaires with more than 10% of items remaining unanswered after this review were excluded from the analysis. Throughout data collection, questionnaire data were entered and checked for accuracy by the research team in weekly batches.

To ensure data entry accuracy, data from the paper-based questionnaires were independently entered into an EpiData 3.0 database by 2 researchers (YY and YoW). Before applying the questionnaire-level exclusion criterion, missingness patterns were characterized descriptively by summarizing the number and proportion of unanswered items per questionnaire and examining their distribution across individual items, scales, and questionnaire sections. No missing value imputation was performed. Statistical analyses were performed using SPSS (version 27.0; IBM Corp). All statistical tests were 2-sided, with P<.05 indicating statistical significance. Categorical variables were summarized as frequencies and percentages. Continuous variables were assessed for normality using the Shapiro-Wilk test and summarized as mean (SD) when approximately normally distributed or as median (IQR) otherwise. Univariable analyses were conducted to compare technophobia scores across participant-characteristic subgroups. Homogeneity of variance was assessed using the Levene test. When the assumptions of approximate normality and homogeneity of variance were met, independent-samples t tests were used for variables with 2 categories and 1-way ANOVA for variables with more than 2 categories. Otherwise, the Mann-Whitney U test or Kruskal-Wallis H test was used, as appropriate.

Correlation analyses were conducted to examine the associations of continuous and ordinal variables with technophobia. Pearson correlation analysis was used for approximately normally distributed continuous variables, whereas Spearman rank correlation analysis was used for ordinal variables or continuous variables that departed from normality. The correlation matrix heatmap was generated using GraphPad Prism (version 10.6; GraphPad Software).

A multivariable linear regression model was fitted with the total technophobia score as the dependent variable. Variables associated with technophobia at P≤.10 in either the univariable or correlation analyses were entered simultaneously into the model using the enter method. Categories for the independent variables were prespecified based on previous literature. Categorical variables were dummy coded, with the lowest category used as the reference for ordered variables to permit comparisons with higher categories; continuous variables were entered on their original scales. The complete coding scheme is provided in Multimedia Appendix 2. Residual normality was assessed using the histogram and normal P-P plot of standardized residuals, and homoscedasticity was evaluated using the scatterplot of standardized residuals against standardized predicted values. Multicollinearity was assessed using tolerance values and variance inflation factors.

Participants in the qualitative phase were selected from older patients with stroke who had completed the quantitative survey and had agreed during the quantitative phase to be contacted for a subsequent interview. In addition, participants were required to be able to describe their experiences of using or avoiding digital health technologies and to be willing to participate in a semistructured interview. A maximum variation purposive sampling strategy was used. Using participant-level data from the quantitative phase, potential interviewees were selected to span the observed distribution of total technophobia scores, while variation in sex, age, and educational attainment was also sought. Researchers contacted potential participants either face-to-face or by telephone to confirm their willingness to participate and to arrange a suitable interview time and location.

Between February 6 and April 30, 2025, semistructured interviews were conducted by 2 female master’s students in nursing who had received formal training in qualitative research. All interviews were conducted individually and face-to-face in a quiet, private office at the study hospital, with no family members, caregivers, clinical staff, or other nonparticipants present. Both interviewers were members of the research team and were familiar with the context of stroke care and rehabilitation, but neither was involved in participants’ direct clinical treatment, nursing care, or care-related decision-making. Before each interview, the interviewers introduced themselves as researchers and explained the study purpose, audio-recording arrangements, confidentiality safeguards, the right to withdraw, and that participation or nonparticipation would not affect the care participants received. Written informed consent was then obtained. With participants’ permission, all interviews were audio-recorded, and no repeat interviews were conducted. After each interview, the interviewers recorded reflexive notes documenting the interview context, participants’ nonverbal cues, and their reflections on the interview interaction. During research team discussions, they critically considered how their professional backgrounds, researcher roles, and interactions with participants might have shaped data generation and interpretation.

