The mGlide intervention has 3 components (Table 3): participant education on the importance of HTN control, training on SMBP and wireless transmission of BP, and responsive antihypertensive medication adjustment by the pharmacist-provider team.

Each participant receives education on the importance of HTN control via an IRB-approved educational video available in English, Spanish, and Hmong. We previously developed this video through a community-engagement process led by community researchers from the Hmong and Latino communities. The study team member then trains the participant on SMBP [18] using a wireless BP monitor provided by the study and a smartphone (participant’s own phone or mobile device such as iPad). The intervention participants are requested to self-monitor their BP daily with specific guidance on timing and proper technique based on the AHA recommendations [18]. The self-monitored BP is automatically transmitted to a provider REDCap database and is used for responsive antihypertensive medication adjustment. The participant BP transmission is automated and facilitated by the participant’s mobile device via an app. Pharmacist teams at each clinic location access the BP data via a web-based user interface that identifies patients whose readings were out of bounds during the prior week. This user interface creates efficiencies for pharmacists reviewing the data and was developed to avoid information overload for pharmacists. Pharmacists adjust medications based on collaborative practice agreements with primary care providers (PCP) and mGlide protocols. Pharmacists adjust medications and communicate with the patient and their PCP as often as every 2 weeks or as needed to reach the BP goal.

Our intervention is a multilevel model with a 3-member team: frontline research team member who educates the participant and trains them on self-monitoring of BP, pharmacist who provides responsive HTN management, and PCP. This team communicates via the electronic medical record as part of clinical care.

Each study participant randomized to the clinical care comparison group receives the same education as intervention participants on the importance of HTN control via the IRB-approved educational video. The comparison group participants receive a digital BP monitor and are trained on BP self-monitoring with proper technique. Participants are encouraged to measure their BP daily and follow up with their PCP regarding their BP, as requested by the PCP as part of their clinical care. Participants, PCPs, and clinic teams are responsible for communicating as they would in usual clinical care. In other words, there is no predetermined follow-up schedule with the PCP specified by the study, and there is no automatic transmission of participant BP readings to providers; rather, participants are responsible for sharing their BP and working with their care team to manage their BP.

The main outcome will be the rate of HTN control at 6 months; the primary outcome will be defined as a binary indicator of uncontrolled HTN or death versus controlled HTN at 6 months. The BP is measured in-person in the clinical research center by a trained staff member who is blind to the patient’s assignment. The BP is measured 4 times using a standard digital monitor using proper technique, and the average of the last 3 BP measurements denotes the BP outcome. HTN control is defined as systolic blood pressure (SBP) <140 mm Hg, which is consistent with American College of Physicians, American Association of Family Physicians, and AHA primary and secondary stroke prevention guidelines [12,13,20]. The 2017 American College of Cardiology/AHA guidelines recommend a SBP goal of <130 mm Hg. The HTN treatment goals and target BP were discussed with the PCPs and clinical pharmacists at all sites prior to study recruitment. Everyone agreed to target a SBP goal of <140 mm Hg for intervention arm participants and requested that lower SBP goals be left to the discretion of the study pharmacists managing the patient. Study pharmacists have access to patient medical records. At 12 months, the BP outcome is assessed in the same manner as at 6 months. This allows the assessment of the secondary outcome of uncontrolled HTN or death versus controlled HTN at 12 months. Additional secondary outcomes include patient-reported medication adherence, medication use self-efficacy, self-reported health status, and quality of life that we assess with validated questionnaires at each of the 3 study visits. We ascertain participant adverse events and health care utilization during monthly phone calls during the first 6 months of the study and bimonthly in months 7-11. We also assess care providers’ experience of the mGlide system usability.

