KQ 1. Behavioral Programs for T1DM and Behavioral, Clinical, and Health Outcomes; Diabetes-Related Health Care Utilization; and Program Acceptability

There was moderate SOE showing reduction in hemoglobin A_1c (HbA_1c) at 6-month postintervention followup with percent HbA_1c reduced by 0.31 for individuals who were enrolled in behavioral programs compared with those receiving usual care. For all other timepoints, there was no significant difference in HbA_1c; the SOE was low due to risk of bias and imprecise effect estimates. For followup timepoints of 12 months or longer, the 95% CIs included our threshold for clinical importance such that we cannot rule out benefit for behavioral programs based on the available evidence. For individuals who were enrolled in behavioral programs compared with those receiving an active control, there was moderate SOE showing a clinically important reduction in HbA_1c of 0.44 percent at 6-month postintervention followup. There was no difference in HbA_1c at other timepoints, however the SOE was low and we cannot rule out a benefit for behavioral programs.

There was low SOE showing no difference in adherence to diabetes self-management (i.e., frequency of blood glucose checks or overall self-management behaviors) at end of intervention and 6-month followup for comparisons with usual care. For comparisons with active controls there was insufficient SOE for adherence to diabetes self-management at all followup timepoints. There was moderate SOE of no difference at the end of intervention for generic HRQL, and insufficient evidence at longer followup. In comparisons with usual care, there was insufficient SOE to assess whether there was any effect on diabetes-specific HRQL at any timepoint, and low SOE of no difference for diabetes distress at end of intervention and 6-month followup. There were no data on HRQL for comparisons of behavioral programs with active controls. No trials reported on micro- and macrovascular complications or on all-cause mortality. The SOE grading was highly influenced by the moderate or high risk of bias (ROB) of individual studies, the imprecise estimates of effect, and (for insufficient SOE grades) the limited amount of data.

Evidence was insufficient to determine whether behavioral programs increased or decreased the number of diabetes-related hospital admissions, emergency department admissions, episodes of severe hypoglycemia, or episodes of severe hyperglycemia. Behavioral programs appear to be acceptable to patients with T1DM based on a proxy measure; our meta-analysis showed a 21 percent increased risk of attrition usual care compared with behavioral programs.

KQ 2. Subgroups for Effectiveness in T1DM

For the KQ, we examined the differential effect of patient characteristics on the effectiveness of behavioral programs for T1DM. In comparisons with usual care, results were consistent with those from KQ 1 when combining all studies of youth and adults. At 6 months, behavioral programs reduced HbA_1c in youth by a statistically significant 0.28 percent and in adults by a non-statistically significant 0.38 percent. At end of intervention, the point estimates indicated greater benefit within the adult subgroup (0.28) than the youth subgroup (0.00), although neither of these values reached statistical significance. None of the point estimates exceeded the a priori established clinically important difference of 0.4 percent HbA_1c.

For subgroups based on age in comparisons with active controls, the small number of studies (and sample sizes) led to wide pooled 95% CIs which in some cases included values of clinical importance both for and against behavioral programs; the SOE was thus graded as insufficient in all but two cases. In studies of youth with followup to 12 months, there was low SOE of a clinically important (reduction by 0.52) benefit for behavioral programs; in studies of adults with 6-month followup, there was low SOE of no difference in HbA_1c.

KQ 3. Potential Moderation of Effectiveness for T1DM: Components, Intensity, Delivery Personnel, Method of Communication, Degree of Tailoring, and Level of Community Engagement

To assess whether program factors (i.e., intensity, delivery personnel, method of communication, degree of tailoring, and level of community engagement) moderated the effectiveness of behavioral programs for T1DM, we performed univariate meta-regressions for comparisons between behavioral programs and usual care at longest followup. Program intensity, including duration, contact hours, and frequency of contacts, appeared not to influence program effectiveness; individual delivery appeared more favorable than group delivery of programs but the results did not reach statistical significance. We did not have enough studies to perform multivariable analysis, neither did we have enough to perform the univariate regressions for outcomes other than HbA_1c.

KQ4. Harms for T1DM

No studies reported on the associated harms (i.e., activity-related injury) of behavioral programs.

