This meta-analysis assessed how successfully Diabetes Self-Management Education (DSME) interventions help people with type 2 diabetes achieve and maintain healthy blood glucose levels. We included 52 DSME programs with 9,631 participants that reported post-intervention A1c levels in randomized controlled trials. The training conditions resulted in significant reductions in A1c levels compared to control conditions. However, the impact of intervention was modest shifting of only 7.23% more participants from diabetic to pre-diabetic or normal status, relative to the control condition. Most intervention participants did not achieve healthy A1c levels. Further, few DSME studies assessed long-term maintenance of A1c gains. Past trends suggest that gains are difficult to sustain over time. Our results suggested that interventions delivered by nurses were more successful than those delivered by non-nursing personnel. We suggest that DSME programs might do better by going beyond procedural interventions. Most DSME programs relied heavily on rules and procedures to guide decisions about diet, exercise, and weight loss. Future DSME may need to include cognitive self-monitoring, diagnosis, and planning skills to help patients detect anomalies, identify possible causes, generate corrective action, and avoid future barriers to maintaining healthy A1c levels. Finally, comprehensive descriptions of DSME programs would advance future efforts.
Diabetes afflicts approximately 25.8 million people in the United States, or 8.3% of the population. Type 2 diabetes, or non-insulin dependent diabetes mellitus (NIDDM), accounts for 90 to 95% of all diagnosed cases of diabetes in adults [
Type 2 diabetes complications stem from the inability of the body to use insulin properly, resulting in heightened blood glucose levels [
The U.K. Prospective Diabetes Study (UKPDS) [
This research addresses the question of how effectively current DSME interventions help patients with type 2 diabetes achieve sustained control of their blood glucose. Physicians and other health care professionals can prescribe effective medications, provide optimal dietary guidelines, and support needed life style modifications. In the end, however, it is the patient and their caregivers who must make the daily decisions needed to control blood glucose.
There is a wide array of support strategies for patients. Web sites maintained by The National Institute of Health (NIH) [
With the growing burden of diabetes on health care systems and the plethora of medical, pharmaceutical, and human factors advancements, it is critical that DSME programs increase their effectiveness, sustainability, and scalability. This meta-analysis of interventions started with six reviews of DSME [
Ellis et al. [
Ismail et al. [
While these reviews reported modest but statistically significant reductions in A1c levels among intervention participants when compared to control participants, most reductions did not reach healthy A1c levels. Intervention participants remained at risk from elevated blood glucose levels as reflected by A1c or other standards metrics. Finally,Norris et al. [
More disheartening in these reviews, reductions in glycemic control were often not sustained over time. The meta-analysis conducted by Norris et al. [
These six overview studies encountered a variety of difficulties and limitations. First, behavioral change techniques have generally lacked standardized definitions and taxonomies [
Another research problem has been attrition rates. Sarkisian et al. [
Because of these methodological problems, we share Gary et al.’s [
A second reason for our analysis was to provide an initial exploration of the dynamics of self-management. Lippa et al [
We reviewed the studies from the six earlier referenced reviews addressing the effect of DSME interventions on blood glucose control [
The titles and abstracts of search results were assessed for relevance and retrieved if appropriate. When the same data were used in multiple publications, we included only one of the publications in our analyses.
All of the studies in our analysis met the following criteria. Intervention participants completed a DSME intervention designed to increase adherence and only data collect Per Protocol (PP) was included. Participants were adults with type 2 diabetes as defined by NIH [ A1c values were available as both baseline and post-intervention measures and data were sufficient to define the means and standard deviations for the A1c. All studies used randomized controlled trials meeting at least one of the following criteria: Random assignment of participants from a single pool (e.g., treatment center, unified recruitment method). Study specified as a randomized trial (unless evidence suggested otherwise, such as significantly different participant baseline characteristics). Study sites were randomly assigned and equivalent (with sufficient evidence). Groups were matched on baseline measures.
A total of 186 unique articles were retrieved. Of these, 134 studies were excluded for the reasons provided in Figure
Excluded article chart.
Table
Description of included studies and interventions.
