Univariate analysis indicates that multiple variables are associated with glycemic control. Age, race/ethnicity, disease duration, medication, number of Project Dulce visits, duration in Project Dulce, total cholesterol, microalbumin-to-creatinine ratio, BMI, insurance status, and the number of diabetes classes attended were all significant. However, after controlling for baseline A1C, time and other demographic, disease severity, health status, and access/quality of care factors, only age, insurance status, disease duration, pharmacotherapy, and total cholesterol were significant in the final model with main effects or two-way interaction terms.
The association of insurance status with glycemic control contradicts previous studies. Harris et al. [9
] did not find an association between glycemic control and insurance coverage or socioeconomic status using NHANES III data, a representative sample of the U.S. population. Similarly, the Michigan community study [12
], a study of blacks and whites in South Carolina [16
], and a study of whites and Mexican-Americans in Texas [17
] did not find an association between glycemic control and socioeconomic status. Our population, however, is distinct. It is a multiracial/ethnic population and all have a low socioeconomic status. Within this low-income population, the uninsured had poorer glycemic control. They represent a subgroup of patients that struggle to find care for a disease that requires close monitoring. Their disease state may have been out-of-control before entering Project Dulce and thus more difficult to gain control. They may also have lacked self-care skills or basic knowledge of diabetes since their care in the past was likely sporadic. Perhaps, factors such as dietary practices, physical exercise, and education level were important predictors and differed in this subgroup [18
]. Providers may have difficulty procuring medication and equipment for this group of patients. All of these factors could help explain the discrepancy but were not accounted for in this study.
Study findings have differed on the association of glycemic control and disease duration. Similar to Blaum et al. [12
] and contradictory to Nichols et al. [11
], we found that the longer someone had been diagnosed with diabetes, the harder it was to maintain glycemic control. Although self-care skills could improve with longer duration of disease, resistance to medication and the need for higher doses or additional medications increase over time. Insulin use is also a factor of disease severity and was a predictor of poorer glycemic control in this study. The mean A1C value (8.3%) of insulin users in our study was equivalent to the mean value of insulin users in the NHANES III data [9
]. They also found insulin users to have poorer glucose control.
Among health status factors, high total cholesterol was associated with poorer glycemic control. Since patients with diabetes are already at high risk for cardiovascular disease, this finding reinforces the need to aggressively screen and treat elevated cholesterol.
Although other health status factors were not associated with glycemic control in multivariate analysis, it is important to assess the health status of Project Dulce patients compared to other populations. Harris [20
] studied health status and outcomes using NHANES III. Blood pressure was elevated (greater than or equal to 140/90 mm Hg) in 55% to 65% of the population compared to Project Dulce's 30% for SBP (greater than 130 mm Hg) and 12% for DBP (greater than 80 mm Hg). Total cholesterol was greater than or equal to 200 mg/dl in 62% to 69% of the NHANES III population compared to 32% of the Project Dulce population. Cigarette smokers compreised 18% to 24% of NHANES III and 13% of Project Dulce. The prevalence of microalbuminuria and obesity, however, was higher in the Project Dulce population than NHANES III, 36.4% vs. 26% to 30% and 57% vs. 34% to 54%, respectively. The fact that a higher proportion of Project Dulce patients compared to a representative sample of the U.S. population were meeting ADA blood pressure and cholesterol recommendations suggests the positive impact and importance of community disease management programs in low-income, multiracial/ethnic communities.
Prior studies have demonstrated race/ethnicity as a predictor of glycemic control with higher proportions of poorly controlled patients among black women and Mexican-American men [9
]. In univariate analysis, our study found that Asians had better glycemic control than Hispanics, blacks, and whites. However, this relationship disappeared in multivariate analysis after taking other factors into account. Perhaps the reason why our finding differs from other studies is that regardless of race/ethnicity, all study patients were of low socioeconomic status. Prior studies examined race/ethnicity in populations with differing socioeconomic status levels.
Similar to results from Shorr et al.'s study [10
] using NHANES III data, our study found that in multivariate analysis, age was not a significant main effect in predicting glucose control. However, the significant age by time interaction term (age*month) indicates that A1C patterns over time differed between age groups. Figure shows that while the 50 to 65 and 65 and over age groups' A1C values fluctuated over time, the younger age group's A1C values steadily rose. Nichols et al. [11
] also found poorer metabolic control among the younger age group. Since our longitudinal analysis accounted for fluctuations in A1C values, we were able to study A1C pattern differences. It would be interesting to see if this same A1C pattern difference among age groups exists in a representative sample of the U.S. population.
The strength of the current study was the use of mixed effects models. This is the first study that used a longitudinal approach to find factors associated with glycemic control. Incorporating repeated measures over time accounts for fluctuations in glucose control and maximizes the amount of information that can be drawn from the data. Another advantage was the size and diversity of the population which included large numbers of Hispanic, Asian, and white patients, far more diverse than studies using representative samples of the U.S. population.
While the race/ethnic population was diverse, the socioeconomic status of the population was not. Most of the patients were of very low income which limits the generalizability of the study results. Missing data was also a limitation. Missing quarterly A1C values was common but mixed effects models still yield unbiased estimates provided that the missing data was missing at random (MAR) [14
Finally, multiple factors affect glycemic control. The mixed effects model incorporated demographic, disease severity, health status, access/quality of care, and behavioral factors but these are just some of the possible factors that affect glycemic control. Psychological and biological factors, self-care skills, knowledge of disease and education level, diet, exercise, other comorbid diseases, etc. were not explained by this model. Nichols et al.'s [11
] study found that only 9% of the variability in glycemic control was explained by the factors in their model and suggested that personal characteristics may not explain a lot of differences in glycemic control among patients with Type 2 diabetes.
This study identified patients with poorer glycemic control in Project Dulce. The findings should not be generalized to all patients with Type 2 diabetes but can be applied to racial/ethnically diverse, low-income populations. Those who were uninsured, had diabetes for a longer period of time, used insulin or multiple oral agents, or had high cholesterol had poorer glycemic control. The younger population also lagged behind others. Secondarily, this study showed that a high proportion of the patients were meeting ADA's blood pressure and cholesterol recommendations, suggesting that community disease management programs in low-income populations can be effective and may contribute to improved health outcomes.
This study provides a useful methodology to assess disease management systems that collect longitudinal data. It does not provide answers to why patients are not optimally controlled but does provide a starting point from which to investigate and address the obstacles that prevent patients with diabetes from reaching their metabolic targets.