Discussion of Key Findings for T1DM

Overall, behavioral programs seem to have some benefit in T1DM for reducing HbA_1c, when followup extends beyond the immediate postintervention period up to 6 months. The delay in benefit may in part reflect the time required for this marker of glycemic control, indicating control over the past 2-3 months, to demonstrate change. Notable though, is the large diversity in program duration whereby end of intervention was anywhere between 1.5 and 25 months. Another contributor may be that a period of time is needed to integrate newly learned self-management behaviors into one's life; however, our findings of no differences in self-management behaviors at any followup timepoint when behavioral programs were compared to usual care do not support this hypothesis. The beneficial findings for HbA_1c at 6 months appear to be tempered by the findings of no difference at longer followup timepoints, although we are unable to confidently rule out benefit at long-term followup. An argument that the findings of benefit could be an artifact of differential attrition between groups—with those more motivated to or more successful in making positive changes returning for followup assessment—appears to be unlikely because of the lower (21%) attrition rate found for behavioral programs compared to usual care.

There are at least a couple reasons why our findings may underestimate the effect of these programs should they be implemented in routine practice. The usual care group in several studies received some form of attention from the investigators (e.g., periodic telephone calls to maintain contact and encourage study participation), and this may have resulted in improved glycemic control for the comparator group and reduced the relative effects of the behavioral program. Participants (or their providers) in the usual care or active control groups (not being blinded to group assignment in most studies) may have become more motivated to practice better self-management (including blood glucose regulation using insulin titrations), which could also attenuate differences between groups. Differences in the “usual care” provided may have also played a role, although this affect may be minimal considering recent evidence that variations in standard care in studies of behavioral interventions for youth with T1DM did not significantly impact study results.²⁹⁵

Our finding of a statistically significant and clinically important reduction by 0.44 percent HbA_1c at 6-month followup for comparisons between behavioral programs and active controls is notable. As per our operational definition, behavioral programs consisted of interactive programs having a duration ≥4 weeks with the inclusion of behavior change techniques; because of this, traditional, didactic educational^{91,92,107,108} or support interventions⁸⁷ were considered comparators rather than interventions. By offering an intervention to both study arms, these studies may have introduced less potential bias from lack of allocation concealment and blinding. Although quite promising, when drawing conclusions regarding the overall benefits of behavioral programs, this finding needs to be interpreted in light of results showing no differences for HbA_1c at other timepoints and insufficient evidence to make conclusions about several other outcomes.

Many of the included studies were directed at adolescents. Self-management of T1DM during adolescence is complex, often characterized by personal challenges and uncertainty, transitions to adult care, less frequent health care visits, and diminished parental involvement; consequently, glycemic control deteriorates over the course of childhood and adolescence for many youth with T1DM.^296-299 For these reasons, many of the studies included in this review aimed to prevent deterioration of glycemic control rather than to improve it. The statistically significant reductions in HbA_1c at 6-month followup (versus usual care), and the clinically important reductions in HbA_1c at 6- and 12-month followup (0.60 and 0.52 percent, respectively) in comparisons with active controls in youth lend substantial support for these programs. Likewise, incorporating more demanding self-management behaviors may negatively impact social and emotional functioning, such that our findings of no difference in generic HRQL at end of intervention may be interpreted as positive.

Most studies for T1DM were undertaken in populations with baseline glycemic control ≥8.5 percent HbA_1c. While this may affect the applicability of the findings to some extent, clinicians may view this as highly relevant to their patient population of which many—particularly in their pubertal years—are struggling to achieve optimal control. Furthermore, the Diabetes Control and Complications Trial (DCCT)²⁰ found that these individuals receive the greatest benefit from HbA_1c reduction.

For T1DM, there was evidence that effectiveness appears not to be moderated by program intensity (i.e., duration, contact hours, or frequency of contacts), and that delivery to individuals compared with groups may be beneficial. We were unable to undertake any analysis to comment on the difference between educational and lifestyle programs, or the addition of a support component to DSME programs. Many individuals with T1DM under good glycemic control may have other risk factors (e.g., overweight, hyperlipidemia, hypertension) for which lifestyle programs may be warranted. Although some behavioral programs were of fairly long duration with highly intense contact with patients,⁸⁸ only one explicitly incorporated a support component.⁸²

