Diabetes mellitus (DM) is a progressive chronic disease that can result in serious short- and long-term complications. Patient self-management education and support are fundamental to improving DM management [1-6], and guidelines recommend that every person with DM receive self-management education in a format appropriate for the patient’s specific cultural, socioeconomic, literacy, and numeracy characteristics [2,3]. A growing number of self-management education programs for patients with DM are incorporating information technologies (IT) to improve their effectiveness and reach [7-9].

Nutrition therapy is one of the most challenging components of self-management for many patients with type 2 DM (T2DM). It is therefore particularly important that patients receive dietary education and collaborate with providers to develop individualized eating plans that are both implementable and sustainable and incorporate their preferences and needs [3]. Despite its critical role, few studies have focused on the nutritional education and counseling component of these DM self-management interventions [4-6,9-12] or reported their effectiveness in improving DM-specific dietary knowledge [7,8].

Patient-centered DM dietary education is especially critical for ethnic/racial minority populations, who often bear a disproportionately high burden of T2DM [2,13,14]. The standard dietary education and advice provided by mainstream health care services may not adequately address the daily challenges faced by minority patients as their cultural, social, and dietary norms and socioeconomic realities differ from that of the majority population. IT tools can provide new opportunities to make DM-related dietary education and counseling more relevant and individually tailored for patients [7,8]; however, few existing examples/initiatives have included sizeable proportions of ethnic minority patients [7,15]. There is a need to expand the evidence base for new digital health technologies that can address these needs [16] among the highest-risk, most vulnerable patient populations.

In this paper, we report the development of a dietician-operated, culturally adaptable Interactive Lifestyle Assessment, Counseling, and Education (I-ACE) software. We further report the results of a pilot randomized controlled trial (RCT) evaluating its effect, compared with standard lifestyle advice (SLA), on improving DM-related dietary knowledge, lifestyle behaviors, and glycemic control in a sample of low-socioeconomic status (SES) adults with T2DM from the Arab minority in Israel.

This was an open, parallel-group, pilot RCT randomized controlled pilot trial testing the effect of a 6-month, dietician-delivered, face-to-face diabetes lifestyle (diet and physical activity [PA]) counseling program using the I-ACE software compared with SLA. The I-ACE software was adapted to provide culturally congruent lifestyle counseling to Arab adults with T2DM.

Ethics approval was obtained from the Helsinki committees of Sheba Medical Center and Clalit Health Services, and all participants provided written informed consent before enrollment. The study was registered at ClinicalTrials.gov (NCT01858506). The CONSORT eHealth checklist is provided in Multimedia Appendix 1.

The Arab minority in Israel is an indigenous population that accounts for 20% of the total population. It differs in language, culture, and religion from the majority Jewish population and, for the most part, resides in residentially segregated and economically deprived communities [17]. It is characterized by a lower SES and higher rates of T2DM, poor glycemic control, and diabetic complications, than the majority population [17-20].

A total of 50 eligible participants were recruited from the clinics of Clalit Health Services in 2 Arab towns. Inclusion criteria were: (1) age between 40 and 64 years, (2) diagnosis of T2DM, (3) having T2DM for ≤10 years, (4) body mass index (BMI) of 27 to 43 kg/m², and (5) hemoglobin A_1c (HbA_1c) between 8.0% and 11.3%. Participants were not eligible if they (1) were receiving short-acting insulin treatment, (2) had inadequate control of comorbid conditions, or (3) had factors that would limit adherence to interventions (eg, any medical or physical condition that prohibited participation in PA or following standard diets recommended for people with diabetes, pregnancy, uncontrolled psychiatric condition, significant cognitive impairment, or blindness).

Recruitment was conducted at the local Clalit Health Service clinics, in collaboration with the physicians of the potential participants. The lists of potential participants meeting the inclusion criteria were generated from Clalit electronic medical records and reviewed by the patients’ physicians to identify eligible candidates. Patients with an HbA_1c result measured more than 3 months before eligibility screening were requested to get the test repeated. Baseline evaluation (including initial lifestyle knowledge, dietary and PA assessments, and anthropometric measurements) were completed in the clinic before randomization.

