During the intervention year, patients receive pharmacist videoconferences facilitated by an HC who is in the patient’s home with a tablet (iPad with cellular plan). The initial pharmacist encounter is scheduled after the HC has already conducted 1-2 home visits and lasts 60 minutes. The pharmacist initially reconciles medications via videoconferencing with HC assistance. Subsequent pharmacist encounters vary in frequency based on patient needs and range in length from 30 to 60 minutes. HCs schedule videoconference appointments during dedicated days or times when pharmacists are available. Videoconferences are conducted using VSee software. VSee transmits personal health information securely and is available for use on desktop computers and mobile devices providing real-time person-to-person audio and video communications. However, it is most suitable for health-related communication, as it is Food and Drug Administration registered and HIPAA compliant (using FIPS 140-2 certified 256-bit Advanced Encryption Standard). Finally, it uses peer-to-peer sessions so that information is not stored on a server.

Pharmacist services are based on a standardized pharmacist management protocol. After the initial medication reconciliation, follow-up pharmacist activities include reviewing home glucose and/or blood pressure, monitoring log data obtained by the HC during home visits, identifying therapeutic goals for HbA_1c and blood pressure collaboratively with patients’ primary care providers (PCPs), formulating an approved plan of care, assessing changes in medications, and documenting the plan in the EMR. In addition, pharmacists provide education related to medication (name and purpose of medications and time, strength, and method of administration); drug interactions and side effects; goals of therapy; basic lifestyle modifications; and use of pillboxes, low-literacy visual medication lists, or other adherence aids. Pharmacists educate and encourage lifestyle changes, consistent with the published guidelines [67-69]. They propose medication changes based on algorithms and protocols derived from national guidelines under physician guidance [70-72]. Pharmacists routinely monitor hypoglycemic events, address prevention, and review treatments. This includes 3 steps: (1) addressing hypoglycemia with every patient contact, (2) applying principles of appropriate therapy (education, empowerment, frequent glucose self-monitoring, flexible medication regimen, individualized goals, and professional guidance), and (3) considering risk factors for hypoglycemia. Overall, there is mixed evidence regarding the benefits of aggressive glycemic control [73-75]. In the proposed study, PCPs and pharmacists adopt the American Diabetes Association approach to individualized care, where the general goal for nonpregnant adults is HbA_1c less than 7%. They may decide upon less stringent goals for those with a history of severe hypoglycemia, limited life expectancy, advanced complications, or extensive comorbid conditions [76]. In addition, pharmacists follow the 2018 American College of Cardiology or American Heart Association guidelines for lipid management (eg, calculating 10 year atherosclerotic cardiovascular risk to determine statin intensity) [77].

PCPs and pharmacists communicate routinely regarding patient care and are located in the same area within the medical setting. Pharmacists review EMRs, including blood test results, clinical progress notes, problem and medication lists, drug allergies, hospitalization records, and emergency room reports. Nonurgent communication and electronic progress notes from each pharmacist encounter will be sent to the PCP through inbox messaging and note forwarding within the EMR. Pharmacist progress notes include a detailed list of medications, estimated adherence levels, and home glucose or blood pressure monitoring log information.

Videoconferencing procedures in the intervention follow the American Telemedicine Association practice guidelines [68]: (1) pharmacist and patient/HC identity verification, (2) informed consent, (3) appropriate physical environment (privacy, lighting, and noise), (4) education and training (pharmacist and HC), (5) alternate communication (eg, telephone contact in case of disruption of service), and (6) documentation in UI Health EMR (by pharmacist and HC). All study pharmacists and HCs receive standardized training on the use of videoconferencing, which includes scheduling, preparing the environment (home or pharmacist office), patient education on telehealth, and documentation. All encounters have contingency plans in place for technology problems (eg, inadequate signal for video streaming through iPads), including the use of mobile phones for all HCs.

HCs will introduce themselves to the patients in person at the data collection visit that aligns with the beginning of the patients’ intervention year (ie, baseline visit or crossover visit at the beginning of year 2 depending on randomization). When possible, the first home visit will be scheduled at the initial introductory meeting. If this is not possible, the HC will follow up by telephone within 1 week to schedule the first home visit. All HCs are either African-American or bilingual or bicultural Latino and work with patients who are concordant for race or ethnicity and language. The HCs have an undergraduate college degree in or related to community health with experience conducting home visits. Specific training for the research study includes 80 hours of standardized HC training or retraining. As outlined in Textbox 2, training begins with an overview of the research protocol and discussion of the HC goals. The fundamental components of providing health support to marginalized populations are addressed with the required reading materials, didactic presentations, and discussions [78].