Year |
|
Who |
Intervention |
Mode of |
Duration of |
Time before | |
---|---|---|---|---|---|---|---|
Adolfsson | 2007 | 88 | 1, 3b | AE, RP | G | 30.33 | 52 |
Agurs-Collins | 1997 | 55 | 4, 0 | RP, SS | G, I, S | 26 | 0 |
Amano | 2007 | 39 | 0 | RP | I | 13 | 0 |
Anderson-Loftin | 2005 | 65 | 2n, 4 | RP, SS | G, S, T | 26 | 0 |
Arseneau | 1994 | 40 | 7 | RP, SS | I? | 0.57 | 8.67, 21.67 |
Barnard | 2006 | 99 | 1, 4, 6 | RP | G, I, T | 22 | 0 |
Brown | 2002 | 224 | 3c, 4, 6 | RP, SS | G, S | 52 | 0 |
1: RP. | 1: G, I. | ||||||
Campbell | 1996 | 200 | 3c, 4 | 2: RP, SS. | 2: G, I, S. | 2 | 11, 24 |
3: CC, RP, SS. | 3: I, T. | ||||||
Cheskin | 2008 | 24 | 4 | RP | G, I | 86 | 0 |
Christian | 2008 | 273 | 1, 7 | AE, RP | I, TECH | 52 | 0 |
D'Eramo-Melkus | 1992 | 49 | 0 | Group 1: CC, RP. |
Group 1: G, I. |
12, 18 | 8, 14 |
Deakin | 2006 | 291 | 4 | AE, CC | G | 6 | 11.33, 54.67 |
Engel | 2006 | 50 | 0 | AE | GINS, T, TECH | 24 | 0 |
Faridi | 2008 | 30 | 3a, 7 | RP | TECH | 13 | 0 |
Fornos | 2006 | 112 | 3 | RP | I, O | 56.33 | 0 |
Franz | 1995 | 179 | 2d | RP | I | 6 | 7, 20 |
Gabbay | 2006 | 332 | 3c | CC, RP | I, T, TECH | 52 | 0 |
Gaede | 2001 | 149 | 1, 3c, 4 | AE, RP, SS | G, I, S | 197.6 | 0 |
Gallegos | 2006 | 45 | 3c | RP, SS | G, I, T | 50 | 0 |
Glasgow | 1992 | 97 | 4, 5, 6 | CC, RP, SS | G | 13 | 0 |
Glasgow | 2000 | 277 | 2n, 4, 5, 6, 7 | RP | O, T, TECH | 26 | 13 |
Goudswaard | 2004 | 50 | 3b | RP | I | 26 | 6, 52 |
Gucciardi | 2007 | 61 | 3c, 4, 5 | CC, RP, SS | G, I | 13 | 0 |
Janssen | 2009 | 491 | 1, 3c | RP? | G, I | 52 | 0 |
Kim & Jeong | 2007 | 51 | 3c | RP | I, TECH | 26 | 0 |
Kim & Song | 2008 | 34 | 3c | RP | TECH | 26 | 0 |
Ko | 2007 | 308 | 1, 2n/d, 4, 5 | AE, CC, RP, SS | G, S | 0.71 | 25, 51, 103, 155, 207 |
Krousel-Wood | 2008 | 76 | 7 | RP | TECH | 13 | 0 |
Kulzer† | 2007 | 181 | 5 | 1: AE, CC. |
G, C: G, I. | 13 | 0, 52 |
Ligtenberg | 1997 | 51 | 1, 4 | AE, RP | G, I, T | 26 | 0 |
Lujan | 2007 | 141 | 6 | AE, RP | G, T | 26 | 0 |
McKibbin | 2006 | 57 | 6 | AE, RP | G | 24 | 0 |
Ménard | 2005 | 61 | 0 | RP | I, O, T | 52 | 0, 26 |
O'Kane | 2008 | 184 | 3a, 4, 6 | RP | G | 52 | 0 |
Pederson | 2007 | 122 | 0 | RP | I, O | 26 | 0 |
Pibernik-Okanovic | 2004 | 108 | 4, 5 | AE, CC | G | 6 | 7, 20 |
Piette | 2000 | 248 | 3c, 7 | RP | T, TECH | 52 | 0 |
Rachmani | 2005 | 110 | 0 | AE, RP | G | 208 | 0, 208 |
Rosal | 2005 | 25 | 3c, 4, 6 | CC, RP | G, I | 10 | 3, 16 |
Schwedes | 2002 | 223 | 1, 3c, 6 | CC, RP | G?, I | 24 | 0 |
Shea | 2007 | 1355 | 6, 7 | CC, RP | TECH | 52 | 0 |
Sone | 2002 | 1973 | 3c | RP | I?, T | 156 | 0 |
Steed | 2005 | 106 | 3b, 4 | CC, RP | G | 5 | 0 |
Sturt | 2008 | 202 | 3c | AE, CC, RP | I, O, T | 12 | 14 |
Sun | 2008 | 146 | 1, 4 | RP | GINS | 24 | 0 |
Trento | 1998 | 96 | 1, 5 | CC, RP, SS | G, S | 52 | 0 |
Trento | 2002 | 90 | 1, 6 | CC, RP | G, I+ | 208 | 0 |
Tsujiuchi | 2002 | 26 | 6 | AE | G | 17.33 | 0 |
Uusitupa | 1993 | 82 | 1, 3b, 3c, 4 | RP | G | 65 | 0, 117 |
Wattana | 2007 | 147 | 3c | RP | G, I, O | 24 | 0 |
Yoo | 2008 | 57 | 3c, 7 | AE | G, I, TECH | 13 | 0 |
Yoon & Kim | 2008 | 51 | 6, 7 | RP | TECH | 52 | 0 |
Note. *Studies with multiple intervention lengths or multiple follow-ups are indicated by lengths separated by commas; †A1c values not provided in text—values estimated from a bar graph.