Eligible participants were randomly assigned in 1:1 ratio either to the I-ACE or SLA study arm using a permuted block design, central computer–generated randomization process, with even-numbered blocks of 2 to 6 participants. The randomization was performed at the Gertner Institute. Allocation concealment was maintained until after the provision of informed consent and randomization. Although the intervention type was known to the participants, dieticians, and study coordinator, the study statisticians were blinded to group allocation until the primary study outcomes analyses were completed. Group assignment was masked from all health care service providers other than the dietician.

The recruitment ran from August 2014 to January 2015. The participant follow-up and data collection from the electronic medical records was completed in March 2016. The pilot trial ended, according to the protocol, after all the participants had been followed up for 1 year.

The sample size calculation was based upon the reported differences in the nutritional knowledge change (percentage of correct responses) between the intervention and comparison groups in 2 nutritional educational interventions, which ranged from 8.9% to 11.5% [21,22]. A sample of 50 participants was needed to provide 90% power to detect a statistically significant difference of this magnitude at the 5% level using a 2-sided test between the experimental and comparison groups, allowing for a dropout rate of 10%.

The study included 2 active intervention arms: (1) the I-ACE experimental arm and (2) the SLA comparison arm. The experimental arm of the intervention used the I-ACE software.

For the primary outcome of DM-related dietary knowledge, the mean (SD) values were reported. A linear mixed regression model for repeated measures was used to evaluate the change in dietary knowledge over time, with an interaction term to determine whether the study groups differed from each other over time. There were missing data in this outcome at the different evaluation periods (34% [17/50] at 3 months, 48% [24/50] at 6 months, and 10% [5/50] at 12 months); however, as there were no missing observations at baseline, we compared the mean baseline DM-related dietary knowledge scores in participants with observed values and those with missing values at each evaluation period. Similar averages imply noninformative missingness (ie, missing at random), supporting the appropriateness of using a linear mixed model for repeated measures with maximum likelihood for this analysis.

For continuous secondary outcomes (eg, food and nutrient intakes, BMI, WC, and HbA_1c), mean (SD) values were reported. Multiple linear regression models were used to test for the differences between the study groups at 12 months for changes in lifestyle behaviors. Linear mixed regression models for repeated measures were used to evaluate within-person change over time in dietary intakes in the I-ACE arm as repeated measures from each counseling session were available. WC and BMI, which had repeated measurements in both study arms, were evaluated with linear mixed regression models for repeated measures, with interaction terms for study group by time. For the repeated outcome measure HbA_1c, we applied a linear mixed model on the log scale (owing to its non-normal distribution), including a random intercept and potential fixed effects (gender, study arm, age, and baseline DM drug therapy), using the nlme package in R. We considered several functions for modeling the effect of time at knots around 6 months as HbA_1c was not measured exactly at 6 months, including (1) 2 intervals with the cut-point of 6 months, (2) 2 intervals with the cut-point of 8 months, (3) natural cubic splines with 1 knot, and (4) natural cubic splines with 2 knots. We compared the mean square error of the models resulting from the different functions using cross-validation. The predicted values for a typical subject from each study arm were presented in a graph for each time. The expected difference in HbA_1c from baseline to 6 months and 12 months and their 95% CIs were calculated from the chosen model.

For binary secondary outcomes (eg, any LPA and recommended LPA level), raw count (number, %) was reported. Multiple logistic regression models were used to test for the differences between the study groups in these outcomes at 12 months.

Sex was forced into all multivariable models. Other key variables (age, study town, DM therapy, SKILLD score, and education) were entered into the models and only those with P<.10 were retained. Given the small sample size, it was important to keep the models as parsimonious as possible.

All analyses were performed using SAS version 9.4 SAS Institute, except for the HbA_1c repeated measures outcome, which was performed with the open-source statistical software platform R [35]. Simple tabulation and narrative description were used to report the participant and dietician feedbacks on the utility and satisfaction with the counseling and the I-ACE software.

The participant flow chart is presented in Figure 1. A total of 195 potential participants were identified from Clalit electronic medical records, of whom 123 did not meet the inclusion criteria, 22 refused to participate, and 50 were randomized (25 to the I-ACE arm and 25 to the SLA arm). All those randomized to the I-ACE arm received the allocated intervention (n=25), and all but 1 participant (who was diagnosed with cancer in the first month after randomization) assigned to the SLA arm received the allocated intervention (n=24). One patient with confirmed diabetes and baseline HbA_1c of 6.1% was included in the study by mistake and randomized to the I-ACE arm. This patient was included in the data analysis. Furthermore, 3 participants, 1 in the I-ACE arm and 2 in the SLA arm, were lost to follow-up and 1 participant in the SLA arm withdrew consent.