Interviews were scheduled when participants’ health status permitted and sufficient time was available. The semistructured interview guide was jointly developed by the research team on the basis of the quantitative-phase findings and relevant literature, with the quantitative findings also used to identify topics requiring further explanatory probing. The guide was reviewed by the principal investigator (YL), who had formal training and experience in qualitative research. Before formal data collection, the guide was pilot-tested with 2 patients who met the eligibility criteria. Data from the pilot interviews were not included in the formal analysis, and the wording of selected questions and follow-up prompts were refined in response to participant feedback. The semistructured interview guide covered five areas: (1) experiences of using digital health technologies and their perceived role in health care, rehabilitation, and health self-management; (2) tension, concerns, and difficulties encountered during use; (3) understanding and evaluating online health information, confidence in completing related tasks, and responses to difficulties; (4) contextual conditions, including social support, device and internet access, costs, and poststroke functional changes; and (5) suggestions for improving digital health technologies and services delivered through them. Questions were phrased in plain, open-ended, and nonjudgmental language and were sequenced from general topics to specific experiences, allowing participants to describe their experiences at their own pace. If a participant showed a marked emotional reaction, the interviewer paused discussion of the current topic or moved to another topic to respect the participant’s wishes and minimize discomfort. Each interview lasted 20-45 minutes. The complete interview guide is provided in Multimedia Appendix 3.

Qualitative data collection and preliminary analysis proceeded concurrently. After each interview, the research team conducted a preliminary review and discussion of the interview transcript and reflexive notes to assess whether the accumulating data continued to broaden or deepen the experiences represented or provided additional explanations relevant to the quantitative findings. Data adequacy was considered to have been reached when successive interviews no longer substantially extended the experiences represented in the data or contributed new explanatory information. This point was first judged to have been reached after the 13th interview. Two additional participants were subsequently interviewed to assess whether further interviews would extend the experiential or explanatory scope of the data. These additional interviews added detail to the existing accounts but did not materially broaden the experiences or explanations already represented. The final qualitative sample therefore comprised 15 participants.

Interview data were analyzed using reflexive thematic analysis as described by Braun and Clarke [54,55]. All interview recordings were transcribed verbatim within 24 hours of each interview and checked against the original audio recordings for accuracy and completeness. The transcripts were then imported into NVivo (version 12.0; QSR International Pty Ltd) for data management.

Two researchers (YY and YoW) repeatedly read all interview transcripts to become familiar with the dataset and independently generated initial codes. At successive stages of coding and theme development, they compared their coding decisions, analytic reflections, and interpretations of the data. When different interpretations arose regarding the meaning, boundaries, or organization of codes or themes, the 2 researchers first returned to the relevant interview extracts and considered each interpretation in relation to the research question and the dataset as a whole. Interpretive differences that could not be resolved through this initial discussion were discussed within the 5-member qualitative research team, which also reviewed the developing thematic structure. During these discussions, the team collectively examined the relevant data extracts, their broader context within the interviews, and the analytic rationale underlying the alternative interpretations. The corresponding codes or themes were revised, merged, separated, or renamed through discussion until consensus was reached. This collaborative process was used to enrich interpretation through reflexive dialogue rather than to establish intercoder reliability.

The analysis combined deductive and inductive elements: the quantitative findings served as sensitizing context for areas requiring further explanation, but no fixed coding framework or themes were specified in advance. The analysis proceeded recursively through familiarization with the data, generating initial codes, generating initial themes, reviewing potential themes, defining and naming themes, and producing the report. Candidate themes were reviewed in relation to the coded extracts and the full dataset to assess their coherence, distinctiveness, and relevance to the research question. Coding and theme development were treated as iterative and interpretive processes grounded in sustained engagement with the data. The research team met regularly to examine emerging interpretations and reflect on how analytic assumptions shaped the analysis. YL provided methodological oversight throughout the qualitative analysis and reviewed the final thematic structure.

All interviews were conducted in Chinese, and transcription, coding, and theme development were also undertaken in Chinese. The full interview transcripts were not translated into English. After completion of the qualitative analysis, the finalized theme names and narrative descriptions of the qualitative findings were translated into English for reporting in the manuscript. Selected participant quotations used to illustrate the themes were also translated into English. Translation was performed by a professional translator assigned by an independent translation company. The translator was a native English speaker proficient in Chinese. YL and YY verified each translated component against the original Chinese analytic materials and interview transcripts and subsequently reviewed the translations with the translator to ensure semantic accuracy, contextual fidelity, consistency with the original analysis, and preservation of the participants’ intended meanings. Any discrepancies in wording or interpretation were resolved through discussion with reference to the original Chinese materials. No generative AI or machine translation tools were used at any stage of the translation or verification process.

