Improved Glycemic Control in Adults With Serious Mental Illness and Diabetes With a Behavioral and Educational Intervention

This study was approved by the Partners Healthcare institutional review board. Participants demonstrated competence to consent via a written assessment of protocol knowledge and participation requirements, and they gave informed consent prior to participating in study procedures. Data were collected between February 1, 2016, and October 31, 2018.

We identified adult outpatients with serious mental illness at an urban community mental health center and nearby clubhouse day program who had HbA1c ≥6.5 (N=43) or HbA1c >6.0 and were taking metformin (N=24), or had a diagnosis of type 1 or type 2 diabetes in the medical record (N=13). Those who did not speak English or were deaf were excluded from study participation. The study was supported by a 1-year grant that was renewed for a second year of funding. Thirty participants were enrolled at the beginning of each year of grant funding and randomly assigned to initial or delayed (4 months later) start (see the online supplement for a CONSORT diagram). Mean±SD age of those enrolled (N=60) was 55±11.4 years; 70% (N=42) were male; and 53% (N=32) were white, 28% (N=17) black, 17% (N=10) other race, and 2% (N=4) Hispanic.

The intervention consisted of 16 weekly 60-minute group meetings. Sessions were co-led by a psychologist specializing in health behavior change in those with serious mental illness and by a psychiatry resident. Guest coleaders included a diabetologist, an internist with behavioral weight management expertise, and a registered dietician. Five manualized content modules were developed by a team that included physicians from departments of psychiatry and general internal medicine and the Diabetes Center. Modules covered basic diabetes disease education and management, nutrition, exercise, stress management, and positive psychology topics and were modeled on content from the Diabetes Prevention Program (see the online supplement for the curriculum). A pedometer capable of storing 7 days of step counts was provided. During each group meeting, participants received a healthy lunch and $3 to defray public transportation costs. Group outings to nearby grocery stores, the hospital cafeteria, and Department of Mental Health gym provided participants with education on nutrition and physical activity available in accessible community settings and enabled participants to practice disease self-management skills.

Core features of the group intervention curriculum included interactive education and problem solving informed by challenges presented by each participant’s home community (i.e., tendency to shop at convenience stores, access to refrigeration, high availability of fast food establishments). The intervention was informed by the self-determination theory of motivation (6), emphasized resilience enhancement, and focused on increasing access to positive emotions and maximizing self-confidence, autonomy, and connection, with the goal of increasing intrinsic motivation, persistence, and goal acquisition (7). Group leaders used a directive approach to assist participants in setting a weekly SMART (specific, measurable, achievable, realistic, timed) behavioral goal (8) related to the weekly discussion topic (i.e., achieving a daily step count goal during exercise modules, eliminating sugar-sweetened beverages during diet modules). Each group included follow-up on each participant’s goal from the prior week and setting an individualized behavioral goal for the coming week.

The primary outcome was change in HbA1c from baseline to week 16. Secondary outcomes included BMI (weight was measured weekly in light clothing without shoes), systolic and diastolic blood pressure (seated, at rest, measured weekly), serum lipid concentrations (low-density lipoprotein [LDL], high-density lipoprotein [HDL], triglycerides [TG], and total cholesterol [TC]) measured at baseline and at week 16, daily step count (measured by study-provided pedometer), and diabetes knowledge and self-management (Short Diabetes Knowledge Instrument, Summary of Diabetes Self-Care Activities, and Problem Areas in Diabetes, assessed every 4 weeks). Because most participants did not fast prior to their lipid panel measurement, limiting TG and LDL measurement validity, we report only TC and HDL. Three independent clinicians assessed and averaged the Clinical Global Impressions Scale (CGI) severity score (range 1–7) for each patient at baseline. (See the online supplement for assessment citations.)

