Clinical Trials

Twenty-three RCTs^{83-90,92,93,96-104,106,108,110,111} and six associated publications^{117-119,121,122,125} examined the effectiveness of behavioral programs among youth; only one study examined children, hence our use of the term youth to categorize these studies. Most RCTs were two-arm trials and focused on DSME compared with usual care. One RCT compared a DSME program delivered in person compared with delivery using Skype⁹⁰ and another compared delivery of DSME in person compared with a telephone support active control.⁸⁷ Two three-arm trials compared a DSME program with usual care and an active control (basic education program),^83,108 although the authors of one combined the two control groups for their analyses.⁸³ Sixteen trials were conducted in the United States;^{83,84,86-88,90,92,96-100,103,104,108,110} six were conducted in Europe,^{85,89,93,101,102,106} and one was conducted in Australia.¹¹¹

The mean age of the youth participants ranged from 9.7–15.4 years (median=13.4). One study did not report age.⁹⁷ The percentage of males ranged from 0–63 percent (median=47). The proportion of nonwhite participants was between 2–82 percent (median=23.5); nine trials did not present information on race or ethnicity.^{90,93,98-101,106,110,111} For most trials, the mean HbA_1c was >7 percent and ranged from 7.4–15.7 percent (median=9.6 percent). One trial did not report absolute baseline HbA_1c.¹⁰³

All trials in youth recruited patients/families from outpatient clinical settings providing usual care throughout the study period. Clinical settings mostly consisted of diabetes/endocrinology clinics located at university-affiliated hospitals, and care was commonly described to include quarterly clinic visits with a multidisciplinary team of providers offering education and additional consults as needed. One study's usual care included eight visits over a one-year period.⁹³ Some studies reported additional components including: regular adherence assessments,^83,98 in-clinic goal setting and a daily phone hour with education provided between visits,^87,88 access to an emergency hotline,⁸⁹ and basic care coordination with clinic reminders and assistance with scheduling appointments.^96,104 Three trials reported that usual care included more advanced education,^87,88,108 and one multicenter trial's exclusion criteria for study centers included the availability of a group education program.⁸⁵

A basic description of the behavioral programs delivered to youth is provided in Appendix F (Table F1). Although all studies included in the review evaluated programs which, as reported, met our operational definition of a behavioral program, there was considerable diversity in terms of the program content and delivery. Some programs were designed to coincide with office/clinic visits; however, there was variability in the degree of integration with medical care and in program intensity. Some programs were fully integrated into the clinic visit and were delivered by the clinic's health care personnel.^93,101,102 Other programs were delivered by non-clinic staff (e.g., trained research assistant, internists) either prior to or after the patient was seen by the health care team.^{83,92,96,98,104} One study combined in-clinic goal setting with automated weekly delivery of tailored education and support messages.⁸⁹ Two office-based programs had relatively high intensity with more than 10 contacts.^93,96 The majority of office-based programs were delivered to the family, with a focus on family teamwork, conflict, and coping.^{83,92,96,98,101,102,104} Programs that did not coincide with clinic/office visits largely consisted of weekly or monthly sessions incorporating various behavioral approaches such as problem-solving, coping, and empowerment training.^{84,86,97,99,100,103,106,110,111} Some also offered a more therapeutic approach together with some degree of self-management training (i.e., behavioral family systems therapy,^90,99,108 motivational enhancement therapy combined with solution-focused therapy,⁸⁵ and multisystemic therapy^87,88). Many programs were targeted at adolescents,^{83,84,87-90,92,93,97,99,102,103,106,108,110,111}while others were tailored to children,¹⁰⁰ or offered to mixed age groups.^{85,96,98,101,104} Below, we present a summary of the program delivery factors.

The total duration of the behavioral programs ranged from 1.2–25 months (median=5.6). The number of contact hours ranged from 1–48 hours (median=9.5). Four trials did not report enough information to calculate the number of contact hours.^84,89,98,103

Five trials delivered the programs to youth only;^{84,86,89,103,106} 16 delivered the programs to both youth and their parents or family members.^{83,85,87,88,90,92,93,96-102,104,108} Four trials delivered the program in person to groups of youth only,^{84,86,106,111} and two trials delivered the program to youth using a mix of in-person sessions supplemented by telephone calls¹⁰³ or text messaging.⁸⁹ Eight trials delivered the program in person to individual pairs of youth and family members.^{83,87,88,90,93,98,108,110} Six trials delivered the program in person to groups of youth and family members.^{85,97,100-102,106} Three trials delivered the program to individual pairs of youth and family members using a mix of in-person sessions supplemented by telephone calls.^92,96,104 Two trials delivered the program to individuals using telehealth⁹⁹ and Skype.⁹⁰

For eight trials, the program was delivered by a single health care professional (e.g., nurse, psychologist, registered dietitian).^{84,86-88,90,106,110,111} Six trials engaged two or more health professionals,^{85,89,93,101,102,108} seven trials used non-health professionals (e.g., research assistants, health-related students or trainees),^{83,92,96,98,99,103,104} and one trial used a combination of a health professional and a trainee.⁹⁷ One trial did not report this information.¹⁰⁰

All of the behavioral programs had some degree of tailoring in terms of their content (e.g., individualized goal setting, topics based on age group) and/or delivery (e.g., coinciding with office visits, number of visits determined based on needs assessment). Several had a moderate–to–high level of tailoring in both content and delivery.^{87,88,90,92,93,96,97,99,103,104,108,110} Four interventions included some degree of community engagement, including involvement of peers and/or school personnel.^85,87,88,90

Observational Studies

Two controlled before-after studies explored the effectiveness of behavioral programs delivered to youth and their parents or families. One study compared a DSME intervention with usual care;¹¹⁵ the other compared a lifestyle intervention with usual care.¹¹⁴

The study by Viner et al.¹¹⁵ was conducted in the United Kingdom. The target population was youth with poor glycemic control (HbA_1c >8.5 percent). The mean ages were 13.0 and 13.1 years for the intervention and control groups, respectively; mean HbA_1c was 10.2 and 10.0 percent for the intervention and control groups. The 1.5-month program was delivered in person to groups of youth (6 meetings) and, separately, to groups of parents (1 meeting). The program was based on motivational and solution-focused techniques, with elements of cognitive behavioral therapy. The content of the program was tailored to youth with adherence issues and also targeted changes at self-identified behaviors. No information was reported for community engagement.

The study by Thomas-Dobersen et al.¹¹⁴ examined a lifestyle program that targeted overweight adolescents; body mass index ranged from 22–36 kg/m². The study was conducted in the United States. The mean ages were 13.9 and 15.2 years and mean HbA_1c was 12.2 and 13.1 percent for the intervention and control groups, respectively. The 3-month program was delivered by a multidisciplinary team in person to groups of adolescents and, in separate group sessions, to their parents. Program content was tailored to adolescents with diabetes although there was minimal tailoring in the delivery of the structured group sessions. No information was reported for community engagement.