Data on the primary outcome (DM-related dietary knowledge) were collected from all 50 participants at baseline (25 in the I-ACE arm and 25 in the SLA arm), but was missing for 17 participants (34%) at the 3-month evaluation (7 in the I-ACE arm and 10 in the SLA arm), 24 participants (48%) at the 6-month evaluation (13 in the I-ACE arm and 11 in the SLA arm), and 5 participants (10%) at the 12-month evaluation (1 from the I-ACE arm and 4 from the SLA arm). To check whether the missing data pattern was informative, we compared the average baseline DM-related dietary knowledge score of participants with observed values to those with missing values at each evaluation period. We did not find significant differences at any of the periods, implying noninformative missingness (see Multimedia Appendix 6).

Information on between-group change in dietary behaviors from baseline to 12 months was missing for 9 participants (4 from the I-ACE arm and 5 from the SLA arm). The analysis of within-group change in dietary behaviors across the intervention counseling visits included all I-ACE participants with more than 1 counseling visit (n=25). The LPA outcomes were missing for 5 participants (1 from the I-ACE arm and 4 from the SLA arm). All 50 participants (25 in each arm) were included in the WC analysis and HbA_1c analyses. Counseling satisfaction and utility questionnaires were missing for 5 participants (2 from the I-ACE arm and 3 from the SLA arm). All analyses were conducted according to the originally assigned study groups.

Table 1 presents the baseline characteristics of the participants, none of which differed significantly by study arm. The average age of the participants at baseline was 53 years, and over 60% were treated with basal insulin, with or without additional oral hypoglycemic therapy. The mean baseline HbA_1c was above 9.0% in both study arms.

Table 2 presents the baseline diabetes knowledge and lifestyle behavior data. Participants did not differ on the overall general diabetes knowledge score as measured by the SKILLD scale, which was below 50 on a scale of 100. There were also no significant differences in baseline lifestyle (diet and PA) knowledge or behavior variables by study arm. Participants had suboptimal intakes of dietary fiber, vegetables, and whole grains and very low participation in LPA.

To construct the DM-related lifestyle knowledge score, we excluded the knowledge questions for which ≥85% of participants gave correct answers at baseline, to focus the score on items in need of and amenable to modification by the intervention. The items included in the score are noted in Table 2.

The mean (SD) number of counseling sessions attended (out of a maximum of 4) did not differ between the I-ACE and SLA arms (3.44 [SD 0.96] vs 2.92 [SD 1.22] respectively; P=.10) Over two-thirds (68%) of the participants in the I-ACE arm attended all 4 counseling sessions and another 16% attended 3 sessions, representing a high adherence of 84%. In the SLA arm, 48% and 16% of the participants attended 4 and 3 counseling sessions, respectively, totaling 64% with high adherence. Session attendance adherence tended to be higher among women than men, but this varied by study arm. Similar proportions of men and women attended at least 3 sessions in the I-ACE arm (88% vs 82%, respectively; P=.46); whereas the proportion tended to be lower for men than women in the SLA arm (54% vs 75%, respectively; P=.03; see Multimedia Appendix 7 for the number of visits and qualitative information about adherence barriers). The digital visit records saved in the I-ACE software for the experimental arm indicated the intervention adherence (ie, the use of the software for assessment and counseling) in 100% of first visits and 85% of follow-up visits.

The primary study outcome, change in DM-related lifestyle knowledge from baseline, differed between the study arms over time (Figure 2; P value for study arm×time interaction=.02). Multimedia Appendix 8 shows the results of the linear mixed model for repeated measures testing the within-group and between-group differences in the knowledge score from baseline to 3, 6, and 12 months. In the I-ACE arm, the mean knowledge score was significantly higher at 3 months than the baseline score and a significant difference was maintained through 12 months. In the SLA arm, the accrual of knowledge occurred more slowly and did not differ significantly from baseline until 6 months. The difference in the slope of the knowledge scores between the study arms was significant at 3 months (higher for I-ACE, P=.03). From the sixth month onward, both groups had approximately equal levels of DM-related lifestyle knowledge, which remained significantly higher at 12 months than their mean baseline scores. The increase in knowledge was significantly greater for women than for men (mean [SE] difference=5.14 [SE 2.28]; P=.03), whereas, the median (IQR) educational level was lower for women than for men (9 [IQR 8-12] years vs 12 [IQR 10-15] years, respectively; P=.049).

