We conducted a three-group pilot randomized controlled clinical trial comparing the efficacy of a behavioral lifestyle intervention modified for underserved populations using either (1) mobile or (2) paper-based tools for self-monitoring of diet, PA, weight, and blood glucose and (3) usual diabetes care and education on glycemic control and weight loss at 3 and 6 months. We used a mixed-method design with quantitative measures to evaluate the feasibility and preliminary efficacy of the intervention and conducted focus groups to assess participants’ acceptability of the intervention. We are reporting the quantitative study findings in this paper. The study is approved by the institutional review board at the University of Texas Health Science Center at Houston. Consolidated Standards of Reporting Trials (CONSORT) of Electronic and Mobile HEalth Applications and onLine TeleHealth was used to guide the reporting of this study; a checklist was uploaded as Multimedia Appendix 1 in this paper [25].

Participant inclusion and exclusion criteria are shown in Textboxes 1 and 2.

We used the age criteria from the Look AHEAD clinical trial and limited the exclusion criteria to include a general clinical population with a goal to be pragmatic in the nature of this study. To obtain a more representative sample of the underserved population, we included patients with amputations as long as they were able to perform regular activity such as walking. Due to the nature of this study being a pilot and feasibility study, we aimed to recruit 30 patients with 10 patients in each group.

Participants were recruited from an American Diabetes Association certified diabetes education program located in a community health center primarily serving uninsured or underinsured individuals living in Harris County, Texas. Flyers were distributed to the patients attending diabetes self-management education classes by their diabetes educators.

If patients showed interest in this study and approached the study team for more details, they were screened for eligibility based on the inclusion and exclusion criteria; their diagnosis of T2DM was confirmed via chart review. Informed consent was obtained from all eligible study participants before enrollment. The study participants were aware of the differences among the three randomization groups during the consenting process. After consent and enrollment, we assigned each patient with a study ID number. The study statistician generated a randomization sheet with the group assignment for each study ID. Study participants were then randomly assigned to one the three study groups based on the randomization sheet.

A checklist was developed and used for each group and individual session to track the content delivered. The principal investigator (PI) attended at least 80% of the group sessions for both paper and mobile groups to ensure treatment fidelity. Training of the two lifestyle counselors (their backgrounds were in public health and kinesiology) occurred 4 months before the study. Mock sessions were conducted on weekly meetings where lifestyle counselors developed PowerPoint slides, delivered mock intervention sessions, and reviewed the checklist for each session.

The recruitment, feasibility, and tracking forms were collected and stored in the PI’s office at the Cizik School of Nursing at The University of Texas Health Science Center at Houston for data processing. Oracle (version 9i, Oracle Corporation, Redwood Shores, California) was used for data management. Form design, data entry, and data verification were performed in TeleForm (version 10.0, Verity Inc, Sunnyvale, California) for automated data entry or verification. All forms were precoded to minimize coding errors. During data collection, forms were screened upon receipt for completeness of response. Once verified, data were exported to the Oracle database for further data processing before being exported to SAS (SAS Institute) for data analysis.

Intention-to-treat (ITT) analyses were performed on the primary and secondary outcomes. Descriptive statistics (frequency and percentage for categorical variables and mean and SD or median and interquartile range for continuous variables) were reported for retention at 6 month of the intervention, attending group sessions, and adherent to the multiple-behavior self-monitoring. For continuous variables with skewed distribution (eg, retention rates and group session attendance rates), nonparametric Mann-Whitney U tests were conducted for comparison between the mobile and paper groups. For the primary outcome, the percentages of weight change over time were compared by Kruskal-Wallis test, and the percentages of HbA_1c change over time were compared by analysis of variance. Sensitivity analyses were conducted using last observation carried forward (LOCF) method to impute the missing data for participants who withdrew or were lost to follow-up.

Demographic characteristics of the sample by randomization group are presented in Table 2. The average age of the participants was 56.4 years, and the average years of education were 12.15 years (SD 1.22). A total of 62% (16/26) of the sample were female, and 69% (18/26) were African Americans. The BMI ranged from 27.4 to 51.1, with average of 38.1 at baseline. The majority of the sample had no health insurance or received only Medicare or county-assisted insurance in Harris County, Texas. All of the study participants were uninsured or underinsured. The household income for all study participants was below US $30,000, and 92% (24/26) had a household income lower than US $20,000. Age (P=.04) and gender (P=.07) differed significantly among the three randomization groups, but no statistically significant differences were found among other demographic variables.