HC training highlights the unique qualification of HCs to provide culturally sensitive support, tailor visits based on the personal preferences of participants, and assist patients in navigating the health care system [79,80]. Diabetes education follows the Diabetes Education Empowerment Program, which targets literacy and cultural awareness [81], and the Training Curriculum for Health Coaches [52,82]. Training includes shadowing a clinical pharmacist to understand medication use, adherence, glucose and blood pressure monitoring, insulin injection, and medication reconciliation. The study investigators provide training in videoconferencing (eg, scheduling and practice encounters), text messaging (eg, practice with the custom text messaging platform and standardized safety procedures), and conducting home visits in a culturally sensitive manner. Ongoing training is provided periodically on topics that reinforce and expand the initial training (eg, motivational interviewing and insulin management). HCs receive routine orientation to clinical operations and staff at their primary care locations. Standardized safety procedures related to text messaging and home visits are also addressed. To evaluate clinical skills, the HCs demonstrate reliable measurement of blood sugar, blood pressure, and administration of insulin (though HCs do not administer insulin to patients).

The HC component involves monthly home visits with ongoing telephone support, including the facilitation of all pharmacist videoconferences. The HCs work with the pharmacist to evaluate adherence (eg, check label instructions and fill and expiration dates of pill bottles), assist in medication reconciliation, review home glucose and/or blood pressure monitoring data, and provide reinforcement of proper medication use. Finally, HCs debrief patients after pharmacist encounters to reinforce and clarify recommendations and plans. HCs and pharmacists communicate with each other between videoconference encounters by phone or secure email, as needed, to coordinate efforts.

Home visits include the HC and patient. Family members are permitted by patient requests. Through open discussion and reflective listening, HCs encourage patients to explore their emotions and share their concerns [83]. Overall, HCs provide diabetes self-management education and support (DSME/S), consistent with the recommendations of the American Diabetes Association’s position statement [83]. Specifically, HCs provide DSME/S that includes the following over the course of the 1-year intervention: engagement, information sharing, psychosocial and behavioral support, integration with other therapies, and coordination of care. Initial HC encounters focus on relationship building and gaining an understanding of the patient’s individualized diabetes education needs. HCs provide diabetes education and consider realistic and achievable food choices, portion sizes, and cooking preparation; discuss relationships between medications, meals, and glucose levels; and help patients integrate movement into their daily lifestyles. Diabetes education is facilitated by our culturally appropriate multimedia education iBook, Living Well with Diabetes/Viviendo Bien con Diabetes [84]. This iBook (available in English and Spanish) provides patients with video testimonials and various interactive educational experiences. HCs present specific chapters on an iPad to reinforce specific diabetes self-management concepts [84]. To promote behavioral change, HCs use motivational interviewing techniques to assist patients in setting specific health behavior change goals. Goals set jointly by HCs and patients follow the specific, measurable, achievable, relevant, and time-specific (SMART) framework [85]. Providing social support while helping patients engage with their existing support systems offers the potential of long-term behavior change beyond the duration of the research period. Educating patients on how to effectively navigate the health care system and use existing resources promotes continued self-efficacy. In summary, HC interactions are culturally tailored to individual needs, preferences, and resources to provide DSME/S [86].

By the end of the second month, HCs will facilitate a videoconference encounter with a pharmacist and patient to conduct a complete medication reconciliation. HCs work with patients to identify adherence barriers and assist in problem solving or referrals for resources aimed at overcoming recognized challenges in medication use. Finally, HCs document summaries of each encounter (with or without videoconferencing) in the EMR and forward notes electronically to PCPs and pharmacists.

Between home visits, HCs will communicate with patients through telephone calls and text messaging. All text messages will be tailored by HCs and sent through a custom software app (mytapp). We developed mytapp for community-based health behavior research. mytapp sends messages immediately or at a scheduled date or time, recurrent messages (daily, weekly, or weekdays), group messages, or multiple question surveys. HCs receive training on mytapp and schedule messages for their patients regarding appointments and medications. HCs send messages to maintain motivation, elicit feedback on progress, and screen for barriers that may reduce the chance of success (Table 1). For example, a morning text message may ask a patient how they did on their goal the previous day.

No more than 7 messages are delivered weekly, except for optional medication reminders. For safety, patients are reminded that text messages are sent in an automated fashion by a computer. Furthermore, they are reminded that urgent health questions should be directed to their PCP and not sent in a text message. HCs monitor patient responses through their study phones.

Texts are monitored by the HCs with additional oversight by the project coordinator.

All participants spend 1 year in the usual care condition, either year 1 or 2, depending on the randomization. During receipt of usual care, they receive health care from their usual providers without the support of a pharmacist or HC. There are no home visits or pharmacist telehealth videoconferencing encounters with usual care. In addition, participants receive a 1-page list of clinic resources with names and direct telephone numbers (eg, social worker or clinical pharmacist) along with a low-literacy, paper-based diabetes education pamphlet [87]. Usual care reflects the type of health care that patients receive outside of any participation in research.

