Closing the Gap: Effect of Diabetes Case Management on Glycemic Control Among Low-Income Ethnic Minority Populations: The California Medi-Cal Type 2 Diabetes Study

Between 1 July 1995 and 30 June 1999, we conducted a randomized, controlled study of individuals ≥18 years who had type 2 diabetes of at least 1 year duration. Individuals with type 2 diabetes were recruited at clinical sites in three counties (Santa Barbara, Los Angeles, and San Diego) serving racial/ethnic minority, low-income Medi-Cal populations. The three clinical sites were selected to participate in the study via a competitive award process. One study site was a community-based program within a county-wide managed care plan for Medi-Cal recipients. The other two study sites were university-based centers. One of these recruited fee-for-service Medi-Cal recipients from within and outside its health care system. The other university-based site recruited participants from both the fee-for-service and managed care plans within hospital-based and outlying clinics served by its health care system.

Informed consent was obtained from all prospective participants using Institutional Review Board–approved forms.

Participants with HbA_1c levels greater than 7.5% were recruited to assess the effectiveness of case management intervention on participants with suboptimal glycemic control (Table 1); 188 were randomized to the intervention group and 174 individuals were randomized to the control group (Fig. 1). Intensive diabetes case management was provided to the intervention group in addition to primary care (Table 2). Individuals randomized to the control group continued to receive usual care from their primary health care provider. Study staff met with the participant at study entry and exit to assess overall health status, glycemic control, diabetes self-care behaviors, and presence of diabetes-related complications. Blood for HbA_1c was collected at 6-month intervals in the control group, and interim contact between study staff and participants was generally limited to that needed to assure collection of HbA_1c samples. Primary care providers responsible for the care of control participants were, in many cases, also responsible for the care of intervention participants within the same clinical setting.

The study staff at each site, consisting of registered nurses and registered dietitians working in close collaboration with an endocrinologist, provided diabetes case management to the intervention group only. Evidence-based practice guidelines and algorithms for medication and insulin initiation and/or adjustment were used in a collaborative practice model with the primary care provider (18) (Table 2). Potential barriers to care were identified, often influenced by the demographics, and socioeconomic status of the participant population and individualized treatment and education strategies was designed to address as many of these barriers as possible (19–21).

The study protocol included basic guidelines for glucose and medication management. American Diabetes Association goals for treatment of diabetes, hypertension, and dyslipidemia served as the basis for the development of protocol guidelines. Treatment goals and targets for therapy were uniform across sites. However, each clinical center had flexibility to use individualized treatment algorithms and strategies for glucose and medication management. All participants in the intervention group were provided with the same glucose meter and received individualized education regarding use and data recording. In addition, diet, exercise, and self-care behaviors were assessed by diabetes educators, and strategies to improve self-care education and management were used throughout the study in the intervention group. These strategies considered participant’s level of education, literacy and functional understanding, treatment goals, general health status, cultural beliefs, and support network. Interactions between the participant and study staff occurred in person at the clinic site and via telephone between visits as needed.

Diabetes care appointments were scheduled and completion monitored; missed appointments were rescheduled. Transportation issues were addressed to improve visit completion. The need for ancillary medical evaluations and/or services (for example, ophthalmologic, podiatry, home health) was identified with subsequent follow-up to ensure receipt of services, results retrieval, and communication of results to the primary care provider.

Level of glycemic control, as measured by serial HbA_1c measurements, was the primary outcome measure. HbA_1c levels were obtained on a quarterly basis in the intervention group and every 6 months in the control group. HbA_1c was assessed by high-pressure liquid chromatography method in the laboratories used by the three centers. The normal range for these laboratories was 4.9–6.1% (percent coefficient of variation less than 3%) and was calibrated to the Diabetes Control and Complications Trial standard (22).

The sample sizes for the two comparison groups were chosen to detect an anticipated HbA_1c reduction in the intervention group that exceeded the corresponding change in the control group by 1.0%. Allowing for a 30% drop-out rate within the planned initial sample size of 150 in each group yielded 105 for analysis; this resulted in a power of slightly greater than 95% with a two-sided significance level of 0.05, assuming a standard deviation for HbA_1c change of 2.0%. Potential gains in power from use of longitudinal analysis, which could accommodate participants with incomplete follow-up, were not considered in the sample size planning.