Integration was undertaken at 3 stages, following established guidance for mixed methods integration [56]. The mixed methods study flow and integration process are summarized in Figure 1. First, the quantitative findings informed purposive sampling for the qualitative phase and refinement of the semistructured interview guide. Second, the qualitative findings were used to explain and contextualize the quantitative associations identified in the first phase. Third, the quantitative and qualitative strands were compared in the Results section to assess convergence, complementarity, and divergence. During interpretation, the socioecological perspective was used to organize the integrated findings across individual, interpersonal support, and technology use context levels.

This study was performed in accordance with the Declaration of Helsinki and received ethics approval from the Shanghai University of Traditional Chinese Medicine Ethics Committee (approval number 2024-1-16‐06). All participants were informed of the study objectives and significance, voluntarily signed informed consent forms, and were advised that they could withdraw from the study at any time without providing a reason, without any impact on their medical care or legal rights. The collected data were anonymized or coded and securely stored in accordance with the General Data Protection Regulation guidelines.

Of the 360 questionnaires distributed, 350 were returned, whereas 10 were not returned. Seven returned questionnaires were excluded: 5 because more than 10% of the items were unanswered and 2 because they were duplicate submissions. Examination of the 5 incomplete questionnaires indicated that missing responses were dispersed across items, scales, and questionnaire sections, without a consistent concentration in any particular item, scale, or section. Accordingly, descriptive examination did not identify an evident systematic pattern of missingness. After these exclusions, the final quantitative analytic sample comprised 343 participants, with no missing values in the final dataset (Figure 2). Among the 343 participants, the median age was 68.0 (IQR 64.0‐73.0) years, and 58.3% (200/343) were male. Most participants had ischemic stroke (266/343, 77.5%), and the median mRS score was 2.0 (IQR 1.0‐3.0). Regarding digital health technologies, 80.2% (275/343) reported an intention to use digital health technologies, and sometimes use was the most commonly reported frequency (144/343, 42.0%). Detailed participant characteristics are presented in Table 1.

Data analysis revealed that the median TS total score among older patients with stroke was 25.0 (IQR 22.0‐33.0). The median scores for the subscales were 10.0 (IQR 8.0‐17.0) for technology tension, 10.0 (IQR 8.0‐12.0) for technology fear, and 6.0 (IQR 5.0‐8.0) for privacy and security concerns (Table 2).

Descriptive statistics were used to summarize factors associated with technophobia toward digital health technologies among older patients with stroke. The median eHealth literacy score was 30.0 (IQR 18.0‐34.0), the median perceived social support score was 66.0 (IQR 52.0‐72.0), and the median stroke self-efficacy score was 83.0 (IQR 64.0‐89.0). Detailed results are presented in Table 3.

Univariable analyses showed that total technophobia scores differed significantly by educational level, employment status, monthly per capita household income, mRS score, number of stroke episodes, current self-rated health status, intention to use digital health technologies, frequency of digital health technology use, use of health-related online media, and use of home-based health management devices (P<.05; Table 1).

Spearman correlation analysis showed that technophobia toward digital health technologies among older patients with stroke was positively correlated with age (ρ=0.133; P=.01) and negatively correlated with eHealth literacy (ρ=−0.632; P<.001), perceived social support (ρ=−0.645; P<.001), and stroke self-efficacy (ρ=−0.511; P<.001). The detailed correlation coefficients and P values are presented in Table 4, and the direction and relative magnitude of the correlations among the study variables are visualized in Figure 3.

