We conducted a 2 × 2 × 2 factorial randomized trial embedded in the eFHS. The trial was registered at ClinicalTrials.gov (NCT03516019).

This study was approved by the Boston University Medical Campus Institutional Review Board (H-36586).

FHS Third Generation (Gen 3; n=4095), Omni Group 2 (n=410), and New Offspring Spouse (n=103) cohorts were enrolled from 2002 to 2005 and have been examined every 6 to 8 years at the FHS research center. Participants in the eFHS were enrolled from the Gen 3, Omni Group 2, and New Offspring Spouse cohorts during the in-person examination 3 (2016-2019) at the FHS Research Center in Framingham, Massachusetts, United States [10]. eFHS eligibility criteria were as follows: English-speaking individuals who owned an iPhone (Apple Inc) with compatible iOS (version 9 or higher) or Android phone; residence in the United States; provision of permissions for notifications and data sharing with the research center; and provision of signed and dated informed consent (2-step consent as part of the research examination and within the eFHS mobile app). eFHS participants using an Android phone or who did not choose to use both devices were not eligible for the messaging trial.

At the time of the research exam, the study technician invited participants to download the eFHS app. The eFHS app allows communication with participants through notifications, permitting us to send the messaging intervention notifications defined below. Participants were asked to answer health surveys administered through the app at baseline and every 3 months. In addition, participants with iPhones were provided with choices to pair the eFHS app with a Withings-Nokia BP cuff and an Apple Watch (series 0; Apple Inc). Participants were trained in the use of the devices and asked to wear the watch daily and send a BP measurement once per week. Participants were provided with written instructions if they preferred remote enrollment. The app automatically transmits data, such as HR or BP measurements, to the FHS research center if participants have paired their devices with their iPhone.

We enrolled participants in the messaging trial between April 13, 2018, and February 8, 2019, and completed the follow-up of all enrolled participants by March 23, 2019, for the primary outcome. We invited all new eFHS enrollees to participate in the messaging trial during their in-person examination 3 at the FHS research center. In addition, we contacted eligible participants who had been enrolled in the eFHS prior to the onset of the messaging trial (n=586) by email and phone to invite them to participate (Figure 1 and Multimedia Appendix 1). A consent form for participation in the trial was sent via smartphone app 1 month before trial initiation. Participants who did not return the consent survey received up to 3 reminder notifications. Participants no longer participating in the eFHS did not receive requests to consent for the messaging trial. The current trial enrolled 205 participants when they attended their in-person examination as part of the parent study and 450 participants who were previously enrolled in the eFHS.

For the primary outcomes, we assessed if participants transmitted at least one BP or HR measurement within 7 days of each notification 3 months after randomization (after 12 notifications). In secondary analyses, we examined the transmission of BP and HR measurements at 1 month and 6 months after random allocation (after 4 and 24 notifications, respectively). We also performed a longitudinal analysis of the repeated weekly transmissions.

Although the notifications did not directly encourage adherence of survey responses (they only referred to HR and BP data transmission), we hypothesized that participants who are more engaged with the devices are more likely to return the 3-month survey. As a consequence, we also examined the effect of the messaging interventions on survey completion at 3 months. We conducted this exploratory analysis only among the newly enrolled eFHS individuals in this trial, because these participants were eligible to receive the first 3-month smartphone survey. Randomization was stratified according to enrollment status (new to eFHS vs previously enrolled).

Based on preliminary data from June 2016 to July 2017, the proportion of participants transmitting data at 3 months from the watch and the cuff was about 50% [10]. With a 2-sided test with an α of .05 and anticipated proportions of 50% and 70% of participants transmitting data in the groups without and with any intervention, respectively, 650 participants would give >99% power for the main effect of each intervention (Multimedia Appendix 3). This sample size assumed no interaction effect between intervention groups. We powered the trial for the three main comparisons: (1) using personalized motivational reminder messaging versus standard messaging; (2) messaging sent on a weekend versus weekday; and (3) messaging sent in the morning versus the evening, when no interaction is present. We did not adjust for multiplicity because the factorial analysis “at the margins” addresses distinct hypotheses, and statistical simulations support that no adjustment is needed in this situation [20]. If interactions existed, 650 participants also provided 77% power to detect a 2-way interaction odds ratio of 2.33 or more with a 2-sided test and an α of .05. Power to detect a 3-way interaction was limited (Multimedia Appendix 3).

We randomly allocated each participant to 1 of 8 groups. Randomization was stratified according to sex, age (≤55 years vs >55 years), and enrollment in the eFHS prior to the beginning of the current messaging trial. A statistician (LT) generated blocked randomization lists for each stratum. Randomization was implemented centrally by CareEvolution through the eFHS app, ensuring allocation concealment. Research staff enrolled participants in the FHS research center before random allocation to one of the intervention groups, which occurred afterward. eFHS researchers, statistician, research assistant, and research center staff were masked to group assignment. As part of the eFHS, support staff helped participants complete their remote monitoring and data transmission. The support staff used standardized scripts to interact with participants over the phone. To limit the possibility of encouraging participants to use the device, we trained the support staff not to ask participants about the notification messaging content.

