The prevalence of diabetes is rapidly rising at both a global [1] and national level [2]. Individuals with diabetes are at a higher risk of heart disease, stroke, kidney disease, and premature mortality [1,3-5]. Type 2 diabetes (T2D) can be managed and its complications avoided by engaging in self-care, including healthy diet, exercise, self-monitoring of blood glucose (SMBG), and taking medications as prescribed [6]. However, multiple barriers impede self-care adherence for patients with T2D [7-10]. Racial/ethnic minorities and people with low socioeconomic status (SES) tend to experience more barriers to diabetes self-care [11,12] and, in turn, have worse self-care adherence [13,14], more complications [13,15], and worse glycemic control (ie, hemoglobin A_1c [HbA_1c]) [16,17].

Basic mobile phone technology (text messaging and phone calls) presents viable opportunities to reach and support adults with T2D to improve self-care adherence and HbA_1c [18-20]. The vast majority of American adults (95%) own cell phones [21]; however, non-whites and those with less education and income are less likely to own a smartphone [21]. Text messaging and phone calls do not require a smartphone, and text messaging is the most common cell phone activity among all mobile phone users [22]. This ubiquity suggests potential to reach patients with low SES and racially/ethnically diverse patients [23,24]. Moreover, text messages can deliver tailored content and address modifiable barriers to diabetes self-care.

Involving human support as part of a diabetes mobile phone intervention may enhance efficacy [25,26] and improve participant engagement [27], particularly among disadvantaged or vulnerable patients [28]. In a recent 6-month randomized controlled trial (RCT), participants were assigned to receive health coaching along with access to a diabetes support app or only health coaching [29]. Although both groups had improved HbA_1c levels, the coaching group showed accelerated improvements [29]. A handful of other health promotion interventions in general populations have compared text messaging alone against text messaging plus human counselors, but the samples in these interventions have been small and therefore more research is needed [30].

In summary, adherence to diabetes self-care remains a prevalent problem and sustainable real-world solutions for diverse patients are needed [31]. Automated text messaging interventions can be resource- and cost-effective and have improved adherence and HbA_1c among underserved groups up to 6 months [32-34]; however, few have been evaluated in long-term trials with diverse samples [35]. Furthermore, none to our knowledge have assessed sustainability of effects after text messaging ends. Finally, it remains unclear whether a human coach or educator in concert with automated text messaging would be more effective for improving diabetes outcomes than text messaging alone.

In response to these gaps in knowledge, we are conducting a 3-arm RCT to evaluate the effects of mobile phone–based diabetes self-care support interventions on self-care adherence and HbA_1c among adults with T2D who are diverse with respect to SES and race or ethnicity. The trial consists of 2 intervention arms and a control group. Intervention arms are (1) Rapid Encouragement/Education And Communications for Health (REACH) and (2) REACH + Family-focused Add-on for Motivating Self-care (FAMS). Both interventions were previously developed and tested for usability among racially/ethnically diverse and predominantly low-SES samples recruited from Federally Qualified Health Centers (FQHCs) [36,37]. REACH provides text messages tailored to address patient-specific barriers to medication adherence based on the Information-Motivation-Behavioral skills (IMB) model [38,39], whereas FAMS provides monthly phone coaching with related text message content focused on family and friend barriers to diet and exercise adherence [37].

The study is designed to evaluate the effects of REACH (either intervention arm) on HbA_1c relative to the control group, while assessing the additive effects of FAMS and effects of either intervention on adherence to self-care behaviors and diabetes self-efficacy. We will also explore the effects of each intervention arm on the psychosocial mechanisms targeted by each intervention and effect modification by race/ethnicity and SES.