Clinical Trials

Seven RCTs^{82,91,94,95,105,107,109} with four associated publications,^{116,120,123,124} and one non-RCT¹¹² examined the effectiveness of behavioral programs among adults. Two RCTs included participants with T2DM. One RCT presented results for HbA_1c separately for T1DM and T2DM and is included in both sections of this report.¹⁰⁷ The other study did not report results separately for T1DM or T2DM; however, the majority (>75 percent) of participants had T1DM so we have included it in this section of the report.⁹⁵ Six trials focused on DSME compared with usual care,^{82,94,95,105,109,112} two examined DSME compared with one⁹¹ or two¹⁰⁷ active controls, and one compared a lifestyle intervention with usual care.¹⁰⁵ Six of the trials were conducted in European countries,^{82,91,94,95,109,112} one was conducted in the United States,¹⁰⁷ and one was conducted in New Zealand.¹⁰⁵

The mean age of participants ranged from 30–49 years. The percentage of males ranged from 35–62 percent. The proportion of nonwhite participants was between 4.5–25 percent in two trials;^94,107 the other trials did not present information on race or ethnicity. For all trials, the mean HbA_1c was >7 percent and ranged from 7.7–9.6 percent. The mean BMI ranged from 24.8–27.6 kg/m²; three trials did not report BMI.^95,109,112

Similar to the trials in youth, usual care was usually provided by out-patient diabetes clinics/centers from which the participants were recruited. Usual care was not described by Karlsen et al.⁹⁵ who took a different approach by recruiting survey respondents, and may have been diverse in the trial of Perry et al.¹⁰⁵ which supplemented clinic recruitment with that from radio and newspaper advertisements. Visit frequency was described less often, but for half of the studies was biannually to quarterly.^{94,105,107,112} The usual care in one trial included provision of and training in a continuous glucose monitoring system.⁸²

A basic description of the behavioral programs delivered to adults is provided in Appendix F (Table F2). Several of the programs incorporated elements of cognitive behavioral therapy,^82,94,95,107 with one combining cognitive behavioral therapy with motivational enhancement therapy.⁹⁴ In one study authors described their program as taking an empowerment approach,⁹¹ another incorporated guided self-determination group training,¹⁰⁹ and one offered self-management training using an ongoing self-help group style.¹¹² The program presented by Amsberg et al.⁸² included a 9-month maintenance period during which telephone support calls were provided; this study also incorporated training using a continuous glucose monitoring system. Below, we present a summary of implementation factors.

The total duration of the behavioral programs ranged from 1.5–12 months (median=6 months). The number of contact hours ranged from 9–52 hours (median=16). One trial included an intense phase (2 months) followed by a 9-month support period.⁸² Five trials delivered the program in person to groups of participants,^{91,95,107,109,112} two delivered the program in person to individuals,^94,105 and one trial used a mix of individual and small group sessions that were delivered in person and by telephone.⁸² For three of the trials, the program was delivered by a single health care professional (i.e., nurse, registered dietitian, physician).^91,94,112 Four trials engaged two or more health professionals,^{82,105,107,109} and one trial used a health care professional and a peer (with diabetes and trained in program delivery) who served as coleader. All reports described the programs to have a moderate-to-high degree of tailoring of content to the participants' individual needs; fewer had mechanisms (e.g., telephone followup, collaborative delivery by professional and participants) to tailor the delivery of the program.^{82,95,109,112} One trial incorporated community engagement through the use of a peer coleader;⁹⁵ the remaining trials either involved no community engagement or did not report this information.

Observational Studies

One controlled before-after study explored the effectiveness of a DSME program among adults (≤65 years) who were receiving intensive insulin therapy.¹¹³ The study was conducted in Italy. Baseline HbA_1c was ≥7.5 percent in 59 and 63 percent of the intervention and control groups, respectively. The 4-month intervention was an education program including empowerment group teaching and situation simulation, and comprised eight 2-hour group sessions led by a physician or dietitian. There was some tailoring of the content towards patients receiving intensive therapy; no information was reported for community engagement.

HbA_1c: Behavioral Programs Compared With Usual Care

Figures 5-7 present our meta-analyses and forest plots of trials reporting HbA_1c stratified by age (youth and adults). A negative MD represents a greater reduction in percent HbA_1c for the behavioral program compared with usual care. We present separate forest plots for different timepoints—end of intervention, 6-month postintervention followup, and 12-month postintervention followup. We provide a narrative summary of the four RCTs that reported outcomes for longer followup timepoints.

Figure 5

Behavioral programs for type 1 diabetes compared with usual care: HbA_1c at the end of intervention. CI = confidence interval; HbA_1c = hemoglobin A_1c; n = number of participants; SD = standard deviation

Figure 6

Behavioral programs for type 1 diabetes compared with usual care: HbA_1c at 6-month postintervention. CI = confidence interval; HbA_1c = hemoglobin A_1c; n = number of participants; SD = standard deviation

Figure 7

Behavioral programs for type 1 diabetes compared with usual care: HbA_1c at 12-month postintervention (youth only). CI = confidence interval; HbA_1c = hemoglobin A_1c; n = number of participants; SD = standard deviation

At the end of intervention for youth and adults combined, our meta-analysis (16 trials, 1,155 subjects) found no difference in percent HbA_1c between individuals receiving a behavioral program and those receiving usual care (MD, -0.11; 95% CI, -0.33 to 0.11).^{82-84,89,93-96,98,99,101,105,106,108,110,112} There was no difference between groups for youth (11 trials, 653 subjects)^{83,84,89,93,96,98,99,101,106,108,110} or for adults (5 trials, 502 subjects)^{82,94,95,105,112}—MD = 0.00 (95% CI, -0.33 to 0.33) and MD = -0.28 (95% CI, -0.57 to 0.01), respectively.

At the end of 6-month postintervention followup for youth and adults combined, our meta-analysis (12 trials, 1,463 subjects) showed that HbA_1c improved for persons who received a behavioral program compared with those receiving usual care (MD, -0.31 percent; 95% CI, -0.47 to -0.15).^{84,86,88,93,94,100,102-104,108,109,111} The reduction in HbA_1c was not clinically important. For youth (10 trials, 1,213 subjects),^{84,86,88,93,100,102-104,108,111} the difference between groups was statistically significant, but it was not clinically important (MD, -0.28 percent; 95% CI, -0.51 to -0.05). For adults (2 trials, 250 subjects), there was no difference between groups.^94,109

At the end of 12-month postintervention followup for youth, our meta-analysis (7 trials, 1,333 youth) found no difference in HbA_1c between individuals receiving a behavioral program and those receiving usual care (MD, -0.22 percent; 95% CI, -0.49 to 0.05).^{83,85,102-104,108,111}

Four studies provided data at longer followup timepoints (data not shown). Three RCTs (2 youth,^103,104 1 adult;⁹⁴ 671 subjects) reported data at more than 1 year, but less than 2 years; there was no difference in HbA_1c between groups (MD, -0.40; 95% CI, -0.92 to 0.12). Two trials (1 youth,⁸⁵ 1 adult;⁹⁴ 467 subjects) reported outcomes at 24 months and found no difference in HbA_1c (MD, -0.08; 95% CI, -1.96 to 1.8).

