Dulce Digital: An mHealth SMS-Based Intervention Improves Glycemic Control in Hispanics With Type 2 Diabetes

Type 2 diabetes is growing in epidemic proportions in the U.S. and worldwide. The International Diabetes Federation estimates that by 2040 there will be 642 million people living with diabetes worldwide, an increase of >50% compared with the present day (1). The U.S. has the highest prevalence of diabetes among developed nations (i.e., 11% of the population between 20 and 79 years of age) (1), and individuals of ethnic/minority and low socioeconomic status are disproportionately affected (2,3). A recent study (4) in the 26 states and District of Columbia that expanded Medicaid under the Affordable Care Act found that diabetes diagnoses increased by 23% in 2014 compared with the previous year. Hispanic individuals in the U.S. experience higher rates of type 2 diabetes and, once diagnosed, exhibit poorer glycemic control than non-Hispanic white individuals (3,5).

Diabetes self-management education (DSME) and support is an effective method to improve clinical and cost outcomes (6,7) and can be successfully tailored for ethnically diverse populations (e.g., Philis-Tsimikas et al. [8]). However, many at-risk individuals are unable to access DSME and support because of practical (e.g., work, transportation, caregiving) and health care access barriers (9–12). In fact, in 2012 only 4.7% of the 21 million people with diagnosed diabetes accessed any accredited DSME program (9). To improve patient and practice performance outcomes, alternative methods must efficiently and effectively extend the reach of the care team to those in need of additional support to reach clinical targets. The widespread adoption of mobile phone technologies, including among low-income and older adults (13), highlights the potential for mobile health (mHealth) technology to circumvent the practical barriers inherent to traditional (e.g., face-to-face) visits.

Short messaging service (SMS), or text messaging, is among the most frequently used mobile communication methods and has been adopted by an estimated three-quarters of mobile users worldwide (14). Text messaging is simple to implement and may provide one of the most rapid methods to overcome the recently implicated limitation of the “digital divide” (15) to improve care. In the U.S., texting among adults in 2011 was higher among Hispanics (83%) and African Americans (76%) than among non-Hispanic whites (70%) (16). Ninety-nine percent of received text messages are opened, and 90% are read within 3 min of receipt (17). Thus, text messaging represents an opportunity to provide frequent, daily, low-cost, and interactive communication that could prove beneficial for population-level diabetes interventions.

Recent research syntheses have shown that mHealth interventions improve adherence and clinical control in patients with type 2 diabetes (18–20). However, most studies were small and nonrandomized and resulted in limited clinical improvements (20–23). Further, few studies have examined the implications of integrating these mHealth technologies into care or the feasibility and acceptability of such approaches in underserved populations (19,24). The current study addresses these gaps by investigating the glycemic benefit and acceptability of a culturally tailored, SMS-based DSME and support intervention (Dulce Digital) among underserved Hispanics with poor control in federally qualified health centers in Southern California.

Between October 2012 and February 2014, 126 individuals consented and enrolled into the Dulce Digital study (Fig. 1). The sample included Spanish-speaking and English-speaking Hispanic men and women, 18–75 years of age, who were uninsured or underinsured (Medicaid) and had type 2 diabetes and poor glycemic control (as indicated by an HbA_1c level of ≥7.5% [58 mmol/mol]). Individuals with plans to move outside the region and those with a severe physical or mental condition that would interfere with participation were excluded. Participants were recruited from clinic sites within Neighborhood Healthcare, a network of federally qualified health centers in San Diego and Riverside counties that serves predominantly low-income individuals of an ethnic/racial minority. All procedures were approved by the Scripps Health Institutional Review Board.

The intervention was tested using a parallel-groups, nonblinded, randomized design. Blocked random assignment with equal allocation was used to assign participants to Dulce Digital or usual care (UC), using a randomly generated numbers sequence. Participants were informed of group assignment after the baseline assessment.

At the baseline visit, all participants viewed a 15-min diabetes educational video developed by Scripps. All participants received a blood glucose meter (OneTouch Verio Meter; LifeScan, Inc., Milpitas, CA), testing strips, and instructions on use. A physical assessment with fasting venous blood draw and study questionnaires were completed at baseline, month 3, and month 6. Assessments were performed by trained, bilingual research assistants at clinic sites in English and Spanish. Participants received an incentive at each assessment and continued to receive UC at the clinic for the study duration. UC services available to all patients included visits with a primary care physician, certified diabetes educator, and group DSME, although the use of the services was dependent on physician and patient initiative.

