This was a 12-week, nonrandomized quasi-experimental parallel-group trial.

Patients were recruited from the cardiology wards of a tertiary general hospital in Hangzhou, Zhejiang Province, China, between July 2022 and March 2023. Potentially eligible participants were identified through daily review of ward admission records. Using consecutive sampling, trained research staff screened all patients admitted to the cardiology wards who had undergone PCI during the recruitment period against the prespecified eligibility criteria. Eligible patients received written information about the study and provided written informed consent before enrollment. Eligible participants met all the following inclusion criteria: (1) age of 18 to 75 years, (2) diagnosis of CHD according to World Health Organization criteria and status after PCI, (3) patient and family willingness to participate and cooperate, and (4) ownership of a mobile phone with the ability to use WeChat and telephone functions. Participants meeting any of the following exclusion criteria were not enrolled: (1) advanced heart failure (New York Heart Association class IV); (2) severe comorbid disease involving major organs (eg, brain, liver, or kidney) or other serious somatic conditions; (3) illiteracy or visual or hearing impairments precluding reading SMS text messages; and (4) cognitive impairment, communication disorders, or major psychiatric illness. Participation was voluntary and uncompensated.

On the basis of the pilot 8-item Morisky Medication Adherence Scale (MMAS-8) data (Multimedia Appendix 1), an expected between-group mean difference of 0.78 points at the 1-month follow-up and a pooled SD of 1.01 were assumed. Using PASS (version 15.0; NCSS, LLC), the sample size was calculated for a 2-tailed independent-sample t test with an α level of .05, power of 80%, and a 1:1 allocation ratio. The calculation indicated that 56 participants were required to complete the study, with 28 participants per group. After allowing for an anticipated 10% loss to follow-up, the recruitment target was set at 64 participants, with 32 participants per group. Ultimately, 92 participants were enrolled, including 46 (50%) participants in each group, thereby exceeding the prespecified recruitment target.

Participants were allocated to the intervention or control condition according to the cardiology ward to which they were admitted, with one ward assigned to each study condition. Ward-based allocation was used to maintain separation between study conditions, minimize intervention contamination, and preserve routine clinical workflow. Individual randomization within the same ward was considered less appropriate because participants shared the same clinical environment and ward-based care team, potentially increasing contamination across study conditions. Cluster randomization was not feasible because only 2 eligible wards were available, resulting in a single cluster per condition. Given the behavioral nature of the intervention, participants and treating clinicians could not be masked. Outcome assessors and data collection personnel were masked to allocation; coded participant identifiers were used, intervention records were inaccessible to assessors, and intervention delivery and outcome assessment were conducted by separate study personnel to minimize assessment bias.

Participants in the control group received standard postoperative counseling on medication, physical activity, and diet together with a printed education booklet on post-PCI care (Multimedia Appendix 2).

In addition to usual care, participants in the intervention group received a BCW-informed WeChat-based messaging intervention initiated immediately after hospital discharge. The program had 2 components. First, medication reminders were sent approximately 30 minutes before each participant’s usual dosing time based on baseline self-reported regimens [31]. Second, 1 nonreminder message was delivered daily at 8 AM for 12 weeks (Figure 1).

Nonreminder messages were drawn from a standardized message library (Multimedia Appendix 3) [26,27]. Development was guided by the BCW. Adherence barriers were first identified using the COM-B framework. Each barrier was then mapped to appropriate intervention functions and associated BCTs. Candidate messages were drafted and refined by an expert panel according to these mappings. Wording and usability were informed by patient feedback from pilot-testing. The final library covered medication information, management of adverse effects, benefits of adherence, self-monitoring strategies, behavioral techniques, and prompts for social support. Detailed COM-B–intervention function–BCT mapping and representative message examples are provided in Multimedia Appendix 3 [26,27]. The complete message library [32], which was previously published as part of our intervention development work, is also included for transparency and reproducibility.