We classified the content based on the description provided in the program. In some cases, several content areas were mentioned.
Rules and Procedures (RP) was the most commonly mentioned content and focused on explicit guidelines, such as specific rules regarding diet and exercise. An example of a procedure would be how to perform blood glucose monitoring. Rules and procedures can include the use of a journal for recording data but do not typically provide support for translating blood glucose readings into effective decision making.
Affective and Emotion (AE) focused on emotion, motivational encouragement, empowerment, and/or confidence building.
Social and Situational (SS) focused on managing social and situational factors that impede effective diabetes self-management. These strategies might include holiday meal planning and selecting restaurant meals.
Complex Cognition (CC) focused on mental models or other complex cognitive strategies designed to use conceptual understanding of diabetes to moderate blood glucose levels. This goes beyond the simple application of rules and procedures to the use of a mental model to detect anomalies and identify causes, and to generate corrective and preventive strategies.
The overall results showed that the DSME interventions significantly reduced A1c levels (Table
Gain score comparison.
Baseline |
Posttreatment |
A1c |
Gain score | Est. % below A1c 6.4 at baseline | Est. % below A1c 6.4 posttreatment | % |
Impact Score | |
---|---|---|---|---|---|---|---|---|
Control | 8.70 (1.48) | 8.18 (1.43) | −3.66** | −0.52 | 11.65% | 14.53% | 3.15** | 2.88 |
Intervention | 8.70 (1.47) | 7.61 (1.34) | −8.29** | −1.09 | 12.73% | 22.84% | 6.96** | 10.11 |
Significance |
|
|
|
Note.
The overall intervention effect was to reduce mean A1c levels from 8.70% to 7.61%, as shown in Table
To put these findings into context, we computed an “Impact Score” reflecting the proportion of participants whose A1c levels were at or below 6.4%. The NIH guidelines set 6.5% A1c as the threshold for diabetes, and so the Impact Score measured the proportion of participants that had moved below this threshold [
These caveats aside, the Impact Scores are revealing. The intervention groups had an estimated 22.84% of participants classified as having A1c values below 6.5%. Thus, almost a quarter of the participants who received the interventions would no longer be considered to have type 2 diabetes. However, at the intervention baseline, 12.73% of participants already had A1c values below 6.5%, so the improvement, while statistically significant, was only 10.11% of the participants. That is, 7.23% of the intervention participants who started with A1c scores of 6.5% or greater achieved a safe level of A1c as a result of the intervention, a result that was significant at the 0.01 level.
Further, the control groups also showed a statistically significant improvement of 2.88%. Therefore, the overall treatment impact (difference in Impact Score between control and intervention groups) was 7.23%. An independent-samples
Finally, we looked at the overall benefit of DSME interventions for participants. Figure
Post-intervention A1c levels
Table
Post-intervention A1c levels.
Control | Intervention | ||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Length of intervention |
|
No. of studies | No. of tests | A1c | A1c | Percent difference |
|
Mean weighted |
Est. % below A1c 6.4 | Control change in est. % | Intervention change in est. % | ||||
|
SD |
|
SD | ||||||||||||
13 weeks |
5,319 | 17 | 32 | 8.22 | 1.28 | 7.70 | 1.29 | 6.40% | 2.13* | 0.46 | 15.68% | 4.49% | 10.58% | 6.09 | |
14–26 weeks | 2,247 | 17 | 20 | 8.08 | 1.58 | 7.52 | 1.40 | 6.89% | 1.84* | 0.49 | 21.19% | 1.27% | 10.74% | 9.47 | |
27 weeks |
6,241 | 19 | 22 | 8.20 | 1.52 | 7.56 | 1.37 | 7.79% | 2.03* | 0.23 | 19.86% | 2.02% | 8.84% | 6.82 |
Note.
