Effectiveness of Self-Management Training in Type 2 Diabetes: A systematic review of randomized controlled trials

The English-language medical literature published between January 1980 and December 1999 was searched using the MEDLINE database of the National Library of Medicine, the Educational Resources Information Center (ERIC) database, and the Nursing and Allied Health database (commenced in 1982). The medical subject headings (MeSH) searched were “Health Education” combined with “Diabetes Mellitus,” including all subheadings. Abstracts were not included because they generally contain insufficient information to assess the validity of the study by the criteria described below. Dissertations were also excluded because the available abstracts contained insufficient information for evaluation and the full text was frequently unavailable. Titles of articles extracted by the search were reviewed for their relevance to the effectiveness of diabetes education, and if potentially relevant, the full-text article was retrieved. Because automated databases are incomplete (13–15), the following journals, believed to have the highest relevance, were searched manually: Diabetes Care, Diabetes Educator, Diabetes Research and Clinical Practice, Diabetologia, and Diabetic Medicine.

Only randomized, controlled trial reports were selected because this type of study design generally supports maximum validity and causal inference (16). We reviewed only studies in which all or most subjects had type 2 diabetes. If the type of diabetes was unclear, then the study was included when the mean age was >30 years. It was believed that the educational techniques and social influences (especially family and peers) relevant to children and adolescents with either type 1 or type 2 diabetes were sufficiently different to warrant a separate review. To examine as broadly as possible the effectiveness of diabetes education, we included studies of subjects with type 2 diabetes >18 years of age, with any degree of disease severity and with any comorbidity. Interventions in all settings were included. Education could be delivered by any provider type, could involve any medium (written, oral, video, computer), could be individual- or group-based, and could be of any duration and intensity. Studies with multicomponent interventions were included only if the effects of the educational component could be examined separately.

Self-management training interventions were classified into one of the following categories by primary educational focus: knowledge or information; lifestyle behaviors, including diet and physical activity; skill development, including skills to improve glycemic control such as self-monitoring of blood glucose (SMBG), as well as skills to prevent and identify complications (e.g., foot care); and coping skills (to improve psychosocial function), including interventions using empowerment techniques or promoting relaxation or self-efficacy. Studies with a focus on knowledge or information were subclassified by primary type of educational approach: didactic or collaborative. Didactic education occurred when the patient attended to the information but did not interact with the instructor or participate actively in teaching sessions. Collaborative education occurred when the patient participated actively in the learning process, including group discussions or hands-on practice, or when teaching techniques included empowerment (17), individualized goal-setting, biofeedback, or modeling. The other three categories of lifestyle, skill development, and coping skills education were generally all collaborative to some extent; therefore, these types of interventions were not subclassified.

Data extracted from eligible studies included descriptive information, analysis methods, and results. Extraction was not blinded, because there is no evidence that blinding results in a decrease in bias in the conduct of systematic reviews and meta-analyses (18,19).

Quality assessment was determined by what was reported in each article, and internal validity was assessed using Cochrane methodology (20) for four types of bias (Table 1). These biases are believed to have significant effects on measured outcomes in intervention studies (21), and if present in an article, note was made in the tables.

These criteria for bias were modified from those used in Cochrane methodologies, because not one study in the literature reviewed fulfilled all definitions for the absence of bias. To avoid selection bias, ideally one requires concealment of the allocation schedule so that neither patient nor researcher can influence assignment sequence (22). However, because most studies in this review did not comment on method of allocation, beyond stating that subjects were randomized, allocation concealment was not used as a necessary criteria for the absence of selection bias. To avoid performance bias, blinding of patients to the intervention is required, which is impossible in diabetes education studies; therefore, patient blinding was not used as a validity criterion. Attrition was noted as a potential bias when more than 20% of initially enrolled subjects dropped out before data collection, and dropouts were not compared or were not found equivalent to completers at baseline.

External validity was also assessed and was considered adequate if the accessible population reasonably represented the target population and study subjects were either a random sample of the accessible population or consecutively referred patients, or if no significant differences between participants and nonparticipants were demonstrated at baseline. Studies with populations that consisted of volunteers, that were convenience samples, or were otherwise selected by the researchers may not be generalizable to target populations; therefore, the nature of these study populations is indicated in the tables.