The study biostatistician prepared a separate computer-generated time-blocked allocation sequence for each of the three clinical centers. Each center’s sequence was provided in individual sealed envelopes sequentially labeled with participant study codes and containing randomization assignment. Treatment group assignment was unknown until the day of randomization.

Neither participants nor caregivers were masked to HbA_1c measurements or other laboratory results. HbA_1c values were used by study staff to guide treatment decisions in the intervention group.

A longitudinal analysis was used to assess changes in HbA_1c. The trend of mean HbA_1c over time for each group was estimated using a normal linear mixed model with the robust standard error option. The dependence between the participants’ repeated HbA_1c measurements was modeled using the group-specific compound symmetry covariance model. This analysis was implemented with PROC MIXED in SAS 8.1 (23). This approach allowed for inclusion of varying numbers and times of follow-up assessments. Thus, participants having at least one HbA_1c value beyond baseline were included in the analyses. Because an initial sharp decrease in the HbA_1c followed by a linear decline at a slower rate was observed in both groups, a piecewise linear model (also known as a “hockey stick” model) (24) was used to model the trend in mean HbA_1c over time. Because individuals with HbA_1c <7.5% were not eligible for participation, the sharp initial drop in both groups is partly due to regression to the mean, which should be comparable in both groups. The breakpoint for the fitted piecewise linear model was determined using the Akaike Information Criterion (25) to choose from a spectrum of breakpoint options. For graphic and tabular presentation in Fig. 2 and Table 3, estimated mean HbA_1c values with standard errors from baseline to 36 months by 6-month intervals were obtained from the fitted model.

We analyzed the percentage of participants who achieved specified HbA_1c goals (<6.5, <7.0, or <7.5%) on at least one occasion during the study. Figure 3 shows that the percent of HbA_1c values that achieved target was consistently greater in the intensive group compared with the control group at the level of ≤6.5, ≤7.0, and <7.5%. By design, the frequency of HbA_1c tests for participants in the intervention group was higher than in the control group (once every 3 months for the intervention versus once every 6 months for the control). On average, there were 6.1 HbA_1c tests per participant in the intervention groups vs. 4.4 in the control group. An HbA_1c test schedule with higher frequency might result in a higher percentage of participants appearing to achieve HbA_1c goals. To compare the difference between the two groups without any bias contributed by the HbA_1c test schedule, we estimated the percentages achieving targets for the intervention group by using the bootstrap method. This method randomly chooses a number of subjects and tests in the intervention group that is set to be identical with the control group. This procedure was repeated 100 times in the intervention group to provide an HbA_1c test schedule that simulates the HbA_1c schedule of the control group (26).

Demographic characteristics were obtained in the intervention and control group (Table 1). For continuous characteristics, groups were compared using means and standard errors. However, because some variables were not normally distributed, the nonparametric Wilcoxon’s rank-sum test method was used to assess significance of observed differences between groups. For categorical variables, descriptive comparisons were based on proportions in each category, and statistical significance was tested with χ² tests. Analysis of baseline data included 186 intervention and 172 control participants. For the longitudinal analysis, all participants with baseline and at least one follow-up HbA_1c were included. Thus, the cohort for HbA_1c analysis included 171 and 146 participants in the intervention and control groups, respectively. All analyses were performed using the intent-to-treat model.

Changes in secondary outcome variables (e.g., weight, BMI, blood pressure, and lipid parameters) were analyzed using the Generalized Linear Model with SAS PROC GLM (23). Participants included in the analysis for a secondary outcome variable needed a baseline and an ending value with corresponding measurement dates associated with the outcome variable. In the case of missing ending measurements, the participant was excluded from that analysis. This method of exclusion resulted in different numbers being analyzed for each of the secondary outcome variables and medications.

There were no statistically significant differences in mean age, duration of diabetes, HbA_1c, weight, BMI, blood pressure, or fasting lipid parameters between groups at baseline (Table 1). The presence of diabetes-related complications after randomization was similar in both groups. Other demographic and socioeconomic indicators were similar between the two groups. The majority of participants represented ethnic/racial minorities. Approximately 40% had 8 years or less of formal education and ∼80% of this population did not have any education beyond the twelfth grade in high school. Level of education and literacy were not included as eligibility criteria.