One trial in adolescents did not report sufficient data to be included in our meta-analysis; the authors found no statistically significant difference between groups at 6-month followup.⁹⁷

Three observational studies (2 youth,^114,115 1 adult;¹¹³ 148 subjects) provided data on HbA_1c at 12-month followup. One youth study (41 subjects) reported a statistically significant and clinically important improvement in HbA_1c for the group receiving the behavioral program (MD, -1.2; 95% CI, -2.24 to -0.16).¹¹⁵ The other youth study (17 subjects) found no difference between groups (MD, 0.67; 95% CI, -1.47 to 2.81).¹¹⁴ The study that was conducted in adults (90 subjects) reported a statistically significant and clinically important improvement in HbA_1c for the group receiving the behavioral program (MD, -0.70; 95% CI, -1.31 to -0.09).¹¹³ These results should be interpreted with caution because of concerns with bias and confounding in observational studies; the only study assessed as having low risk of bias found no difference.¹¹⁴

HbA_1c: Behavioral Programs Compared With Active Control

Figures 8-10 present our meta-analyses of trials reporting HbA_1c for youth and adults in comparisons with active controls. We present the results by followup timepoint (end of intervention, 6-month followup, 12-month followup) and age group. One trial in adults was a three-arm trial comparing a behavioral program to two different active controls (didactic education to either groups or individuals); these arms were combined for the meta-analysis.¹⁰⁷

Figure 8

Behavioral programs for type 1 diabetes compared with active control: HbA_1c at end of intervention. CI = confidence interval; HbA_1c = hemoglobin A_1c; n = number of participants; SD = standard deviation

Figure 9

Behavioral programs for type 1 diabetes compared with active control: HbA_1c at 6-month postintervention. CI = confidence interval; HbA_1c = hemoglobin A_1c; n = number of participants; SD = standard deviation

Figure 10

Behavioral programs for type 1 diabetes compared with active control: HbA_1c at 12-month postintervention. CI = confidence interval; HbA_1c = hemoglobin A_1c; n = number of participants; SD = standard deviation

At the end of intervention, our meta-analysis for youth and adults (4 trials, 419 youth^87,92,108 and 110 adults¹⁰⁷) found no difference between behavioral programs and active controls for HbA_1c (MD, -0.32; 95% CI, -0.97 to 0.33). When examining the results by age subgroups, similar results were found for youth (MD, -0.33; 95% CI, -1.65 to 0.99; I²=69%).^87,92,108 and adults (MD, -0.35; 95% CI, -0.81 to 0.11).¹⁰⁷

At the end of 6 months postintervention, our meta-analysis for youth and adults combined (4 trials [259 adults,^91,107 208 youth^92,108]) showed that HbA_1c improved for those receiving a behavioral program compared with those receiving an active control (MD, -0.44; 95% CI, -0.69 to -0.19); this reduction in HbA_1c is clinically important. For youth, the difference was not statistically significant (MD, -0.60; 95% CI, -2.56 to 1.36);^92,108 for adults, the difference was not statistically significant and the effect size was not clinically important (MD, -0.38; 95% CI, -0.93 to -0.17).^91,107

At the end of 12-month followup, our meta-analysis for youth and adults combined (3 trials [110 adults,¹⁰⁷ 195 youth^92,108) found no difference in HbA_1c (MD, -0.44; 95% CI, -1.04 to 0.16). For youth, the difference was statistically significant and clinically important (MD, -0.52; 95% CI, -1.04 to 0.00); the behavioral program studied by Weinger et al.,¹⁰⁷ failed to demonstrate any difference (MD, -0.14; 95% CI, -0.61 to 0.33).

HbA_1c: Comparative Effectiveness of Two Behavioral Programs

One RCT (72 youth) examined the same DSME program delivered in person compared with delivery by Skype.⁹⁰ There was no difference in HbA_1c between groups at the end of intervention (MD, -0.04; 95% CI, -0.87 to 0.79) or at 6-month followup (MD, -0.24; 95% CI, -1.10 to 0.62).

Adherence to Diabetes Self-Management: Behavioral Programs Compared With Usual Care

This section presents the results from trials that reported on adherence to diabetes self-management. This outcome was measured in a number of ways and we report them separately. The most common measure was self-monitoring of blood glucose (SMBG) and was most commonly reported as the frequency of blood glucose testing over 1 day.^{84,86,88,96,104} Two studies reported the frequency of testing over the past week;^93,109 we converted this to the number of tests per day. We present separate forest plots for different timepoints (end of intervention, 6 month followup). We provide a narrative summary of the one RCT that reported outcomes for longer followup.

At the end of intervention (Figure 11), our meta-analysis (4 trials, 282 youth) found no difference in frequency of SMBG between youth receiving a behavioral program and those receiving usual care (MD, 0.15; 95% CI, -0.54 to 0.84).^84,88,93,96

Figure 11

Behavioral programs for type 1 diabetes compared with usual care: self-monitoring of blood glucose (tests per day) at end of intervention. CI = confidence interval; n = number of participants; SD = standard deviation

At the end of 6-month postintervention for youth and adults combined (Figure 12), our meta-analysis (5 trials [4 youth,^84,86,88,93 1 adult¹⁰⁹], 252 subjects) found no difference in SMBG between individuals receiving a behavioral program and those receiving usual care (MD, 0.40; 95% CI, -0.36 to 1.16). Adults receiving the behavioral program in the trial of Zoffmann et al.¹⁰⁹ increased their frequency of SMBG (MD, 1.42; 95% CI, 0.11 to 2.75).

Figure 12

Behavioral programs for type 1 diabetes compared with usual care: self-monitoring of blood glucose (tests per day) at 6-month postintervention. CI = confidence interval; n = number of participants; SD = standard deviation

One trial (390 youth) reported SMBG at 24-months postintervention.¹⁰⁴ The results showed individuals receiving the behavioral program performed more poorly than those receiving usual care (MD, -0.36; 95% CI, -0.69 to -0.03).

