The systematic review and meta-analysis study protocol was developed prospectively and reported using Preferred Reporting for Systematic Review and Meta Analyses (PRISMA) guidelines [19]. The systematic review and meta-analysis procedures used were developed in consultation with the Cochrane Handbook for Systematic Reviews of Interventions [20]. The protocol was registered at the international prospective register of systematic reviews (PROSPERO) (ID: CRD42017057282) [21].
Search strategy
The search strategy, including databases used and search terms, was developed in consultation with a medical librarian. An initial search was developed for OVID MEDLINE using keywords, medical subject (MeSH) terms and publication types based on the PICO framework (participants, comparison, intervention, and outcomes). Participants were African Americans (“African Americans,” “African Americans”[MeSH] with type 2 diabetes (“type 2 diabetes,” “type 2 diabetes mellitus,” “diabetes,” “T2DM”(type 2 diabetes mellitus), “Diabetes Mellitus”[MeSH], “Diabetes Mellitus, Type 2”[MeSH], “NIDDM” (Non-insulin dependent diabetes mellitus), or “Non-insulin dependent diabetes mellitus.”) The intervention was DSME (“diabetes self-management education,” “self management education,” “DSME,” “education.” “health education,” “diabetes education,” “Patient Education as Topic”[MeSH], or “Self Care”[MeSH]); the comparator was a control group in a randomized-controlled trial or quasi-experimental study with matched controls (“randomized controlled trial.” “controlled clinical trial,” “randomly,” “randomized,” “trial,” “control,” “groups,” or “quasi-experimental”). Outcomes were HbA1c (“HbA1c,” “A1c,” “glycemic control,” “Hemoglobin A, Glycosylated”[MeSH], or “hemoglobin A1c protein, human”[MeSH]) and QOL (“HRQL,” “QoL,” “health-related quality of life,” “Quality of Life”[MeSH]), “QOL tools OR questionnaires OR surveys,” “SF-36,” “WHOQOL,” “DQOL.” “well-being,” “psychological well-being,” or “emotional well-being”). A sample OVID MEDLINE search strategy may be found in Additional file 1.
Databases searched were OVID MEDLINE, Ovid Eric, PsycINFO, Scopus, CINAHL EBSCO, and the Cochrane Central Register of Controlled Trials. To minimize the potential omission of relevant studies, the citation lists of included studies were reviewed to identify additional studies for potential inclusion. Additionally the tables of contents for selected journals (Diabetes Care, The Diabetes Educator, Annals of Internal Medicine, and Annals of Family Medicine) were hand-checked. The search strategy also included grey literature sources such as non peer-reviewed government and nonprofit publications (the Agency for Healthcare Research and Quality, the ADA, and the Centers for Disease Control and Prevention).
Inclusion and exclusion criteria
All citations were reviewed against pre-determined inclusion and exclusion criteria for eligibility in the systematic review. Included study designs were randomized-controlled trials or quasi-experimental studies with a matched control group comparing DSME to usual care. The inclusion of quasi-experimental study designs was consistent with the Cochrane Consumers and Communication Review Group standards for evaluation of complex interventions [22]. “Usual care” could consist of usual primary care, assignment to a wait-list, or a minimal educational intervention. The definition of DSME was based on the ADA and AADEs’ National Standards for Diabetes Self-Management Education and Support; e.g., a program to “facilitate the development of knowledge, skills, and abilities that are required for successful self-management of diabetes” [13]. Further, the intervention needed to support at least one of the AADE7 Self-Care Behaviors: healthy eating, being active, monitoring, taking medications, problem solving, healthy coping, and reducing risks [23].
Participants were African-American adults with type 2 diabetes mellitus; to be included, interventions either needed to have exclusively African-American participants, or to report the outcomes for African-American participants separately. All potential settings (clinics, hospitals, community settings, virtual/telehealth/phone, or combinations) were included. Studies selected for the systematic review were eligible for inclusion in the HbA1c meta-analysis if they measured HbA1c mean and standard deviation both pre- and post-intervention; similarly, studies included in the systematic review were eligible for inclusion in the QOL meta-analysis if they measured QOL mean and standard deviation both pre- and post-intervention.
Studies were excluded if: 1) the study population was not exclusively African-American or results for African-Americans are not reported separately; 2) the study had participants with type 1 diabetes, unless type 1 and type 2 diabetes results are reported separately; 3) the study control groups received anything other than usual care; 4) the intervention targeted providers or systems, rather than patients; 5) the intervention was a diabetes disease management or care management intervention, rather than DSME (for example, studies focusing exclusively on medical nutrition therapy or disease management); or 6) the study did not measure either HbA1c or QOL as an outcome. A study was defined as measuring QOL if it used one or more general or diabetes-specific QOL measures, which were pre-specified through a comprehensive literature search using keywords and phrases related to quality of life and frequently-used synonyms (diabetes and “quality of life” or “health-related quality of life” or “psychosocial adjustment” or “distress”). There were no study exclusions based on participant age or sex, article language, or publication date.