To continuously evaluate the fidelity of both conditions, we review logs maintained by the HCs weekly, including visit dates, length of visits, visit content, and disposition (eg, visit completed, and patient unreachable). Weekly group meetings with HCs and the research team provide an opportunity to monitor intervention delivery. Monthly lunch meetings with clinical pharmacists, HCs, and investigators ensure active collaboration between HCs and pharmacists. All technology-related difficulties are reported to the coordinator immediately and discussed in weekly meetings with the investigators. One investigator, a health psychologist, maintains weekly contact with HCs to provide emotional support and incorporate structured training opportunities. Text messages sent and received via mytapp are reviewed monthly.

Trained, blinded RAs collect physiological and self-report data through interview administration within the clinical research center at UI Health at 5 time points: baseline and every 6 months for 2 years. RAs are matched to patient language preferences in English or Spanish. Laminated cards with Likert-type scale responses are provided as a visual reference to the patients. Interview data are entered directly into laptop computers using Research Electronic Data Capture (Vanderbilt University) electronic data capture web application [88]. The baseline survey requires an average duration of 60 minutes with subsequent follow-up visits lasting 30 minutes. Subjects receive US $30 as compensation for travel and time at each of the 5 data collection points (plus US $50 if at least one videoconference is completed in the prior 6 months during the intervention period). Public transit cards and parking passes are provided when needed. Participants are informed of their randomization assignment at the end of the baseline data collection.

Sociodemographics include age, gender, self-reported race and ethnicity, country of origin, income, highest level of education, employment status, global health status [89], and insurance. Diabetes and medical history include self-reported time since diabetes diagnosis, prior receipt of diabetes education, current therapy, known diabetes complications, and comorbid conditions. Health literacy is assessed using 3 screening questions with high discriminatory power among English- and Spanish-speaking populations [67,90,91]. Mobile phone use and comfort sending text messages is also assessed using 5 items [68].

Intermediate variables are also collected at each of the 5 time points. Perceived severity of diabetes and perceived susceptibility are assessed using 2 items adapted from the study by Bradley et al [92]. The perceived benefits of therapy are measured by a 5-item survey related to the benefits of therapy [71]. Diabetes distress is measured using the brief Diabetes Distress Scale [73]. Depression is measured using the Patient Health Questionnaire [75,76]. Social support is measured using an assessment of the amount of total support received and satisfaction with support from family, friends, and the health care team [93]. Self-efficacy is measured using the Stanford 8-item self-efficacy for diabetes survey [94]. Contextual data include an environmental survey that addresses loneliness, social cohesion, and neighborhood safety as well as the identification of stressful life events [95-97].

HCs record the dates of patient contacts in REDCap (eg, phone calls and home encounters). Clinical progress notes are completed in the EMR after every participant contact (by phone if longer than 15 minutes and in-person). Information includes mode, time, location (home vs clinic), content of contact, results of glucose or blood pressure self-monitoring, goals, life events, pharmacist interactions, and interventions. Intensification of therapy will be defined as the number of dosage increases of antihypertensive, hypoglycemic agent, or insulin or the addition of a new agent since the baseline visit [98-100]. Chart review will define the number of PCP visits and pharmacist videoconferences as well as the number of pharmacist- or physician-initiated medication changes.

Physiologic outcomes will be collected at 5 time points (0, 6, 12, 18, and 24 months). The research staff will perform phlebotomy, blood pressure, weight, and height recordings. HbA_1c and fasting lipid profile (total cholesterol, high-density lipoprotein, LDL, and triglycerides) are obtained via phlebotomy. A calibrated digital scale measures weight. A height stadiometer measures height, with subjects removing their shoes (for BMI assessment). Blood pressure is assessed in subjects sitting down for at least 5 minutes, following a standard procedure. Health-related quality of life is measured using the EuroQol Group 5D and Diabetes Distress Scale [73].

The revised Summary of Diabetes Self-Care Activities Measure captures basic diet, exercise, blood sugar testing, foot care, and smoking with 11 core items [101]. Additional questions address skipping medications and insulin injections to evaluate adherence, as well as taking aspirin regularly. Alcohol misuse is assessed using the Alcohol Use Disorder Identification Test—Concise [102].

The sample size calculation is powered to detect the primary outcome, which is the change in HbA_1c. Reviews of published studies suggest that successful education programs lower A_1c levels by 0.4% to 1.7% [3,103]. On the basis of our previous experience with the patient population, we estimated a mean baseline HbA_1c level of 10% with a SD of 1.8 and an effect size of 0.56 for aim 1. The cross-time correlation was estimated to be 0.30, with a compound symmetry structure. We adjusted for clustering and assumed an intraclass correlation coefficient of 0.01, with 5 clusters. This yielded a design effect of 1.34. Two-sided alpha of .05 and 80% power were assumed. Allowing for a 20% dropout rate, 220 patients were required [104].