Two trials in adults measured adherence of blood glucose testing using an item from the Summary of Diabetes Self-Care Activities (SDSCA) questionnaire.¹²⁶ This self-report measure assesses the number of days in the previous week that SMBG was practiced. At the end of intervention one trial (74 adults) found that those in the behavioral program reported performing SMBG 1.4 days (95% CI, 0.35 to 2.43) more than those receiving usual care.⁸² At 6-month postintervention, one trial (244 adults) found no difference between groups (MD, -0.06; 95% CI, -0.60 to 0.48).⁹⁴

Four trials in youth used the Diabetes Self-Management Profile (DSMP)¹²⁷ to assess adherence to the diabetes regimen at different timepoints. At the end of intervention, Wysocki et al.¹⁰⁸ (54 youth) reported a clinically important improvement in the overall DSMP score for those who received the behavioral program compared with those receiving usual care (MD, 5.00; 95% CI, 0.60 to 9.40). This difference had disappeared by 12-month postintervention. Two studies assessed adherence at 6-month postintervention followup; we did not pool the results as the studies reported different summary measures. In 2012, Nansel et al.¹⁰⁴ (390 youth) found no difference between groups (MD, 1.31; 95% CI, -1.12 to 3.74). In an earlier study, Nansel et al.¹⁰³ (81 youth) reported the proportion of adherence to an optimal diabetes regimen using the modified DSMP. They found no difference between groups (MD, -0.03; 95% CI, -0.06 to -0.01). The fourth study reported that there was no difference between groups on the DSMP at end of intervention; however, the authors did not provide any data.⁹⁹

Two trials reported on adherence to medication. One trial (190 youth) used a questionnaire item to assess the number of times youth skipped an insulin dose in the past month.⁸⁵ The authors reported that the odds of skipping one or more doses compared with no doses of insulin at 12-month followup was 0.82 (95% CI, 0.48 to 1.38) and at 24-month followup was 1.30 (95% CI, 0.78 to 2.17) for the group receiving the behavioral program. One trial in adults (74 adults) used the medication item of the Diabetes Self-Care Inventory¹²⁸ and found no difference at the end of intervention between those receiving the behavioral program and those receiving usual care (MD, 0.22; 95% CI, -0.60 to 1.04).⁸²

Adherence to Diabetes Self-Management: Behavioral Programs Compared With Active Control

One trial (149 adults) found no difference in frequency of SMBG between groups at 6-months postintervention (MD, -0.20; 95% CI, -0.76 to 0.36).⁹¹ The same trial measured adherence to several diabetes self-care activities using the SDSCA and found no difference between groups at 6-month postintervention (MD, 0.00; 95% CI, -0.35 to 0.35).⁹¹

One trial (54 youth) used the DSMP to assess adherence to the diabetes regimen.¹⁰⁸ At the end of intervention and 12-month followup, Wysocki et al.¹⁰⁸ found no difference between the group that received the behavioral program compared with those receiving an active control—MD = 2.40 (95% CI, -2.46 to 7.26) and MD = 2.00 (95% CI, -3.78 to 7.78), respectively).

One trial (149 youth)⁹² used the Diabetes Behavior Rating Scale, which reflects the frequency of routine diabetes care behaviors over the previous week.¹²⁹ No data were provided; however, the authors reported that at end of intervention, and 6- and 12-month followup, those receiving the behavioral program performed more poorly that than those in the active control group.

Adherence to Diabetes Self-Management: Comparative Effectiveness of Two Behavioral Programs

One RCT (71 youth) studied the same DSME program delivered in person compared with delivery by Skype.⁹⁰ The authors used the DSMP to assess adherence and found no difference between the groups at the end of intervention or at 6-month followup (MD, 0.85; 95% CI, -4.56 to 6.26 and MD, 0.74; 95% CI, -4.97 to 6.45, respectively).

Other Clinical and Behavioral Outcomes

Table 4 summarizes the results for other clinical and behavioral outcomes. For most outcomes results were reported in single trials.

Table 4

Other clinical and behavioral outcomes for type 1 diabetes.

Health-Related Quality of Life: Behavioral Programs Compared With Usual Care

Studies reporting on HRQL assessed this using generic and diabetes-specific quality of life measures. Generic HRQL was measured by a number of tools (e.g., World Health Organization Well-Being Index,¹³⁰ Pediatric Quality of Life [PedsQL],¹³¹ Wellbeing Questionnaire¹³²), as was diabetes-specific HRQL (PedsQL diabetes module,¹³¹ Pediatric Diabetes Quality of Life, Well-being Enquiry for Diabetes¹³³). A group of studies reported on diabetes distress/stress (tools included Problem Areas in Diabetes¹³⁴ and Diabetes Stress Questionnaire⁸⁴), for which we analyzed separately from diabetes-specific HRQL. For all analyses we present the results as SMD. Figure 13 presents our meta-analyses of trials, stratified by age (youth and adults), that reported generic HRQL at end of intervention. Longer-term followup results were reported for generic HRQL and are summarized in Table 5. The meta-analysis results in Figure 14 for diabetes-specific HRQL at end of intervention were not stratified by age. Figures 15 and 16 present the meta-analyses for diabetes distress at end of intervention (stratified by age) and 6-month followup, respectively.

Figure 13

Behavioral programs for type 1 diabetes compared with usual care: generic health-related quality of life at end of intervention. CI = confidence interval; n = number of participants; SD = standard deviation

Table 5

Behavioral programs for type 1 diabetes compared with usual care: generic health-related quality of life at 6-, 12-, and 24-month postintervention.

Figure 14

Behavioral programs for type 1 diabetes compared with usual care: diabetes-specific health-related quality of life at end of intervention. CI = confidence interval; n = number of participants; SD = standard deviation; Std = standardized

Figure 15

Behavioral programs for type 1 diabetes compared with usual care: diabetes distress/stress at end of intervention. CI = confidence interval; n = number of participants; SD = standard deviation; Std = standardized

Figure 16

Behavioral programs for type 1 diabetes compared with usual care: diabetes distress at 6-month postintervention followup. CI = confidence interval; n = number of participants; SD = standard deviation; Std = standardized

At the end of intervention for youth and adults combined (Figure 13), our meta-analysis (7 trials [5 youth,^93,96-98,110 2 adult^82,95], 474 subjects) found no difference in generic HRQL between individuals receiving a behavioral program and those receiving usual care (SMD, 0.10; 95% CI, -0.18 to 0.38). The lack of difference remained for the subgroups of adults (2 trials, 137 subjects; MD, 0.35; 95% CI -1.93 to 2.63)^82,95 and youth (5 trials, 337 subjects; MD, 0.01; 95% CI -0.33 to 0.35).^93,96-98,110

Three RCTs in youth reported on generic HRQL for longer followup timepoints (Table 5).^85,93,98 There was no difference in HRQL between groups at any of the timepoints.

Diabetes-specific HRQL was reported by three trials at the end of intervention (Figure 14). Our meta-analysis of these trials (2 youth,^97,110 1 adult,¹¹² 212 subjects) found no difference between behavioral programs and usual care (SMD, 0.08; 95% CI, -1.44 to 1.60; I²=73%). One observational study in adults (90 subjects) found no difference between groups at 12-months postintervention (SMD, 0.03; 95% CI, -0.39 to 0.45).¹¹³

Distress/stress was reported for six trials; negative scores represent reduced distress. At end of intervention (Figure 15), our meta-analysis for youth and adults combined (4 trials [2 youth,^84,93 2 adults^82,95], 209 subjects) found no statistically significant difference in diabetes distress for behavioral programs compared with usual care (SMD, -0.31; 95% CI, -0.83 to 0.21). Stratified by age, there was no difference for the studies of youth (SMD, -0.21; 95% CI, -2.84 to 2.60) or adults (SMD, -0.41; 95% CI, -3.78 to 2.96; I²= 57%). At 6-month followup for youth and adults combined (4 trials [3 youth,^84,93,111 1 adult¹⁰⁹], 236 subjects), changes to diabetes distress did not differ for behavioral programs compared with usual care (SMD, -0.28; 95% CI, -0.94 to 0.38) (Figure 16).

