Effectiveness of Multilevel and Multidomain Interventions to Improve Glycemic Control in U.S. Racial and Ethnic Minority Populations: A Systematic Review and Meta-analysis

Racial and ethnic disparities in type 2 diabetes outcomes are a critical health care issue (1). For example, Black people with diabetes have a twofold-increased risk of amputation and end-stage renal disease compared with White people with diabetes (2). Similarly, both Black and Hispanic people with diabetes are at an increased risk of developing diabetic retinopathy compared with White people (3). Despite decades of efforts to improve these disparities, there has been a striking lack of progress (4). This is due in part to social and structural determinants, such as food insecurity, poverty, housing instability, and racial discrimination, that contribute to racial and ethnic health disparities (5–8).

Diabetes self-management education (DSME) programs are a prominent type of diabetes intervention in which participants work with instructors on goals related to medication adherence, physical activity, nutrition, and other topics. They receive ongoing support as they incorporate skills and strategies learned during the program into their regular self-management (9). Most DSME programs focus on individual behaviors and self-care, and these programs have demonstrated effectiveness in improving clinical outcomes, such as hemoglobin A_1c (HbA_1c) and fasting blood glucose levels (10). DSME programs have been found to be modestly effective in reducing HbA_1c in racial and ethnic minority populations (11). The effectiveness of DSME programs may be limited by a failure to address the complex systemic causes of diabetes health disparities. There is a strong theoretical foundation to support this proposition: the biopsychosocial model posits that health is an interconnection between biological, psychological, and socioenvironmental factors, and the social-ecological model describes individuals being nested within relationships, communities, and society (12,13). Therefore, more recently, interventions to improve outcomes among racial and ethnic minority populations have focused on interventions that target a more comprehensive set of barriers (multilevel or multidomain interventions) (14–18). These studies not only incorporate DSME in their interventions but also include strategies that affect a broader sphere of influence (i.e., multiple levels, namely, interpersonal, community, and society levels) and more domains of a person’s life (i.e., sociocultural issues, health care system, and physical environment). Some examples of these interventions include DSME interventions that leverage peer support (interpersonal level) or provide culturally tailored recommendations (sociocultural domain) (19–21).

Evidence has been promising for multilevel interventions for people with diabetes in the general population. For example, a meta-analysis demonstrated that multilevel DSME using peer support was effective in reducing HbA_1c (22). However, there is not a systematic review or meta-analysis providing evidence for multilevel or multidomain interventions among racial and ethnic minoritized populations. Thus, we conducted a systematic review and meta-analysis of DSME interventions within racial and ethnic minority populations to understand the effectiveness of multilevel and multidomain interventions compared with single-level and single-domain interventions. We predicted that multilevel and multidomain DSME interventions would have greater effectiveness in improving glucose control compared with single-level and single-domain DSME interventions because of their greater scope of influence and individual tailoring.

We conducted a systematic review and meta-analysis of randomized controlled trials of diabetes interventions that were focused on racial and ethnic minority groups to summarize their effects on HbA_1c. This review was registered with the International Prospective Register of Systematic Reviews (PROSPERO) (CRD42019122625) database and has been conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Supplementary Table 1).

This systematic review on DSME interventions selected articles from a general systematic review on type 2 diabetes interventions conducted in the U.S. among adults (18 years or older) from racial and ethnic minority groups (defined as >50% racial or ethnic minority) or interventions that reported results by race and ethnicity. We performed an electronic search of research databases (PubMed, Scopus, CINAHL, and PsycINFO) from 1985 through 2019 with the assistance of a biomedical librarian. If articles were unclear on the type of diabetes, we used these assumptions: 1) if articles included participants with both type 1 diabetes and type 2 diabetes, we included articles if at least 90% of participants had type 2 diabetes, and 2) if articles did not specify the type of diabetes, we assumed that the population had type 2 diabetes, except if recruitment was conducted in settings with large proportions of adults with type 1 diabetes (e.g., diabetes camps or endocrinology clinics). We required articles to have a minimum follow-up of 3 months. Because of the broadness of the search, the search terms included concepts of “diabetes,” “study design,” “language,” “race/ethnicity,” and “disparities” (Supplementary Table 2); inclusion criteria according to the Population, Intervention, Comparison, Outcomes, and Study design (PICOS) framework are outlined in Supplementary Table 3. Once we completed the search, we eliminated duplicates, reviewed the title and abstract of each article for inclusion criteria, and reviewed the full article if necessary. Two independent researchers reviewed each article. Similarly, two independent reviewers collected data (study design, intervention and participant characteristics, control group, and clinical outcomes) from each article, and discrepancies were checked by a third reviewer. From this large systematic review, we selected articles that examined DSME interventions and measured postintervention HbA_1c (Fig. 1). We defined DSME studies as those that involved educating participants and/or directly providing them with skills to help them manage their diabetes in an ongoing process. This often included education on topics such as physical activity, diet and nutrition, or glucose self-monitoring.