After randomization, participants assigned to Dulce Digital (n = 63) were provided with instructions on how to receive and send text messages. Participants who did not have a cell phone with texting capability were provided one (Kajeet, Inc., McLean, VA) (n = 22) at no cost for the duration of the study. Participants using their own phones had the costs of the additional texts covered by the study ($12/month). Content for the text messages was primarily derived from our culturally appropriate DSME curriculum (Project Dulce), which has been shown to improve clinical, behavioral, and cost outcomes in this population and others (8,25,26). In addition to the core educational messages derived from Project Dulce (e.g., “Use small plates! Portions will look larger and you may feel more satisfied after eating.”), the Dulce Digital intervention provided ongoing support via motivational messages (e.g., “It takes a team! Get the support you need—family, friends and support groups can help you to succeed.”), medication reminders (e.g., “Tick, tock. Take your medication at the same time every day!”), and blood glucose monitoring prompts (e.g., “Time to check your blood sugar. Please text back your results.”). All content was converted into 119 brief, ≤160 character, text message–friendly format and sent out via a contracted patient health management technology platform (Rip Road LLC, New York, NY). Two to three messages a day were sent at study start, with frequency tapering over 6 months. Message timing was standardized across all participants and correlated with traditional meal or testing times. Blood glucose–monitoring prompts encouraged participants to text message in their next observed value; one value ≥250 or ≤70 mg/dL or three values between 181 and 250 mg/dL prompted a bilingual study coordinator to call the participant to assess possible reasons for hyperglycemia/hypoglycemia and to encourage as-needed follow-up with providers. The study coordinator also contacted the participant if there was no blood glucose value sent in for 1 week. Medical management was not provided by the coordinator.

Participants self-reported sociodemographic characteristics. Information regarding prescribed medications was extracted from electronic health records. HbA_1c and lipids (total cholesterol, LDL cholesterol, HDL cholesterol, and triglycerides) were conducted by the laboratories of Quest Diagnostics (West Hills, CA), which adhere to guidelines set forth by the College of American Pathologists. Systolic and diastolic blood pressure were measured with a standardized protocol according to guidelines using a standard digital sphygmomanometer (HEM-907XL; Omron). Body weight and height were measured using a traditional balance scale and stadiometer to the nearest 0.1 lb and 0.2 inch, respectively. Finally, Dulce Digital participants (only) completed self-report items at the month 6 assessment visit to evaluate intervention feasibility and acceptability. To obtain further detail regarding participants’ perceptions of Dulce Digital, two 90-min focus groups were conducted with a randomly selected 20% of intervention group participants.

Data analysis was performed using IBM SPSS Statistics for Windows, version 23.0 (IBM Corporation, Armonk, NY) and Hierarchical Linear and Nonlinear Modeling software (HLM7; Scientific Software International, Lincolnwood, IL) by A.L.F. Descriptive statistics were obtained, and distributions were examined for normality. The triglyceride variable was significantly skewed as was the natural log transformed to normalize the distribution; however, because no appreciable differences between analyses using transformed versus untransformed variables were observed, results are presented for untransformed data only.

Mixed models were used to examine whether the two groups evidenced differential rates of change over time for HbA_1c and secondary outcomes (i.e., time-by-group interactions). To evaluate a possible dosage effect on the primary outcome in Dulce Digital, the number of text messages and the number and duration of study coordinator phone calls were examined as predictors of month 6 HbA_1c level while controlling for baseline HbA_1c level. All analyses controlled for age and sex.

The majority of patients were middle-aged, female, born in Mexico, and uninsured and reported less than a ninth-grade education (Table 1). At baseline, the overall sample (N = 126) exhibited poor glycemic control (mean HbA_1c 9.5% [80 mmol/mol], SD 1.3; fasting plasma glucose 187.17 mg/dL, SD 64.75); mean lipid values were close to target, and blood pressure averages were in the normal range. No between-group differences were observed in clinical outcomes at baseline (P values >0.10).

Thirteen (10.3%) participants were lost to follow-up (Fig. 1). At baseline, these participants reported higher annual incomes (P = 0.002) and were less likely to own a cell phone (P = 0.04) than those who completed at least one follow-up assessment; no other statistically significant differences were observed (P values >0.05).

