The Txt2Prevent study is a mixed method, assessor-blinded, randomized controlled trial with a parallel group design. The study protocol and intervention development have been previously reported [26]. This study is reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) eHealth checklist [27] and is registered at ClinicalTrials.gov (NCT02336919).

Patients with a diagnosis of ACS, as identified by clinical staff, were recruited from St. Paul’s Hospital, a tertiary care hospital, in Vancouver, Canada between June 2015 and October 2016. Patients were eligible to participate if they had ACS (unstable angina or any type of myocardial infarction) as their primary admitting diagnosis, had daily access to a phone with SMS text messaging capabilities, were able to provide informed consent, and were able to read and understand English. Exclusion criteria comprised having coronary artery bypass graft surgery as a treatment for the ACS admission, having a prescheduled surgery within the study period, an expectation that the individual would not survive the duration of the study due to non-CVD reasons, being discharged to a long-term care center, or living outside the province of British Columbia. As this was a pilot study, the sample size was based on convenience. We aimed to recruit 76 participants as we previously estimated this was feasible over 6 months of recruitment. All participants provided written informed consent. Ethics and institutional approvals were obtained from Providence Health Care Research Ethics Board and Simon Fraser University’s Office of Research Ethics.

Baseline questionnaires, which included demographic information as well as measures of self-management, health-related quality of life, and cardiac self-efficacy, were administered in-person in the hospital when possible or within 7 days after discharge. Clinical information was gathered from the participant’s medical record.

After participants completed the baseline questionnaires and were discharged from the hospital, they were randomly assigned to either the intervention (Txt2Prevent plus usual care) or usual care. A statistician not associated with the study generated a random allocation schedule, which randomized participants in a 1:1 ratio using variable block sizes, stratified by sex. A research assistant not involved in recruitment or outcome assessment accessed a secure randomization database to obtain allocations for each patient and informed participants of their group assignment.

Participants in the Txt2Prevent group received 48 unique, automated, one-way messages over 60 days following randomization in addition to usual care. An additional 4 messages relating to study administration (eg, indicating the end of the study and requesting participants to inform us if they were readmitted; see SMS text messages sent on days 7, 26, 45, and 60 in Multimedia Appendix 1) were also sent during the study period. The SMS text messages were delivered at a time of day specified by the participant, began after the participant was randomized, and were sent daily for the first 36 days and then every other day until day 60. We had received feedback from patients in the design stage that it would be helpful to have more support in the beginning, but that daily messages may not be required for the whole period. We primarily covered time-sensitive information about recovery in the first 36 days. Once we covered the primary recovery topics suggested by our clinical advisory committee, we then switched to focus more on healthy living SMS text messages for the remainder of the study period. We chose 60 days for the program length because the initial period after discharge is the highest risk for readmission [4,28]. In addition, many patients will have seen their cardiologist or started a cardiac rehabilitation program by then [29] as well as may be starting to adjust to their new normal or have returned to work [30,31]. SMS text messages covered a range of topics, from time-sensitive information regarding their recovery (eg, timely follow-up with their health care professional) to general healthy living advice (eg, SMS text messages regarding physical activity, diet, and psychosocial health), and were delivered in a prespecified order (Table 1 and Multimedia Appendix 1). Participants received different SMS text messages on 2 occasions based on their smoking status; no other aspects were personalized. The usual care group did not receive any SMS text messages or have any contact from research staff during the study period.

The primary outcome was follow-up scores (controlled for baseline scores) in self-management domains as measured by the Health Education Impact Questionnaire (heiQ; version 3) [32]. The heiQ comprises 40 questions that cover 8 domains in total. All domains were measured and reported on separately: Health-directed behavior (4 questions), Positive and active engagement in life (5 questions), Emotional distress (6 questions), Self-monitoring and insight (6 questions), Constructive attitudes and approaches (5 questions), Skill and technique acquisition (4 questions), Social integration and support (5 questions), and Health service navigation (5 questions). As per the questionnaire’s scoring instructions, each domain score was calculated by averaging Likert scale responses (scaled from 1 to 4). Higher values are desirable, except for the Emotional distress domain. The heiQ was developed using item response theory and structural equation modeling, and the subscales have “acceptable” to “high” internal consistency (Cronbach α ranging from .70 to .86, depending on the domain) [32].

