This information is for reference purposes only. It was current when produced and may now be outdated. Archive material is no longer maintained, and some links may not work. Persons with disabilities having difficulty accessing this information should contact us at: https://info.ahrq.gov. Let us know the nature of the problem, the Web address of what you want, and your contact information.
Please go to www.ahrq.gov for current information.
Part 1. Impact on Management of Patients with Diabetes Mellitus (continued)
Details of each study are provided in the evidence tables (Appendix D). Below, we provide a
summary of the characteristics of these studies and a tabulation of study data. Study results are
tabulated in terms of proportion reporting statistically positive results (with regard to either
glycemic control or risk factor status) and in terms of effect size (for studies of glycemic control
and weight reduction). Note that because health events such as hospitalization were rarely
described, these were excluded from the tables summarizing proportion of studies reporting
statistically significant intervention effects.
Characteristics of the Studies Included in the Detailed
Of the 61 included studies, 48 (79 percent) compared a behavioral
therapy group to a non-behavioral therapy control group. The remaining 13 studies (21 percent)
conducted head-to-head comparisons of different behavioral interventions, treatment intensities,
or other aspects of treatment that might modify the effectiveness of behavioral therapies. These
latter studies are described in a separate section on effect modifiers.
Types of patients
While all included studies selected patients based on diagnostic criteria for
diabetes mellitus (both insulin dependent and non-insulin dependent), there was no evidence that
trials sought to systematically include patients whose glycemic control was more or less difficult
Types of treatment
Of the 48 studies reviewed in detail, the majority (56 percent) evaluated
the efficacy of cognitive-behavioral treatment. Of the remainder, six studies (13 percent)
examined relaxation-based interventions, 14 studies (29 percent) tested behavioral interventions
aimed specifically at diet and/or exercise, and two studies (4 percent) examined blood glucose
Intensity of treatment
Thirty-four of the 48 studies (71 percent) reported on the frequency of
treatment sessions. Of these 34 studies, the majority (24, 62 percent) reported that treatment
sessions were conducted at least weekly, 1 (3 percent) reported treatment sessions were
conducted biweekly, 2 (6 percent) reported treatment sessions were conducted monthly, 4 (12
percent) reported treatment sessions were conducted bi-monthly, and 3 (9 percent) reported that
that treatment sessions were conducted every three months.
Forty-seven of the 48 articles (98 percent) reported on the duration of the treatment phase of the
study. The mean duration of treatment was 33.2 weeks, with the length of treatment varying
from 1 to 260 weeks. The mean duration is high due to several studies that had long treatment
phases (treatments that lasted 208 weeks to 260 weeks). The duration of treatment for most
studies was in the 10 to 20 week range.
Per the inclusion criteria, health outcomes were reported in all of the
studies. Further, all studies included some measure of metabolic control. Of the 48 articles
included, 19 (40 percent) used HbA1c as the primary measure of metabolic control, 7 (15
percent) used HbA1, 16 (33 percent) used GHb, and the remaining 6 studies used fasting blood
glucose as the primary metabolic outcome measure.
Thirty of the articles (63 percent) measured weight as an additional health outcome. Of risk
factor status measures, 21 (44 percent) reported cholesterol and 9 (19 percent) report blood
pressure. Smoking was reported uncommonly (in 2 studies, 4 percent). Other health events were
also rarely reported, including health care utilization (reported in 3 studies, 6 percent), and
morbidity and mortality (reported in 2 studies, 4 percent).
Quality of life/general health, adjustment, self-efficacy, stress/hassles, distress, and mood
measures were also reported in a number of the studies. Of the 48 studies, 6 (12 percent)
reported on quality of life and general health, 1 (2 percent) reported on adjustment to diabetes, 1
(2 percent) reported on self-efficacy, 4 (8 percent) reported on stress and hassles, and 2 (4
percent) reported on distress. Regarding mood, 3 (6 percent) reported on anxiety and 3 (6
percent) reported on depression.
Quality of studies
The majority of included studies provided details of the patient
characteristics (93 percent), the number of withdrawals and dropouts (83 percent), and described
the intervention well enough to allow replication of the study (78 percent). Fewer studies (17, 35
percent) relied on a manual-based treatment protocol or described the methods used to train the
individuals administering the intervention (10, 21 percent).