We are conducting a 15-month, 3-arm RCT with 2 treatment arms and 1 control arm. We are using an effectiveness-implementation hybrid design to evaluate the effectiveness of the interventions while planning for and collecting information about implementation potential (Type 1 approach) [40]. This paper focuses primarily on the protocol for evaluating effectiveness, but REACH was designed to be sustainable [36], and our community-based research methods lay the groundwork to explore barriers and facilitators to implementation in FQHCs (briefly described in the Discussion section). For the trial, interested and eligible patients with T2D were recruited from primary care clinics. We designed our recruitment approach to overrepresent racial/ethnic minorities and patients with low SES. Participants in each arm complete study measures at baseline and 3, 6, 12, and 15 months post baseline (Figure 1, top panel). Participants in either intervention arm receive intervention exposure for 12 months; sustained changes are assessed with a 15-month follow-up.

The same study measures are administered to all participants, regardless of condition. The schedule of measures is shown in Table 1. In the section below we focus on those measures central to the analyses outlined in this paper.

During enrollment, RAs explain to participants that all study participants receive a mobile phone–based program with different types and frequencies of text messages and phone calls. RAs also tell participants that a member of the research team will call them in a few days to explain more about what to expect based on their assigned condition. After enrollment, participants are randomized to one of the 3 arms using optimal multivariate matching to ensure better balance in the primary outcome and important covariates across arms [60]. The variables we use to match participants include baseline HbA_1c, insulin status, race, age, duration of diabetes, gender, income, and education. We give diabetes duration (rank value to correct for skew) and HbA_1c greater weight [61]. Twice as many participants are assigned to the control condition (n=250) as those assigned to REACH only (n=125) or REACH + FAMS (n=125). To accommodate this 2:1:1 design, patients are matched and then randomized to control or REACH and those randomized to REACH are matched and randomized to REACH only or REACH + FAMS. This helps ensure covariate balance among all 3 arms. For those assigned to REACH + FAMS, the FAMS intervention components last for the first 6 months only. All participants in the intervention arms receive REACH only for the latter 6 months of the exposure period (Figure 1, top panel).

Within a week of enrollment, participants are randomized, and a member of the research team calls each participant to explain what to expect from the mobile phone program to which they are assigned and obtain any information needed specifically for their assigned condition (eg, preferred times to receive daily text messages if assigned to either intervention arm). If we are unable to reach participants for this condition explanation within 3 weeks, they are administratively withdrawn; we still include these participants’ baseline data in our analyses but discontinue attempts to contact them. This run-in period ensures that the initiation of the study experience aligns with baseline data and identifies individuals who may be difficult to contact and therefore not good candidates for the 15-month trial. During the condition explanation we reiterate and assess participants’ understanding of the intervention components available to them, based on their condition. We do not use the terms “intervention” or “control” to explain the assigned conditions. Each condition is described briefly below and in Figure 1, bottom panel; the intervention components are described in more detail in the respective development papers for REACH [36] and FAMS [37].

We have implemented several fidelity checks to ensure that participants receive the interventions as intended. First, text messages are automated to help ensure users have the intended experience. Second, we monitor text messages to identify and correct errors and make contact with participants who stop responding to address any technical issues. MEMOTEXT securely collects and stores all text message data (eg, date and time text messages are sent and received, participants’ text message responses). Our team performs weekly checks on these data to ensure the text messages are delivered and monitor participants’ responses. As part of the REACH intervention, participants receive a daily adherence text message that asks them whether they have taken all of their diabetes medication that day. Participants are asked to reply either Yes or No. If they answer No, participants receive a follow-up text message asking them to please tell us why, with several response options (ie, “1=forgot, 2=sick, 3=clinic told me to, 4=ran out of meds, or type out a reason”). We monitor responses to these messages weekly, and if a participant does not respond to any adherence text messages for 14 consecutive days, a team member will call the participant to determine whether he or she is having any technical problems. To avoid coercing participants into responding to text messages, we only ask whether they have had any problems lately with receiving or responding to their text messages. The date of the call and the participant’s response are documented, and we subsequently troubleshoot as needed. We do not make repeated calls if the participant remains nonresponsive but may contact a participant more than once if periods of consecutive nonresponse are separated by periods of responsiveness.

