Multidisciplinary Diabetes Care in a Safety Net Clinic: Lessons Learned from a Quality Improvement Initiative
Analysis
Data abstracted from de-identified patient records included HbA1c values, interventions received, and sociodemographic data. Generalized linear mixed modeling (GLMM) was used to examine changes in patient HbA1c levels over time [12]. All models included a random intercept to account for correlated observations within patient. All analyses were performed using Proc GLIMMIX in SAS v. 9.3 [13].
Results
A total 271 patients with HbA1c above 9 were included in the analysis: 71 in the intervention group and 188 in the usual care group. The intervention group was further differentiated by month of enrollment: October (n = 37), November (n = 27), and December (n = 9). Mean patient age in the overall sample was 51.6 years. 
In the intervention group, most patients received patient education 56% (n = 40), almost half had a clinical pharmacy visit, but only 17% (n = 12) received a dietitian consultation. Overall, there was a 1.4% decrease in HbA1c in the intervention group, compared to a 1.3% HbA1c decrease in the usual care group.
GLMM was used to examine differences in HbA1c levels according to month of intervention enrollment (October vs. November vs. December) in the intervention group over time. Figure 1 shows predicted HbA1c values over time with trend lines fit for each of the three subgroups. 
Preliminary analysis showed that potential contamination (diffusion of the treatment) would be likely to attenuate differences in the outcomes between the intervention and usual care conditions. Further analysis by subgroups were conducted to describe the intervention potential “spillover” to the usual care group participants not intended to receive the intervention. GLMM also examined differences in HbA1c levels between the intervention and usual care groups over time. The interaction between each treatment group and time was not statistically significant (F(2,268) = 1.34, P = 0.26), indicating that changes in HbA1c over time were not related to treatment group. A statistically reliable main effect for time (F(1,268) = 44.33, P < 0.001) indicated that in all groups, HbA1c values significantly decreased over time.
Follow-up analyses utilized GLMM to examine differences in HbA1c levels among patients in both groups who received at least one of the interventions (visiting a dietician, clinical pharmacist, education session, and clinical case manager). The interaction between intervention and time was not statistically significant for visiting the dietician, receiving education, or being assigned a case manager. The interaction between time and visiting a clinical pharmacist was statistically significant (F(1,204)= 7.78, P = 0.01) such that patients visiting the clinical pharmacist had lower HbA1c values over time relative to those that did not (Figure 2). 
Discussion
HbA1c decreased significantly among intervention patients with uncontrolled diabetes over a 3-month period, regardless of which month they entered the study. However, there was no significant difference in HbA1c reduction between patients who received all 4 multidisciplinary interventions, one intervention, or those who received usual care. Patients in the intervention who attended clinical pharmacist visits had significantly greater HbA1c reduction than patients who did not, as did patients who attended a diabetes education session by a patient educator.
