Web-based interventions were included and defined as web-based if participants received information and directly interfaced with the internet, but they were not required to input data to a website [24]. In addition, studies were included if they were published in English; were RCTs or pilot RCTs; included an assessment of nutrition, diet, or dietary behavior; included adult participants (≥18 years) with diagnosed T2DM. Exclusion criteria were studies using non–web-based digital interventions; studies including participants with prediabetes or type 1 diabetes; studies including a combination of T2DM and participants with other types of diabetes or where outcomes for multiple chronic diseases were assessed; and studies focused on diabetes prevention.

A systematic literature search of four electronic databases, Medline, Embase, The Cochrane Library, and CINAHL for relevant papers published between January 2013 and May 2019, was conducted. Papers published before January 2013 were excluded as previous systematic reviews have reported on web-based, computer-based, and digital interventions in people with T2DM up to this date [22,25,26]. From these reviews, we extracted papers that met our inclusion criteria. Additionally, we conducted an in-depth exploration of reference lists for related papers and searched grey literature, including Google Scholar.

Keywords used in the search were (“Type 2 Diabet*” OR “diabetes mellitus, type 2” OR “T2DM” OR “T2D”) AND (“web-based” OR “internet” OR “online” OR “digital” OR “information technology” OR “IT” OR “computer-assisted” OR “computer-based” OR “computer interface” OR “ehealth”) AND (“diet*” OR “nutrition” OR “self-management” OR “lifestyle modification”).

Search results were merged using reference management software (Endnote 8) and duplicate papers removed. Screening of the titles and abstracts for individual studies were conducted in duplicate, independently, by two authors (JD, SI) with disagreements resolved by consensus. Articles deemed eligible for full-text review were assessed for eligibility by two authors, independently (JD, SI), and disagreements for inclusion were reached via consensus. Corresponding papers from the same study were merged to extract relevant outcomes. The following parameters were extracted from included studies: author/date, study design, sample size, total study period and length of follow up, population characteristics (including location, age, and comorbidities), behavioral change theory or model, digital intervention characteristics (including the type of digital intervention and duration of exposure), intervention providers, type of dietary intervention or guidelines administered, the dietary assessment used, dietary behavior change outcomes, and significant T2DM clinical outcomes including glycemic control as indicated by blood glucose levels or HbA_1c, and biomarkers weight, waist circumference, and CVD biomarkers. We also extracted intervention components, website usability rates, modes and methods of delivery, and process evaluation measures.

The risk of bias was assessed by two researchers independently using the Cochrane risk-of-bias tool for randomized trials (RoB 2) [27]. RoB 2 provides an in-depth framework structured into five domains to assess the risk of bias in RCTs. These five domains assess the risk of bias arising from the randomization process, deviations from intended interventions, missing outcome data, outcome measurements, and selection of reported results. During the assessment, each domain is given a rating of low, high, or unclear. Conflicting assessment ratings were resolved by consensus. Results were then calculated to reach a quality assessment rating of poor, fair, or good quality.

For qualitative analysis, differences in end intervention measures between groups and change between groups were reported, depending on the outcomes reported for individual studies. We also reported significant within-group changes. Data were considered statistically significant if the reported P value was <.05.

The included studies provided results from a total of 1056 participants representing three countries—France [28], Malaysia [29], and the United States [30-32]. The number of participants ranged from ≤26 in the pilot study [30] to ≤463 in the RCTs [32]. The mean age of participants ranged from 50.5 years [29] to 59 years [32]. Three studies had an intervention period of 4 months [28,30,32], one was 3 months [31], and one was 6 months [29]. Two studies included a 12-month follow up [29,32], one had a 10-month follow up [31], and another had a 32-week follow up [30]. The Malaysian Dietary Intervention for People with Type 2 Diabetes: An e-Approach (myDIDeA) [29] focused solely on dietary behavior by providing participants with a personalized intensive dietary intervention. The remaining studies assessed dietary behavior but included dietary recommendations alongside physical activity [28,30], mindfulness training [30], social support and self-management information [31,32], and face-to-face contact [32]. Three studies included health practitioner-assistance [29,31,32], ranging from contact with a nutritionist [29,32] through to availability of health care coaches and professionals [31,32], and physicians [32]. The Accompagnement Nutritionnel de l’Obésité et du Diabète par E­coaching (ANODE) study [28] was a fully automated intervention aside from providing technical assistance. Three studies [29,31,32] utilized behavioral theories or models, which varied widely. Only myDIDeA [29] applied user-centered design theory to support intervention development.