The secondary outcomes were health-related quality of life, cardiac self-efficacy, medication adherence, and health care resource use. Health-related quality of life was measured with the EQ-5D-5L [33], using health state valuations derived from a representative sample of the Canadian adult general population [34]. Self-reported health status was also captured using the EQ-5D-5L Visual Analog Scale (EQ VAS), a 0-100 VAS with anchors defined as “the best health you can imagine” (100) and “the worst health you can imagine” (0) [35]. The EQ-5D has been validated and used around the world. We used the 5-level scale as it has higher discriminatory power, interobserver reliability, and test–retest reliability than the 3-level EQ-5D [33,36]. The 3-level scale was validated in an ACS population, and had high usability, reasonable criterion validity when compared with other quality of life scales, and good test–retest reliability [37]. Cardiac self-efficacy was measured with a modified Sullivan Cardiac Self-Efficacy scale (CSE) [38], such that scores were calculated for the 2 domains (Control symptoms and Maintain function) as well as for the total by averaging Likert scale responses (0-4). For the modifications, we combined the first 4 questions regarding symptom control into 2 questions, as well as added 3 questions about diet and emotional well-being. We calculated the total for the original questions (Total) as well as a total including the additional questions (Total plus). The original Sullivan CSE scale has high internal consistency (Cronbach α of .90 and .87 for the 2 scales, respectively), and good convergent and discriminant validity when compared with other distress and disability scales [38]. For our modifications, Cronbach α were .71, .76, .79, and .82 for “Control symptoms,” “Maintain function,” “Total,” and “Total plus,” respectively. Medication adherence was measured with the Morisky Medication Adherence Scale (MMAS-8) [39-41]. As per questionnaire documentation, we calculated an adherence score on a scale from 1 to 8 and categorized participants as having low (<6), medium (6 to <8), or high adherence (8). This 8-item medication adherence scale has good internal consistency (Cronbach α=.83) and reliability when assessed in a hypertensive population [39]. It has good sensitivity (93%) and moderate specificity (53%) [39]. We also assessed how many participants at follow-up reported taking the recommended medications for post-ACS treatment [42]: acetylsalicylic acid, ticagrelor/clopidogrel, a statin, a beta blocker, and an angiotensin-converting-enzyme inhibitor/angiotensin receptor blocker for those with reduced vascular function. Health care resource use over the 60-day follow-up period (eg, visits to health care practitioners, visits to hospitals, cardiac rehabilitation program participation) was self-reported through a questionnaire developed by the research team. Any self-reported hospitalizations were verified with hospital records. Two blinded assessors (MM and MT) categorized hospital readmissions as cardiac or noncardiac.

Study feasibility was assessed through descriptive statistics on recruitment rates, follow-up rates, questionnaire completion rates, method of questionnaire completion (eg, postal mail, phone), percentage of participants randomized within 7 days, and percentage of participants who completed follow-up within 6 weeks after finishing the 60-day study period. In addition, study staff kept a log of barriers encountered. Acceptability was measured via two 5-level Likert scale survey questions that asked how satisfied participants were with the program (strongly disagree to strongly agree) and whether they thought the program helped them manage their condition. Acceptability was also assessed via 2 questions in semistructured phone interviews—specifically whether they would recommend the program to other patients with ACS and whether they read the SMS text messages. Participants with a range of demographic characteristics who were randomized to the Txt2Prevent group were invited to participate in the semistructured interviews after the 60-day study period. Detailed findings from the interviews, which covered participants’ experiences with the program and gathered feedback on program attributes, will be presented in a separate paper.

Follow-up questionnaires were administered 60 days after randomization primarily via postal mail, except for the health care resource use questions, which were completed over the phone for most participants due to its complex branching. All surveys were administered at baseline and at follow-up, except for the medication adherence scale and health care resource use questionnaire, which were only administered at follow-up.

As this was a pilot study, we undertook similar analyses to what is anticipated for a full trial and considered them exploratory in nature. Descriptive statistics are presented as the mean and SD, or count data with percentages. Analyses were conducted following the intent-to-treat principle. Only complete cases were analyzed. For continuous data from the heiQ, EQ-5D-5L, EQ VAS, and CSE questionnaires, analysis of covariance (ANCOVA) was used to test the effect of group assignment on the follow-up scores when controlling for baseline scores. We then reran the ANCOVA adjusting for age and sex as prespecified covariates as well as previous CVD status for the heiQ, and previous CVD status and marital status for the cardiac self-efficacy, due to their prognostic value [43]. We tested for the following assumptions for ANCOVA tests: independence of covariate and treatment effect, homogeneity of regression slopes, linearity, normality of residuals, homoscedasticity, and homogeneity of variances. The CSE scales, the EQ-5D-5L, and the EQ VAS were negatively skewed, which was primarily driven by outliers. We conducted the analyses with and without the outliers and present results of both analyses in situations where the outlier impacted the conclusion. We also calculated Cohen d effect sizes (mean difference score from the Txt2Prevent group minus mean difference score for the usual care group divided by pretest pooled SD [44]) to provide more context for the continuous data results. For our questionnaires, a negative Cohen d effect size implies the Txt2Prevent group had worse outcomes than the usual care group.