Our pair-wise meta-analyses used the Hartung-Knapp-Sidik-Jonkman random effects model^73-75 that typically provides a more conservative estimate of the 95% CI around pooled effect sizes than the common DerSimonian and Laird approach; the latter approach has been shown to lead to too many statistically significant results especially in the face of heterogeneity and few studies. The effect of our approach is that some results—especially those pooling few studies—are found statistically nonsignificant when another approach may find significance; moreover, the 95% CI in some cases spreads wider than those of the individual studies. For example, our reported 95% CI for the effect on HbA_1c for youth receiving a behavioral program compared with an active control at 6-month followup is -2.56 to 1.36 (not significant due to inclusion of 0 [no effect]), although the DerSimonian and Laird approach provided an estimate of -0.95 to -0.25 (significant). This factor also applies those findings for T2DM on the overall effectiveness of behavioral programs across all outcomes.

Effectiveness of Behavioral Programs Across Outcomes

There is evidence showing a beneficial effect of behavioral programs, compared to both usual care and other active interventions, at end of intervention for glycemic control; however, at longer followup results were only statistically significant at 6 months for comparison with active controls, and none of the results were considered to be clinically important based on our threshold of a 0.4 percent change in HbA_1c. There was substantial statistical heterogeneity in these pairwise meta-analyses, supporting our subsequent analysis for KQs 5 and 6 to determine which program factors, and population characteristics, influence (and optimize) the effects.

Behavioral programs showed some benefits in terms of reducing BMI (0.21-0.92 kg/m² to 12-month followup), weight (1.3-1.68 kg; end of intervention) and waist circumference (3.2 cm; short term), and daily energy intake (64-150 kilocalories per day to 6 months)—mainly when compared with usual care. There was little evidence around the outcomes related to changes in physical activity and medication adherence, and findings were consistently of no difference.

Health-related quality of life was reported by fewer studies than anticipated. On average, findings of no difference were found for most studies and outcomes, except for Diabetes Distress where results favored behavioral programs compared with usual care at end of intervention but not at longer followup. Effects on diabetes complications were only reported for one study. Diabetic retinopathy was reduced by 14% and very-high-risk kidney disease by 31% in participants receiving a ≥8 year-long intensive lifestyle program compared with didactic education and support in the largest trial, conducted by the LookAHEAD research group.^278,292 Mortality between behavioral programs and active control groups (5 comparisons; 6,050 participants) was 14 percent lower for those receiving behavioral programs (RR, 0.86; 95% CI, 0.77 to 0.96). There was no difference for comparisons with usual care (25 comparisons; 4,659 participants; RR, 1.28; 95% CI, 0.84 to 1.94).

KQ 5. Potential Moderation of Effectiveness for T2DM: Components, Intensity, Delivery Personnel, Method of Communication, Degree of Tailoring, and Level of Community Engagement

In a network meta-analysis with usual care serving as the main reference, programs demonstrating effect sizes for HbA_1c at or above our threshold for clinical importance (i.e., 0.4 percent HbA_1c difference between groups) represented all three major program component categories of diabetes self-management education (DSME), DSME and support, and lifestyle. The effect sizes of minimally intensive DSME programs (≤10 contact hours) were less than our threshold for clinical importance, but were all higher than that of educational interventions not meeting our criteria for a behavioral program (e.g., didactic education programs represented by many active controls). Programs having larger effect sizes and higher probabilities of being best (≥5 percent) were more often delivered in person rather than including technology. All effective programs using some form of technology were of moderate or high intensity.

Lifestyle programs resulted in the largest effect sizes for BMI. Program intensity appeared less important than method of delivery; providing some individual (rather than solely group-based) delivery appears beneficial for improvements in BMI at longest followup.

KQ 6. Subgroups for Factors Moderating Effectiveness in T2DM

All of our results for this KQ relied on between-study rather than within-study comparisons, such that the effect of randomization is removed and the results are considered observational and possibly biased through confounding by other study-level characteristics.

In terms of overall effectiveness at longest followup for HbA_1c, participants with suboptimal or poor glycemic control (≥7 percent HbA_1c) appear to benefit more than those with good control (<7 percent) from behavioral programs when compared to usual care and active controls. The effect sizes were not clinically important for either group. Few differences were evident when evaluating potential moderation by program factors after rerunning the network meta-analysis of KQ 5 with a subgroup of studies having participants with suboptimal or poor baseline glycemic control.