The 12-week message delivery schedule, including reminder timing and the stage-specific distribution of nonreminder messages across capability-, motivation-, and opportunity-focused phases, is detailed in Multimedia Appendix 4. Weeks 1 to 3 focused on capability building, including PCI and medication knowledge, side effect management, and self-monitoring skills. Weeks 4 to 6 emphasized motivation by reinforcing adherence benefits, providing behavioral feedback, and strengthening psychological and social support. Weeks 7 to 12 targeted opportunity through habit formation, environmental cues, planning for travel or busy schedules, and self-reward to support long-term adherence. Participants could pause or opt out at any time and were instructed to call a 24-hour hotline for adverse events or worsening symptoms.

Disease-specific health status was assessed using the 19-item Seattle Angina Questionnaire (SAQ). The SAQ comprises 5 domains: physical limitation, angina stability, angina frequency, treatment satisfaction, and disease-related quality of life. Scores for each domain were transformed to a scale from 0 to 100 using the following formula: (observed score – lowest possible score)/(highest possible score – lowest possible score) × 100. Higher scores indicate better disease-specific health status, including fewer angina-related limitations and better functioning and quality of life. The SAQ has demonstrated acceptable internal consistency, with a reported Cronbach α coefficient of 0.76 [39].

All 92 participants were analyzed according to their initial group allocation using observed data; no imputation was required because outcome data were complete. For continuous outcomes, between-group differences at 12 weeks were estimated using analysis of covariance (ANCOVA), with treatment group as the fixed factor and the baseline value of the corresponding outcome as a covariate. Homogeneity of regression slopes was assessed using the treatment group–by-baseline interaction, and linearity was assessed by adding a quadratic baseline term. Residual normality and homoscedasticity were assessed using the Jarque-Bera and Breusch-Pagan tests, respectively. Because heteroscedasticity was identified in several models, HC3 heteroscedasticity-robust SEs were used for all final models. When homogeneity of regression slopes was not supported, the interaction term was retained, and adjusted group estimates were calculated at the overall mean baseline value. Detailed ANCOVA diagnostic results are presented in Multimedia Appendix 5. Sensitivity analyses included a per-protocol analysis restricted to the 87 participants who completed the full assigned care protocol and a difference-in-differences analysis comparing individual changes from baseline to 12 weeks between groups. The per-protocol analysis used the same ANCOVA models as the primary analysis, whereas the difference-in-differences analysis used HC3 robust SEs. All statistical tests were 2 sided, with a P value below .05 indicating statistical significance. Statistical analyses were conducted using SPSS Statistics (version 25; IBM Corp) and Python (version 3.12; Python Software Foundation).

This study received ethics approval from the ethics committee of the First Affiliated Hospital of Zhejiang University (11T20220185B). Written informed consent was obtained from all participants. All procedures complied with good clinical practice and the Declaration of Helsinki. Only deidentified data were analyzed, and access was restricted to the study team. Data handling complied with national data protection regulations and institutional data security policies.

Between July 2022 and March 2023, a total of 180 consecutive patients were screened for eligibility. Of these 180 patients, 88 (48.9%) were not enrolled: 56 (63.6%) did not meet the eligibility criteria, 23 (26.1%) declined to participate, and 9 (10.2%) were discharged or transferred before enrollment or did not complete the baseline assessment. The remaining 92 participants were nonrandomly allocated by ward admission to the intervention group (n=46, 50%) or the control group (n=46, 50%; Figure 2). Assigned care was completed by 95.7% (44/46) of the participants in the intervention group and 93.5% (43/46) of the participants in the control group. Overall, 94.6% (87/92) of the participants completed the full assigned care protocol. Although 5.4% (5/92) of the participants did not complete the full protocol, all completed the 12-week outcome assessment. Therefore, all 92 participants were included in the outcome analyses. Baseline characteristics were similar between groups (Table 1). The mean age was 61.11 (SD 8.52) years in the intervention group and 61.74 (SD 5.60) years in the control group. Educational attainment was generally low, with 58.7% (54/92) having junior high education or lower and 15.2% (14/92) having a college education or higher. Most participants were retired (57/92, 62%) or employed (35/92, 38%). A total of 16.3% (15/92) had a monthly per capita household income of less than ¥3000 (¥1=US $0.15 as of August 18, 2026), 52.2% (48/92) had an income of ¥3000 to ¥5000, and 31.5% (29/92) had an income above ¥5000. Comorbid chronic disease was present in 59.8% (55/92) of participants, and 53.3% (49/92) were newly diagnosed with CHD.