Health-Related Quality of Life: Behavioral Programs Compared With Active Control

One trial in youth failed to demonstrate a difference in diabetes-related quality of life between a behavioral program and an active control at 12-month followup (130 subjects; insufficient data reported to calculate SMD).⁹²

Diabetes-Related Health Care Utilization: Behavioral Programs Compared With Usual Care

Diabetes-related health care utilization was reported infrequently and only for trials comparing behavioral programs to usual care. We summarize the results in Table 6. One RCT in youth found a reduced risk of diabetes-related hospital admissions at end of intervention and at 6-month followup for those receiving behavioral programs compared with usual care.⁸⁸ The same trial also reported fewer admissions to the emergency department at the end of intervention. Another RCT in youth⁸⁵ and one in adults⁹⁴ found no difference in hospital admission at any timepoint. One trial reported that there was no difference in the number of diabetes-related hospital and emergency department admissions at the 6-month followup; however, the authors did not provide any data.⁹⁷

Table 6

Behavioral programs for type 1 diabetes compared with usual care: diabetes-related health care utilization at end of intervention, 6-, 12-, and 24-month postintervention followup.

Program Acceptability: Behavioral Programs Compared With Usual Care

Figure 17 presents our meta-analysis stratified by age (youth and adults) of trials that reported participant attrition at their longest followup timepoint. Our meta-analysis (21 trials, 2,503 subjects) found a 21 percent increased risk of attrition for individuals receiving usual care compared with those receiving the behavioral program (RR, 1.21; 95% CI, 1.05 to 1.39).^{82-86,88,89,93-95,99,100,102-106,108-111}

Figure 17

Behavioral programs for type 1 diabetes compared with usual care: participant attrition. CI = confidence interval; n = number of participants

Program Acceptability: Behavioral Programs Compared With Active Control

Three RCTs (218 youth^87,108 and 160 adults⁹¹) compared behavioral programs with active comparators. The pooled analysis (data not shown) found no difference between the groups for participant attrition (RR, 1.05; 95% CI, 0.46 to 2.4).

Program Acceptability: Comparative Effectiveness of Two Behavioral Programs

One RCT (72 youth) compared the same DSME program delivered in person compared with delivery by Skype.⁹⁰ There was no difference between the groups in participant attrition (RR, 0.55; 95% CI, 0.28 to 1.11).

Age

In KQ 1, we presented our results by age groups (youth and adults). Behavioral programs appeared to be more effective in reducing HbA_1c for adults than for youth at end of intervention when compared to usual care (Figure 5); the effect size in the meta-analysis for adults^{82,94,95,105,112} was greater in absolute terms than for the youth^{83,84,89,93,96,98,99,101,106,108,110} (MD = -0.28 vs. 0.00 respectively); the results for adults approached statistical significance and the 95% CI contained our threshold for clinical importance. At 6-month followup, the effect sizes for youth^{84,86,88,93,100,102-104,108,111} and adults^94,109 appeared similar (MD = -0.28 vs. MD = -0.38, respectively); only the results for youth reached statistical significance, although the 95% CIs in both groups included a clinically important effect size favoring behavioral programs. No study in adults reported at 12-month followup; the youth results showed no difference (MD, -0.22; 95% CI, -0.49 to 0.05) although the 95% CI included a clinically important effect for behavioral programs.

When compared with active controls at end of intervention, the effect sizes for youth (MD, -0.33; 95% CI -1.65 to 0.99) and adults (MD, -0.35; 95% CI -0.81 to 0.11) were both similar to the overall effect size and nonsignificant with imprecise 95% CIs. At 6-month followup, the effect size was larger for the youth^92,108 than for the adults^91,107 (MD -0.60 vs. -0.38) but both results failed to reach statistical significance. At 12-month followup, results for youth were statistically significant and clinically important (MD, -0.52; 95% CI, -1.04 to 0.00);^92,108 for adults there was no difference at 12-month followup (MD, -0.14; 95% CI, -1.28 to 1.00).¹⁰⁷.

In the studies that included adults only, the mean age across the studies ranged from 30.3–49.2 years. None of the studies reported results separately for young adults or older adults.

Level of Glycemic Control

One RCT (101 youth) conducted a subgroup analysis of 54 youth with suboptimal baseline glycemic control (HbA_1c ≥8 percent).⁹⁶ At the end of intervention, Katz et al.⁹⁶ found that those receiving the behavioral program had greater odds of maintaining or improving their HbA_1c compared with those receiving usual care (odds ratio, 3.4; 95% CI, 1.0 to 11.9). This compares favorably to the overall study results which found no difference in change in glycemic control for the group receiving the behavioral program (MD, 0.30; 95% CI, -0.22 to 0.82). No data were reported for the subgroup of youth with optimal baseline HbA_1c. Subgroup analysis at the study level was not conducted because the mean baseline HbA_1c was >7 percent for all studies.

Other Subgroups

No data were reported for any of our other pre-specified subgroups: race or ethnicity, socioeconomic status, or time since diagnosis.

HbA_1c

Individuals receiving behavioral programs compared with usual control improved their glycemic control (i.e., reduced percent HbA_1c) at end of intervention (66 comparisons; 8,715 subjects; MD, -0.35; 95% CI, -0.56 to -0.14; I²=74%),^{135,137,139,141,142,145,147,151,153,155,160,162,171,173,175-177,179,184,188-191,197,203,205-208,210,213-220,222-226,228,229,231,233,236,239-241,247,249,253-255,257-262,265} but not at 6-month (23 comparisons; 4,138 subjects; MD, -0.16; 95% CI, -0.36 to 0.04; I²=61%)^{136,140,143,146,163,173,178,183,193-196,211,215,229,234,235,241,246, 249,259} or 12-month followup (9 comparisons; 1,494 subjects; MD, -0.14; 95% CI, -0.4 to 0.12; I²=59%).^{146,158,163,173,178,193,223,234} The results were of a smaller magnitude when behavioral programs were compared with active control groups at end of intervention (25 comparisons; 7,518 subjects; MD, -0.24; 95% CI, -0.41 to -0.07; I²=70%).^{107,154,161,164-166,168,169,174,175,184,186,191,192,198,202,230,252,265} For 6-month followup, the effect size was similar but the results reached statistical significance (6 comparisons; 595 subjects; MD, -0.19; 95% CI, -0.37 to -0.01).^{107,156,181,182,201} The estimate was nonsignificant and imprecise at 12-month followup (6 comparisons; 486 subjects; MD, -1.10; 95% CI, -2.56 to 0.36).^{107,164,192,201} No result was clinically important based on our prespecified threshold of 0.4 unit change in percent HbA_1c. The meta-analyses for HbA_1c indicated high heterogeneity in effect between studies across timepoints (I² ranged from 61–98 percent). As described in the Methods, we performed sensitivity analyses to explore this issue; however, none of the prespecified variables reduced the heterogeneity to below 50 percent so we present the original results.