Older adults (≥65 years) did not benefit at longest followup in terms of reduction in HbA_1c from behavioral programs in comparison with usual care or active controls. In adults <65 years, the effect size for behavioral programs compared with usual care was statistically significant (reduction of 0.31 percent) and compared with active controls at longest followup was clinically important (0.43 percent). In a subgroup analysis of our original network meta-analysis of HbA_1c—removing the studies of participants with a mean age ≥65—the most noticeable change was the increase in effect size for active controls incorporating dietary or physical activity interventions, which produced clinically important effects (0.55 percent reduction in HbA_1c). The active controls still showed zero probability of success, perhaps due to the heterogeneity between, or small sample sizes of, the associated comparisons.

In comparison to usual care and active controls, behavioral programs offered to predominantly minority participants (≥ 75 percent nonwhite) appear to provide more benefit for glycemic control than those offered to populations with a lower proportion (<75 percent) of nonwhite individuals. The effect size for minority participants reached clinical importance when comparing behavioral programs to usual care (0.43 percent reduction in HbA_1c). Based on univariate regression analyses for the subgroups based on race/ethnicity, none of the program factors (e.g., intensity, delivery personnel) reached statistical significance for influencing the effectiveness of behavioral programs compared to usual care on HbA_1c. Lifestyle programs appeared favorable over DSME or DSME plus support for the group of studies (n=24) with predominantly white individuals (p=0.07).

Discussion of Key Findings for T2DM

The focus of our review for T2DM was on identifying factors contributing to the effectiveness of multicomponent programs. Our review includes the highest number of studies to date, and focuses on programs meeting current recommendations to change patient behaviors and patient-important outcomes (e.g., HRQL). We relied on strict inclusion criteria to study interactive programs incorporating behavioral strategies aiming to change multiple behaviors, without confounding by changes to medical management (e.g., medication changes, differing frequency of provider visits). Another strength of the review is our analytical approach; the network meta-analysis enabled differentiation of the various comparators, and incorporation of comparisons (e.g., intervention vs. intervention) often not amenable to other strategies. Moderate- and high-intensity (≥11 hours contact time) programs appear to be necessary to provide individuals with clinically important effects on HbA_1c; this outcome may also benefit from in-person delivery rather than using technology. For BMI, providing some individual delivery, rather than solely relying on group formats, appears beneficial.

Our review adds to previous findings in that lifestyle programs—not specifically training people in diabetes related self-care behaviors but focusing more on weight reduction and increases in physical activity—may provide similar or more benefit than DSME programs for improving glycemic control for individuals with T2DM. A feature of behavioral programs that may be particularly attractive to patients is that unlike some common drug therapies used in the management of type 2 diabetes, behavioral programs have the potential to reduce HbA_1c without contributing to weight gain. Our review confirms previous suggestions that programs with an interactive nature, employing behavioral approaches and covering multiple behaviors, are beneficial when compared with didactic educational interventions. Although perhaps not to a clinically important degree for individuals, the burgeoning growth of this disease means that even small gains in glycemic control from behavioral programs may serve as a substantial benefit for public health.

Our finding that single-topic, non-educational interventions (active controls of dietary or physical activity interventions) offer more benefit than do basic education interventions, supports the need to carefully distinguish and account for different comparators during the systematic review process. We used longest followup timepoint for the analyses to answer KQ 5 and 6, which may capture the “durability” of the programs better than restricting the analysis to the immediate postintervention period.

It appears from our network meta-analysis results for HbA_1c, that both individual and group delivery can be beneficial; this agrees with other work in this area³⁰⁰ (also see below section on Findings in Relation to What is Already Known). Our results for KQ6 suggest that other factors (or combination of factors) may influence the effects of this variable; for instance, delivery format may be highly dependent upon the population served and program content. Studies within nodes having high effect sizes which offered programs in groups tended to be those offered to minorities, including Mexican Americans,^150,152,208 where support from peers was incorporated as a key program feature.