Outcomes are summarized in a qualitative fashion to 1) aid in generating hypotheses, 2) detail the categorization of variables for future quantitative syntheses (23), and 3) portray the heterogeneity of the populations, interventions, methodology, study quality, and outcomes in this literature. It was believed that derivation of a single summary statistic would not be meaningful in determining what interventions are effective in what populations. The power of statistical tests of homogeneity is low, and failure to reject a hypothesis of homogeneity does not prove that studies are sufficiently similar to be aggregated (24).

We classified outcomes as 1) process measures including knowledge, attitudes, and self-care skills; 2) lifestyle behaviors, psychological outcomes, and quality of life; 3) glycemic control; 4) cardiovascular disease risk factors; and 5) economic measures and health service utilization. Because a study can have multiple outcomes, each study can be listed one or more times in the results tables, which are classified by outcome. Glycated hemoglobin measures are presented as percentage change in the text and the figure, due to the measurement of different glycated components of hemoglobin in different studies as well as the variability of measurement between laboratories and over time (25).

A total of 72 discrete studies, published in 84 articles, were identified. These studies are heterogeneous with respect to patient population, educational intervention, outcomes assessed, study quality, and generalizability (Tables 2-6,Table 3 —,Table 4 —,Table 5 —,Table 6 —). Review of this literature reveals a number of important generalizations concerning the components and determinants of effective interventions and the outcomes most conducive to improvement.

Most studies measuring changes in diabetes knowledge demonstrate improvement with education (Table 2) (26–46), including those with follow-up of 6–12 months after the last intervention contact (28–30,36,40,43). Seven studies demonstrated improved knowledge for both the intervention and control groups (47–53), suggesting possible contamination due to the infeasibility of blinding participants. A number of studies demonstrated that regular reinforcement or repetition of the intervention seemed to improve knowledge levels at variable lengths of follow-up: Bloomgarden et al. (34) (nine visits in 18 months), Korhonen et al. (35) (one visit every 3 months for 12 months), Campbell et al. (29) (regular reinforcement with visits and telephone calls over 12 months), and Rettig et al. (46) (12 visits in 12 months). Knowledge was measured using a variety of instruments, often specifically developed for the study and lacking in documented reliability and validity (26,30,32,33,35,39,44,47,52,54–56).

Several studies observed increased frequency of, or more accurate SMBG, demonstrated by a decreased discrepancy between measurement by the patient and health-care personnel (40,45,57–59) (Table 2). Several studies examined the relationship between skills teaching and glycemic control. Although three of these studies (40,57,60) noted an increase in frequency of SMBG, no corresponding improvement in HbA_1c was found. Wing et al. (61) taught adjustment of diet and physical activity in conjunction with SMBG, but the patients in this study failed to show improved glycemic control at 1 year.

Several studies examined interventions focusing on foot lesions with mixed results. Litzelman et al. (62) noted a decrease in serious foot lesions at 1 year after an intervention consisting of group education, with three follow-up visits, provider guidelines, and chart reminders. Other studies failed to demonstrate improvements with interventions (41,46,63). Malone et al. (64) found a significant decrease in foot ulcer and amputation rates, although this study had significant methodological inadequacies.

Most studies that examined dietary changes were positive for self-reported changes, including improvements in dietary carbohydrate or fat intake (38,39,65–70) (Table 3), a decrease in caloric intake (39,67), and an increase in consumption of lower glycemic-index foods (71). A few studies demonstrating improved dietary changes found corresponding improvements in weight (38,66,72) or glycemic control (31). Only two studies failed to show improvement in diet: one had an 18-month follow-up and an intervention delivered every 3 months (35), and the other (73) noted improved dietary habits during the intervention but no significant difference at 6 months.

Studies measuring physical activity outcomes had variable results. Hanefeld et al. (65) demonstrated an increase in activity at 5 years with a didactic intervention. Among studies with shorter follow-up duration, Wood 54 noted an increase in physical activity at 4 months, Glasgow et al. (74) found an increase in the number of minutes of activity 3 months after an intensive intervention, and Wierenga (75) found improved physical activity after five intervention sessions at 4 months. Five studies found no changes in physical activity compared with control groups (30,40,69,76,77). It is unclear what factors might account for success in some studies and not in others.

Four studies examined psychological outcomes (Table 3) (33,40,74,78); improvements were noted in problem solving (74) and anxiety levels (33). Quality of life was examined in three studies. Kaplan et al. (79) noted an increase in quality of life at 18 months for an intervention subgroup that received intensive counseling on both diet and physical activity. Two studies of brief interventions failed to demonstrate improved quality of life (60,67).