Overall Efficacy of Treatment
Proportion of positive studies (Table 1). Studies were defined as positive when the behavioral
intervention group(s) showed significant improvements on a specific measure in comparison to a
non-behavioral control group. Of the 48 randomized trials of behavioral interventions compared
to a control, 22 (46 percent) indicated statistically significant improvements in glycemic control.
Studies which examined the effect of treatment on risk factors (i.e., weight, cholesterol, blood
pressure) were somewhat less often positive, but still positive more often than would be expected
by chance alone.
Subjective outcomes were reported for the minority of studies. Proportion of statistically
positive studies for subjective measures are as follows: three of six studies measuring quality of
life, one of one studies of adjustment, one of one studies of self-efficacy, two of four studies of
stress/hassles, zero of two studies of distress, and one of six studies of mood. Thus, overall,
subjective measures do not appear to be more likely to be improved than are health outcomes,
and, at least for mood, may be somewhat less affected.
Proportion of positive studies by study sample size
Treatment outcome was examined as a
function of how many patients were randomized to each of the group conditions (Table 2).
Studies were grouped into two categories:
- Smaller—those that included fewer than 25 participants per group.
- Larger—those that include 25 participants per group or more.
glycemic control, larger studies appear to be more commonly positive than smaller studies,
whereas the opposite is suggested by the data on risk factor status.
Efficacy of treatment on glycemic control
Of the 48 studies, 29 (60 percent) provided
sufficient data to calculate effect size post-treatment (time points less than or equal to 3 months
after the end of treatment), corresponding to 37 active interventions evaluated (Figure 1). The
mean effect size post-treatment was 0.35 (CI, 0.21 to 0.49) which translates into an absolute
decrease in HbA1c of approximately 0.62 percent (CI, 0.32 percent to 0.88 percent). Thirteen of
the 48 studies (27 percent) provided sufficient data to calculate effect size at a followup point
beyond 3 months, corresponding to 19 active interventions evaluated (Figure 2). The mean effect
size at followup was 0.24 (CI, 0.09 to 0.40), which translates into an absolute decrease in
HbA1c of approximately 0.47 percent (CI, 0.18 percent to 0.78 percent). Again, it appears that
larger studies tend more often to indicate a positive effect for behavioral therapy on glycemic
Efficacy of treatment on weight control
Twenty-six studies (54 percent) included data
sufficient to estimate effect size for weight control. As seen in Figure 3, no effect was seen
overall or by sample size for this outcome measure.
Effectiveness vs. time since completion of therapy
To further examine the durability of
behavioral therapy in effecting a change in health outcome, we plotted glycemic effect size vs.
time since completion of therapy (Figure 4), and weight control effect size vs. time (Figure 5).
In both cases there was an unimpressive trend; for glycemic effect there was a negligible trend
towards a decay in effectiveness and for weight control a negligible increase in effectiveness
Efficacy of Treatment by Treatment Characteristics
Efficacy by type of behavioral intervention
Treatment outcome was examined for the four
different categories of behavioral intervention: cognitive-behavioral therapy, relaxation-based
interventions, diet/exercise interventions, and blood glucose awareness training (Table 3).
Proportion of positive studies was similar to the overall results for cognitive-behavioral therapy
and behavioral exercise/diet interventions. However, studies of relaxation therapy are less likely
to report positive results, and information on blood glucose awareness vs. control intervention
Figure 6 illustrates the effect sizes ordered by type of treatment. Here we see that both
cognitive-behavioral therapy and behavioral diet/exercise programs tend to be more effective
than control in improving glycemic control, whereas no clear trend is demonstrated in the more
limited studies of relaxation therapy and blood glucose awareness training.
Efficacy by intensity of treatment
Only one study was identified in which different intensities
of therapy were compared head-to-head (Hendricks, 2000). In this study, the investigator
compared cognitive-behavioral interventions at monthly and once every 3 month intervals. No
significant differences in glycemic control or health care utilization were reported.