A different picture emerged using the mean weighted
An analysis of variance of gain scores (intervention mean minus intervention baseline) by intervention length found significant group differences,
Table
Mean outcome A1c levels for control and intervention groups, by delay from end of intervention to time of test.
Control | Intervention | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Length of delay |
|
No. of studies | No. of tests | A1c | A1c | Percent difference |
|
Mean weighted |
Est. % below A1C 6.4 | Control change in est. % | Intervention change in est. % | Impact Score | ||
|
SD |
|
SD | |||||||||||
No delay | 8,729 | 39 | 43 | 7.92 | 1.45 | 7.42 | 1.29 | 6.29% | 2.31* | 0.29 | 21.46% | 3.16% | 10.58% | 7.42 |
1–13 weeks | 1,291 | 7 | 9 | 8.38 | 1.36 | 7.66 | 1.15 | 8.61% | 1.65† | 0.12 | 13.66% | 1.89% | 7.38% | 5.49 |
14–26 weeks | 1,572 | 9 | 11 | 9.07 | 1.54 | 8.26 | 1.56 | 8.92% | 1.83* | 1.02 | 11.66% | 1.08% | 8.95% | 7.87 |
51 weeks or more | 2,215 | 8 | 11 | 8.13 | 1.30 | 7.65 | 1.47 | 5.82% | 2.50* | 0.32 | 19.76% | 4.41% | 11.63% | 7.22 |
Note.
Three of these four durations resulted in significant differences between control and intervention conditions. The 1–13 week condition did not. Most of the studies relied on the immediate post-intervention measurement. The 14–26 weeks condition had the greatest percent difference between intervention and control groups, 8.92%,
The different analyses varied in their conclusions about sustained A1c reductions although none even approached the 10-year retention intervals associated with health indicators. The Impact Score, at the far right of Table
Analysis of variance conducted between intervention and control means revealed a marginally significant difference by duration,
We had planned to compare the effectiveness of different intervention strategies but found that 21 studies of the 52 studies used only rules and procedures. Twenty-nine used rules and procedures in conjunction with one or more of the alternative training methods. In contrast, only one study used Affective/Emotional as a single approach. Complex Cognition was used in conjunction with alternative intervention approaches in 18 studies. Affective and Emotional was used in conjunction with alternative intervention approaches in 16 studies. Social and Situational was used in conjunction with other approaches in 10 studies. Only 3 of the 52 programs did not rely on rules and procedures, at least in part.
The intervention programs that relied entirely on rules and procedures achieved significant reductions in A1c, from 7.71% in the control group to 7.25% (
We examined three classes of intervention presenters: nurse only, nurse in combination with other professional, and no nurse (Table
Mean Outcome A1c levels for control and intervention groups, by type of professional who delivered intervention.
Control | Intervention | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
Who delivered intervention |
|
No. of studies | No. of tests | A1c | A1c | Percent difference |
|
Mean weighted |
||
|
SD |
|
SD | |||||||
Nurse only | 2,996 | 9 | 10 | 8.18 | 1.39 | 7.58 | 1.34 | 7.24% | 2.32* | 0.17 |
Nurse in combination with others | 3,275 | 14 | 21 | 8.38 | 1.17 | 7.67 | 1.01 | 8.54% | 2.01* | 0.59 |
No nurse | 7,536 | 29 | 43 | 8.08 | 1.57 | 7.59 | 1.5 | 6.07% | 2.32* | 0.34 |
Note.
The present research looks at the daunting challenge of translating medical evidence about Type 2 diabetes self-management into patient decision making, behavioral change, and ultimately blood glucose control. For people with type 2 diabetes, like those with many other chronic conditions, health care providers can prescribe medications, describe optimal dietary patterns, and outline needed life style modifications, but only the patient can implement these critical recommendations. Because adherence depends on patient decisions, we looked at interventions intended to support adherence. We asked: how well are current educational interventions preparing patients to make effective blood glucose control decisions?
First, the good news. Our meta-analysis showed that intervention groups overall showed moderate reductions in A1c from baseline to post-intervention assessment. The average reduction in A1c for the intervention groups was from 8.70 at baseline to 7.61 at the post intervention assessment. The A1c improvements seem fairly robust, 1.09, but must be interpreted in light of the reductions shown by the control participants. The control participants started at the same baseline of 8.70 and reduced it to 8.18, a modest improvement of 0.52. Both experimental and control groups demonstrated a significant (at the 0.01 level) reduction in A1c.