Multimedia Appendix 9 presents the change in lifestyle behaviors from baseline to 12 months. With regard to the dietary behaviors recommended for diabetes management (eg, adequate consumption of nonstarchy vegetables, whole grains, and dietary fiber and limited consumption of fruit and added sugar), the I-ACE arm exhibited a significant reduction in added sugar intake and a significant increase in dietary fiber intake. The SLA arm exhibited a positive change with regard to a significant decrease in fruit intake and a marginally significant increase in dietary fiber intake but a negative change with regard to a significant decrease in vegetable intake. Although there was a trend toward greater improvement in dietary behaviors in the I-ACE arm than the SLA arm at 12 months, none of the differences reached statistical significance, with the exception of a marginally significant lower intake of added sugar in the I-ACE arm.

As the I-ACE software enabled collecting dietary intake data at each counseling session, we were able to evaluate the within-person change in dietary behaviors across the study visits in the experimental arm (Table 3). Significant changes in the desired direction for all dietary variables occurred during the most intensive phase of the intervention (from baseline to 3 months), during which there were monthly counseling sessions. After that, there was a regression toward baseline intakes and only the changes in added sugar and dietary fiber from baseline remained significant at 12 months.

The increase in the percentage of participants in the I-ACE arm reporting any LPA (sex-adjusted odds ratio [OR] 2.80, 95% CI 0.67-11.58; P=.16) or the recommended LPA level (sex-adjusted OR 5.01, 95% CI 0.52-47.92; P=.16) was greater than that in the SLA arm (Multimedia Appendix 9). However, given the small sample size, the differences did not reach statistical significance.

For the repeated HbA_1c measurements, the model with natural splines with 2 knots (at 5 and 8 months) best fit the data. In addition, the final model included gender. The effect of the study arm was nonsignificant (P=.40); therefore, it was omitted from the model. The effect of time was significant (P<.001). According to the model, the mean HbA_1c values were expected to decrease from baseline levels by 11% after the 6-month follow-up (6 months/baseline HbA_1c ratio=0.89, 95% CI 0.88-0.91) and by 7% after the 12-month follow-up (12 months/baseline HbA_1c ratio=0.93, 95% CI 0.92-0.95). For example, the average HbA_1c level at baseline of 9.1% (95% CI 8.8-9.5) was expected to decline to 8.1% (95% CI 8.0%-8.2%) at the 6-month follow-up point and to 8.5% (95% CI 8.4%-8.6%) at the 12-month follow-up point.

Figure 3 presents the predicted HbA_1c (in original units) for a typical subject from the SLA and I-ACE study arms. For the purpose of this figure, the study arm was not omitted from the model, although it was nonsignificant. HbA_1c levels decreased over time during the active intervention period (1-6 months) for both study arms, reaching a minimum value at approximately 6 months. During the postintervention follow-up period, the levels slightly increased but were still lower at 12 months than baseline levels. There was no significant change in WC over time and this did not differ by study arm in multivariable analysis. Crude results were similar for BMI, and owing to the nonnormal distribution of this variable and its close correlation with WC, multivariable analysis was not conducted for BMI.

Over 90% of the participants in both study arms expressed high satisfaction with the dietician and lifestyle counseling. Most participants indicated that they understood and could implement/utilize the information provided on general nutritional recommendations, portion sizes, exchanges portions, sample menus, and their progress over time; however, the proportion who indicated that they could only understand or implement these materials partially, or not at all, tended to be higher in the SLA than the I-ACE arm (Figure 4).