ANCOVA diagnostic analyses supported the linearity assumption for all outcome models. Homogeneity of regression slopes was supported for all outcomes except MMAS-8 and SAQ angina frequency, for which the group-by-baseline interactions were significant (P<.001 and P=.03, respectively). Departures from residual normality or homoscedasticity were identified in several models; therefore, HC3 heteroscedasticity-robust SEs were used throughout. Detailed diagnostic results can be found in Multimedia Appendix 5.

At the overall mean baseline MMAS-8 score of 6.05, the adjusted 12-week mean MMAS-8 score was higher in the intervention group than in the control group (7.40, SE 0.05 vs 6.22, SE 0.10; adjusted mean difference 1.18, 95% CI 0.96-1.40; P<.001; Table 2).

Secondary outcome findings were generally consistent with those for the primary outcome. Compared with the control group, the intervention group had higher BMQ necessity subscale scores (adjusted mean difference 1.62, 95% CI 1.06-2.17), lower BMQ concerns subscale scores (adjusted mean difference −3.25, 95% CI −3.87 to −2.63), and a higher necessity-concerns differential (adjusted mean difference 5.01, 95% CI 4.03-5.98; P<.001 in all cases). SEAMS scores were also higher in the intervention group (adjusted mean difference 4.04, 95% CI 3.21-4.87; P<.001).

For the SAQ, the intervention group had more favorable adjusted scores for physical limitation, angina stability, angina frequency, and quality of life. Treatment satisfaction did not differ significantly between groups (adjusted mean difference 0.40, 95% CI −2.02 to 2.82; P=.74). The SAQ summary score was higher in the intervention group (adjusted mean difference 6.13, 95% CI 4.72-7.53; P<.001).

Sensitivity analyses supported the robustness of the primary findings. In the per-protocol analysis of the 87 participants who completed the full assigned care protocol, the adjusted between-group difference in MMAS-8 score was 1.14 (95% CI 0.92‐1.36; P<.001). The difference-in-differences analysis also favored the intervention (between-group difference in change=1.08, 95% CI 0.75‐1.40; P<.001). Findings for the secondary outcomes were generally consistent with those of the primary analysis, whereas SAQ treatment satisfaction did not differ significantly between groups (95% CI −1.93 to 2.82; P=.71). Detailed results can be found in Multimedia Appendix 6.

In this quasi-experimental study, a WeChat-based messaging intervention was associated with higher medication adherence and more favorable medication beliefs, self-efficacy, and disease-specific health status among patients after PCI. MedText-PCI was designed to address both unintentional and intentional barriers to medication taking, and the observed differences extended beyond adherence to broader behavioral and health outcomes. Unlike generic or reminder-dominant approaches, MedText-PCI explicitly mapped adherence barriers to BCTs and corresponding message components targeting capability, opportunity, and motivation. The assessment of medication beliefs and self-efficacy alongside adherence provided preliminary insights into outcomes aligned with these behavioral targets.