Improvements in control groups are common and typically attributed to a placebo effect. In the current study, it may also have occurred because some of the studies provided the control group with unspecified “standard training” while the experimental group received innovative training. The intervention group improvement was only 0.57 better than the control group. Nevertheless, it was significantly better (
Next, the bad news. According to NIH criteria [
Our findings were more positive than the results of the six previous meta-analyses. Ellis et al. [
Nevertheless, the Impact Score (the proportion of intervention group participants who moved from a level of 6.5 or above to a level of 6.4 or below, from baseline to post-intervention, in comparison to the control group) was only 7.23%. This is a small achievement in the face of the resources that went into the interventions. Less than 8% of the intervention participants moved below the line for diabetes, compared to the control group. We recognize that the 6.5 level is somewhat arbitrary, but nonetheless it provides a yardstick for assessing program impact.
In this study, the intervention groups with the shortest durations had significantly greater gain scores. Interventions tended to work at the beginning, but their effects appeared to attenuate over time. This is consistent with Norris et al. [
This study evaluated DSME interventions. While earlier studies sometimes included people with type 1 diabetes, this study was restricted to people with type 2 diabetes. Unlike some earlier studies, the present sample was restricted to studies using randomized trials. Even with better selection criteria, our outcomes were consistent with earlier research: the benefits of DSME were modest [
This meta-analysis has several limitations. First, the 52 studies included were all submitted to and accepted by professional journals. Authors are less likely to submit null findings and editors are less like to accept them. It is, therefore, likely that our outcomes describe more successful interventions. Second, adherent and successful participants are more likely to complete interventions than are less adherent and unsuccessful participants—the problem of attrition rates. Of the 46 studies that reported beginning and end sample size, 10 studies (22%) had attrition rates greater than 20%. Some of the studies in our sample had very high attrition rates (e.g., greater than 40%). Our outcomes are therefore likely to describe more successful studies and the improvements of more successful participants. Taken together, the outcomes are likely to be biased in support of intervention effectiveness.
The 52 studies we reviewed relied primarily on teaching rules and procedures. A total of 21 programs used rules and procedures exclusively. Only three of the programs did not report using rules and procedures. Our findings show that the rules and procedures approach is effective and its effect is sustained, but modest. The gain score in this category was only a 0.46 reduction compared with that of the control group. The interventions that either blended rules and procedures with other methods, or relied on other methods showed larger improvements over the control group, resulting in a reduction of 0.64. The addition of other strategies, such as complex cognitive or affective interventions might, therefore, serve to enhance the effects of rules and procedures-based methods. Despite the improvement in the intervention groups, the final mean values for both of these conditions were still over 7.0 A1c.
The interventions had some effect but the effect was not strong enough to help most people avoid the threat of the damages associated with type 2 diabetes over the long term. When people are first diagnosed with type 2 diabetes, we have to send them home with sample menus and lists of foods to avoid. We have to inform them of the dangers of excessive sugar and carbohydrates. We have to convey the procedures for measuring blood glucose levels. Rules and procedures are necessary, but do not appear to be sufficient.
Lippa et al. [
We had difficulty in synthesizing different DSME programs because of the lack of standard reporting procedures. Often methodologies and intervention descriptions were too brief and ambiguous to see what actions were actually taken. For example, terms such as “diabetes education program” and “healthy lifestyle” were pervasive and often underspecified. These phrases may involve diet and exercise, but the exact type of education is unknown. Future studies should embed curricula in the text or have links to the material online.
Also commonly lacking were indicators of the intensity, mechanisms, and presenters of training. Some studies may have achieved better results because of extensive preparation for the intervention facilitators prior to the interview. For example, in Adolfsson et al. [
Program variables are important in evaluating the cost versus effectiveness trade-off. By giving more attention to clarifying their methods, future DSME programs can help to promote progress and contribute to Evidence-Based Medicine. Abraham and Michie [
The findings reported in this meta-analysis illustrate the positive but modest gains of existing DSME efforts. There are certainly patients who will have difficultly altering long-term behavioral patterns and others who are simply unwilling to try. Nevertheless, innovative DSME programs that build mental models that help people detect anomalies, identify possible causes, and generate corrective actions hold the possibility of moving more participants to healthy A1c levels. We have come a long way and we have miles to go.
The authors thank Dr. Nathan Bowling for providing support with data analysis; Dr. Bryan Gibson for thoughtful comments on cognitive modeling as a support for behavioral management; and Ashley Wenner for her work with literature search and coding. Support for this research was provided by the Department of Psychology, Wright State University, Dayton, OH, and by MacroCognition LLC, Yellow Springs, OH.