To address missing data, we will examine the data for patterns of missingness and potential bias in missingness. If data are not missing at random, we will apply one of several available imputation methods in a sensitivity analysis based on the nature and extent of missing data [105]. We will determine effectiveness using intention-to-treat principles with actual imputation of missing data [106-109]. This will allow us to appreciate the potential biases inherent in a real-world setting due to dropout and poor adherence to study procedures.

Patients who share a single PCP might exhibit similarities that are not shared with patients cared for by other PCPs. To address this, we will include random effects in the model for clinic site, PCP, and HC (though the small numbers of HCs may call for a fixed effect approach or insufficient replication) [110]. To examine a stricter examination of effect, we will also do a per-protocol analysis of complete cases. Completed cases will not be dependent on intermediate time point data collections (6 and 18 months). Potential selection bias will be corrected by including model covariates that are differentially related to study participation.

Univariate comparisons between the 2 study groups for outcomes and covariates at baseline will be conducted using chi-square tests for categorical variables, Kruskal-Wallis tests for ordinal variables, and t tests for continuous variables. All tests will be two-sided. Nonnormal continuous data will be transformed before the analysis. To provide a comprehensive analysis of our primary hypothesis, we will extend the usual analysis of crossover designs [111] by including a longitudinal trend component in the first year. Thus, we can examine the time course (0, 6, and 12 months) as linear or quadratic over the first 3 measurements. This will allow us to investigate whether changes are made early and at what rate they continue throughout the rest of the period. This analysis permits a comparison of the trends between the 2 conditions. In addition, we will regard subjects as a random effect and will use Gaussian mixed model estimation. We can then substitute treatment by period interactions for the carry-over effects, and the model can be reduced in a recommended sequence (first omit carry-over, then omit treatment, and finally omit period) [111]. We will also explore patterned covariance structures such as compound symmetry and autoregression along with incorporating time-constant and time-varying covariates, such as health literacy or diabetes distress.

The primary analysis of all physiological outcomes jointly in the repeated measures design will be conducted using a general linear model framework. Repeated measures multivariate analysis of variance will be used to explore the simultaneous impact of the treatment on multiple correlated dependent variables, including the use of Roy-Bargmann stepdown F tests and discriminant function analysis as post hoc tests to identify subsets of outcome measures affected [112]. Multivariate analysis of variance (MANOVA) secondary analyses will explore the impact of inhomogeneous baseline variables on the results. Group by time-trend interaction contrasts will be used to explore different group trajectories of change. The potential consequences of medication intensification (eg, initiation of insulin) will be explored by evaluating changes in BMI and quality of life.

Exploratory subgroup analyses will follow a heterogeneity in treatment effects framework [113,114]. We will determine which subjects in the intervention group had the greatest improvement in outcomes, based on multiple prespecified patient characteristics, including race (African-American or Latino), depression, comorbidities, baseline behaviors, health literacy, continuity of care, and social support. These analyses will also consider additional comparisons between subjects above and below the median levels of videoconferencing and text messaging activity. Statistical tests will be implemented as interactions within the full data set.

If the intervention results in improved HbA_1c more than that in the usual care, we will examine whether changes in self-efficacy, health beliefs, or social support serve as mediators for improved outcomes. In addition, we will explore diabetes-related behaviors as well as medication treatment intensification as mediators using MPlus [115-118]. To further identify the relative contributions of videoconferencing and/or text messaging, we will conduct meditational modeling with videoconferencing time, number of text messages, and HC contact time as mediating variables. This may describe any dose-response relationship between mHealth utilization and outcomes. We will compute bias-corrected bootstrap standard errors using MPlus. This offers accurate confidence intervals for mediation coefficients [119]. Due to anticipated relationships between HC activity level and mHealth delivery, we plan to inspect correlations between mediators and incorporate any significant findings in model development. This analysis will demonstrate whether the intervention effects are sensitive to mediators. Finally, given the sample size, observed (rather than latent) variables will be used in the mediation models. To enhance the power of mediation modeling, we adopt an α=.10 type I error criterion to improve the chances of finding promising mediators for future studies.

We expect a limited amount of decline in physiologic outcomes with the transition back to usual care. We will test the hypothesis that improvements in HbA_1c (primary outcome), blood pressure, and LDL cholesterol (secondary outcomes) will be maintained during the maintenance period. The analyses will be conducted in a manner similar to our primary hypothesis and will evaluate changes in the initial intervention group 1 year after intervention completion.