The conceptual basis for this program, detailed in a previous publication 27, combined Social Cognitive Theory 28, Goal Systems 29, and Social Ecological Theory 30. This model has evolved to include multiple system and social-environmental factors, including social support, that influence self-management of chronic illness 30. Participants were able to set personal lifestyle change goals at the start of the intervention, and then received ongoing peer and professional support for their goals throughout the treatment program.

The MLP was delivered by a registered dietitian, an exercise physiologist, a stress-management instructor, and a combination of professional and lay support group leaders. The program started with a 2 1/2-day non-residential retreat, which was followed by 4-hour weekly meetings consisting of 1 hour each of Mediterranean-style potluck, physical activity, stress management, and support groups. The intervention was conducted in four sequential waves, 2 months apart, with approximately 40 treatment condition and 30 control condition participants in each wave. This schedule was necessary to keep the group size manageable within staff and space constraints.

The first 6 months of the intervention were designed to teach the program components and build group cohesion.

Since the MLP intervention targeted different sources of support (e.g., friends, family, health care providers), measures were included to address each of these components. The Brief Chronic Illness Resources Survey (CIRS) 30 provided a profile of an individual's support for behavior-specific disease management, ranging from more proximal support (e.g., family and friends) to more distal factors (e.g., neighbourhood or community). In these analyses, the total score from the CIRS was used to represent received support. The UCLA 37 measures three types of support (i.e., informational, tangible, emotional), three dimensions of support (i.e., amount, satisfaction, reciprocity), and four sources of support (i.e., friends, relatives, partner, organizations). For this project, two other sources of support – people in a support group and medical care providers – were added to the original UCLA. The Total Positive and Total Negative scales were included in these analyses.

The Diabetes Problem-Solving Interview was developed for adults with type 2 diabetes 38. For this study, the interview was modified to ask respondents to write a description of how they would react to nine scenarios presenting potential challenges to program adherence. Coders scored responses to produce an average rating of problem-solving skill. Inter-rater reliability on these scores ranged from r = .57 to r = .90 and averaged r = .72. Six-month test-retest reliabilities using only the control group was r = .51 for overall skill ratings.

Two measures of self-efficacy were used to assess different areas of diet and physical activity-related obstacles. Confidence in Overcoming Challenges to Self-Care was used to assess confidence overcoming obstacles to adhering to diet, physical activity, and stress-management 39. This 49-item instrument assesses confidence in overcoming such factors as cost, time, social pressures, competing demands, and thoughts associated with achieving one's dietary, physical activity, and stress-management goals. Recent analyses of this instrument 39 demonstrated reasonable psychometric properties for a brief scale (test-retest reliability r = .60; internal consistency (α = .50). The total score was used in these analyses. Participants' self-efficacy for achieving their dietary and physical activity goals was assessed using the Sallis Self-Efficacy for Diet and Exercise Behavior instrument 40. This instrument assesses confidence in performing diet and exercise behaviors for at least 6 months. The Eating Behavior scale contains 20 items (test-retest reliability ranged from r = .43 to .64; Cronbach's α .84) and the Self-Efficacy for Exercise scale contains 12 items (test-retest reliability was r = .68; α coefficients ranged from .85 to .93) 40.

Since CHD and diabetes are closely linked to depression 13., the Center for Epidemiologic Studies Depression Scale (CES-D) was administered. The CES-D is a general measure of depressive symptoms that has been shown to have good reliability and validity 41, and has been used extensively in epidemiologic studies.

The Perceived Stress Scale was administered to determine whether perceived stress changed in conjunction with the stress-management intervention. This 14-item measure, based on the transactional stress theory of Lazarus, has been found to be an independent and significant predictor of physical symptoms and health behaviors after controlling for psychological symptoms 42.

The Diabetes Distress Scale (DDS), a diabetes-specific measure of quality of life, also was used. Respondents rated the degree to which common barriers to adherence were problematic for them (lower ratings indicate better perception of quality of life). This scale assessed diabetes-specific overall emotional distress, interpersonal distress, regimen-related distress, and physician-related distress. A recent 32-item revision of the DDS 43 produces subscales on these four dimensions as well as an overall score; lower scores indicate better quality of life. The DDS has good internal reliability with the four subscales (α = 0.87), and has been shown to be responsive to psychosocial intervention 43. Two of the four DDS scales were used: the regimen-related and interpersonal distress scales.

Using methods described in detail elsewhere 44, an economic analysis of the costs associated with the MLP intervention was conducted. The analysis was conducted primarily from the perspective of the potential adopting organization or potential payer, such as Medicare or a health insurance plan. The analysis evaluated the cost of delivering the MLP compared to the UC condition. Using retrospective data collected during the development and implementation of the MLP, estimates were made of total intervention costs, incremental costs associated with the intervention group relative to the comparison condition, costs per participant, and marginal costs per incremental improvement in several of the primary outcomes. Total intervention costs were estimated as the sum of the costs associated with recruitment of participants (in the intervention arm) and staff; labor costs associated with the time spent by educators, dietitians, physicians, exercise physiologists, meeting leaders, and support group leaders; retreats; training, phone charges, and supplies; and rent attributable to facility space needed for group meetings. All assessment costs were excluded.

Prior to conducting the repeated measures MANCOVAs, univariate statistics and correlation matrices were examined for potential additional correlates of the behavioral and psychosocial dependent variables. For the behavioral outcomes, social desirability, income, number of comorbid conditions, and age were all entered as covariates, and none was a significant contributor to the models. Social desirability was entered as a covariate in all of the psychosocial analyses presented.