Fifteen variables that were associated with technophobia at P≤.10 in the univariable or correlation analyses were entered into the multivariable linear regression model. Residual diagnostics did not indicate substantial departures from residual normality or an obvious funnel-shaped pattern in the plot of standardized residuals against standardized predicted values, suggesting no marked heteroscedasticity. All tolerance values were >0.1, and all variance inflation factors were <10, indicating no evidence of problematic multicollinearity among the independent variables. In the adjusted model, monthly per capita household income, frequency of digital health technology use, eHealth literacy, perceived social support, and stroke self-efficacy were associated with technophobia toward digital health technologies (P<.05). Specifically, patients with a monthly per capita household income of Renminbi (Chinese yuan; RMB) 5000‐6999 (RMB 1=US $0.14 as of January 29, 2026) had lower technophobia scores than those with an income of less than RMB 1000 (B=−3.60, 95% CI −6.44 to −0.76). Compared with patients who never used digital health technologies, those who used them occasionally also had lower technophobia scores (B=−3.98, 95% CI −6.94 to −1.03). Higher eHealth literacy (B=−0.45, 95% CI −0.58 to −0.32), perceived social support (B=−0.35, 95% CI −0.42 to −0.29), and stroke self-efficacy (B=−0.08, 95% CI −0.13 to −0.02) were associated with lower technophobia scores. Detailed results for the statistically significant variables are presented in Table 5, and the diagnostic plots and complete multivariable linear regression results are provided in Multimedia Appendix 2.

Participants’ accounts suggested that competence in using digital health technologies was underpinned by both the ability to appraise online information and the confidence to persist when difficulties arose. Together, these resources appeared to shape whether uncertainty was experienced as an unmanageable threat or as a challenge that could be addressed through continued effort.

Quantitative findings were integrated with the qualitative themes using the adapted socioecological perspective to examine convergence and complementarity across the individual, interpersonal support, and technology-use context domains. Figure 4 summarizes this integration; the detailed joint display is presented in Multimedia Appendix 5.

At the individual level, participants who used digital health technologies occasionally had lower technophobia scores than those who had never used such technologies. The qualitative findings provided convergent support for this categorical association, indicating that familiarity acquired through use was accompanied by greater understanding of and trust in digital health technologies. An additional finding that emerged inductively from the qualitative interviews was that some participants expressed willingness to explore AI-assisted health management. Higher eHealth literacy and stroke self-efficacy were similarly associated with lower technophobia scores. Qualitative analysis further indicated that difficulty appraising online health information was accompanied by concerns about information credibility, fraud, privacy, and security, whereas greater confidence in learning and problem solving was reflected in continued efforts to use digital health technologies and less apprehension during use.

At the interpersonal level, higher perceived social support was associated with lower technophobia scores. The qualitative findings further clarified this association, indicating that practical assistance, explanations, feedback, and reassurance from family members and others helped participants navigate operational uncertainty and alleviate fear during technology use. Interviews also revealed that, when such support was limited, some patients maintained a strong willingness to use digital health technologies but remained vulnerable to practical difficulties, often accompanied by emotional distress and frustration.

At the technology use context level, participants with a monthly per capita household income of Renminbi (Chinese yuan; RMB) 5000‐6999 (RMB 1=US $0.14 as of January 29, 2026) had lower technophobia scores than those in the reference category of less than RMB 1000. Qualitative findings concerning affordability helped contextualize this category-specific association: participants experiencing financial strain perceived operational errors as potential sources of additional expense and described heightened vigilance, hesitation, and tension when using digital health technologies. These accounts highlighted the perceived financial consequences of technology use as an important contextual dimension of technophobia toward digital health technologies.

Taken together, the integrated findings showed that the correlates and subjective experiences of technophobia toward digital health technologies spanned three socioecological domains: (1) individual capabilities and prior use experience, (2) interpersonal support, and (3) perceived financial consequences of technology use.

The principal contribution of this study is its integration of quantitative and qualitative evidence to explain technophobia toward digital health technologies among older patients with stroke across three interrelated domains: (1) individual capabilities and prior experience, (2) interpersonal support, and (3) the context of technology use. The median technophobia score was 25.0 (IQR 22.0‐33.0), which was lower than the value reported by Peng et al [19] among community-dwelling older adults. This descriptive difference may be related to the greater willingness to use digital health technologies, greater prior exposure to such technologies, and more immediate access to support among participants in the hospital-based sample.