We were unable to draw any conclusions about the choice of delivery personnel from the network meta-analysis when answering KQ 5; there were too few studies in the categories of DSME and support, and lifestyle to account for this variable when creating the nodes. Drawing from the pair-wise meta-analysis results for those trials comparing two or more interventions (i.e. comparative effectiveness), there may be no difference when program delivery is conducted by heath care professionals or by lay providers (e.g., peers with diabetes, community health workers). Four trials (575 subjects) found no difference (MD, 0.00; 95% CI, -0.23 to 0.23)^{144,172,227,256} in effectiveness when programs were delivered by peers compared with health care professionals. One trial (72 subjects) found no difference when the support phase of DSME was provided by clinic staff compared with diabetes educators (MD, 0.02; 95% CI, -0.60 to 0.64).²²⁷ Most trials reported on extensive training programs for those delivering their programs. One reason why programs delivered by health care professionals were not superior may be that physicians, nurses, and dietitians receive little or no training in behavioral techniques as part of their formal education. This may be particularly true when extensive knowledge and expertise in theoretically guided approaches (e.g. motivational interviewing), or several behavior change techniques are required. Diabetes educators, highly regarded for their thorough knowledge and skills in diabetes education, may require substantial training and supervision when starting to apply advanced behavioral techniques such as motivational interviewing; to date this technique has shown benefit for improved glycemic control in the short term when delivered by clinical psychologists^232,250 but not by diabetes educators.²⁴⁸ It could be speculated that the benefits for glycemic control may improve with time after those delivering the programs gain experience.

Our findings for KQ 6 suggest that people with good baseline glycemic control (<7 percent HbA_1c), advancing age (≥65 years), and white/European ancestry (studies not having a majority of minority participants) may not benefit to the same extent as participants with suboptimal or poor glycemic control, racial/ethnic minorities, and those of younger age. The finding of better success for patients with poorer glycemic control has been found in previous systematic reviews (for one example see Duke et al.³⁰⁰). Intuitively, individuals with good glycemic control may not achieve as much benefit from behavioral programs—there is little room for improvement and good self-management behaviors may already be practiced regularly. Our findings may have been different if we had chosen a different level of glycemic control for subgroup analysis; after consultation with several experts we were unable to define a “poor control” cut-point. Some caution is warranted when considering our findings for the age subgroups; there were limited studies where the average participant age was ≥65 years, as specified for our subgroup analysis. Moreover, we relied on between-study differences for these subgroup analyses rather than within-study analysis for individual programs. Many trials included a broad range of ages up to 72 years, and the median age of the entire sample in this review was 58; the overall applicability of the results for KQ5 appear to apply to middle- and older-aged adults. Results may have differed for other patient-important outcomes such as quality of life; however, there were insufficient data for these analyses.

The findings for ethnicity need to be interpreted in light of our method of analysis and differences in baseline glycemic control between subgroups. Glycemic control appeared to be worse for the minority (HbA_1c=8.80 percent) compared with the majority/white (HbA_1c=7.60 percent) subgroup; it is thus hard to distinguish if ethnicity or glycemic control is more likely to have the greater influence in moderating program effectiveness. Ethnic minority groups have been shown to have higher HbA_1c levels than Caucasian groups; this finding holds after adjusting for factors affecting glycemic control (i.e., age, sex, BMI, duration of disease, mean plasma glucose) and thus may not be influenced by behavioral programs.³⁰¹ Conversely, a systematic review by Nam et al.³⁰² which found benefit for culturally tailored diabetes education, found that lower baseline HbA_1c levels better predicted positive responses to the programs. There are likely additional factors involved. Many investigators enrolling a large proportion of ethnic minorities in the trials included in this review adapted programs in ways to make them more culturally and linguistically acceptable—often including peers in the delivery or social support groups—which may have enhanced their effectiveness. Our reliance on study-level data to create subgroups (i.e., the entire study was delivered to minorities) may have limited our ability to capture differences in effects from programs delivered to a wider population base, which may reflect routine practice in many community health settings.

Although our discussion has centered on our findings related to our KQs, which focus on effect moderation, the important benefits shown by the LookAHEAD research group²⁵² should be highlighted. Reduction in retinopathy by 14 percent and nephropathy by 31 percent in those participating in a long-duration, intensive lifestyle program cannot be ignored.^278,293 Additionally, our findings from pairwise meta-analysis of 14 percent reduced mortality between those receiving behavioral programs and active controls was heavily influenced by the large weight (contributing to >50 percent of the pooled effect) of this study in the analysis.