Using an intent-to-treat approach, all N = 126 participants were included in multilevel modeling analyses examining differences in the rates of change over time between the groups. Group means for all indicators at baseline, month 3, and month 6 are shown in Table 2. A statistically significant time-by-group interaction effect was observed for the primary outcome, HbA_1c (P = 0.03), indicating that Dulce Digital exhibited greater improvement in glycemic control across time compared with UC (Fig. 2A). The Dulce Digital group exhibited a significantly lower mean HbA_1c level compared with UC at both month 3 (8.5 ± 1.2% [69.0 ± 13.1 mmol/mol] vs. 9.3 ± 1.9% [78.0 ± 20.8 mmol/mol], P = 0.03) and month 6 (8.5 ± 1.2% [69.0 ± 13.1 mmol/mol] vs. 9.4 ± 2.0% [78.0 ± 20.8 mmol/mol], P = 0.03), after controlling for baseline HbA_1c. No significant time-by-group interaction effects were observed for any other clinical indicators. Because of significant group differences in the number of individuals prescribed oral (P = 0.03) and injectable (P = 0.02) antihyperglycemic agents at baseline (Table 1), these variables were examined as covariates in HbA_1c analyses. Neither medication status at baseline nor changes in medications over the 6-month period altered the results. Further, the proportion of individuals who received medication changes during the study did not differ significantly between Dulce Digital (n = 37; 51%) and UC (n = 32; 59%; P = 0.38).

Participants in the Dulce Digital intervention received an average of 354.17 text messages (SD 44.94); variation is attributable to a small number of participants (n = 6) who chose to discontinue receiving texts prior to the end of the study and to receiving additional texts during the 1-month window for completing the final study visit. Participants texted back 3–352 blood glucose values (mean 57.77 blood glucose values, SD 60.01). The number of blood glucose values texted in by participants predicted month 6 HbA_1c (P = 0.03); with each additional blood glucose value sent in, an additional 0.006 decrease in HbA_1c level was observed at month 6 (Fig. 2B). The number of text messages sent out to participants was not a significant predictor of month 6 HbA_1c levels (P = 0.28). Nineteen of 63 (30%) participants in the Dulce Digital intervention received phone calls from the study coordinator in response to out-of-range blood glucose values. For these n = 19 participants, the number of coordinator phone contacts ranged from 3 to 13 calls/participant (mean 7.63 calls, SD 2.81); the modal call duration was 5 min. Neither the number nor the total duration of coordinator phone calls per participant predicted month 6 HbA_1c levels (P values >0.05).

In response to the subset of self-report items administered in the Dulce Digital group (only) at month 6, the vast majority of participants indicated that the text messages helped them to manage their diabetes “a lot” (96%), that they would continue receiving Dulce Digital text messages if given the choice (96%), and that they would recommend Dulce Digital to a friend or family member with diabetes (97%). Consistent with these findings, focus group participants (n = 12) indicated high acceptability; common themes that emerged indicated that text messages were sufficient in frequency and easy to understand. However, individuals who were provided with a separate study phone reported that it was inconvenient to carry two phones.

To our knowledge, this is the first randomized controlled trial using a text message–based DSME and support intervention to demonstrate significantly greater improvements in glycemic control compared with UC in a high-risk, underserved, Hispanic population. These findings suggest that, if implemented on a wider scale, simple, low-cost, text message–based mHealth approaches such as Dulce Digital have the potential to achieve a significant public health benefit in diabetes, a chronic health condition that is rapidly increasing in the Hispanic and other underserved populations.

In this study, all patients had an initial HbA_1c level ≥7.5%, with the majority (61.9%) exhibiting an HbA_1c level ≥9%, thus placing them at high risk for diabetes complications. The Dulce Digital group mean did not reach the optimal level of <7.0–7.5%. However, during the 6-month intervention, 68% of Dulce Digital participants achieved reductions in HbA_1c levels, with individual decreases ranging from −0.1% to −4.6%. Overall, Dulce Digital achieved a mean 1% reduction in HbA_1c, which exceeds the average effect size reported in reviews of other mHealth approaches (pooled HbA_1cΔ = −0.5%) (18–20) and the average HbA_1c reduction reported in a recent synthesis of 118 traditional DSME trials (average HbA_1cΔ = 0.74%) (27). Baseline HbA_1c levels in the reviewed studies ranged from near normal (6.5%) to extremely elevated (9.8%), which is similar to the results of our study. In one review from Greenwood et al. (28), it was noted, unsurprisingly, that when study participants’ mean baseline HbA_1c level was at or close to goal, a small intervention effect was seen. The relationship of the baseline HbA_1c level to the intervention effect demonstrates the importance of choosing the optimal population for interventions. With the current broad use of diabetes registries and the ease of identifying higher-risk individuals using population health management methods, it may be worthwhile to focus initial efforts of technology interventions in our higher-risk patients. An HbA_1c reduction of the magnitude found in the current study should not be underestimated and should be considered not only statistically significant but also clinically relevant; it rivals the effects of some glucose-lowering medications (29–31) and has been associated with decreased risk for retinopathy, nephropathy, and neuropathy (32–34).