For count data from the health care resource use questionnaire (eg, number of readmissions), we used negative binomial regression analyses as our data had some overdispersion. For these analyses, we adjusted for age and sex, as prespecified. For binary response data from the health care resource use questionnaire (eg, cardiologist visit within 60 days—yes/no), we used a robust Poisson regression to determine relative risk as our outcomes occurred frequently [45]. For the robust Poisson regression, we adjusted for age and sex as prespecified, as well as geographic region [6] and income [6]. Chi-square tests were used for categorical variables unless there were low expected counts, in which case the Fisher exact test was used. Analyses were performed using SPSS (version 25; IMB Corp). Statistical significance was set at P<.05.

Four hundred patients were assessed for eligibility from June 2015 to October 2016. After excluding those who did not meet inclusion criteria and those who declined to participate, 76 participants were randomized (Figure 1).

The mean age of participants was 60 years (SD 9.3) and 73% (55/75) were male (Tables 2 and 3). Nine participants did not complete the study (2 withdrew, 3 failed to complete any of the follow-up questionnaires, and 4 partially completed the follow-up questionnaires; Figure 1). There were no statistically significant differences in baseline characteristics between those who did not complete the follow-up assessments compared with those who did, including in age (t₇₃=–0.16 [2 tailed], P=.87), sex (Fisher exact test, P=.18), and whether or not they completed education past high-school (Fisher exact test, P=.09; full analyses not shown). Data collection ended in December 2016.

There were no statistically significant differences between groups for the heiQ scores in either the unadjusted (Multimedia Appendix 2) or adjusted model in any of the 8 domains (adjusted mean difference [Txt2Prevent minus usual care] for each domain: Health-directed activity: –0.13, 95% CI –0.39 to 0.13, P=.31; Positive and active engagement in life: 0.03, 95% CI –0.19 to 0.25, P=.76; Emotional distress: 0.04, 95% CI –0.22 to 0.29, P=.77; Self-monitoring and insight: –0.14, 95% CI –0.33 to 0.05, P=.15; Constructive attitudes and approaches: –0.10, 95% CI –0.36 to 0.17, P=.47; Skill technique and acquisition: 0.05, 95% CI –0.18 to 0.27, P=.69; Social integration and support: –0.12, 95% CI –0.34 to 0.10, P=.27; Health services navigation: –0.05, 95% CI –0.29 to 0.19, P=.69) (Table 4). Cohen d effect sizes were all below 0.20, indicating negligible effects, except for the Self-monitoring and insight domain, where the Txt2Prevent group had worse outcomes than the usual care group, estimated at a small negative effect (d=–0.48; Table 4).

There were no statistically significant differences in EQ-5D-5L health state values (P=.51) or EQ VAS scores (P=.71; Table 5). For cardiac self-efficacy, in the adjusted models, there were no statistically significant differences between the 2 groups, except for the “Total plus” domain, where the Txt2Prevent group had worse outcomes (Control symptoms: P=.10; Maintain function: P=.05; Total: P=.05; Total plus: P=.03; Table 5). The statistically significant finding on the “Total plus” scale was due to an influential outlier that impacted the normality assumptions of ANCOVA. When the influential outlier was excluded in the adjusted analysis, the P value for the Total plus scale was no longer significant (P=.05). Depending on the self-efficacy domain, there were small or medium negative effects for the Txt2Prevent group (ie, this group had worse outcomes) based on the Cohen d values for the self-efficacy scores (Table 5).

There were no statistically significant differences in the mean medication adherence scores between the 2 groups (P=.27; Multimedia Appendix 2). When categorized into low, medium, and high adherence, 34% (11/32) of those in the Txt2Prevent group and 42% (15/36) of those the usual care group were classified as high adherers (χ²₂=2.10, P=.35). There were no statistically significant differences between the 2 groups for the categories of cardiac medications they were prescribed (Fisher exact test; P values ranged from .24 for statins to >.99 for beta blockers; Multimedia Appendix 3).

There were no differences between the groups in either the percentage of participants who visited the hospital or the mean number of visits to the hospital for all-cause or cardiac visits (Table 6). There were no differences in whether participants had visited a family physician, joined a cardiac rehabilitation program, or visited a cardiologist over the study period in adjusted analyses (Multimedia Appendix 4).