First, the observed improvement in medication adherence may reflect the intervention’s simultaneous targeting of unintentional and intentional barriers to medication taking. Contemporary reviews indicate that medication nonadherence reflects both unintentional execution failures, such as forgetting and routine disruption, and intentional decisions shaped by medication beliefs and motivation [40]. MedText-PCI was explicitly designed to address both pathways. Timely prompts before dosing served as contextual cues intended to reduce reliance on memory and help integrate medication taking into daily routines, thereby supporting opportunity within the COM-B model [41]. On the intentional pathway, nonadherence may reflect doubts about medication necessity, concerns about adverse effects, and ambivalence toward long-term therapy. Educational messages were designed to improve understanding of PCI medications and the practical management of side effects, whereas motivational content sought to reinforce perceived necessity and address medication-related concerns. The observed widening of the necessity-concerns differential was consistent with changes in medication-related motivation [42]. Improvements in self-efficacy may additionally reflect greater psychological capability for managing medication taking [43].

Comparison with previous SMS text messaging–based cardiovascular interventions provides important context for interpreting our findings. Findings from earlier cardiovascular trials have been mixed. The original Text4Heart trial [19] reported an improvement in self-reported medication adherence, whereas TXT2HEART [21], TEXTMEDS (Text Messages to Improve Medication Adherence and Secondary Prevention) [22], Text4Heart II [23], and the StAR (SMS-Text Adherence Support) trial [44] found no clear improvement in adherence relative to usual care. Against this background, the higher MMAS-8 scores observed with MedText-PCI provide preliminary evidence that a messaging intervention focused specifically on medication taking after PCI may support short-term adherence. Several differences may help contextualize these findings. Unlike interventions that rely mainly on medication reminders or general supportive messages, MedText-PCI combined medication-timed reminders with educational and motivational content. These messages were developed by linking COM-B–identified adherence barriers to BCW intervention functions, BCTs, and specific message components [32]. By comparison, TXT2HEART [21] used a longer intervention with progressively reduced message frequency, TEXTMEDS [22] delivered broader weekly supportive messages after acute coronary syndrome, and Text4Heart II [23] evaluated adherence primarily using pharmacy-based medication possession ratios. These differences in intervention intensity, behavioral targets, adherence measurement, follow-up duration, and usual care conditions may partly explain variation across studies. However, they do not establish that the MedText-PCI design is more effective, particularly because the present study was nonrandomized, relied on self-reported adherence, and assessed outcomes after only 12 weeks.

Second, the favorable pattern observed with MedText-PCI extended beyond medication adherence to theoretically relevant psychological and health-related outcomes. Previous messaging studies have examined similar outcomes, although their findings have been mixed. Text4Heart found no clear between-group differences in self-efficacy or illness perceptions despite improving self-reported adherence [19], whereas Park et al [45] reported no significant improvement in medication self-efficacy following reminder and educational SMS text messages among patients with CHD. Similarly, Txt2Prevent found no significant effects on cardiac self-efficacy or generic health-related quality of life after acute coronary syndrome [46], and neither the StAR trial [44] nor Text4Heart II [23] demonstrated clear benefits for generic health status or quality of life outcomes. TEXT ME (Tobacco, Exercise, and Diet Messages) improved several cardiovascular risk factors and lifestyle behaviors but did not assess medication-specific beliefs or self-efficacy [47]. In MedText-PCI, medication beliefs and self-efficacy were assessed as outcomes proximal to the intervention’s proposed behavioral targets, whereas disease-specific health status represented a potential downstream health outcome. The more favorable scores observed for these outcomes are consistent with the intervention’s theoretical rationale. However, treatment satisfaction did not differ between groups, suggesting that the intervention may have had limited influence on the relational and service attributes captured by this SAQ domain, such as interpersonal care, continuity, and access. High baseline scores and the relatively short 12-week intervention period may have further limited the potential for improvement. These findings suggest that digital adherence support may complement rather than replace clinician-led relational care [48]. Future versions could therefore incorporate bidirectional communication with the care team and assess patient experience outcomes that are more sensitive to communication and partnership [49,50]. Moreover, because these variables were assessed concurrently and were largely self-reported, the findings do not establish that changes in beliefs or self-efficacy mediated the observed adherence difference.