A team of six reviewers, including three of the authors (E.M.F., A.N.D., and N.L.) and three trained research assistants, identified intervention levels and domains. Each intervention was independently classified by two team members. Discrepancies were reviewed by a third team member and discussed to consensus. We used the framework from the National Institute on Minority Health and Health Disparities (NIMHD) to classify interventions by level (individual, interpersonal, community, and society) and domain (biological, behavioral, physical or built environment, sociocultural environment, and health care system) of influence (23). As an example, a DSME intervention that focused on increasing individual self-management behaviors (e.g., medication adherence) would be classified as an individual-behavioral intervention. If that intervention was culturally tailored, it would also be inclusive of the sociocultural domain. If the intervention was expanded to include peer support, then it would also be classified as inclusive of the interpersonal level. Health care system interventions included changes to care delivery, for example, integrating medication management support by pharmacists alongside the primary care team. Community-level intervention provided a new resource in the community, like integration with food banks. We identified levels and domains and then categorized interventions as single-level and single-domain, multilevel, multidomain, or multilevel and multidomain DSME intervention.

We used the Cochrane Risk of Bias for randomized trials tool to determine the level of risk of bias for each intervention (24). Additionally, we used funnel plots to assess publication bias. We used the GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) framework for assessing the quality of evidence overall (25).

We examined the effect of DSME interventions on HbA_1c overall and the effect on HbA_1c for each of the following categories of DSME interventions: 1) single level and single domain, 2) multilevel, 3) multidomain, and 4) multilevel and multidomain. We also conducted post hoc analyses to assess the effectiveness of DSME interventions that address specific levels, domains, and common level and/or domain combinations. Each of these meta-analyses compared DSME interventions versus control treatment (i.e., usual care, waitlist, or attention control). For all meta-analyses, we used the random-effects model. More specifically, first, we calculated a weighted average of the estimated effects within individual studies (a pooled effect estimate). Second, we determined weights using the inverse of the within-study variance (sampling variance) and the between-study variance for each study (26). Lastly, we constructed a z test and its associated 95% CI of pooled intervention effects (26). Heterogeneity was assessed with the I² statistic.

There were 106 unique DSME trials among racial and ethnic minority populations with an HbA_1c outcome for our meta-analysis; in total there were 109 analytic comparisons, because three trials reported results separately by racial and ethnic subpopulations. Supplementary Table 4 provides details on the included trials. Table 1 provides an overview of the combinations of levels and domains addressed by the interventions in these trials as well as an example of each combination. All interventions included at least the individual level and the behavioral domain; 27 (25%) focused only on this single level and single domain. The remaining interventions included multiple levels and/or domains. Common combinations were the following: individual-behavioral, sociocultural (19%, n = 20); individual, interpersonal-behavioral, sociocultural (21%, n = 22); and individual, interpersonal-behavioral, sociocultural, health care system (12%, n = 13). There were no interventions at the society level, only three at the community level, and only one within the physical or built environment domain. Overall, 70 (66%) interventions were multidomain and 54 (51%) interventions were multilevel; 45 (42%) were both multilevel and multidomain.

Across trials, the average age of participants ranged from 45 to 73 years, and most participants were female (Supplementary Table 5). Most trials enrolled >50% of participants from one race or ethnicity: American Indian/Native American (2 trials), Asian/Pacific Islander (8 trials), Black/African American (41 trials), and Hispanic/Latinx (43 trials) participants. Enrollment rates were similar across trials defined by level and domain (single level, mean 45%; multilevel, 41%; multidomain, 43%; multilevel and multidomain, 41%).