Recruitment of the target sample took 17 months (Table 7), which was longer than the anticipated 6 months. As much as 55.8% (223/400) of patients we approached were ineligible (Figure 1), of which 34.1% (76/223) were scheduled for surgery and 33.6% (75/223) did not own a cell phone. Of those eligible, 55.8% (96/172) declined to participate (Figure 1). Our randomization system worked well as 97% (74/76) of participants were randomized within our target of 7 days of discharge and 66% (50/76) were randomized within 2 days (Table 7). Those that were not randomized within the target time frame were due to delayed completion of baseline questionnaires. Almost 91% of participants (69/76) completed follow-up for the primary outcome. We obtained all follow-up questionnaires for 88% (67/76) of participants and at least one follow-up questionnaire for 93% (71/76) of participants; however, one participant was excluded from the analysis because they were admitted for a noncardiac reason. While we developed electronic versions to provide an alternative option, the majority of participants were willing to complete the questionnaires in their default format (60/69 [87%] completed packaged questionnaires by postal mail and 65/69 [94%] completed the health care resource use by telephone; Table 7).

We had a technical problem with our delivery system 11 months into recruitment where 80 SMS text messages were not delivered for 10 days for 28% (10/36) participants. It is suspected an operating system update caused the error as a server reboot fixed the error. Once we fixed the error, all affected participants resumed the SMS text messages where they left off. In order to keep blinding and consistency in the timing of the outcome measurements, follow-up assessments were still scheduled for 60-days after randomization. Of the 10 affected participants, 2 did not complete follow-up, 5 completed the primary outcome assessment between 60 and 70 days after randomization, and 3 completed follow-up after 70 days. After this technical problem, we implemented more regular system checks by the staff involved in randomization to ensure all SMS text messages were being delivered.

Regarding acceptability, over 93% (29/31 [94%]) of participants in the Txt2Prevent group reported they agreed or strongly agreed that they were satisfied with the program. About 74% (23/31) agreed or strongly agreed that it helped them manage their condition. When asked in semistructured interviews, 17/18 participants said they would recommend the program to other patients with cardiac issues. The participant who said he would not recommend the program stated that his recommendation would depend on whether the person took the time to read the SMS text messages. All but 2 interview participants reported reading every SMS text message, and these reported that while they did read most of the SMS text messages, it was possible they did not read all of them. All interviewed participants said they would be willing to use SMS text messaging again for health purposes.

Our pilot study assessed the impact, feasibility, and acceptability of a 60-day SMS text messaging program in supporting patients with ACS following hospital discharge. In exploratory adjusted analyses, we did not find statistically significant differences on follow-up scores (controlling for baseline scores where applicable) between the Txt2Prevent group and the usual care group in their self-management domains, health-related quality of life, medication adherence, health care resource use, and self-efficacy in adjusted models, except for the “Total plus” domain, which was impacted by an influential outlier. The study protocol was generally feasible, as seen by high adherence to the study protocol targets for randomization time frames and questionnaire completion rates, although recruitment took much longer than estimated. In terms of acceptability, participants reported they generally found the program acceptable and believed it helped them manage their condition.

In our pilot study, we failed to demonstrate the effect of SMS text messaging on our questionnaire outcomes, including the heiQ, CSE, and medication adherence. Previously, 2 interventions using apps reported improvements in heiQ domains over the short term [47,48], although this was not the case in 2 web-based interventions [49,50]. We also did not find improvements in self-efficacy scores, which is in contrast to a study that used SMS text messages and phone calls for patients with CVD [51]. In addition, several previous SMS text messaging studies targeting CVD medication adherence had a positive effect [52]. Other studies assessing SMS text messaging in a CVD population have measured specific risk factors, such as blood pressure and cholesterol. Chow et al [24] reported positive effects on low-density lipoprotein–cholesterol, systolic blood pressure, body mass index, physical activity, and smoking in an SMS text messaging program. However, not all studies have reported positive effects [22]. Zheng et al [53], who used a framework similar to Chow et al’s [24], found greater levels of physical activity at 6 months in the intervention group compared with the control group, but did not observe statistically significant (P=.22) effects on blood pressure. Another study investigating the effect of weekly SMS text messages or emails on the primary prevention of CVD risk factors found no improvements after 1 year [54]. Therefore, while many previous studies have reported positive effects, the results are not consistent.