Most participants in the I-ACE arm thought that the I-ACE software was helpful to the dietary counseling (91%) and did not detract from their interaction with the dietician (96%). The study dietician also expressed overall satisfaction with the I-ACE software. She found the pictorial educational materials and the quantitative nutrient information, provided when individually tailoring sample menus, particularly helpful. She observed that younger participants who were technology oriented engaged more with the software in the counseling sessions than older participants. She noted areas that the software did not address (eg, emotional distress of coping with multiple comorbidities or other personal issues and other emotional support issues that may impact lifestyle behaviors and the readiness to make lifestyle changes).

We have no harms or unintended consequences to report.

This pilot trial provided indications that the use of the culturally adapted I-ACE software for dietician-delivered lifestyle counseling can increase the pace of acquiring DM-related lifestyle knowledge and showed a trend toward improving lifestyle behaviors (diet and LPA). It was further associated with improved HbA_1c results, but not to a greater extent than the dietician-delivered SLA.

This study adds to the body of evidence supporting the efficacy of dietician-delivered dietary counseling for improving dietary knowledge and behaviors and health outcomes in patients with T2DM [3,36]. Consistent with our findings in this study, IT-assisted interventions among T2DM patients have resulted in more rapid knowledge acquisition than non-IT-assisted interventions, particularly when they utilize a combination of provider contact and technology [4,36].

This also holds for ethnic/racial minority groups, particularly when the interventions are culturally adapted/specifically tailored for the target groups [37,38]. In this trial, both groups received written materials in Arabic and dietary counseling from a culturally and linguistically congruent dietician. The digital platform used in the I-ACE arm enabled additional enhancement of cultural adaptation and individual tailoring. The study dietician had access to pictures and infographics reflecting ethnic dietary habits and commonly consumed foods to support the counseling process, both for educational purposes and for creating individually tailored sample menus.

In addition, I-ACE facilitated the use of a broad range of behavioral change modeling techniques to enable active interaction and immediate feedback. These included motivational interviewing, goal setting, goal modification, simulation of the effects of specific lifestyle changes, and tracking of progress. The Alive! IT-based lifestyle intervention, conducted in a workplace rather than specifically among T2DM patients, employed a similarly broad suite of behavioral change techniques (including the less common component of simulation) and showed significant lifestyle change compared with a wait-listed control group [39,40].

As an additional clinical counseling support feature, I-ACE makes it feasible to collect and update quantifiable data on changes in dietary behaviors at each counseling session, providing a new resource for tracking behavioral change over time. Furthermore, the documentation of the visit record and the dietary counseling process provides a data resource for quality control and other organizational purposes.

The I-ACE tool differs from existing lifestyle IT tools/applications, which typically rely upon multiple administrations of daily trackers to collect lifestyle data, resulting in a heavy user burden, user fatigue, and disuse [41]. In addition, these tools/apps are generally used independently by patients, and thus, quantified, actionable information about dietary self-management may not be readily accessible and is not a documented component of the clinical visit record [42-48]. Numerous studies have shown lifestyle-related IT tools to be more effective when used in clinical settings with health care provider contact than when used independently [4,36,49].

Few apps developed specifically for T2DM self-management include the assessment of dietary/lifestyle habits. Those that have included this domain either (1) did so in a superficial and nonquantified manner and did not improve dietary behaviors [44] or (2) required detailed recording of daily dietary intake, which resulted in a very high user burden [50] and was thus one of the least used features of mobile T2DM support apps [45]. Other apps/Web-based IT tools included unidirectional messages from the app to the patient or noninteractive Web-based information that did not support any individual tailoring of dietary/lifestyle plans [51,52].

Users’ satisfaction with I-ACE was high, suggesting it may enhance the dietary counseling process. Responses of the I-ACE arm participants to the utility questions also suggest that it may have potential for optimizing the patient’s understanding of and adherence to general nutritional recommendations, portion size and exchange portion recommendations, individually tailored menus, and the tracking of progress over time.

Reviews of the literature show that dietary counseling, particularly among patients with T2DM, improves the glycemic control to the same magnitude as expected from introducing a new drug [3,36-38,53]. In this pilot trial, we found that a short, dietician-delivered dietary counseling series (eg, 4 sessions over the course of 6 months) produced a decrease of 1% point from baseline HbA_1c at the end of the active intervention period and a 0.5% decrease at 12 months (6 months after the active intervention ended), in a sample starting with poor glycemic control.