In three trials (701 subjects) providing comparative effectiveness between DSME delivered to groups compared with delivery to individuals or via a mixture of individual and group delivery, there was a beneficial effect for those individuals receiving DSME in groups at up to 12-months followup (MD, -0.36; 95% CI, -0.63 to -0.08).^192,212,234 In contrast, there was a benefit at end of intervention shown in a trial comparing individual DSME and motivational interviewing with group-based empowerment DSME and supervised group exercise (143 subjects; MD, -0.30; 95% CI, -0.58 to -0.02).²⁴⁴ Several comparative effectiveness studies found no difference in HbA_1c changes between groups. Some examples include the addition of an additional treatment (e.g., problem solving therapy,¹⁶⁸ music therapy¹⁹⁹) or a support aspect to a DSME or lifestyle program;^172,227,230 others include comparisons between peer and health professional delivery of a program component (see Appendix I).^{144,172,227,256}

Six trials reported on HbA_1c but did not provide data suitable for inclusion in the meta-analysis. Five trials comparing a behavioral program with usual care did not find a significant difference between groups.^{149,157,167,187,200} One trial comparing two behavioral interventions with different delivery methods also found no difference between groups.¹⁵⁹

Visualization of funnel plots did not suggest publication bias, and using the Egger test⁷⁸ for this outcome resulted in no significant indication of bias for comparisons with usual care (p=0.25) or active controls (p=0.21) at end of intervention.

Change in Body Composition

Compared with usual care, behavioral programs assisted participants in reducing their BMI (kg·m^-2) at all three timepoints—end of intervention (36 comparisons; 4,280 subjects; MD, -0.51; 95% CI, -0.66 to -0.36),^{135,137,139,145,151,153,155,162,171,175,179,184,189,190,206,208,210,214,215,224,226,233,239-242,246,249,251,255,257,259-61} 6-month followup (14 comparisons, 1,840 subjects; MD, -0.21; 95% CI, -0.32 to -0.1),^{136,143,146,163,183,193,211,215,241,246,249,251,259} and 12-month followup (5 comparisons; 867 subjects; MD, -0.92; 95% CI, -1.44 to -0.4).^{146,157,163,193,238} When compared to active controls, behavioral programs did not reduce BMI at any followup timepoint. Body weight (kg) was reduced at end of intervention in those receiving behavioral programs compared with those receiving usual care (37 comparisons; 4,070 subjects; MD, -1.68; 95% CI, -2.06 to -1.30),^{137,141,145,147,153,160,167,176,178,184,188,190,191,200,203,205,213,214,217,222,224-226,239,246,249,254,258-263,265} or active control (15 comparisons; 6,212 subjects; MD, -1.30; 95% CI, -2.48 to -0.12; I²=78%).^{148,154,165,166,169,174,184,186,191,198,202,252,265} There was no reduction in weight at other timepoints; one trial showed an increase in weight at 12-month followup for the behavioral program compared with active control arm (95 subjects; MD, 3.70; 95% CI, 1.67 to 5.73).²⁰¹ Waist circumference (cm) was reduced at end of intervention (17 comparisons, 1,521 subjects),^{145,153,162,167,190,203,214,215,224,226,241,254,255,259,261} in those comparisons with usual care—MD = -3.17 (95% CI, -4.36 to -1.98; I²=64%). One study found significant reduction in waist circumference at 6-month followup for those receiving a behavioral program compared to an active control (38 subjects; MD, -5.70; 95% CI, -6.54 to -4.86).¹⁵⁶ There was no difference found in two studies comparing behavioral programs to usual care at 12-month followup;^157,163 no data were available at 12-month followup for studies comparing behavioral programs to active control.

One comparative effectiveness trial (99 subjects) found that BMI was reduced (MD, -1.80; 95% CI, -2.51 to -1.09) at end of intervention for individuals receiving a cognitive-behavioral-therapy based lifestyle program including a portion-controlled diet compared with DSME including a meal plan.¹⁷⁰ Participants in this study who received the lifestyle program also reduced their weight and waist circumference more than those receiving the DSME program—MD = -5.10kg (95% CI, -7.22 to -2.98) and MD = -3.60cm (95% CI, -5.33 to -1.87), respectively.

Quality of Life

Outcomes for quality of life were categorized into five subcategories based on their focus (i.e., generic vs. diabetes-specific) and the similarity between studies in measurement scales. Groups of studies reported outcome data based on the SF-36 Health Survey (physical and mental component scores), and the Problem Areas in Diabetes (PAID) scale (0–100; lower score favorable) measuring diabetes distress. Accordingly, three of our subcategories represent these tools (i.e., Quality of Life–SF36 Physical, Quality of Life–SF36 Mental, and Diabetes Distress), for which we present results as MD. Other subcategories were created to combine other generic (Quality of Life–Other; e.g., WHO Quality of Life Brief, W-BQ12, EuroQol 5D) and diabetes-specific (Diabetes-specific Quality of Life; e.g., Diabetes Quality of Life, Diabetes Distress Scale, Appraisal of Diabetes, Diabetes Symptom Checklist) quality of life questionnaires; these results are presented as SMDs.

There was no difference in Quality of Life-SF36 (Physical) or Quality of Life-SF36 (Mental) when measured at end of intervention for comparisons with usual care,^{155,214,222,239} or up to 6-months followup for comparisons with active controls.^169,181,252 There was no difference found for Quality of Life–Other in comparisons (n=7) with usual care up to 6-month followup,^{195,196,206,226,249,253} or in comparisons (n=4) with active controls up to 12-months followup.^154,192 Results favored behavioral programs compared with usual care for Diabetes Distress (8 comparisons, 1,384 subjects) at end of intervention (MD, -1.82; 95% CI, -3.43 to -0.21),^{142,147,211,218,225,226,228,233} but not at longer followup.^146,211,234 The result at end of intervention is not clinically important based on our prespecified threshold of a 0.5 SD using the mean SD of the included studies. One study (167 subjects) evaluating this outcome in a comparison to active controls found no difference at 6-month followup.¹⁸¹ There was no difference in Diabetes-specific Quality of Life at any followup timepoint to 12-month followup when comparing behavioral programs to usual care,^{146,163,175,177,189,215,253} or at end of intervention for programs compared with active controls.^154,168,175

One trial assessed the effects on quality of life when the support phase of a DSME and support program was delivered by peers, clinical practice staff, or health care professionals (diabetes educators). Siminerio et al.²²⁷ found that Diabetes Distress worsened for the group receiving support from peers when compared to the group receiving support from the educators (74 subjects; MD, 24.70; 95% CI, 15.02 to 34.38). This effect is considered clinically important. There was no difference in Diabetes Distress when delivery of nonprofessional clinic staff was compared to that by health care professionals.