We also collect fidelity data on the FAMS coaching sessions. We track the number of FAMS phone coaching sessions completed by each participant and the content of each session. Fidelity data includes the goal set during coaching, the type of family or friend support or barrier discussed, the skill-building exercise employed, the verbal contract (eg, to implement a skill learnt during coaching, such as assertive communication, with a specific friend or family member), the participant’s confidence rating of his or her ability to complete the verbal contract, and, for subsequent sessions, the outcome of the verbal contract from the previous session. Fidelity data will be presented with results to inform the degree to which the intervention was delivered as intended and to provide context for interpretation of study findings. Fidelity data will also serve as a process benchmark for future trials that may seek to reproduce the study findings or implementation studies that engage clinic staff in intervention delivery.

The study is designed to evaluate the effects of REACH (either intervention arm) on HbA_1c relative to the control group (primary analysis), while assessing the effects of FAMS. We will use generalized estimating equation models to estimate potentially time-varying intervention effects while adjusting for the baseline measure of the outcome and the type of HbA_1c test result (ie, lab-drawn at the clinic or by using the mail-in kit). The models use clustered data and allow nonlinear associations between baseline and follow-up outcome measures. A lag 1 autoregressive correlation structure will be used and alternative correlation structures tested to demonstrate the results are robust to model selection. We will use a longitudinal model to evaluate intervention effects. We will use an omnibus test for the intervention effect, then provide point-estimates with confidence intervals for each follow-up, and graphically depict our results.

Analysis will follow a conservative intention-to-treat principle, and participants with missing values will be included along with those with complete data. Multiple imputation will be used to impute missing covariate and outcome values. The analysis with multiple imputation assumes Missing-at-Random (ie, the model properly handles missing data by including covariates associated with reasons for dropout). A sensitivity analysis for the impact of the imputation of missing outcome data will exclude the outcome from the imputation process and analyze only the observed outcomes.

All procedures have been reviewed and approved by the Vanderbilt University Institutional Review Board (IRB) and this trial is registered on ClinicalTrials.gov (see NCT02409329 and NCT02481596). All data collected from participants at each assessment period are stored on REDCap’s secure server. Any participant data sent to MEMOTEXT are deidentified and stored on their Health Insurance Portability and Accountability Act (HIPAA)-compliant secure server. In addition, all reporting of text message data by MEMOTEXT and all recorded REACH Helpline voicemails are stored on their HIPAA-compliant Web server, and only IRB-approved study staff can access these voicemail messages using a secure passcode. EHR data are shared with the study team according to the policies of each individual clinic.

We included specific language in the informed consent document outlining our processes for securing participants’ data. We described that REACH Helpline voicemail messages, information shared via text message, and all study forms would be assigned a study number with no personal identifying information and be either password-protected on a secure server or in a locked filing cabinet at Vanderbilt. We explained that research team members would only access personal information for necessary study procedures, such as to issue payment or contact for follow-up appointments. Finally, we explained to participants that, if they share or lose their phone, the study text messages may disclose to others that they have diabetes, take diabetes medications, and/or received an HbA_1c test. All participants were provided with the REACH Helpline number and encouraged to call to ask questions about the study.

This study will be one of the first RCTs to deliver a long-term, theory-based, text messaging intervention to promote self-care adherence among racially/ethnically and socioeconomically diverse adults with T2D. We designed the interventions to use basic mobile phone technology (text messaging and phone calls) and provide an experience that is individually tailored and interactive for adult patients with T2D. We developed both interventions with input from racially diverse patients with low SES [36,37] and designed our recruitment strategies for the RCT to oversample racial/ethnic minorities and patients with fewer resources. Moreover, our study will explore the feasibility of a family-focused intervention delivered via mobile phones, and allows exploratory analyses comparing the effects of text messaging alone versus text messaging plus phone coaching. We will also be the first to provide information on the feasibility and acceptability of inviting members of a patients’ social support network to receive text messages about how to support the patient with his or her T2D, based on the 125 participants in our sample given the option to do so as part of FAMS.