At the same time, the limited number of dietary consultations in this pilot trial did not appear to be adequate to ensure fully sustained improvements in dietary behaviors and glycemic control. Systematic reviews of diabetes self-management education and nutritional interventions among adults with T2DM have shown a higher success rate in interventions with over 10 hours of contact with providers [4]. Interventions that provided dietary counseling encounters for more than 6 months reported that the improvement and continued reduction of HbA_1c was maintained for up to 2 years [3].

This pilot study has several strengths and limitations. The study adds to the very limited literature on culturally adapting and testing IT tools for low-SES ethnic minority populations and suggests that such tools can improve the efficiency of DM-related dietary education. The RCT design adds also to the limited body of RCTs evaluating the IT-supported lifestyle and educational interventions in low-SES minority communities, which typically have disproportionately high T2DM rates and few well-designed studies targeting the key self-management domain of nutrition among T2DM patients.

Although the study was adequately powered to test the primary outcome of DM-related dietary knowledge, it had limited power to test the secondary outcomes. Nevertheless, we observed trends toward greater improvement in lifestyle behaviors in the I-ACE arm than the SLA arm. This is particularly impressive as, from the outset, we aimed to recruit a challenging group of patients who were overweight or obese and had poor glycemic control, suggesting that despite having free access to health care services and subsidized medications, their engagement with and benefit from existing health services was suboptimal. Our experience with the intervention indicated that adequate resources and maximum institutional flexibility and accessibility are needed to insure successful implementation, particularly in vulnerable, low-SES populations. Nevertheless, we observed reasonably good adherence to the study intervention, with a suggested trend of improved engagement among men in the I-ACE arm so that it equaled that of women; whereas in the SLA arm, the men’s engagement was lower than that of women. Given the small sample in our pilot RCT, these suggested trends should be confirmed in a larger efficacy trial.

We used the same dieticians to deliver dietary counseling in both study arms to eliminate the possibility of differences between study arms because of different dieticians. The use of the I-ACE counseling approach may have affected the way the dietician delivered standard care in the comparison arm. In addition, the study was conducted in community clinics serving local neighborhoods comprised extended families. As the randomization was conducted on an individual basis, neighbors and extended family members who were assigned to different study arms may have had shared what they learned from the experimental intervention approach and materials outside of the clinic setting. Both these possible sources of contamination, however, would have led to an underestimation of the I-ACE experimental effect and thus do not detract from the significance of our findings. We recommend that a definitive RCT should consider a cluster randomized design, particularly if conducted in neighborhood clinics with very strong, extensive social networks.

Finally, the intervention in this pilot study was of limited duration and the findings suggest that a longer/more intensive intervention is important to assure sustainability.

This study was conducted in a low-SES ethnic minority community at high risk of T2DM and its complications and showed potential for improving self-care, promoting healthy lifestyle behaviors, and improving surrogate health outcomes (HbA_1c). The findings may be informative for planning definitive RCTs to evaluate IT-based clinical interventions in other similar populations.

The use of the I-ACE software took more time than is currently allocated in clinical care in Israel, particularly for the initial dietary counseling session (35-45 min), although the median time for follow-up visits (17 min) did not differ substantially from the 15-min slots allotted in routine follow-up care. This time differential, which is largely the result of the structured task flow, the additional tools at the dietician’s disposal, and the associated documentation, needs to be taken into account when considering application to clinical care. One possible accommodation would be to model routine clinical care differently. For example, the national health insurance basket of services in Israel includes 14 visits per year to a dietician for patients with DM, which, in practice, are rarely fully utilized. An alternative model worth testing would be the integration of I-ACE in a series of fewer but longer dietary counseling sessions, with a comparison of its effectiveness with the usual care with multiple short visits. In addition, further technological development, such as making the assessment and other features of the I-ACE software available for self-use by patients via mobile or Web apps, could reduce the time required for documentation in the face-to-face consultations with the dietician.

This pilot trial supports the potential of the dietician-delivered, culturally adapted I-ACE dietary counseling intervention to increase the efficiency of DM-related lifestyle education and to improve lifestyle behaviors, compared with the usual dietary counseling in a minority population, although confirming that both types of counseling, coupled with active outreach, improved glycemic control. These findings provide support for conducting a large-scale trial to evaluate the effectiveness and applicability of the I-ACE software in routine clinical care among ethnically diverse populations.