Micro- and Macrovascular Complications

Authors of the LookAHEAD trial (5,145 subjects) studied outcomes of myocardial infarctions, stroke, heart failure, diabetic nephropathy, diabetic retinopathy, and diabetic neuropathy. Diabetic retinopathy was reduced by 14% (hazard ratio, 0.86; 95% CI, 0.75 to 0.98) in participants receiving their intensive lifestyle program compared with an active control (didactic education and support) over a median of 8 years.²⁷⁸ A secondary analysis of nephropathy using a post hoc outcome of very-high-risk chronic kidney disease—a combination of the a priori outcomes albuminuria and estimated glomerular filtration rate, found a lower incidence of nephropathy for the intensive lifestyle program at the 8 year end-of-intervention timepoint (risk difference 0.27 cases per 100 person-years; hazard ratio, 0.69; 95% CI, 0.55 to 0.87).²⁹³ Results for the other outcomes in this trial did not reach statistical significance—myocardial infarction (RR, 0.86; 95% CI, 0.70 to 1.05), stroke (RR, 1.06; 95% CI, 0.79 to 1.44), heart failure (RR, 0.83; 95% CI, 0.64 to 1.08), and diabetic neuropathy (RR, 1.13; 95% CI, 0.92 to 1.38).

All-Cause Mortality

One study examined all-cause mortality as an pre-specified outcome;²⁵² there were enough data in 27 reports to calculate a difference in all-cause mortality for the associated comparisons. There was no difference in all-cause mortality between participants receiving behavioral programs and usual care (25 comparisons; 4,659 subjects; RR, 1.28; 95% CI, 0.84 to 1.94); mortality between behavioral programs and active control groups (5 comparisons, 6,050 subjects) was 14 percent lower for those receiving behavioral programs (RR, 0.86; 95% CI, 0.77 to 0.96).

HbA_1c

Accounting for all variables of program components and delivery variables (Table 3) when creating the network was deemed not appropriate for various reasons. When choosing which variables to use, we prioritized them by considering factors including the: reliability and specificity with which we could categorize programs in each variable based on extent of reporting, overlap in meaning between variables, and the ability to inform those individuals making decisions to implement these programs in community settings. Deciding between program duration (months) and intensity (contact hours), the latter was chosen because it accounts for duration to some extent, aligns with our focus on interactive programs, and better enables one to estimate resource requirements in terms of personnel and space. Degree of tailoring was not chosen because every program incorporated this to some extent and categorizing this (e.g., minimal versus moderate in terms of content and delivery) was considered unreliable based on study reporting. Moreover, the use of technology (captured in the delivery method variable) was also considered a way to tailor the program to individuals, particularly in cases of poor access due to travel or time constraints. The level of community engagement was also not used because, when incorporated, this was largely via use of lay or peer providers which was captured in the delivery personnel variable. The remaining variables were placed in order (program components, program intensity, method of communication, method of delivery, and delivery personnel) and we then created nodes trying to incorporate as many variables as possible without having numerous nodes either empty (a theoretical grouping of variables that did not represent a studied program), or with only one or two programs. Dividing the data by the first variable of program components (DSME, DSME and support, and lifestyle) resulted in a relatively large number of DSME comparisons. For this group, we were able to use all five variables to create 24 potential nodes (18 which contained comparisons). We did not capture the variable of delivery personnel for the DSME and support, and lifestyle groups because most nodes would in this case contain at most one comparison.

When interpreting the results, we relied primarily on the relative ranking of the nodes, and looked for trends in the findings based on program variables that appeared to determine whether the effects would offer clinical benefit. Some nodes had very few studies, small sample sizes, and/or wide credibility intervals, thus we did not make any firm conclusions for a single node (or for differences in 561 potential comparisons) but rather from looking across nodes with similar features.

The results of the network meta-analysis indicated that, in comparison to the reference of usual care, 14 nodes produced MDs which fell at or above our clinically important threshold (0.4) for change in percent HbA_1c. Four of these nodes represent DSME, five represent DSME and support, and five represent lifestyle programs. Six nodes represent medium-intensity programs (11–26 contact hours), six represent high-intensity programs (≥26 contact hours), and two (one DSME and support, and one lifestyle) represent low-intensity programs (≤10 contact hours). The mean contact hours for the programs represented by these effective nodes was 26.4 (range 7-40.5 hours); the mean total program duration was 8 months (range 2-12). None of the nodes representing low-intensity DSME programs showed clinically important effects; all had greater impact on HbA_1c than basic educational controls, but lower impact than a stand-alone dietary or physical activity intervention. Three of four nodes representing DSME programs with MDs showing clinically important effect were delivered by health care professionals.

Eleven of the 14 nodes representing clinically important effects were delivered in person rather than incorporating some form of technology. Behavioral programs in the nodes with the highest PB (36 and 10.7 percent, respectively) were delivered in person rather than by incorporating technology. Similar observations were noted for the other four nodes having PB ≥5 percent, of which three were delivered in person and one was delivered using some form of technology; the latter group of studies provided supportive telephone calls between in-person sessions during lifestyle interventions tailored to minorities.^143,160 All effective nodes representing some use of technology were of moderate or high intensity.

An outlier having an MD of 2.80 (95% CI, 1.14 to 4.48) represented a study by Brown et al.¹⁵² which found greater HbA_1c reduction at end of intervention in a group receiving DSME compared with one receiving DSME with the addition of a care manager.

Body Mass Index

We created nodes using four variables for BMI (i.e., program component, program intensity, method of communication, and method of delivery). Of the 39 plausible nodes (each differing by only one level of one variable), there were studies with data to populate 26 nodes.

Averaging the baseline values in the studies, BMI at baseline was similar for programs classified as DSME (32.4 kg·m²), DSME and support (33.0 kg·m²), and lifestyle (32.9 kg·m²). The effect sizes for BMI from behavioral programs relative to usual care ranged between -1.77 kg·m² and 3.29 kg·m². The node with the most beneficial MD only represented one study¹⁵⁷ evaluating a low-intensity lifestyle program with multiple brief contacts over 6 months. Nodes with rank orders 2 and 3 were both lifestyle programs of low and medium intensity, respectively. The node having the most studies (n=12) represented a DSME program of medium intensity (11–26 hours) which was delivered in person to groups; the results indicated this program to have 0 percent PB. One difference between the programs in this node and those with higher PB is that the higher PB all offered some individual delivery, rather than relying only on group delivery. Likewise, the majority of nodes having the highest MDs (i.e., 8 of the highest 10) offered some individual delivery.