Type 1 Diabetes

The inclusion criteria for most studies did not specify a minimum HbA_1c level; however, for all studies the mean HbA_1c was over 7 percent. For most (70 percent), the mean HbA_1c was over 8.5 percent. The results of this report may only be applicable to individuals with suboptimal and poor glycemic control.

For studies targeting youth, the mean age across most studies ranged from 12 to 15 years. Therefore, the results should be generally applicable to older children and adolescents. One trial targeted younger children (8 to 12 years);¹⁰⁰ it is unclear whether the results of this report are applicable to younger children.

For studies targeting adults, the mean age across studies ranged from 30 to 49 years. No studies specifically targeted older adults (≥65 years), therefore it is unclear if the results are applicable to older adults.

Approximately 50 percent of studies specified that participants have a minimum duration of T1DM of ≥1 year. For studies that targeted youth, the mean duration of diabetes ranged from 2.7 to 7.3 years. The results of this report may only be applicable to children and adolescents who have been diagnosed with T1DM for at least 2 years. For studies that targeted adults, the mean duration of diabetes ranged from 7.5-23 years. It is unclear whether the results of this report are applicable to adults whose T1DM has been recently diagnosed.

We did not find evidence to confirm or refute whether behavioral programs are more or less efficacious for other subgroups, including sex or racial or ethnic minorities.

All of the studies targeting adults were conducted in the United Kingdom, Europe, or New Zealand. It is unclear whether the results from these studies are applicable to community health settings in the United States. For youth, most studies (70 percent) were conducted in the United States; the remaining studies were conducted in Europe and Australia. Despite potential differences in settings and health systems, results were similar across the studies.

The studies were conducted primarily in outpatient diabetes clinics affiliated with a secondary or tertiary care hospital. Our findings are generally applicable to these settings in the United States.

Type 2 Diabetes

The range of baseline HbA_1c in the included RCTs was 6.3-12.3 percent (median=8.0) which would appear to make the results of this review applicable to the majority of people enrolling in behavioral programs. We conducted subgroup analyses for KQ 6 based on baseline glycemic control (<7 vs. ≥7 percent HbA_1c) at the study level, which provided some insight into the relative effectiveness based on this level of glycemic control. This analysis may be limited by the small number of studies in the <7 percent subgroup (n=9 RCTs) and because the analysis was based on between-study rather than within-study variability in glycemic control which may not accurately reflect differences for individual programs. The results of this report are therefore most applicable to people having HbA_1c levels ≥7 percent.

The range of mean ages in the included studies was 45-72 years (median=58), therefore the results of the pairwise meta-analyses on overall effectiveness and of the analysis for KQ 5 are most applicable to middle- and older-aged adults. Our subgroup analysis for KQ 6 based on age (<65 vs. ≥65 years) provided some data on the relative effectiveness for these age groups, but similar to that for baseline HbA_1c, may be limited by the small sample of studies on older adults (n=9) and our analytical approach. Our exclusion criteria related to duration of diabetes (mean <1 year)—implemented in order to capture programs providing training in ongoing self-management and lifestyle behaviors—limits the relevance of this review for newly diagnosed patients. The mean duration of diabetes ranged from 1-18 years with a median of 8.1 years. No study performed subgroup analysis based on duration of diagnosis (≤1 vs. >1 year) and we were unable to perform this at the study level because the mean in all cases was above 1 year. The results appear to be applicable to both men and women, and for people on a variety of diabetes treatment regimens (19.2 percent were on insulin). Overall, there was fairly good representation of individuals reporting a minority racial/ethnic background. Subgroup analysis based on those studies reporting of race/ethnicity (24 comparisons for <75 percent minorities vs. 33 comparisons for ≥75 percent minorities) was conducted to increase the relevancy of the findings to these population groups.

The results seem applicable to community health settings in the United States. The majority (63 percent) of trials were conducted in the United States, and based on our inclusion criteria related to Human Development Index⁶² all studies were performed in countries of similar development status. Some trials were conducted in academic settings in health fields—thought to have application in community health settings—although there may be some differences if these programs were delivered in different settings. Although details were reported inconsistently, health systems differences (i.e., usual care) may vary widely between study populations and could potentially influence the results obtained from behavioral programs. The effect from this difference should be minimal for this review, since we limited our results to changes from baseline between groups randomly assigned and judged to receive similar medical care.