Sample size was calculated to address outcomes of failure to achieve HTN control or death at 6 months and 12 months after randomization. We based sample size calculations on (1) reported 1-year stroke survivor mortality rate of 4% [21] and assuming similar mortality among primary care participants with elevated CVD risk, (2) pilot results that showed a primary outcome rate of HTN control or death at 6 months of 15% in the intervention and 50% in the comparison groups [17], and (3) a 20% attrition rate. We identified that a sample size of 450 would provide at least 85% power, with an alpha of .05 to detect a 15% effect size. We planned for a more modest intervention effect size than achieved in our pilot results due to the longer period of observation and more diverse health systems and patient populations when compared to the pilot study. Similarly, while the loss to follow-up in the pilot study was very modest at 4%, we planned for a 20% attrition rate due to the economically stressed primary care population in our low-resource health systems. Nationally reported rates of uncontrolled HTN in stroke survivors are ~50% [7]. FQHC primary care rates of uncontrolled HTN are ~38%-40% [22]. Our sample size will allow valid subgroup analysis among the subgroup of primary care patients in all low-income sites (n=200) and the stroke survivor subgroup (n=100).

Data collection is standardized by the use of a detailed manual of operations. The REDCap data entry system also has built-in logic to check data at the time of entry and minimize errors. Monthly data quality reports ensure that data are validated and data entry is completed in a timely fashion. If participants drop out, we are documenting specific reasons for drop out.

We will use a Cochran-Mantel-Haenszel test stratified by the randomization groups to test whether the odds of HTN control at 6 months differs between the intervention and control study arms. Statistical significance level will be set at the P<.05 level. We also will perform the same analysis at 12 months to examine if the difference between the intervention and control groups is sustained. Secondary analyses using the repeated BP assessments obtained at 6-month and 12-month visits are planned. We will use longitudinal generalized estimating equation models, with BP control modeled as a binary variable and average BP modeled continuously in separate models, with relevant covariates. Interactions will be tested to evaluate potential differential effects by baseline demographic and clinical characteristics.

We have planned for missing data. We are recording reasons for study drop out and will examine patterns of missingness. For the primary analysis, we will (multiply) impute the binary HTN control outcome. We are collecting a rich amount of covariate information on participants to inform the multiple imputation models.

We plan the following analyses to examine secondary outcomes related to system usability for care providers. The Welch t test will be used to examine the number of antihypertensive medication changes (dose adjustment, addition of new medications) per patient during the 6-month intervention period for the mGlide intervention arm. We will examine the provider mGlide experience using the Computer System Usability Questionnaire overall score as well as scores along the 3 principal factors identified on the Computer System Usability Questionnaire: System Usefulness, Information Quality, and Interface Quality. We also will collect qualitative feedback from providers using focus groups. These data will be analyzed using a qualitative framework including classic content analysis and microinterlocutor analysis [23,24].

Patient-reported outcomes (Table 2) will be examined by comparing whether there are group differences on the Patient Activation Measure [25], patient medication self-efficacy measure (Medication and Self-Efficacy Scale - Revised) [26], medication adherence (Hill-Bone Medication Adherence Scale) [27,28], self-reported health status (EQ-5D-3L) [29,30], and patient satisfaction with health systems and providers (Consumer Assessment of Healthcare Providers and Systems adult survey) [31] using a 2-sample t test separately at 6 months and 12 months after baseline. Finally, we will compare the rates of adverse events between the 2 trial arms using both adjusted and unadjusted negative binomial regression models with intervention as the key independent variable.

We plan a cost-effectiveness analysis that will follow the recommendations of Cost-Effectiveness in Health and Medicine, 2nd Panel [32] to present the analysis from both the societal and health care perspectives and to include an inventory of the non-health care impacts of the mGlide intervention. We will measure effectiveness by the differences in quality-adjusted life years (as derived from the EQ-5D-3L quality-of-life weights) across the 2 trial arms. Costs will represent the intervention cost and any differences in downstream health care and other costs for the societal perspective (productivity or time, informal care, travel) across the 2 arms. We will extrapolate quality-adjusted life years and cost differences using a 10-year stochastic event model. Thus, the cost-effectiveness analysis will meld both the trial follow-up experience of participants and their modeled experience for the remaining years, based on differences in HTN control rates between the 2 trial arms. Specifically, we will determine downstream health care utilization using a state transition (Markov) model using Monte Carlo microsimulation. Our team includes an experienced health economist (JN).