Three of the studies indicated the type of dietary recommendations prescribed [28-30], which were standard diabetes dietary guidelines modified to suit the Malaysian population [29], the National Nutrition and Health Program guidelines of France [28], and a very low carbohydrate ketogenic diet (VLCKD) [30]. The remaining two studies, computer-assisted self-management/computer-assisted self-management plus social support (CASM/+) [32] and the Diabetes Network (D-Net) [31], provided self-administered and/or personalized tailored dietary instruction from health professionals with general healthy eating recommendations and goal setting to reduce fast foods, fried foods, or sugar-sweetened beverages, and increase fruit and vegetable consumption. The tools and scales used for dietary assessment varied widely across all five studies (Table 2). myDIDeA [29] used a validated Malaysian 36-item Dietary Knowledge, Attitude, and Behavior (DKAB) score. ANODE [28] used the International Diet Quality Index (DQI-I) with food frequency questionnaires from both 24-hour and 3-day diet recalls. CASM/+ [32] used dietary questionnaires measured by two validated scales, the 20-item Kristal Fat and Fiber Behavior scale (FFB) and the 15-item Block/National Cancer Institute (NCI) Fat Screener. D-Net [31] also used the Block/NCI Fat Screener along with the 8-item “Starting the Conversation” scale. The VLCKD [30] used unvalidated self-reported dietary intake obtained from participants’ entries on a consumer-based website.

Four of the studies reported a statistically significant dietary behavior change in the intervention group (Table 2) [28-30,32]. Compared to a control group receiving usual care, ANODE [28] found significant improvements in dietary quality with participants choosing healthier foods overall and improvements in saturated fat and sodium intake. Similarly, compared to an enhanced care control group, CASM/+ [32] found that participants’ overall eating habits improved along with reductions in overall fat intake. Compared to a control group prescribed the ADA’s “create-your-plate” diet, participants following a VLCKD [30] demonstrated adherence with decreased consumption of carbohydrates within the prescribed range of 20-50 grams per day and decreased consumption of added sugar. Compared to a control group prescribed usual care, myDIDeA [29] showed that web-based interventions could be a feasible option for supporting dietary behavior change for people with T2DM in developing countries such as Malaysia, with participants in the intervention group achieving a 12.2-point difference in DKAB score. CASM/+ [32] demonstrated dietary behavior change in individuals with lower literacy and numeracy and diverse and higher-risk populations such as American Indian/Alaska Native, Asian, Black and African American, and Latino.

Changes in clinical outcomes were inconsistent and differed across studies (Table 3). Two studies reported statistically significant improvements in glycemic control for HbA_1c between groups compared to usual care [28] and the ADA’s “create-your-plate” diet [30]. One study found significant between-group differences in fasting blood glucose and HbA_1c [29] compared to usual care, benefits that were only observed in the intervention group at 12-months follow-up. However, clinical outcomes were not reflected across all four studies that reported significant changes in dietary behavior. CASM/+ [32] reported no significant clinical improvements. Reductions in weight or waist circumference were seen in two studies [28,30]. Twenty percent of the ANODE study’s [28] intervention participants achieved >5% weight loss, and 90% achieved at least 5% weight loss in the VLCKD intervention [30]. Weight reductions could be explained by improved overall diet quality and decreased fat intake [28], decreased sugar and carbohydrate consumption [30], greater calorie deficits in both intervention groups compared to control groups, along with recommendations for physical activity included in both interventions. Additionally, following a VLCKD yielded significant reductions in triglycerides [30].