The clinical relevance of the observed between-group differences, however, requires separate consideration. No validated minimal clinically important difference (MCID) for the MMAS-8 has been established specifically for patients with CHD. Although the adjusted between-group difference was 1.18 points, both adjusted means remained within the conventional medium-adherence range [51]. Thus, the finding indicates improved self-reported adherence, but its clinical importance cannot be established based on an MMAS-8 MCID [52]. For the SAQ, the mean differences in physical limitation (8.21 points) and the summary score (6.13 points) met commonly cited clinically important thresholds of approximately 8 and 5 points, respectively [53,54]. The other SAQ domain differences did not reach their corresponding thresholds. However, because these thresholds primarily describe within-person changes, caution is warranted when applying them to between-group differences. Better medication adherence may plausibly improve long-term cardiovascular outcomes. Nevertheless, because this 12-week pilot study did not assess cardiovascular events, the observed improvement in MMAS-8 scores cannot be translated into a specific cardiovascular risk reduction. Larger, longer-term randomized trials using objective adherence measures and adjudicated cardiovascular outcomes are therefore needed.

Delivery through WeChat may have facilitated access because the platform was familiar to participants, but this context also limits generalizability. Although the behavioral content could be adapted to SMS text messaging, WhatsApp, patient portals, or other platforms, its effectiveness may vary according to platform functions, costs, accessibility, cultural expectations, and integration with health care systems [55]. Digital exclusion also warrants attention, particularly among older adults and people with limited digital literacy, restricted internet access, or shared devices [56]. Future studies should therefore evaluate the intervention across platforms and settings while considering simplified interfaces, user training, caregiver support, and alternative delivery options.

Commercial messaging platforms also raise privacy and data security concerns because notifications displayed on locked screens or shared devices may disclose medication use or health conditions [57]. Implementation should minimize sensitive content; provide users with control over notification previews and opting out; and establish appropriate procedures for consent, access, data retention, and deletion. Platform-specific governance, regulatory compliance, and users’ privacy preferences should also be evaluated [58]. Given the 12-week follow-up, larger and longer-term trials should examine the durability of intervention effects, equity, privacy, cost-effectiveness, and integration with clinical workflows and caregiver support.

This study has several limitations. First, despite enrollment exceeding the prespecified minimum, the nonrandomized design, modest sample size, and single-center setting limit causal inference and generalizability. Ward-based allocation and the inability to mask participants and clinicians may have introduced ward-level confounding, selection and performance biases, Hawthorne effects, and differential attention despite the use of common institutional care procedures and separate outcome assessment. Second, the 12-week intervention and follow-up periods precluded evaluation of the durability of the observed differences or their association with longer-term clinical outcomes. Third, medication adherence was self-reported and may have been affected by recall and social desirability biases; the absence of objective adherence measures also prevented corroboration of the MMAS-8 findings. The psychological and health status outcomes were assessed concurrently and were largely self-reported. These assessments cannot determine whether changes in these outcomes mediated the observed adherence difference. Finally, the uniform message library and delivery schedule provided limited scope for personalization, and privacy perceptions and data security outcomes were not formally assessed. Generalizability may also be limited among individuals with lower digital literacy or restricted digital access and in settings where WeChat is not routinely used.

In this quasi-experimental study, a WeChat-based messaging intervention was associated with improvements in medication adherence, medication beliefs, self-efficacy, and disease-specific health status among post-PCI patients. By addressing cognitive, motivational, and social support–related barriers in addition to forgetfulness, the program extended beyond reminder-only approaches and may offer a low-intensity strategy for supporting secondary prevention. Larger, adequately powered multicenter randomized trials with longer follow-up are needed to evaluate the intervention’s effectiveness, sustainability, cost-effectiveness, and implementation across diverse settings using objective adherence measures. Future research could examine whether AI can identify patient-specific barriers to medication adherence and use this information to select and deliver the most appropriate messages from an established message library.