A modified intention-to-treat (ITT) analysis (9) was performed that included all participants who attended at least one group meeting (N=35). For dependent variables, a linear model with a random intercept for participants was fit to the data, thereby controlling for dependencies due to repeated measures. Fixed effects included a linear trend to estimate the change during intervention and coefficient to estimate change postintervention where applicable. Bayesian mixed-effects modeling was conducted with the R package “rstanarm” (version 2.13.1, Stan Development Team, 2016) to estimate the effect of the intervention on outcomes of interest. Effects were considered significant if p values were less than 0.022 (the lower boundary for a two-sided test after controlling for Monte Carlo approximation error). Missing data were assumed to be missing at random and thus were handled by the hierarchical nature of the mixed- effects models. In instances in which model assumptions (e.g., normally distributed residuals) were violated, a mixed-effects model robust to outliers (based on the Student’s t distribution) was used instead. (See the online supplement for assessment of model performance.) For participants with a delay of 4 months between enrollment and study start, change in HbA1c was assessed via a paired-samples t test, to serve as a control.

Two-hundred sixty-two individuals were screened via medical record review by community mental health center staff (S.M.) for study eligibility. Eighty potential participants met inclusion criteria; 60 participants provided informed consent and were enrolled. Thirty-five individuals attended one or more group sessions and were included in the modified ITT analysis. (See the online supplement for CONSORT diagram.)

In the modified ITT sample of participants who attended at least one group, average attendance was 11.5±6.5 group sessions (72% of total sessions); two participants attended one group, eight attended from two to eight groups, and 20 attended from nine to 16 groups. Mean±SD age was 52.9±10.9 years, 77% (N=27) were male, 46% (N=16) were white, 34% (N=12) black, 20% (N=7) other race, and 3% (N=3) Hispanic. Ninety-seven percent (N=34) were taking any antipsychotic medication, 63% (N=22) were taking clozapine or olanzapine, 66% (N=23) were taking metformin, 34% (N=12) were taking other oral diabetes medication, and 26% (N=9) were taking insulin. Mean baseline HbA1c was 7.5%±1.6%, and mean BMI was 33±3.8 kg/m². Thirty-five percent had systolic blood pressure ≥130 mmHg, and 51% had diastolic blood pressure ≥80 mmHg. The mean CGI severity rating was 4.8±0.9. There were no noted significant differences between the modified ITT sample of participants who enrolled and attended at least one group and those who enrolled but attended no groups. (See baseline characteristics table in the online supplement.)

In the modified ITT sample (N=35), HbA1c was 7.5%±1.6% at baseline and 7.1%±1.4% at week 16 (β=−0.6, SD=0.25, p=0.01). In the sample with a delay of 4 months from enrollment to start of study procedures (N=16), HbA1c values did not differ significantly between enrollment (7.8%±2.3%), and the start of intervention (7.6%±1.7%) which was a difference of −0.2 (t=−0.02, df=15, p=0.99). For participants who enrolled but attended no groups, HbA1c was 6.5%±1.1% at baseline and 6.4%±0.8% after 16 weeks (β=−0.46, SD=0.47, p=0.156).

Mean BMI was 33.3±3.8 kg/m² at baseline and 32.9±4.1 kg/m² at week 16 (β=−0.76, SD=0.17, p<0.001). Diabetes knowledge improved from 7.2±2.7 at baseline to 8.5±2.7 at week 16 (β=1.75, SD=0.42, p<0.001), as did days spent observing general (p=0.003) and specific (p=0.001) diet self-care and exercise self-care (p=0.015). Total cholesterol, HDL, systolic and diastolic blood pressure, steps per day, foot self-care, and diabetes distress did not change.

Nine of the 15 participants who had enrolled in year 1 who attended at least one group remained in the clinic for 1 year after the end of the group intervention and provided 1-year follow-up data. HbA1c in this subgroup was 7.1%±1.3% at end of treatment and 7.1%±1.7% after 1-year follow-up.

Adults with serious mental illness and diabetes who participated in at least one session of a 16-week group behavioral and educational intervention for diabetes self-management (N=35) had significant improvement in glycemic control. Although this was an open-label intervention that should be replicated in a controlled trial, confidence in this result is increased by the observations that HbA1c did not significantly change for individuals prior to intervention start, that enrolled participants who attended no group sessions had no significant change in HbA1c in the 4 months after enrollment, and that reduction in HbA1c during the intervention appeared to endure for 1 year after enrollment. Significant improvements in BMI, weight, diabetes knowledge, and diet and exercise self-care are also notable.

These improvements occurred in a study population who were psychiatrically stable but significantly symptomatic; 63% were taking clozapine or olanzapine, antipsychotic medications with the greatest metabolic adverse effects. This finding, if replicated, suggests that diabetes education and behavioral interventions in a psychiatric care setting could significantly improve glycemic control and associated medical risk factors in a vulnerable population with an almost 30-year mortality disparity.