Descriptive findings on HbA_1c and weight outcomes at each study data collection time point (baseline, 3 months, and 6 months) are summarized in Table 3. At baseline, there were no statistical significant differences on HbA_1c among the three randomization groups; at 6 months, a statistical significant difference on HbA_1c was found among the three groups, with mobile group having an average HbA_1c level <7%, whereas the paper group and control group had an average HbA_1c level around 9%. Results from the ITT analysis on the primary outcome of HbA_1c showed that there were no statistical significant group differences on HbA_1c level change over 6 months (P=.44); however, a medium effect size of Cohen d=0.28 was detected for HbA_1c changes. At 6 months, participants in the mobile group had an average weight loss of 1.8%, whereas the paper group had an average of 4% weight gain, and the control group had an average of 1.6% weight gain. There were no statistical significant differences among the three groups on weight changes over time (P=.20). A medium effect size of Cohen d=0.40 was found for changes on weight outcomes over time. Sensitivity analysis using LOCF for imputations did not show any statistically significant differences on the HbA_1c and weight outcomes.

To our knowledge, this pilot study is the first to report the feasibility and acceptability of using mobile and connected tools to enhance an evidence-based behavioral lifestyle intervention for the underserved community. We compared the efficacy of standard diabetes care and education with behavioral lifestyle interventions enhanced with either using smartphone apps and a Bluetooth-connected glucometer for self-monitoring of multiple behaviors or paper diaries on improving glycemic outcomes among overweight or obese adults with T2DM living in underserved communities. The feasibility and acceptability of the study were demonstrated by the high retention rates at 3 and 6 months and high rates of patient engagement in using the mobile apps. In fact, our retention rates of 96% at 3 months and 92% at 6 months were higher than those reported in most of the previous behavioral lifestyle interventions mediated by technology in obesity and T2DM [26,27], including those in medically underserved communities [28].

The comparative findings revealed the mobile group participants had higher group session attendance and higher patient engagement and adherence to self-monitoring of multiple behaviors than the paper group, which was consistent with previous studies reporting higher adherence to self-monitoring rates using electronic diaries compared with paper diaries among overweight or obese populations [29,30]. As compared with the diabetes population, a study planning to use a mobile app to support patient self-management did not recruit enough patients [31]. We recruited our 27 patients in less than 1 month, and those who participated in the mobile group had high adherence rates to all components of the intervention over 6 months. This may suggest that a mobile app alone does not interest patients as much as a hybrid of face-to-face sessions using mobile apps to support self-monitoring.

Although the previous literature comparing electronic diaries and paper diaries for self-monitoring of diet and PA did not reveal significant differences in weight loss outcomes [21,32], our study not only showed significantly better adherence to self-monitoring in the mobile group but also a trend for greater weight loss and glycemic control with medium effect sizes in the mobile group. Furthermore, the mobile group had significantly lower HbA_1c levels at 6 months than the paper group. A meta-analysis of lifestyle weight loss interventions in overweight and obese adults with T2DM revealed that the majority of the trials did find <5% weight loss; however, they did not reveal significant beneficial effects on glycemic control [33]. The self-monitoring intervention in these lifestyle weight loss interventions focused on self-monitoring of diet and PA only, with a few of them adding self-monitoring of weight, whereas our study used a holistic approach to introduce SMBG and weight, along with self-monitoring of diet and PA behaviors to help patients understand the relationship between their behaviors and outcomes.

To our knowledge, our study is the first to combine self-monitoring of diet, PA, and weight using a mobile app that are used in behavioral weight loss interventions along with a connected glucometer to help patients learn their behavioral patterns in association with their weight and blood glucose outcomes. Previous studies had used either connected glucometers along with access to a live certified diabetes educator coach [34] or personalized feedback messages based on connected glucometer results [35] for general T2DM patients, not specifically targeting overweight or obese T2DM patients from underserved communities.

Several limitations to this study should be acknowledged. First, the study sample was recruited from an underserved community in an urban setting, so the study findings may not be generalizable to underserved communities in rural areas. Second, the focus of this study was feasibility and acceptability; thus, the study did not have sufficient power to detect group differences. Third, we provided smartphones and Bluetooth-enabled glucometers to the participants because none of the study participants reported owning a smartphone; the adherence to self-monitoring may be different for those who previously owned a smartphone. Fourth, our measure on adherence to self-monitoring of PA depended on patient adherence to the recommended PA behaviors. Although this approach has been used in several other studies [29,36], it may underestimate the actual adherence to self-monitoring of PA. For example, our adherence to self-monitoring of PA was lower than adherence dietary self-monitoring, which could suggest that participants did not exercise at all on that particular day, they did not bother to enter 0 for exercise minutes, and instead, they left it blank. Future research should examine the difference between adherence to self-monitoring and adherence to the actual behavior separately. Fifth, our study only looked at the short term outcomes; maintaining long-term effect may be a different challenge that future studies should consider examining.

Delivering a behavioral lifestyle intervention enhanced with multiple-behavior self-monitoring using smartphone apps and a connected Bluetooth glucometer in an underserved community is feasible and acceptable, and using mobile tools including smartphone apps and connected glucometers has the potential to increase patient adherence to self-monitoring of multiple behaviors and improve glycemic control among underserved populations. A full-scale randomized controlled trial is needed to confirm the findings of this feasibility trial.