The major study end point was glycemic control, as measured by HbA_1c change from baseline. Results of the analysis using the mixed effects model showed that the intervention group had a significant improvement in HbA_1c compared with the control group (Table 3). The intervention and control groups had similar mean HbA_1c values at baseline, 9.54 ± 0.12 and 9.65 ± 0.13%, respectively (P = 0.58). A progressive reduction in HbA_1c over the study period was seen in both groups, from 9.54 to 7.66% (a reduction of 1.88%) in the intervention group and from 9.66 to 8.53% (a reduction of 1.13%) in the control group. However, the reduction in the intervention group was significantly greater at each time period (P < 0.01). The difference between the estimated HbA_1c means was greater in the intervention group than in the control group by at least 0.65 at all time periods and progressively enlarged to 0.87 by study end (Fig. 2). We found no relationship between duration of diabetes and HbA_1c changes from baseline to the end of the study or between duration of diabetes and end HbA_1c in either the intervention or control groups.

We analyzed the percentage of participants who achieved specified HbA_1c goals (≤6.5, ≤7.0, or ≤7.5%) on at least one occasion during the study. Figure 3 shows that the percent of HbA_1c values that achieved target was consistently greater in the intensive group compared with the control group at the level of ≤6.5, ≤7.0, and ≤7.5%. The differences in percent of values achieving target between groups were significant at all three levels of HbA_1c (P < 0.01).

Diabetes medication usage was quantified using a simple scoring system, assigning one point for each medication used and dividing medications into oral agents or insulin. As shown in Table 4, the greater reduction in HbA_1c observed throughout the study in the intervention group was accompanied by an increased use of oral hypoglycemic agents without any significant change in the number of subjects using insulin. In contrast, no significant change in medication use occurred in the control group.

Measures of weight, blood pressure, and lipid status were also analyzed in the intervention and control groups (Table 5). In the intervention group, there were significant changes from baseline to the end of the study with decreases in diastolic blood pressure (P = 0.012), LDL cholesterol (P < 0.001), and total cholesterol (P = 0.018) and an increase in HDL cholesterol (P = 0.012) that were associated with case management. In the control group, there were similar but nonsignificant decrements in diastolic blood pressure, LDL cholesterol, and total cholesterol; however, the increase in HDL cholesterol was statistically significant (P = 0.04). Despite the differences observed between baseline and end of study (up to 36 months of observation) within the intervention group, when the results at the end of the study from the intervention and control groups were compared with each other, there were no significant differences between the two groups for change in weight, BMI, systolic blood pressure, or triglyceride levels.

Frequency of severe hypoglycemia was assessed. Severe hypoglycemia was defined as requiring the assistance of another person, which is comparable with that used in the Diabetes Control and Complications Trial (1). The number of severe hypoglycemic events was small and frequency similar across study sites. The incidence of severe hypoglycemia was greater in the intervention group compared with the conventional group (6.95 vs. 3.55%); however, this difference was not statistically significant (P = 0.28). Likewise, the difference in the rate of hypoglycemic events per 100 patient-years in the intervention and conventional groups was not significant (P = 0.22).

The goals of the case management intervention used during this study focused on improving diabetes care and associated outcomes, with glycemic control being the major outcome measure. Interventions were designed to enhance individual participant self-management efforts, improve access to needed supplies and services, identify and address associated risk factors, minimize participant loss to follow-up, and facilitate access to needed community resources. The study protocol included basic guidelines for glucose and medication management, as well as goals for treatment of hypertension and dyslipidemia. Treatment goals and targets for therapy were uniform across sites. However, each clinical center had flexibility to use individualized treatment algorithms and strategies to achieve these goals. Participant interactions were designed to address glycemic control, lifestyle adjustments, learning deficiencies and needs, and overall adaptation to self-management. Ongoing collaboration with the primary care provider was crucial to the participant’s care.

This study provides evidence that diabetes case management is feasible and can substantially improve glycemic control in a racial/ethnic minority and/or low-income Medi-Cal population. A progressive reduction in HbA_1c over the study period was seen in both groups, from 9.54 to 7.66% (a reduction of 1.88%) in the intervention group and from 9.66 to 8.53% (a reduction of 1.13%) in the control group. The difference between the HbA_1c means as estimated by the mixed effects model was greater in the intervention group than in the control group by at least 0.65% at all time periods and progressively enlarged to 0.87% by the study end. This improvement in HbA_1c in the intervention group was most likely due to the addition of the case-management strategy and not just medication induced.