Studies that focused on glycemic control are described in Table 4 and Fig. 1. Both control and intervention study groups tended to have improved glycated hemoglobin measures (29,31,32,36,48,49,60,66,68,74,78,80–83) (Fig. 1). All studies were unblinded. In 14 studies, an improvement was noted in glycemic control in the intervention group compared with the control group (26,28,32,33,47,48,50,65,71,76,79,84–87). Percentage change in glycated hemoglobin ranged from –26 to +4% in the intervention groups and from –33 to +15% in the control groups. In three studies, glycated hemoglobin decreased more in the control group (61,80,83), although the difference was significant in only one study (80).

Length of follow-up after completion of an intervention seemed to have a major effect on outcomes, and studies with a follow-up period of ≤6 months tended to demonstrate greater effectiveness (31–33,48,50,71,76,84). Few studies had follow-up periods longer than 1 year after the last intervention contact, and these showed mixed effects on glycemic control. The positive studies were either very intensive interventions (79) or had a high attrition rate, leaving a very select group at follow-up (28). Studies with prolonged interventions (follow-up periods >1 year and regular contacts with the intervention subjects during that time) also had mixed results. Two studies (47,65) demonstrated improved glycemic control, although generalizability of these studies is difficult due to a low participation rate (65) and a lack of information on study participation (47). Ten others produced no significant effects, despite regular patient contact (29,34,35,67,69,82,86,88–90).

For knowledge and information interventions, the method of delivery seemed to have a relationship to glycemic control. Compared with didactic interventions, collaborative interventions produced somewhat more favorable results, particularly if interventions were repetitive and ongoing (26,28,48,50,76,84,86).

Most studies focusing on changes in lifestyle generally failed to show improvements in glycemic control compared with control groups (36,39,43,49,66,67,70,72–74,77,78,81–83,88,90–95), but a few studies (31,71,79,84) showed improved glycemic control in researcher-selected or volunteer populations with follow-up <6 months. Improved glycemic control was associated with weight loss in some studies (28,47,48,76,79) and not others (31,65,71,84). Increased physical activity levels were associated with improved glycemic control in one study (65), although another study noted no changes in physical activity despite improvements in glycemic control (76).

Improved glycemic control and increased knowledge were not consistently correlated. Although a number of studies demonstrated an increase in knowledge with an improvement in glycemic control (26–28,31–33,50), others demonstrated improved metabolic control with no change in knowledge (47,76), and eight studies demonstrated increased knowledge but no significant improvement in glycemic control (29,34–36,40,49,80,88). Two of three studies focusing on coping-skills training produced improvements in glycemic control (85,86); these involved frequent group support meetings.

Computers have been used recently as an educational tool in a number of studies, and effects on glycemic control have been mixed: positive results in three studies (32,39,50) and negative results in another study (67,68). Additionally, videotapes have been used as adjuncts for teaching, with positive (31) and negative (91) results.

A large number of studies examined the effects of diabetes self-management training on risk factors for cardiovascular disease, including body weight, serum lipid levels, and blood pressure (Table 5). Thirteen studies demonstrated positive effects on weight loss; the average weight loss for these studies was ∼2 kg (range 1.3–3.1) (28,36,38,47,66,72,74,76,80,82,84,89). Most studies with positive results involved regular contacts or reinforcement sessions (38,47,66,76,82,84) or very short follow-up periods (72,74), although four studies had follow-up periods of ≥5 months (36,38,80,82). All other studies with follow-up of ≥6 months after the end of the intervention failed to show significant differences in weight loss between control and intervention groups (30,31,61,65,71,73,77,79,84,87,88,90,91). A number of other studies with shorter follow-up periods also had negative results (29,34,39,59,75,78,82,92,96–99). Only three studies involved didactic interventions (34,47,65), and only one of these studies showed a decrease in weight (47).

A large number of studies examined the effects of self-management training on lipid levels, and some produced improvement in total cholesterol (range −0.9 to −0.07 mmol/dl) (66,68,81,83,93), LDL (−0.4 mmol/dl) (100), and HDL (+0.1 mmol/dl) (100). Others found initial positive results but no significant difference from baseline at final follow-up (69,82,101). Positive studies involved interactive, generally individualized, repetitive interventions. Some studies have shown no beneficial effects on lipids (29,34,47,65,76,88,91,92). Of the three didactic studies (34,47,65), none resulted in improved lipid profiles.