All analyses were performed two ways. First, a complete-case approach was used. Participants with missing follow-up data on the outcome variable of interest were excluded from the analysis. Second, identical analyses were conducted after missing data were imputed using EMCOV 45. Significance and conclusions from imputed analyses were mostly identical or stronger compared to complete-case analyses. For ease of understanding, the nonimputed results are presented in the tables, with the imputed results, where different, described in the text.

Women randomized to the MLP intervention were similar to women in the UC condition. The baseline characteristics of women assigned to UC vs. MLP have been reported in Toobert et al. (2003) 24 and are presented for convenience in Table 1. As can be seen, this was a high-risk group of women: Their average BMI exceeded 35 kg/m² and most had multiple chronic illnesses.

Twenty-four-month follow-up data were collected for 237 (85%) of the randomized participants. Analyses of the characteristics of dropouts vs. participants present at 24-month follow-up revealed no significant main effects or interactions with treatment condition on baseline characteristics of age, weight, waist/hip ratio, body mass index, smoking status, type of medication, income, education, living arrangement, diabetes complications, or number of comorbidities. Significant differences were found as follows: Those who remained in the study at 24 months were older at diabetes diagnosis (mean age of diabetes diagnosis for dropouts was 49.6 vs. 53.4 years for those who remained in the study), and had taken medications for less time (mean years taking diabetes medications for dropouts was 7.3 vs. 4.5 years for those who remained in the study). There was also a significant difference in employment status between those retained and those who dropped out at 24 months. Specifically, 34.3% (N = 81) of those who remained in the study were employed compared to 55.8% (N = 24) of dropouts.

ICCs were computed for key dependent variables to determine whether there was significant clustering by wave of the study participants. All ICCs were less than .003 (median = .001), indicating an absence of wave effects.

Of 23 meetings offered over the first 6 months, the mean (SD) number of sessions attended for the MLP participants was 12.4 (5.7); the range was 0–20. Of 39 meetings offered from 6–24 months for the 63 weekly meeting participants, the mean (SD) number of sessions attended was 19.4 (13.2), or 50%; the range was 0–38. Of the four sessions offered from 6–24 months for the 82 personalized computer-assisted condition participants, the mean (SD) number of sessions attended was 2.5 (1.8), or 63%; the range was 0–4.

Analyses revealed few significant and no meaningful differences between the two experimental maintenance conditions; given the small sample size, these two conditions were merged and compared to the UC condition in all further analyses.

MLP participants showed significantly greater improvement overall, at all follow-up assessment points, on percent of calories from saturated fat than did the UC group. FFQ data indicated that the intervention produced significant 24-month between-condition effects and consistent improvements in mean percent calories from saturated fat (average reduction in percent calories from saturated fat was 3% for MLP vs. 1% for UC).

Physical activity improvements also were maintained over 24 months. A significant between-condition effect was found on the repeated measures ANCOVA for the kcals/kg/hr moderate-intensity activity measure derived from the CHAMPS, showing the MLP conditions to be superior to the UC condition in frequency, duration, and intensity of physical activity.

Stress-management results from the 7-day self-monitoring log indicated that MLP participants significantly increased the number of daily minutes they practiced stress-management techniques (yoga stretches, breathing, guided visualization, and meditation) compared to UC participants, and these effects maintained across all time points. The imputation analysis yielded similar results with the exception that the condition-by-time effect became significant.

There was a significant between-condition effect for the flexibility repeated measures MANCOVA, which included both the sit-and-reach and range-of-motion tests. For the sit-and-reach test, flexibility significantly improved in the MLP group and worsened in the UC condition. Range-of-motion flexibility was greater at each time point in the MLP compared to the UC; however both conditions decreased their range-of-motion flexibility over the course of the study.

Repeated measures MANCOVAs revealed a significant improvement in perceived social support (see Table 3), F(3,195) = 13.68, p < .001. Follow-up ANCOVAs indicated improvements in the total perceived Positive and Negative Support scales measured by the UCLA Social Support Inventory. Imputation analyses revealed an additional condition-by-time effect for the UCLA Positive scale and a loss of the significant between-group effect for the UCLA Negative scale. The total score of the brief CIRS also showed a significant long-term treatment effect in favor of the MLP participants on use of community and social environmental resources.

A significant overall effect was found on the repeated measures ANCOVA for problem-solving skills. The effect was due primarily to significantly larger improvements among the MLP participants on the quality of the problem-solving strategies employed.

MLP participants made and maintained modest but significantly greater improvements than UC participants on measures of dietary self-efficacy, exercise self-efficacy, and confidence in overcoming challenges to illness management.

As shown in Table 3, the separate repeated measures MANCOVAs revealed only one significant effect for perceived stress and no significant effects for depression or self-monitored stress.

As shown in Table 3, despite a pattern of greater improvement for the MLP condition on the regimen-related subscale, the overall repeated measures MANCOVA for the quality of life (DDS) regimen-related and interpersonal distress scales did not reach the p < .05 level of significance for the MLP compared to UC.

Total direct and indirect intervention costs were estimated to be $409,165, or $2,510 per MLP participant relative to UC for the 24-month period ($309,302 direct costs). Using the same data employed to estimate results in Table 2, change scores were calculated for measures demonstrating statistically significant improvements. For the behavioral improvements, there was a cost of $221 per unit improvement in calories derived from saturated fat as measured by the FFQ, $1,434 per kilocalorie increase in physical activity, and $1,168 per unit improvement in daily stress management.