Both the intervention group and the control group achieved a sizeable drop in HbA_1c levels during this study. However, we had not expected the extent of the improvement seen in the control group. Factors that may have influenced the change in this group include the following: 1) primary care providers responsible for the management of intervention participants also provided care for control group participants and may have altered their intervention strategies based upon observations of unmasked outcomes in the intervention group; 2) control participants were required to have HbA_1c testing every 6 months, and the frequency and results of these data may have had an effect on improving glycemic control; and 3) regression toward the mean in participants with elevated HbA_1c levels at baseline may explain part of the drop in HbA_1c observed in both treatment groups.

The secondary outcome variables showed a response to case management intervention that was similar but less dramatic than the HbA_1c response. In the intervention group, improvement in most risk factors was observed when compared with baseline; this included significant decreases in diastolic blood pressure, LDL, and total cholesterol, accompanied by a significant increase in HDL cholesterol. As with HbA_1c, the control group showed improvements from baseline that were similar but smaller than the intervention group. However, when the improvements observed within each group were compared between intervention and control groups, no significant advantage to being in the case management group was found. This is likely due to the fact that improvement occurred in both groups although it was not generally significant in the control group. We conclude that with respect to our secondary outcome variables, case management was successful but did not manage to create an improvement greater than the modest effect achieved by the combination of improved standards of care and new, more powerful pharmaceuticals in the control group. Also, as with the HbA_1c effect discussed earlier, the design of the study may also have influenced diabetes care practices in the control group as a result of the provider education that accompanied intervention practices, diminishing the differences between the two groups.

In our present study, baseline parameters tested were not predictive of glycemic control outcomes. It is of interest to note that there was a positive correlation between the frequency of HbA_1c testing and the reduction in HbA_1c levels in the intervention group. This issue warrants further study. The observed glycemic control improvement was sustained over the 36 months of the study. This improvement persisted without weight gain or a significant increase in the frequency of severe hypoglycemia, two barriers that often interfere with participant willingness to strive for tight glycemic control (7,9,10).

Eliminating disparities that exist in health status among minority groups has become a national priority (20). Our study has demonstrated that diabetes case management is a feasible treatment approach that can substantially improve glycemic control in disadvantaged populations. The U.K. Prospective Diabetes Study demonstrated that improving glycemic control reduces the risk of the microvascular complications of type 2 diabetes (2). The Center for Disease Control Diabetes Cost-Effectiveness Group has shown that the costs per quality-adjusted life-year for intensive glycemic control are comparable with other commonly funded interventions, especially for individuals age <65 years (3). If disparities in the diabetes health status among racial/ethnic minority and low-income populations are to be reduced or eliminated, this study provides a basis to suggest that additional resources should be invested in providing systematic diabetes case management for these populations (27,28).

Sansum Medical Research Institute (Santa Barbara, CA): Lois Jovanovic, MD; Alison Okada Wollitzer, PhD; Kimberley Yorke, RN, CDE; Joanne Bishop, RD, CDE. Harbor UCLA Medical Center (Los Angeles, CA): Eli Ipp, MBBCh; Constance Vadheim, PhD; Molly Solares, RN; Margarita Enriquez, RN, CDE. Department of Medicine, University of California (San Diego, CA): Orville Kolterman, MD; Gayle M. Lorenzi, RN, CDE; Monica Giotta, MS, RD, CDE. California Diabetes Prevention and Control Program, California Department of Health Services (Sacramento, CA): Curt Weidmer, MD, MPH; Ann Albright, PhD, RD; Guozhong He, PhD.

The State of California Medi-Cal Managed Care Division and Center for Disease Control and Prevention funded the study. This study also received support from the GCRC at Harbor-UCLA Medical Center grant number MO1-RR00425.

The authors wish to acknowledge Lifescan for their generous donation of glucose reflectance meters and supplies and the following individuals who contributed to the study: (Santa Barbara) David Bearman, MD; Mario Shi; (Harbor UCLA) Cindy Mayeda, RN, CDE; Colleen Cortez RN, CDE; Juan Olmos RN; (UCSD) Georges Argoud, MD; David Bazzo, MD; Susan Cahill, RN; Juan Calderon, MD; Ray Fink, MD; Eric Gold, MD; Guadalupe Heredia, RN, CDE; Tyson Ikeda, MD; Juanito Perez, MD; Jorge Robles, MD; and Chi Tran, MPH.