Notably, the number of blood glucose values texted in by participants predicted month 6 HbA_1c values. Individuals who sent in more blood glucose values may have been more actively engaged with the intervention (and were thus reading more of the DSME messages) and/or more active participants in their health. Nonetheless, additional research is needed to reach conclusions about the mechanism underlying this relationship. Results did not indicate a significant “dosage” effect between text message frequency and HbA_1c level. However, further research is warranted to determine the “optimal” frequency of texts, acknowledging that optimization will likely need to be individualized according to each patient’s unique needs and preferences. Similarly, the number of phone contacts with the study coordinator (in response to blood glucose values) did not predict month 6 HbA_1c levels. This finding is not surprising given the brief, safety-focused nature of these calls; the coordinator did not provide health coaching or in-depth clinical guidance but rather assessed for emergent safety risks and recommended physician follow-up as appropriate.

No statistically significant intervention effects were observed for secondary clinical indicators. However, this may be due to the fact that, on average, blood pressure and lipids were close to target levels at baseline, leaving little room for improvement. Further, the Dulce Digital content was not designed to address lipid or blood pressure control.

The current results should be interpreted in the context of several limitations. First, although attrition was comparable to that observed in prior studies (35), attrition was higher in the intervention group relative to the UC group. Thus, it is possible that participants who remained in the study were more engaged. A worst-case scenario sensitivity analysis is presented in Supplement A. Second, it is unknown how much of the intervention content was received (i.e., read or comprehended) by participants. Nonetheless, the number of blood glucose values texted in may be construed as an indirect indicator of intervention engagement. Third, 44% of potentially eligible individuals were enrolled, with the majority reporting practical barriers that are commonly experienced in underserved populations (e.g., no transportation, work conflict, caregiving responsibilities, other time conflict). Thus, it is expected that by introducing the Dulce Digital program as part of routine clinic care, program reach could be expanded to individuals who could not attend additional (research-required) visits. Finally, this trial was not designed to examine cost-effectiveness. However, with respect to sustainability and scalability of an intervention such as Dulce Digital compared with other low-resource settings, the promise of text message–based programs is great. There is no additional cost for the technical infrastructure whether it is delivered to 400 or 4,000 individuals. The cost to the patient/user is only related to the text messages, and for Dulce Digital (in particular) a smartphone is not required. In other research, a significant population-level cost savings was attributed to a text-messaging program designed to facilitate diabetes care coordination in a predominantly African American population (22). This study included a care management component facilitated by nurses or medical assistants.

Dulce Digital offers a potential solution to the burgeoning primary care demand-capacity imbalance to better address the complex needs of the growing number of individuals with type 2 diabetes. Text-messaging approaches are attractive as a chronic disease public health intervention for a number of reasons, including their frequent use, enormous reach, low cost, and relative simplicity. Mobile phone and texting use is high among Hispanics (16), a group that experiences disparate diabetes prevalence and outcomes. Moreover, the present investigation indicated that the Dulce Digital approach was highly acceptable in this population. This model is flexible, lending itself to adaptation for other chronic conditions (e.g., arthritis, chronic pain) and for delivery by other personnel to address the health needs of underserved populations across the nation. Future investigations should examine the sustainability of the improvements in glycemic control beyond 6 months; expanding intervention content to target additional populations at risk for diabetes and other cardiometabolic indicators that are central to diabetes control (e.g., blood pressure, lipids); and individualizing text message content and delivery timing and frequency to each patient’s unique needs and progress.

Acknowledgments. The authors thank the staff and participants of the Dulce Digital study for their important contributions and the administration and staff at Neighborhood Healthcare for their partnership in this investigation. The authors also thank peer educator Magdalena Hernandez, an employee of Scripps Whittier Diabetes Institute, for recruitment and cohort maintenance support.

Funding. The current research was supported by McKesson Foundation grant 115M803379 and National Center for Advancing Translational Sciences grant NCATS 1UL1 TR001114-01. The Investigator-Initiated Study Program of LifeScan, Inc., provided glucose testing meters and strips for all participants.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.L.F. researched and analyzed data and wrote and reviewed the manuscript. L.C.G. contributed to the research development, conduct of research, and data analysis and wrote and reviewed the manuscript. M.I.G. contributed to the conduct of research and data analysis and reviewed the manuscript. M.T. contributed to the research data and analysis and reviewed the manuscript. J.A.E., T.C., and J.Sk. contributed to the research analysis and reviewed the manuscript. M.R., S.D.-S., and J.Sc. contributed to the research development and conduct of research and reviewed the manuscript. A.P.-T. contributed to the research development, conduct of research, and research analysis and wrote and reviewed the manuscript. A.P.-T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 74th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 13–17 June 2014.