Barriers to self-care adherence are personal, multidimensional, and change over time [66,67]. Findings from other studies suggest that helping patients overcome their unique barriers may improve adherence and HbA_1c [68]. For instance, in a 12-month RCT, intervention participants received phone calls from diabetes educators who provided tailored strategies for coping with self-care barriers [69]. HbA_1c decreased more among intervention participants than control participants, suggesting content addressing modifiable self-care barriers can be effective. However, study participants were predominantly white and well-educated [69], limiting the generalizability of the results. Not only will our diverse patient sample provide more generalizable results, but including measures that assess patients’ barriers to medication adherence and family and friend involvement in self-care will allow us to determine whether improvements in the psychosocial mechanisms targeted by the interventions explain improvements in outcomes.

Findings from the RCT will advance understanding of the health benefits of mobile phone–based interventions, with generalizability to racial/ethnic minorities and persons with low SES with chronic conditions such as diabetes [70,71]. The REACH intervention was designed to be incorporated into routine clinical care at FQHCs to support diabetes self-care adherence with minimal time investment from providers and staff. As a Type 1 effectiveness-implementation hybrid design, the primary focus of this study is to evaluate the intervention’s effectiveness. Therefore, we had research staff execute protocols to ensure a structured test of effects. For the secondary goal of assessing facilitators and barriers to implementation, we will invite intervention participants who have finished the trial, as well as FQHC providers and administrators, to participate in interviews to collect qualitative and quantitative data on their perceptions of REACH and FAMS. These interviews will focus on strategies for uptake and sustainability in clinic settings. This information will be used to develop recommendations for implementing and evaluating mobile phone–delivered interventions, like REACH and FAMS, in FQHC settings.

Limitations of this study include reliance on self-report measures of adherence. Compared with objective measures, self-report measures are subject to social desirability and recall bias. However, each measure of adherence has drawbacks. Self-report measures are inexpensive, brief, and unobtrusive, and we have selected validated measures with balancing strengths and weaknesses. Another challenge is participants changing their cell phone plans and numbers; however, the REACH Helpline (where participants can inform us of changes in their contact information), requesting secondary contact information (eg, a work number, a family member’s or friend’s number to use if we cannot reach them), calling participants after 14 consecutive days of nonresponse, and regular follow-ups help us maintain contact with participants. Our study is powered to examine the effects of receiving REACH on HbA_1c; therefore, analyses examining the effects of other outcomes (ie, self-care behaviors, self-efficacy) and comparing the effects of either intervention arm are potentially very informative but may be underpowered. Because the trial does not include a separate FAMS condition (ie, without REACH), we are not able to evaluate the effects of FAMS only. Finally, the interventions are currently only available in English, which was necessary to enhance feasibility of successfully completing this initial trial; however, translation to Spanish is a goal, should they prove effective.

We anticipate this study will help determine the effectiveness of a tailored text messaging intervention for supporting diabetes self-care adherence and reducing HbA_1c among racially/ethnically and socioeconomically diverse patients. Additionally, we aim to determine whether (1) tailoring IMB model-based content to user-specific medication adherence barriers is effective for improving medication adherence behavior and HbA_1c, thereby supporting the IMB model as an appropriate framework for interventions to promote medication adherence in diabetes and (2) basic mobile phone technology is a feasible and potentially effective medium for family-focused interventions and for engaging family members and friends in adults’ self-care efforts. Beyond these primary aims, we will be able to examine data on users’ responses to text messages throughout the trial, the choice to receive fewer text messages after 6 months, and participant characteristics associated with either. Findings will inform the design and length of future text message–delivered interventions in similar populations.