Differences between our findings and others may be due to our intervention’s design. Guided by an advisory committee that included clinicians (a cardiac nurse, a family physician, a community pharmacist, and 2 cardiologists), researchers, and 2 patient partners with lived experiences of CVD, the SMS text messages were education based and included prompts that aligned with hospital messaging and current guidelines [55]. The messages were revised according to feedback from patient focus groups. We wanted to test whether a simple program design (ie, one-way delivery, prespecified order of SMS text messages) was effective before considering more complicated interventions. SMS text messages were designed to broadly apply to patients and did not include tailoring, such as personalization, feedback, or content matching. Incorporating tailored messages could be beneficial as systematic reviews identified that mobile health studies with positive effects often used tailoring [56,57]. In addition, by having a multifactorial focus, we may not have covered topics frequently enough to instigate change. For example, our medication adherence results contrast previous studies reporting positive effects [22], but only 7 of our 48 SMS text messages covered medication-related topics (see SMS text messages sent on day 1, 7, 9, 17, 18, 22, and 24 in Multimedia Appendix 1). The messages were also 1 way only, in part because this required fewer resources to implement. This may be a limitation as some meta-analyses and reviews have reported 2-way messages were more effective, although this is not consistent [57-59]. Chow et al [60] incorporated behavior change techniques such as prompting consequences and self-monitoring of behavior. It is unclear which behavior change techniques are effective, but a future study could consider including more behavior change techniques [61,62].

Regarding the feasibility of the study protocol, we required 17 months to recruit 76 participants instead of the anticipated 6 months. Six months was estimated because there were approximately 750 ACS discharges in the previous year, and a feasibility survey showed 50% (14/28) of patients owned a mobile phone. We assumed 40.0% (300/750) of patients would be eligible and of those 50.0% (150/300) would agree to participate. However, we missed approaching many patients due to restrictions imposed by our ethics board. The research assistant had to obtain bed numbers of patients with ACS from the clinical nurse leader. They then asked the bedside nurse to confirm with the patient if they could explain the study. This required forming strong relationships with clinical staff. Evening recruitment visits also helped as patients were often discharged shortly after returning to the ward from the catheterization laboratory. Ultimately, 43.0% (172/400) of approached patients were eligible and 44.2% (76/172) of eligible patients agreed to participate. More patients declined to participate in our study compared with the 10%-30% refusal rates reported by other CVD SMS text messaging studies that recruited from hospitals or outpatient clinics [23,63]. Many patients refused when initially approached by clinical staff, so we could not document their reasons. Having brief, standardized wording to provide clinical staff could increase uptake. Although the focus groups and feasibility survey indicated patients were interested in this type of program, the hospital environment could have created challenges as patients may be overwhelmed, unsure of their postdischarge needs, or be wary of committing to a research project [64]. However, recruiting outside the hospital would contradict with the time-sensitive nature of the program. The randomization process worked well, although delays happened when questionnaires could not be completed before discharge. Our follow-up rates were slightly lower than previous CVD SMS text messaging studies, which were often between 97% and 91% [53,63,65,66]. Having 2 different questionnaire formats (telephone for 1 questionnaire and mail for the remaining questionnaires) likely caused some of the partial completions.

Our study has several limitations to consider. As this was a pilot study, we did not determine our sample size based on power calculations and were likely underpowered to detect clinically important differences. We chose the heiQ as it covers potential proximal and intermediate outcomes of self-management programs [32]. Self-management is important as it is linked to improved health behaviors and reduced costs and health care visits [67-69]. Previously, a 10-week proof-of-concept study (n=35) evaluating a peer-support app reported improvements in heiQ domains, indicating changes are possible in small samples over the short term [48]. Other SMS text messaging studies frequently use clinical measures, such as blood pressure, making it difficult to compare our results directly. In addition, as our measures were self-reported, there may have been biases (eg, social desirability bias or recall bias); however, we primarily used validated questionnaires and confirmed self-reported hospital visits with hospital records. For some measures, a clinically meaningful change has not yet been determined, so we calculated Cohen d effect sizes for better comparison [70]. Participants in the intervention group may have been impacted in ways not captured by the questionnaires or had a different perspective at follow-up (eg, been more aware they were not meeting recommendations). While we cannot know this, it is possible as participants in the interviews and acceptability survey provided positive feedback about the program.

In our exploratory analyses, we did not demonstrate any positive effects of the SMS text messaging intervention in terms of self-management, medication adherence, health-related quality of life, cardiac self-efficacy, or health care resource use. The Txt2Prevent program had an intentionally simple design and was acceptable to participants, but design changes may be needed before proceeding to a larger study. The study protocol was feasible to implement, although improvements to the recruitment process are likely required. Future work should investigate the effect of tailoring, multifactor versus single-factor interventions, 2-way versus 1-way SMS text messaging, and the effectiveness of behavior change techniques.