An overview of intervention components, attrition, usability, and modes and methods used to deliver the interventions is shown in Multimedia Appendix 1. The attrition rate varied among studies, with the highest dropout rate seen in the CASM/+ interventions [32], losing 34.2% of intervention participants randomized to two complex intervention arms, compared to 19.5% of the control group. The lowest attrition rate was seen in myDIDeA, losing 4.8% of intervention participants and 10.6% of the control group, perhaps because the intervention focused solely on providing a structured dietary intervention and implemented a user-centered approach. Across all web-based interventions, the highest website usage was reported in the first month, followed by declined usage over time. Three studies reported login rates as a measure for usability [28,31,32]. Only one study [29] reported both login rates and time spent on site. Various modes and methods of delivering intervention components were reported, including providing content on a website, which was used in four of the five studies [28,29,31,32], while one sent content via email with text, videos, and links to various web resources [30]. Four studies provided some form of self-monitoring and feedback, whether automated or assisted by a health care professional [28,29,31,32]. Three of the five interventions [29-31] provided updated intervention materials to participants ranging from biweekly to bimonthly. No associations could be drawn between these methods and participant adherence and intervention outcomes due to heterogeneity in the methods of delivery and a lack of detailed reporting.

Postintervention process evaluation measured levels of adherence, usability, acceptability, and program satisfaction (Multimedia Appendix 2). The most common statistic used to measure adherence or intervention usage was website login rates [28,29,31,32], with studies consistently demonstrating higher login rates early in the intervention (Multimedia Appendix 1). Two studies included a more comprehensive process evaluation [28,29], both of which provided participants with self-reported feedback questionnaires. myDIDeA [29] usability rates were 72%, acceptability 62%, and program satisfaction 64%. ANODE’s [28] satisfaction rate was 70%. D-Net [31] indicated an implementation percentage of 100% related to their dietary assessment component but did not provide overall satisfaction rates.

Figure 2 summarizes the risk-of-bias assessment [27]. Four studies had a reasonable sequence generation described [28,30-32]. It was unclear in three studies if allocation concealment was adequate [29,30,32]. Only one study sufficiently blinded participants and personnel [29], one study was unclear [31], while the remainder did not blind [28,30,32]. However, blinding in dietary behavior studies is often not feasible. Three studies sufficiently blinded outcome assessors [28-30], while two remained unclear [31,32]. Only one study failed to provide complete outcome data [30]. Two studies reported full datasets [28,31], two were unclear [30,32], while one [29] had several outcomes included in study protocols [36] that were missing from the final study outcomes. Other biases were low [28,31,32] or unclear [29,30]. Based on these assessments, only one study [28] was assessed to be of fair quality, while the remaining four were assessed to be of poor quality.

Only five studies met the eligibility criteria, and these were heterogeneous in terms of dietary recommendations, focus on diet alone or with additional behavioral components, behavioral theories and models applied, and target population. The modes and methods of delivering the web-based interventions also differed, and the duration and follow up periods varied widely. While most studies report statistically significant improvements in dietary behavior change, the use of different dietary assessment measures made it difficult to compare study outcomes and generalize results. Web-based interventions could improve and expand reporting of website statistics to help inform patterns of participant behavior, and consideration of measuring adherence to diet as a means of determining if greater adherence leads to greater improvement in clinical outcomes is warranted. Web-based interventions could be a cost-effective way to provide education and support to individuals with T2DM, broadening access for a greater number of people, including those who have location or mobility constraints and cannot access face-to-face services. More research is needed to explore web-based dietary interventions for these diverse populations, including younger adults and the elderly. Web-based dietary interventions must be studied in larger cohorts, for longer durations, and with more clearly defined dietary recommendations. Studies must also explore intervention content, modes and methods of delivery, and whether these enhance participant engagement or contribute meaningfully to the expected outcomes.

This review provided evidence that web-based interventions may be an effective way to improve dietary behavior in people with T2DM. The results also suggest improvements in glycemic control and clinical outcomes may be possible, although the studies in this review yielded inconsistent results. While this preliminary evidence showed promise of a positive effect, the small number of studies and the fact they are highly heterogeneous makes it difficult to draw any firm conclusions. The field requires more well-designed web-based dietary interventions that report dietary prescription and adherence in people with T2DM to confirm their effectiveness in optimizing dietary behavior and improving clinical outcomes.