Studies examining blood pressure control also revealed mixed results. Some studies demonstrated a decrease in systolic blood pressure (−4 mmHg) (28) and diastolic blood pressure (−3 to –8 mmHg) (27–29,76), whereas others showed no significant changes (34,73,82,89).

Only two studies examined cardiovascular disease events or mortality, one of which found no significant difference in cardiovascular disease or mortality events after 5 years of visits every 3 months (65); the other study found no significant difference in mortality 13 months after a 1-h group didactic educational session (64).

Most studies examining economic outcomes and health-care utilization (Table 6) failed to demonstrate improvements in measured parameters (34,46,60), except the study by Wood (54), which demonstrated a decrease in emergency room visits 4 months after a short-duration intervention. Glasgow et al. (68) calculated that the cost of a social cognitive theory–based lifestyle intervention, effective in decreasing cholesterol and in improving food habits, was $137 per patient. Franz et al. (102) found the per-patient cost-per-unit change in glycohemoglobin to be lower for control subjects than for intervention patients. They also demonstrated (102) a cost-effectiveness ratio (direct costs only) of $56.26 per percent change in HgA_1c for results achieved at 6-month follow-up. No cost-benefit analyses of diabetes education were identified.

A large number of randomized controlled trials of the effectiveness of self-management training in individuals with type 2 diabetes have been performed. Despite limitations in methodology and heterogeneous population characteristics, settings, interventions, outcomes, and lengths of follow-up, a number of generalizations can be made from these studies (Table 7).

In reviewing the literature, it is clear that diabetes self-management training has evolved from the primarily didactic interventions of the 1970s and 1980s into the collaborative, more theoretically based “empowerment” models of the 1990s (12). Didactic interventions focusing on the acquisition of knowledge and information demonstrate positive effects on knowledge but mixed results on glycemic control and blood pressure and no effect on weight. Collaborative interventions focusing on knowledge tend to demonstrate positive effects on glycemic control in the short term and mixed results with follow-up >1 year. Effects of collaborative interventions on lipids, weight, and blood pressure were mixed.

It is apparent that factors other than knowledge are needed to achieve long-term behavioral change and that this may account for the lack of a consistent positive relationship between knowledge and glycemic control. It has been suggested that 1) although intensive treatment can improve metabolic control, the role of patient education in that process is uncertain (34); 2) changes in attitude and motivation are needed to achieve metabolic control (35); 3) integrating education with other therapies, such as intensified insulin treatments, is important in improving glycemic control (60); 4) a minimum threshold of diabetes knowledge is required; and 5) improved personal attitudes and motivations are more effective than knowledge in improving metabolic control (110). Many have also noted the lack of a relationship between SMBG and glycemic control for subjects with type 2 diabetes (111–116), although several randomized controlled trials have shown a relationship in type 1 diabetes (117,118).

The literature is divided regarding the relative merits of group versus individual therapy, and in our review, both types of delivery demonstrated mixed results for interventions that focused on knowledge, lifestyle, or skills. Lifestyle interventions were generally more effective in group settings, with positive outcomes noted for weight loss (8,36,47,48,72,74,76,77,94) and glycemic control (31,36,71,76,79), although two studies of lifestyle interventions in individual settings had positive effects on weight (38,80). Both individual (38,39,66–68) and group (72,75,93) lifestyle interventions had positive effects on diet and self-care behaviors. It is notable that skills teaching was effective in both group (41,62) and individual settings (45,58).

Others have drawn conclusions similar to ours about effective interventions in diabetes self-management training. Brown’s meta-analyses (9,10) support the effectiveness of diabetes education, with positive effect sizes (from largest to smallest) for the outcomes of knowledge, dietary compliance, skill performance, metabolic control, psychological outcomes, and weight loss. Padgett et al. (11) reviewed the effectiveness of diabetes education in 1988 and found diet instruction and approaches based on social learning theory to be the most effective interventions; physical outcomes and knowledge were most improved. A qualitative review of diabetes self-management education concluded that behavior change strategies were much more effective than didactic methods and that patient education was most effective when combined with health-care provider medication adjustment and reinforcement of educational messages (5). Anderson (119) noted that effective diabetes-management programs must be noncomplex, individualized to a person’s lifestyle, and reinforced over time, and they must respect an individual’s habits and routines and incorporate social support. Similar generalizations are found in reviews of chronic disease care. Von Korff et al. (120) concluded that effective programs in chronic disease care include collaborative problem definition; targeting, goal setting, and planning; a continuum of self-management training and support services; and active and sustained follow-up. Wagner et al. (121) stated that chronic illness programs require psychoeducational programming, and they emphasized the importance of responding to patients’ individual needs, readiness to change, and self-efficacy. Mullen et al. (122) noted that the most beneficial components of educational interventions in chronic diseases were individualization, relevance, feedback, reinforcement, and facilitation.