The findings further revealed an important tension within the concept of technophobia. Patients’ concerns about additional costs, inaccurate or misleading information, privacy breaches, and online fraud were grounded in real-world risks, while poststroke cognitive, executive, and communication impairments may increase their digital vulnerability [57,58]. Research involving people with acquired brain injury has likewise shown that cognitive and executive impairments are associated with susceptibility to online fraud and its financial and psychological consequences [59]. Technophobia may therefore comprise intertwined rational and irrational elements and may, in turn, increase patients’ risk of digital exclusion and their tendency to avoid digital health technologies [60]. Future research should deepen the theoretical interpretation of technophobia by building on its existing conceptualization as an irrational fear.

At the individual level, technophobia should not be equated with unwillingness to use digital health technologies. Although perceived usefulness was not directly assessed, the coexistence of technology-related concerns and willingness to use digital health technologies may be interpreted within the Technology Acceptance Model, which identifies perceived usefulness as an important contributor to behavioral intention [61]. eHealth literacy and stroke self-efficacy may represent conceptually distinct capability-related resources relevant to technophobia: the former concerns the ability to access, appraise, and apply digital health information, whereas the latter concerns confidence in performing poststroke activities and self-management tasks [62,63]. However, stroke self-efficacy is conceptually distinct from self-efficacy for using digital health technologies; their respective independent associations with technophobia require further investigation.

The unadjusted findings for age and self-rated health should be interpreted with caution. Although age is an important demographic correlate of technophobia, previous studies have reported inconsistent findings regarding the relationship between age and digital health literacy [64,65]. In this study, the weak association observed in the unadjusted analysis did not remain statistically significant after multivariable adjustment, suggesting that age may serve primarily as a marker of accumulated differences in experience with technology, capabilities, and support resources. Similarly, participants with better self-rated health had higher technophobia scores in the univariable analysis, but this difference was no longer statistically significant after adjustment. Previous research has also shown that the association between internet use and self-rated health may be influenced by socioeconomic and health-related factors [66,67], indicating a complex pattern of selection effects and confounding. In addition, patients with better self-rated health may have engaged with digital health technologies more independently and evaluated the associated risks themselves, whereas those in poorer health may have used these technologies less directly or relied more heavily on others.

At the interpersonal level, the observed association between perceived social support and technophobia was consistent with previous research [26]. This association may be understood through the stress-buffering hypothesis of social support [68]. When individuals encounter stressors related to digital health technologies, lower perceived social support may weaken the buffering effects of internal cognitive regulatory processes [69], intensify threat appraisal, and consequently exacerbate technophobia toward digital health technologies [70]. These findings suggest that technophobia toward digital health technologies among older patients with stroke should not be understood solely as a problem of person–technology fit but should also be considered within the context of the interpersonal support environment in which patients are embedded. The digital health equity framework further suggests that the availability of such support depends not only on patients’ family or social networks but also on whether health care systems incorporate accessible technology assistance into digital care pathways [71,72].

At the technology use context level, economic circumstances emerged as one of the factors associated with technophobia toward digital health technologies among older patients with stroke [21]. Ongoing medical and rehabilitation expenditures after stroke may increase the financial burden on patients with lower incomes [1], while the use and maintenance of digital health technologies themselves require financial investment and these technologies undergo rapid updates. Economic stratification within the older population may therefore contribute to intragenerational disparities in access, increasing the psychological burden associated with engaging with digital health technologies and thereby exacerbating technophobia toward these technologies [22].

In the clinical implementation of digital health technologies, technophobia should not be used as an independent criterion for determining whether patients are suitable candidates for digital stroke care. Instead, it should be considered alongside eHealth literacy, prior technology use experience, stroke self-efficacy, and social support to identify the type and intensity of support required by individual patients. Assessment findings may inform the development of tiered support models, including independent use by patients, use with guidance from health care professionals, and shared use by patients and caregivers. Training should be centered on specific stroke rehabilitation or self-management tasks and incorporate practical demonstrations, repeated practice, and timely feedback. Digital health platforms should also adhere to principles of accessibility and usability, incorporating adaptive designs that accommodate poststroke impairments in language, cognition, vision, and motor function, while prioritizing information credibility, privacy protection, and safety as core design requirements [57].