Although these results are derived from an open-label trial and should be interpreted cautiously for that reason, they stand in contrast to those from two randomized controlled trials of behavioral and educational interventions of which we are aware for individuals with schizophrenia and diabetes, which did not report significant HbA1c reduction (4, 5). Sajatovic and colleagues compared usual primary care with Targeted Training in Illness Management (TTIM), consisting of 12 weekly educational group sessions delivered in a primary care setting followed by 48 weeks of brief telephone maintenance sessions, for 200 participants with serious mental illness and diabetes. Those assigned to TTIM demonstrated increased diabetes knowledge and improved psychiatric symptoms and functioning but no significant effect on HbA1c, BMI, or blood pressure (5). McKibbin and colleagues (4) assigned 57 participants with diabetes and schizophrenia-spectrum illness to receive Diabetes Awareness and Rehabilitation Training (DART), a 24-week group diabetes education, nutrition, and lifestyle exercise intervention, or usual care plus three educational brochures on diabetes. As in our intervention, DART included weekly weigh-ins, pedometers, healthy food samples, and reinforcements for attendance and behavior change. Consistent with our results, participants assigned to DART had significantly reduced weight, improved diabetes knowledge and self-efficacy, and increased physical activity scores that were sustained at 12-month follow-up but did not demonstrate significant improvements in HbA1c (10).

Among those who attended at least one group, average attendance was 11.5 of 16 sessions (72%), indicating good retention of those who engaged in any sessions. Over half of those who gave informed consent (58%) attended at least one group. Retention was likely improved by colocation of the group intervention in the community mental health center where participants were regularly attending behavioral health appointments, as opposed to primary care (4) or research settings (5), and receipt of incentives for being a part of the group program (e.g., healthy free lunch, pedometer, and $3 for attendance).

In contrast to using classical motivational interviewing approaches, group leaders in our intervention used a directive, at times prescriptive, approach to assist with goal setting, providing clear recommendations and praise for behavior change efforts and group engagement. This—together with active weekly behavioral goal follow-up for progress on evolving, individualized, environmentally informed, actionable goals and goal evolution—may have affected participant success. The intervention took into account barriers identified by participants to purchasing and preparing healthy food (e.g., lack of food storage options, tendency to shop at convenience stores) and to exercising (e.g., paranoia, unsafe neighborhoods) and addressed these barriers in concrete ways through group problem-solving exercises. A qualitative study is underway to identify factors affecting engagement in this population, determine key messages retained, and identify areas for improvement.

There are important limitations to consider in evaluating this study. This was an open-label study of the effect of fixed-enrollment groups with no treatment-as-usual control. Nearly 60% of those who enrolled participated in at least one group. Because approximately 40% of those who consented attended no groups, results are generalizable to individuals who are willing to at least initiate participation in an intervention rather than to all with serious mental illness and diabetes. We speculate that the delay of up to 20 weeks between consent and initiation of the intervention for half the cohort due to funding and staffing limitations likely contributed to attrition. Participants who attended at least one group may have been more motivated to make positive health behavior changes. Although medication optimization was not an intervention in the study, participants may have sought intervention from their primary care team as a result of study participation; thus we cannot definitively isolate the effects to the behavioral intervention. The intervention was delivered by a highly trained psychologist and psychiatrist; future studies will evaluate our ability to replicate these results with alternative staffing models (e.g., community health workers or peer specialists).

A 16-week behavioral and educational group intervention improved modifiable health behaviors that underlie the large morbidity and mortality disparity in those with serious mental illness and diabetes. Improvements were noted in objective measures of glycemic control, BMI, weight, and diabetes knowledge and self-care among adults with diabetes and serious mental illness who attended at least one group. Encouraging evidence for maintenance of this effect over 1 year was also observed. These results warrant further testing in relation to a control with the goal of discovering an intervention that is effective and scalable to mitigate the large public health impact of suboptimal diabetes care in this vulnerable population.

The authors thank North Suffolk Mental Health Association for its community partnership and Drs. David Schoenfeld, Christine Ulysse, Gladys Pachas, and Trina Chang for their technical assistance.