Altogether, compared with control treatment (i.e., usual care, waitlist, or attention control), DSME interventions decreased HbA_1c by a percentage point change of −0.36 (95% CI −0.44, −0.29; I² = 63%; number of analytic comparisons = 109, total number of participants or patients [total n] = 20,132) (−3.9 [95% CI −4.8, −3.2] mmol/mol). The individual-behavioral trials decreased HbA_1c by a percentage point change of −0.34 (95% CI −0.46, −0.22; I² = 33%; number of analytic comparisons = 27; total n = 5,851) (−3.7 [95% CI −5.0, −2.4] mmol/mol) (Fig. 2 and Table 2). Interventions that were both multilevel and multidomain decreased HbA_1c by a percentage point change of −0.43 (95% CI −0.55, −0.31; I² = 69%; number of analytic comparisons = 47; total n = 7,150) (−4.7 [95% CI −6.0, −3.4] mmol/mol). Interventions that were either multilevel or multidomain decreased HbA_1c by similar percentage point changes: −0.40 (95% CI −0.51, −0.29; I² = 68%; number of analytic comparisons = 56, total n = 8,619) (−4.4 [95% CI −5.6, −3.2] mmol/mol) and −0.39 (95% CI −0.49, −0.29; I² = 68%; number of analytic comparisons = 73; total n = 10,748) (−4.3 [95% CI −5.4, −3.2] mmol/mol), respectively.

Subgroup analyses by level-domain combinations showed that multilevel and multidomain trials that included the interpersonal level and the sociocultural domain (in addition to individual-behavioral) had the largest percentage point decrease in HbA_1c: −0.46 (95% CI −0.60, −0.33; I² = 64%; number of analytic comparisons = 38; total n = 5,257) (−5.0 [−6.6, −3.6] mmol/mol). Multilevel and multidomain trials that included the sociocultural and health care system domains decreased HbA_1c by a percentage point change of −0.41 (95% CI −0.62, −0.20; I² = 68%; number of analytic comparisons = 12; total n = 2,892) (−4.5 [−6.8, −2.2] mmol/mol). Similarly, multilevel and multidomain trials that included the interpersonal level and health care system domain had a percentage point decrease in HbA_1c of −0.40 (95% CI −0.57, −0.24; I² = 68%; number of analytic comparisons = 24; total n = 4,362) (−4.4 [−6.2, −2.6] mmol/mol). Among specific levels and domains included in multilevel and multidomain trials, inclusion of the sociocultural domain had the largest percentage point decrease in HbA_1c (−0.43 [95% CI −0.54, −0.33], I² = 63%, number of analytic comparisons = 59, total n = 8,009) (−4.7 [−5.9, −3.6] mmol/mol), followed by the health care system domain, which decreased HbA_1c by −0.35 (95% CI −0.51, −0.18; I² = 71%; number of analytic comparisons = 29; total n = 5,208) (−3.8 [−5.6, −2.0] mmol/mol). However, overlapping CI indicate differences may not be significant. The three multilevel and multidomain trials that included the community level had a nonsignificant decrease in HbA_1c of −0.07 (95% CI −0.58, 0.43; I² = 33%; total n = 559) (−0.8 [−6.3, 4.7] mmol/mol).

In the quality review, 76 trials had low concerns of bias, 24 trials had some concerns of bias, and 6 trials had high concerns of bias (Supplementary Table 6). There were no concerns regarding randomization, outcome measurement, or selection of the reported result. However, regarding deviations from the intended intervention, 12 trials had some concerns and 1 trial had high concern. There were some concerns for missing outcome data for 17 trials and high concerns for 5 trials. Overall, the quality of evidence was moderate to high across all trials (Supplementary Table 7). Funnel plots were generally symmetric (Supplementary Fig. 1).

Given the persistence of racial and ethnic disparities in type 2 diabetes outcomes as well as complex contributing factors, interventions that target more than just the individual level are a potential solution. Our meta-analysis found that most DSME randomized controlled trials in the U.S. for racial and ethnic minority populations have included Black/African American and Hispanic/Latinx populations and have been multilevel and/or multidomain. All interventions addressed the individual level and behavioral domain, and most trials combined interpersonal, behavioral, and sociocultural or health care system domains. However, few trials aimed at changing the higher community level of influence, and no trials aimed at changing the society level (e.g., social norms, societal structure, and policies). Within these DSME trials, interventions that were at the individual and interpersonal levels and behavioral and sociocultural domains of influence had the largest decrease in HbA_1c, and sociocultural interventions were the most promising. While we did not directly compare intervention types, DSME interventions that were multilevel and multidomain had effects on glycemic control similar to those of individual-behavioral interventions.

This finding is not consistent with expectations from theory. The biopsychosocial model’s framework suggests that DSME interventions including a sociocultural environmental domain would have an added benefit in improving diabetes outcomes beyond that of individual behavioral interventions. Instead, the intervention types all affected HbA_1c similarly in our meta-analyses. We hypothesize that the major reason is that trials to date have focused on lower levels of the sociocultural environment, focusing on relationships rather than communities and society. Other studies have also noted the lack of larger-scale diabetes disparity interventions. For example, one literature review of multilevel diabetes interventions for U.S.- and Canada-based native peoples found only three interventions that addressed environmental or policy-level factors (27). Furthermore, the articles in our study did not address factors such as discrimination and structural racism, which are an important and growing area of study in health disparities (28,29).