There are important limitations in execution of many of these studies. Internal validity was frequently threatened by 1) lack of blinding of the assessor, 2) infeasibility of blinding study subjects, 3) high attrition, 4) contamination of the control group, 5) unintended cointerventions, 6) lack of detail on allocation concealment (20), 7) response-set bias whereby intervention group participants report dietary and other habits that match the goals of the intervention rather than actual behavior (123), and 8) deficits in the reliability and validity of the instruments used to measure knowledge, self-care, and dietary habits. Brown (124) has previously noted that the measurement of knowledge is seriously flawed. More recent studies have demonstrated little improvement. In addition, most studies compare a more intensive intervention to basic care and education, as it is generally considered unethical to randomize a group to receive no education, thus minimizing measured effects of the intervention.

There was frequently an inadequate description of study interventions and participants, including the representativeness of study populations. Generalizability was also frequently limited by the volunteer nature of the study populations. Glasgow and Osteen (125) noted similar deficiencies in information on the representativeness of study populations in diabetes self-management training studies, as well as in the reporting of patient characteristics.

The behavioral theories on which interventions were based are documented in a few studies (29,40,60,67,68,79,93,96), as were the behavioral tools (27,30,46,48–50,72,73,75,76–78,91,92,94). However, data are insufficient to determine which behavioral tools and theories are most advantageous.

Although only randomized, controlled trials were reviewed, there is an important body of literature with other study designs. It is more difficult to draw conclusions about causality from nonexperimental designs than from an experimental design (16). Nonetheless, nonexperimental designs, if methodologically sound, reveal important information about the effectiveness of interventions (126). Randomized, controlled trials in this area of research are not always feasible, or even desirable, particularly when examining community educational interventions. Glasgow et al. (127) note the increasing importance of recognizing the complexity of disease determinants and multilevel system interventions. Classic randomized, controlled trials emphasize efficacy, to the exclusion of factors influencing effectiveness, such as adoption, reach, and institutionalization (127).

This review supports concerns expressed by others that researchers may not be measuring the most important outcomes (125,127). Glasgow and Osteen (125) reviewed Brown’s 1990 meta-analysis (10) and concluded that “Program evaluations to date have focused too narrowly on assessing knowledge and GHb outcomes to the exclusion of other important variables.” They stated that process and mediating variables (such as self-efficacy, problem-solving, and coping skills) and quality-of-life outcomes must receive much more attention in intervention research. Unfortunately, our review suggests that little has changed in the past 10 years, as researchers have continued to focus on knowledge and glycemic control to the exclusion of outcomes reflecting a more holistic view of patient function, longevity, and quality of life.

There are clearly many gaps in the literature on effectiveness of diabetes self-management training in type 2 diabetes (Table 7). More work must be done to identify the predictors and correlates of glycemic control, because knowledge levels and SMBG do not correlate well with blood glucose. Behavioral theory must have a more explicit role in future studies to improve the understanding of behavior change in the self-management of chronic illness. The role of electronic media in diabetes self-management training, the role of nontraditional health-care providers, and the optimal training of health educators has yet to be determined. The role of individual needs assessment within the context of group teaching has not been clarified. Quality-of-life outcomes must be brought to the forefront of future research.

The objectives for ideal self-management interventions in diabetes are clear: behavioral interventions must be practical and feasible in a variety of settings; a large percentage of the relevant population must be willing to participate; the intervention must be effective for long-term important physiological outcomes, behavioral end points, and quality of life; patients must be satisfied; and the intervention must be relatively low-cost and cost-effective (68). How best to achieve these objectives is not entirely clear. There are some well-designed and -executed studies that support the effectiveness of self-management training for patients with type 2 diabetes, particularly in the short term. The challenge is to expand upon this current knowledge to achieve all of the objectives of ideal self-management. Further research of high methodological quality in diverse study populations and settings and using generalizable interventions is needed to assess the effectiveness of self-management interventions on sustained glycemic control, cardiovascular disease risk factors, and ultimately, microvascular and cardiovascular disease and quality of life.

We thank Frank Vinicor for his thoughtful comments on the manuscript and Kristi Riccio for technical support.