Caregivers may be incorporated into the guidance and support processes for digital health technology use, provided that patients’ autonomy is respected and informed consent is obtained. Existing evidence suggests that telerehabilitation, remote training, and remote support are feasible approaches and may improve certain aspects of caregiver competence, self-efficacy, and health-related outcomes; however, findings regarding caregiver preparedness and caregiver burden remain inconsistent across studies, underscoring the continued need for technical assistance, digital skills training, and professional health care support [73,74]. Caregiver involvement may assist patients in understanding how to use digital health technologies, identifying trustworthy health information, and addressing problems encountered during use. For patients who lack family support or face a higher risk of digital exclusion, alternative pathways, including professional guidance, digital navigation support, in-person assistance, or other nondigital service options, should be considered to avoid making family support an implicit prerequisite for accessing digital health services [72].

The perspectives of older patients with stroke and their caregivers should be incorporated throughout the design, testing, and evaluation of digital health technologies. Patient advisory groups and participatory approaches can help identify needs that may not be captured through conventional usability assessments. Future research could integrate multimodal data, including behavioral logs, speech, images, and interaction trajectories, to characterize technology use under real-world conditions [75]. AI-assisted remote monitoring and virtual guidance may be prioritized for exploration among patients who have both the willingness and the requisite capacity to use these approaches, provided that professional oversight, algorithmic explainability, and clearly defined lines of responsibility are ensured [76].

At the health policy level, digital stroke programs should not be limited to providing patients with technology platforms but should incorporate caregiver training, community-based digital support, professional navigation services, and ongoing technical assistance into program design, workforce planning, and funding arrangements [73,77]. Governments, health care organizations, and payers should also consider the burdens imposed on patients by devices, internet connectivity, maintenance, and technical support, and explore sustainable reimbursement mechanisms that support digital follow-up, telerehabilitation, and digital navigation support. Procurement and regulatory criteria for digital health products could be further expanded to include stroke-specific accessibility, platform interoperability, privacy and data security, and AI governance requirements [78]. Beyond overall enrollment and usage rates, program evaluations should examine initial adoption, sustained engagement, and discontinuation patterns among patients with different levels of income, digital health literacy, functional status, and social support to determine whether digital services introduce new disparities in access.

This study has several limitations. First, recruitment from a single tertiary hospital in Shanghai may have introduced selection bias and limited the generalizability of the findings to community-dwelling stroke survivors and other care settings. Future multicenter studies should recruit participants from diverse hospital- and community-based settings across different geographic regions and health care contexts. Second, participation required individuals to understand the study content and to communicate their responses to the questionnaire or interview either in writing or orally. As a result, patients with severe cognitive, language, or communication impairments may have been underrepresented, and the findings may not have fully captured the features of technophobia among those unable to express their experiences with digital health technologies independently. Third, although overall poststroke disability and activities of daily living were assessed and analyzed using the mRS and Barthel Index, these measures may not adequately capture more specific functional dimensions closely related to digital health technology use. In particular, the severity of cognitive impairment among eligible participants, upper limb function, degree of care dependence, and access to digital devices were not systematically assessed as structured quantitative variables or incorporated into the statistical model. The absence of these variables may have resulted in residual confounding and limited the explanatory depth of the model with respect to technophobia and differences across patients with varying functional status. Future research should use multicenter, large-sample longitudinal designs and incorporate more specific stroke-related functional indicators and measures of digital device access to further validate and refine the present findings.

This study showed that technophobia toward digital health technologies among older patients with stroke was inversely associated with monthly per capita household income, frequency of digital health technology use, perceived social support, eHealth literacy, and stroke self-efficacy. Qualitative findings further suggested that participants who used digital health technologies more frequently expressed greater acceptance of AI apps and a stronger willingness to use them. In addition, some participants who reported lower perceived social support still expressed a strong willingness to use digital health technologies. Future multicenter, large-sample longitudinal studies that include additional variables are needed to further validate and refine these findings.