Glycemic Control

• In terms of overall effectiveness at longest followup for HbA_1c, participants with suboptimal 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 in a subgroup of studies having participants with suboptimal baseline glycemic control. Of the two nodes representing low-intensity programs that were found to have clinically important effects in the original network analysis, one was shown not effective for participants with suboptimal glycemic control. Active controls of dietary or physical activity interventions were not as effective for participants with suboptimal control.

Age

• 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 active controls at longest followup was clinically important.

Race/Ethnicity

• Subgroup analysis of our meta-analyses comparing behavioral programs to usual care and active controls indicated that programs offered to predominantly minority participants (≥ 75 percent nonwhite) appear to provide more benefit than those offered to populations with a lower proportion (<75 percent) of nonwhite individuals. The effect size for minority participants reached clinical importance.
• 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. The subgroup of majority/white participants appeared to benefit more from lifestyle programs than from DSME or DSME plus support programs.
• Glycemic control appeared to be worse for the minority (HbA_1c=8.8 percent) compared with the majority/white (HbA_1c=7.6 percent) subgroup.

Glycemic Control

Initially, we conducted a subgroup analysis on the outcome of HbA_1c by baseline glycemic control (HbA_1c <7 vs. ≥7 percent) using the pair-wise meta-analysis results for HbA_1c at longest followup timepoint (data not shown). For behavioral programs compared with usual care, our meta-analysis showed a small benefit (MD, -0.12; 95% CI, -0.22 to -0.01; I²=3%) for HbA_1c for participants with a baseline HbA_1c <7 percent (6 trials, 1,239 subjects);^{194,196,223,246,249,260} the analysis showed greater benefit (although not clinically important) for participants with a baseline HbA_1c ≥7 percent (76 trials; 11,086 subjects; MD, -0.32; 95% CI, -0.42 to -0.21; I²=71%). There was no difference in change in HbA_1c for persons with baseline HbA_1c <7 percent receiving a behavioral program compared with an active control (3 trials, 169 participants; MD, -1.43; 95% CI, -3.57 to 0.71; I²=99%);^174,186,201 persons with HbA_1c ≥7 percent at baseline had greater reduction in HbA_1c after receiving behavioral programs compared with an active comparator (20 trials, 7,709 subjects; MD, -0.18; 95% CI -0.30 to -0.06; I²=38%), but this was not clinically important.

To explore potential moderation of effect based on the factors of interest, we performed a subgroup analysis of our network meta-analysis described in the section for KQ5. We removed the studies in which baseline HbA_1c was <7 percent (n=9)^{174,186,194,196,201,223,246,249,260} and repeated the analysis for a subgroup with baseline HbA_1c ≥7 percent; there were an insufficient number of studies with baseline HbA_1c <7 percent to run the analysis using these studies, or to perform meta-regression analysis. The results are presented in Table J1 in Appendix J. The categorization of all nodes remained the same in relation to the variables of interest. The changes in this subgroup analysis include: 1) the effect sizes for nodes ranked 1 and 13 reduced substantially to ranks of 31 and 23 (from -1.37 to 0.09 and from -0.45 to -0.15, respectively), and 2) the active (dietary or physical activity) control became less effective (MD -0.14 vs. -0.39) for participants having ≥7 percent HbA_1c.

Age

The same set of subgroup analyses performed for baseline glycemic control was conducted for our age subgroups; the study population in nine studies reporting on HbA_1c had a mean age ≥65 years.^{147,155,166,196,203,218,221,223,230,236} We first performed subgroup analyses by age group (≥65 years vs. <65 years) using the pair-wise meta-analyses results for HbA_1c at longest followup timepoint in comparisons between behavioral programs and both usual care and active control (data not shown). For behavioral programs compared with usual care, the meta-analysis for participants <65 years indicated that HbA_1c reduced to a statistically significant extent at longest followup (76 comparisons; 11.491 subjects; MD, -0.31; 95% CI, -0.42 to -0.21; I²=72%); for older adults the results indicated no difference (7 comparisons; 734 subjects; MD, -0.24; 95% CI, -0.50 to 0.03; I²=55%). For comparisons with active controls for participants <65 years, the benefit of behavioral programs was statistically and clinically significant (26 comparisons; 7,669 subjects; MD, -0.41; 95% CI -0.70 to -0.12; I²=93%). For older adults, behavioral programs compared with an active control (3 comparisons, 206 subjects) failed to reduce HbA_1c (MD, -0.23; 95% CI, -0.60 to 0.14; I²=0%).

Subsequently, we performed a subgroup analysis for populations <65 years by removing the data from the studies (n= 9)^{147,155,166,196,203,218,223,230,236} having mean age ≥65 from our network meta-analysis described in the section for KQ5. The results are presented in Table J2 in Appendix J. The categorization of all nodes remained the same in relation to the variables of interest. The only notable change in this subgroup analysis was that the effect size for the active control of a dietary or physical activity intervention became clinically important (MD, -0.55) although the PB remained at 0 percent.

Race/Ethnicity

We conducted subgroup analyses based on race/ethnicity (i.e. ≥75 percent nonwhite [minorities] and <75 percent nonwhite participants) for the outcome of HbA_1c at longest followup for behavioral programs compared to usual care and active controls (data not shown). Using the pairwise meta-analysis for HbA_1c when comparing behavioral programs to usual care, there was a clinically important effect for minority participants (33 comparisons; 4,774 participants; MD, -0.42; 95% CI -0.56 to -0.27; I²=55%)^{137,141,143,151,153,162,171,179,188,189,195,197,205-208,210,215-219,222,228,229,231,233,240,246,247,257,262}which was greater than that seen for the comparisons with <75 percent minorities (24 comparisons; 5,110 participants; MD, -0.16, 95% CI -0.31 to 0.00; I²=75%).^{139,142,147,160,175-177,183,184,194,196,214,220,224,234,235,239,249,253,254,258,259} For comparisons between behavioral programs and active control groups, there was no statistically significant reduction in HbA_1c among minorities (5 comparisons, 400 participants; MD, -0.32; 95% CI -0.67 to 0.04; I²=0%);^{164,173,182,198,230} studies with a larger proportion of white participants also showed no difference (10 comparisons, 6,214 participants; MD, -0.50; 95% CI -1.24 to 0.23; I²=99%).^{107,168,169,175,184,201,202,252} Glycemic control at baseline appeared to be worse for the minority (8.8 percent HbA_1c) compared with the majority/white (7.6 percent HbA_1c) subgroup.

We also conducted univariate meta-regressions for each race/ethnicity subgroup. For this analysis, we used outcome data for changes in HbA_1c at longest followup in comparisons between behavioral programs and usual care. Table 14 shows the results for each variable examined. No statistically significant finding was generated. The subgroup of majority/white participants appeared to benefit more (with a difference near our threshold of change in HbA_1c) from lifestyle programs compared with DSME or DSME plus support, but the results did not reach statistical significance.

Table 14

Results for race/ethnicity subgroups using univariate meta-regressions analyzing the association between different program factors and the effectiveness of behavioral programs compared to usual care in improving HbA_1c for T2DM.