Extant literature suggests that the higher spheres of influence may be key to greater progress. For example, a systematic review of obesity interventions among people from low socioeconomic backgrounds demonstrated that individual behavioral change interventions had no effect, but community-level interventions (e.g., school based) and those that focused on polices to change the built environment were effective (30). Similarly, a multilevel, multidomain intervention throughout the state of Delaware to address Black-White colon cancer disparities eliminated disparities within a few years by focusing on policy issues related to access to screening and treatment, which involved a community-wide coalition and health care systems (31).

Thus, to eliminate health care disparities in diabetes, it is likely necessary to engage higher domains of influence than current clinical trials do. Indeed, quasi-experimental studies have demonstrated that public policies like soda taxes or produce prescriptions may lead to major improvements in diabetes outcomes, but randomized trials of policy-level interventions can be very challenging to design (32,33). Similar recommendations were made by the National Clinical Care Commission to the U.S. Congress to focus on social factors and environmental exposures (e.g., food, schools, housing, green spaces, neighborhoods, and clean water), public health (e.g., breastfeeding, maternity leave, and food labels), health care (e.g., DSME and support and virtual care), and policy (e.g., coverage of HbA_1c for screening and National Diabetes Prevention Program insurance coverage) (34).

This study has several limitations. Classification of the interventions into levels and domains of influence was challenging and necessarily subjective. Our study also only assessed the effect of interventions on HbA_1c, a surrogate diabetes marker in individuals. It is possible that there are advantages of multilevel or multidomain interventions in influencing other outcomes both at the individual and population levels that we did not capture in our design. Examples could include the impact of interventions on factors such as quality of life or strength of social networks. We also restricted the meta-analysis to randomized controlled trials, which may have resulted in the exclusion of highly effective diabetes self-management interventions evaluated by other study designs. It is also possible that DSME interventions of all kinds have a ceiling effect, and further innovations that anchor on the individual level and behavioral domain of influence may have small additional benefits to HbA_1c. Future work could examine studies published after our review period, which ended in 2019.

Improving HbA_1c with DSME alone has been a challenge among racial and ethnic minoritized populations, because diabetes control is affected by multiple negative social drivers of health (e.g., disproportionate poverty and unemployment, structural racism, and disparities in neighborhood-level health-promoting and health-harming factors). Expanding DSME to address some of these factors (e.g., peer coaches and connections to community-based resources) may modestly increase diabetes control. Analyses by domain show that sociocultural DSME interventions had the greatest effect size, which is promising but not conclusive. Contrary to expectations, results did not show multilevel, multidomain interventions significantly outperforming individual behavioral interventions. More work to investigate the impact of addressing the larger drivers of health will be needed to assess the impact of diabetes (prevention and) control on socially marginalized populations such as racial and ethnic minority groups.

Acknowledgments. The authors thank Alexander Rodriguez, Raj Shetty, and Jielu Yu (University of Chicago) for their contributions to article review and data preparation.

N.L. is an editor of Diabetes Care but was not involved in any of the decisions regarding review of the manuscript or its acceptance.

Funding. The primary funding for this work was National Institutes of Health grant R01MD013420. This study also was supported by other grants from the National Institutes of Health (2P30DK092949, P30 DK092924, P50MD017349, and K24AG069080). This research was also supported by University of Chicago Medicine’s Center for Healthcare Delivery Science and Innovation (HDSI).

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

Author Contributions. E.M.F. researched data, contributed to discussion, and wrote the first draft of the manuscript. E.M.S., A.D., W.W., H.H.M., A.J.K, and N.L. researched data, contributed to discussion, and reviewed and edited the manuscript. S.U. and N.T. researched data and reviewed and edited the manuscript. E.S.H. and M.E.P. contributed to discussion and reviewed and edited the manuscript. All authors approved the final version of the manuscript. N.L. 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. Portions of this work were previously presented at the AcademyHealth 2022 Annual Research Meeting, Washington, DC, 4–7 June 2022, American Diabetes Association 82nd Scientific Sessions, New Orleans, LA, 3–6 June 2022, and the 2023 Midwest Society General of Internal Medicine conference, Chicago, IL, 19–20 October 2023.

Handling Editors. The journal editor responsible for overseeing the review of the manuscript was M. Sue Kirkman.