Study selection
Two independent reviewers (AC and DC) conducted the selection process through each phase of the review. All citations identified through the search were imported into a shared bibliography, and duplicate records of the same report were removed. The reviewers independently extracted information from the abstracts into structured evidence tables based on the pre-determined inclusion and exclusion criteria. Based on these criteria, they independently assessed the abstracts’ eligibility for full-text review. The two reviewers compared their results and reached consensus; a third reviewer (ML) served as a tiebreaker when needed. From this process, articles were selected for full-text review. The two reviewers independently read and assessed the full-text articles using the inclusion and exclusion criteria and met to compare results and reach consensus, with the third reviewer serving as a tiebreaker. Through this full-text review, the reviewers identified the final set of articles eligible for inclusion in the systematic review.
Data extraction
For the articles included in the systemic review, the two reviewers extracted further study data for inclusion in a structured evidence table. Descriptive categories included source citation, number of participants, mean participant age, percentage of participants who were African-American, and study design. Reviewers also recorded whether the intervention was group or individual-based, intervention content, presence of cultural tailoring (according to the studies’ authors), the intervention’s definition of usual care, duration, number of contact hours, provider type, DSME topics addressed, and attrition rate. The HbA1c and QOL measures used, HbA1c/QOL measurement frequency, and results were also recorded.
Bias and quality appraisal
Risk of bias was examined as an outcome across studies using the Cochrane Collaboration’s Risk of Bias tool, which assesses the presence of biases that pose threats to internal validity [24]. Types of bias examined in the Cochrane Risk of Bias tool included selection bias (random sequence generation and allocation concealment), performance bias (blinding participants and researchers to the intervention a participant receives), detection bias (blinding of outcome assessment from knowledge of what intervention a participant received), attrition bias, reporting bias, and other bias [24]. Studies were judged to have a low, high, or unclear risk of bias for each of these criteria. Quasi-experimental studies were automatically designated to have a high risk of bias on the random sequence generation item of the tool [22]. Two reviewers (AC and DC) independently assessed study bias and then met to compare results and reach consensus. Although assessment of publication bias was included in our protocol, due to the small number of studies in our HbA1c meta-analysis, publication bias could not be assessed. When fewer than ten studies are included in a meta-analysis, tests for forest plot asymmetry are not recommended due the low power to detect a real asymmetry [25].
The overall quality of included studies was assessed using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) criteria. In the GRADE system, evidence can be rated as high, moderate, low, or very low. Randomized controlled trials begin with a rating of high quality, and observational studies with a grade of low quality. Factors that can lower a quality rating include limitations in design and execution, heterogeneity (inconsistency of results), indirectness (research does not measure desired intervention or outcomes), imprecision (few patients or events), and publication bias. Factors that can increase a rating include a large magnitude of effect, a dose-response continuum, and plausible residual confounding in observational studies [26]. Two reviewers assessed study quality independently (AC and DC) and met to reach consensus.
Meta-analysis
All analyses were performed in Review Manager version 5.2 [27]. For studies containing both pre-and post-intervention HbA1c levels, these values were extracted as mean ± standard deviation. First, a meta-analysis was conducted to assess possible baseline HbA1c differences between intervention and control groups. Next, the mean HbA1c for both intervention and control groups at the conclusion of the intervention was transformed into a weighted mean difference (WMD), in which the contribution of each study to the mean difference is weighted by its sample size, and 95% confidence intervals (CIs) were calculated and combined in a random-effects meta-analysis. A random-effects meta-analysis is appropriate when combining studies with differences in the treatment effect [20]. A forest plot was also generated for the HbA1c WMD.
Study heterogeneity was explored using Cochran’s Q and I 2, with p < .05 for Cochran’s Q and I 2 ≥ 50% indicating substantial heterogeneity [28]. In addition, several subgroup analyses were conducted for HbA1c. First examined was the impact of culturally-adapted versus non culturally-adapted DSME based on the authors’ descriptions of their interventions. Additionally, subgroup analyses were conducted based on intervention contact hours (< 10 versus ≥10), given that 10 or more contact hours has been shown to lead to better DSME outcomes; DSME provider type(s) (e.g., individual (physician, nurse, dietician, pharmacist, health educator), or multiple provider types), individual, group, or combination individual/group DSME, and attrition rate. For QOL, studies with pre-and post-intervention QOL mean ± standard deviation were eligible for inclusion in a meta-analysis.