For trials in which behavioral therapy was compared to a non-behavioral control, studies were
grouped into two intensity categories based on number of sessions: low intensity, those that
included less than 14 weekly sessions; and high intensity, those that included 14 or more weekly
sessions. The mean number of sessions for the low intensity group was 7.9 and the mean for the
high intensity group was 22.8. As seen in Table 4, there was a trend toward a benefit from a
greater number of sessions for both glycemic control and risk factor status.
In Figures 7 and 8, effect size for glycemic control within 3 months of completing therapy and
beyond 3 months of completing therapy is plotted against numbers of interventions. (Note that
the results were similar for glycemic control vs. duration of therapy or frequency of therapy (data
not shown).) A negligible negative correlation between number of interventions and effect of
treatment is noted in the short-term outcomes and a negligible positive correlation is noted in the
Since type of therapy appeared to influence efficacy, in Figure 9 we examine intensity level
within therapy type. There is no evidence that specific therapies are more or less likely to be
effective at higher levels of intensity. (Note: a similar plot for effectiveness by intensity and
treatment type for glycemic control beyond 3 months of completing therapy also showed no
trend (data not shown).)
Interaction of glycemic effect and weight effect
Weight control was measured as a risk factor
in 30 of the 48 non-behavioral control studies. To examine whether interventions that were more
effective in controlling glycemic control were associated with improvements in metabolic
control, we plotted the effect size of both, for outcomes within 3 months of completing therapy
(Figure 10) and beyond 3 months of completing therapy (Figure 11). A modest positive
correlation suggests that the two effects tended to be related.
Mode of calorie restriction
Five studies conducted head-to-head comparisons of modes of
calorie restriction. Wing, Blair, Bononi et al. (1994) showed a significant decrease in fasting
glucose means for very low calorie diet (VLCD) when compared to low-calorie diet (LCD) at the
end of the 12-week treatment period, but followup data at 15 weeks post-treatment showed no
significant differences between the two intervention groups. A similar study, Wing, Blair,
Marcus, et al. (1994 ) conducted over a 50-week intervention period showed no significant
differences at the end of treatment between VLCD and LCD groups in glycemic or weight
control. Williams (1998) compared standard behavioral therapy to interventions of 1-day and 5-day
VLCD over a 20-week period, with no significant differences in HbA1c change means or
cholesterol means after treatment. Both Williams (1999) and Wing (1996) compared varying intensities of VLCDs with weekly
meetings over 12- and 20-week periods respectively, neither reporting significant results in
metabolic or weight control.
Other potential effect modifiers
Seven studies investigated other modifiers of effect on
glycemic control and risk factors, with only one study, Glasgow (2002), reporting significant
effect differences between treatment groups.
Large studies of other effect modifiers
Glasgow (2002) compared 4 treatment conditions—basic
goal setting, community resources, telephone followup and combined condition—on a group of
320 patients over a period of 12 months. Telephone followup showed significant effect when
compared to other groups for both glycemic control and lipid ratio (p<0.05).
Glasgow (2003) conducted a large study (N=320) comparing three online interventions—tailored
self-management, peer support and information only—over a period of 10 months. No
significant differences were reported in glycemic control or lipids ratios between the three
Rickheim (2002) evaluated 170 patients, comparing group vs. individual education, with 4
intervention sessions constituting 5-7 hours of education. There were no significant differences
in effect on metabolic or weight control between the groups.
Wing (1985) compared a standard behavioral weight control program to a weight control plus
glucose monitoring program, each set of interventions conducted weekly on 25 patients. There
were no significant differences between the intervention groups on glycemic or weight control.
Small studies of other effect modifiers
The effect of treatment alone vs. treatment together with a
spouse was investigated in Wing (1994) using a 20-week cognitive-behavioral therapy
intervention on a total of 49 patients and 49 spouses. No significant differences of effect were
reported between the groups for glycemic or weight control.
Lamparski (1989) looked at the effect of current vs. non-current feedback in a blood glucose
awareness training (BGAT) program. Interventions were conducted on two 18-patient groups
over a period of 4 weeks, with no significant differences in effect between the two groups
(current vs. non-current feedback).
Mayer-Davis (2001) conducted a small study (N=33) comparing intensive lifestyle interventions
with and without formal evaluation. The 8-week study did not report comparative data between
the two groups.
Return to Contents
Proceed to Next Section