Baseline characteristics of participants are presented as mean (SD) for continuous variables or percentage for categorical variables. Characteristics were obtained using standard protocols and definitions from examination 3 (Multimedia Appendix 4). We performed analyses “at the margins” to assess the main effect of each intervention, meaning that, for example, we determined the effect of personalized versus standard notifications by comparing outcomes among all participants randomly allocated to personalized notifications (regardless of days and times of messaging) with those of all participants randomly allocated to standard notifications [21]. We also evaluated 3-way and 2-way interactions between the 3 interventions (messaging content, day, and time) on the transmission of BP and HR data. For the analyses at any specific time point (1 month, 3 months, and 6 months), we used logistic regression models. We estimated the absolute difference in the proportion of participants adequately transmitting data as well as the odds ratio. For the longitudinal analysis of the repeated weekly transmissions, we used a logistic regression model with random intercept. For the analysis of 3-month survey completion, we defined complete, partial, and missing response as all, some, or no survey questions answered. We performed this analysis in the subgroup of individuals who had not been enrolled in the eFHS prior to the beginning of the messaging trial. We used chi-square tests to compare the proportion of complete, partial, and missing response between intervention groups “at the margins.” Two-sided tests were performed for all analyses with a significance level of .05.

Of the 791 eligible FHS participants, we received consent from and randomly allocated 655 (82.8%) participants (Figure 1). Table 1 summarizes participant characteristics. Participants were, on average, aged 53 (SD 9) years, and 59.8% (392/655) were women. In all, 450 participants had already been enrolled in the eFHS prior to the beginning of the randomized messaging trial, including 232 participants aged >55 years, and 205 participants were new eFHS participants, including 69 participants aged >55 years. Characteristics of participants in the trial were similar to the eFHS participants who chose to use BP cuffs or smartwatches but were not part of the trial (Multimedia Appendix 5). Overall, the proportion of participants transmitting BP and HR measurements decreased over time. HR transmission via the smartwatch was higher than BP transmission. Participants aged >55 years were more likely to transmit BP and HR measurements (Multimedia Appendices 6 and 7). At 1, 3, and 6 months after randomization, 130/300 (43.3%), 129/298 (43.3%), and 110/270 (40.7%) of participants aged >55 years transmitted BP data compared to 119/353 (33.7%), 91/353 (25.8%), and 63/316 (19.9%) of participants ≤55 years, respectively (longitudinal model P value <.001); 205/300 (68.3%), 205/298 (68.8%), 170/258 (65.9%) of participants aged >55 years transmitted HR data compared to 237/363 (65.3%), 216/353 (61.2%), and 174/304 (57.2%) of participants ≤55 years, respectively (longitudinal model P value <.001).

At 3 months after randomization, the proportion of participants who transmitted BP measurements was larger in the personalized messaging group compared to the standard messaging group (126/324, 38.9% vs 94/327, 28.8%), for an absolute difference of 10.1% (95% CI 2.9%-17.4%; P=.006; Table 2). There was no evidence of a difference in the proportions of participants transmitting BP measurements between messaging sent on a weekend versus weekday (111/324, 34.3% vs 109/327, 33.3%; P=.80) or between messaging sent in the morning (7 AM) versus the evening (7 PM; 114/326, 35% vs 106/325, 32.6%; P=.53) at 3 months. Similar results were observed at 1 month and 6 months (Table 2).

There was no evidence of interaction between the 3 types of messaging interventions (personalized vs standard, weekend vs weekday, and morning vs evening; Multimedia Appendices 8 and 9). We observed an odds ratio of 2.03 (95% CI 1.27-3.24) versus 1.24 (95% CI 0.78-1.96) for personalized versus standard notification when notifications were sent during the week as opposed to the weekend, which was not statistically significant (interaction P=.14; Multimedia Appendix 9).

In the longitudinal analysis of weekly transmission, there was also evidence that the proportion of participants who transmitted BP measurements was larger in the personalized messaging group compared to the standard messaging group over 24 weeks (P=.04; Figure 3 and Multimedia Appendix 10). Notifications sent in the morning were associated with slightly increased BP transmission over 24 weeks as compared to evening notifications (P=.03; Multimedia Appendix 11). There was no evidence of a difference over time for weekend versus weekday (P=.80; Multimedia Appendix 12).

At 3 months after randomization, there was no evidence of a difference in the proportions of participants transmitting HR measurements for any of the intervention: personalized versus standard notification (212/324, 65.4% vs 209/327, 63.9%; P=.69), morning versus evening notifications (212/326, 65% vs 209/325, 64.3%; P=.85), and weekend versus weekday notifications (210/324, 64.8% vs 211/327, 64.5%; P=.94). Results at 1 month and 6 months are shown in Table 3. The proportions of participants transmitting HR was higher in the personalized versus standard notification (186/281, 66.2% vs 158/281, 56.2%; absolute difference=10%, 95% CI 2%-18%; P=.02).

There was no evidence of interaction between interventions (Multimedia Appendices 13 and 14). In the longitudinal analysis leveraging 24 weeks of weekly messages, the proportion of participants who transmitted HR measurements was larger in the personalized messaging group compared to the standard messaging group over 24 weeks (P=.02; Figure 4 and Multimedia Appendix 15). There was no evidence of a difference over time for morning versus evening notifications (P=.54; Multimedia Appendix 16) and weekend versus weekday notifications (P=.90; Multimedia Appendix 17).

Among 205 participants who had not been enrolled in the eFHS prior to the beginning of the messaging trial, there was no evidence of a difference between intervention groups in terms of survey completion. In the personalized notification group, 58.8% (60/102) and 41.2% (42/102) of participants had complete and missing survey response at 3 months, respectively, as opposed to 57.3% (59/103) and 42.7% (44/103) in the standard notification group (P=.83; Multimedia Appendix 18).