Regional, Geographic, and Racial/Ethnic Variation in Glycemic Control in a National Sample of Veterans With Diabetes

A national cohort of veterans with type 2 diabetes was created by linking patient and administrative files from the VHA National Patient Care and Pharmacy Benefits Management databases. The Pharmacy Benefits Management database includes use information for every prescription filled in the VA. Veterans were included in the cohort if they had type 2 diabetes defined by two or more International Classification of Diseases, Ninth Revision (ICD-9) codes for diabetes (250, 357.2, 362.0, and 366.41) in the previous 24 months (2000 and 2001) and during 2002 from inpatient stays or outpatient visits on separate days (excluding codes from laboratory tests and other nonclinician visits), and prescriptions for insulin or oral hypoglycemic agents (VA classes HS501 or HS502, respectively) in 2002 (18). Veterans identified as having type 2 diabetes by ICD-9 codes were excluded from the cohort if they did not have prescriptions for diabetic medications (HS501 or HS502) in 2002 (see Supplementary Fig. 1 for cohort definition). The datasets were linked using patient scrambled Social Security Numbers. When the data were merged and limited to include complete adherence data, this resulted in a cohort of 690,968 veterans, of whom 72.86% were NHW, 12.85% were NHB, and 5.11% were Hispanic with type 2 diabetes. There were also 9.18% of veterans with missing or unknown race/ethnicity information. For a comparison of the 690,698 in our study population with those veterans with type 2 diabetes who were not using diabetes medication (n = 201,255), see Supplementary Table 1. The study was approved by our institutional review board and local VA Research and Development committee.

The primary outcome was HbA_1c level. In addition, poor control was defined as HbA_1c ≥8.0%.

The primary covariates of interest were 1) race/ethnicity, classified as NHW, NHB, Hispanic, and other/unknown/missing; 2) location of residence (urban and rural/highly rural) (19); and 3) region, which accounts for the five geographic regions of the country based on the VHA Veteran’s Integrated Service Networks (VISNs): Northeast (VISNs 1, 2, and 3), Mid-Atlantic (VISNs 4, 5, 6, 9, and 10), South (VISNs 7, 8, 16, and 17), Midwest (VISNs 11, 12, 15, 19, and 23), and West (VISNs 18, 20, 21, and 22) (20). The race covariate was retrieved from the 2002 outpatient and inpatient Medical SAS datasets. When missing or unknown, the variable was supplemented using the inpatient 2003 Medical SAS dataset fields for race, the outpatient 2004 Medical SAS dataset fields for race, and the VA Vital Status Centers for Medicare and Medicaid Services field for race.

Age, sex, marital status (i.e., single or married), and percentage service-connectedness (i.e., degree of disability due to illness or injury that was aggravated by or incurred in military service) were available.

Veterans were classified as to whether they were taking insulin only, insulin and oral hypoglycemic agents, or oral hypoglycemic agents only. Annual medication possession ratio (MPR) was calculated and defined as the number of days supply divided by 365 days (or if deceased during that year, the number of days until death) over the study period for each veteran for insulin or oral hypoglycemic agents (VA classes HS501 or HS502, respectively). If MPR exceeded one, MPR was set to one. In addition, MPR was dichotomized at >80%.

Variables included substance abuse, anemia, cancer, cerebrovascular disease, congestive heart failure, cardiovascular disease, depression, hypertension, hypothyroidism, liver disease, lung disease, fluid and electrolyte disorders, obesity, psychoses, peripheral vascular disease, and other (AIDS, rheumatoid arthritis, renal failure, peptic ulcer disease and bleeding, weight loss), and were defined according to ICD-9 codes at entry into the cohort.

Models for the relationship between HbA_1c and race, urban/rural residence, and VISN were developed in a sequential fashion. We used a generalized linear model approach (lmer function in R) to fit the models for HbA_1c as a response variable with race/ethnicity, location of residence (urban/rural), and region as main independent variables of interest with adjustment for potential confounders. The general model had the following regression format, E(Y_i |X_i,Z_i) = g⁻¹(X_iβ + Z_iγ_i), where g is a monotone link function and Y_i is the response for the i^th subject with T_i (T_i = 1,…,5) repeated measurements, X_i and Z_i represent vectors of fixed and random effect covariates, respectively, and g is considered as the identity link resulting in a linear mixed model for the continuous HbA_1c outcome and the logit link for the dichotomous HbA_1c outcome. A person-level random effect was included in all models to account for within-individual correlations. This approach accommodates a wide range of distributional assumptions, multilevel data such as longitudinal measurements on subjects, measurement of subjects at different time points, modeling individual level effects, missing data, and time varying or invariant covariates (21). Model fit was assessed using plots of residuals and predicted values. All statistical tests used a two-tailed α = 0.05 level of significance and were performed in R, version 2.11.0.

The study population consisted of 690,968 veterans with diabetes receiving prescriptions for insulin or oral hypoglycemic agents (VA classes HS501 or HS502, respectively) in 2002 that were followed until death, loss to follow-up, or through December 2006. During the follow-up period, 24.65% of individuals in the cohort died. Table 1 shows the characteristics of the sample.

Relative to NHWs, HbA_1c was 0.45% higher in NHBs, 0.38% higher in Hispanics, and 0.57% higher in individuals with other/unknown/missing racial/ethnic group after adjusting for trends over time (Table 2). After further adjusting for demographic factors, including location of residence and geographic region, racial/ethnic HbA_1c differences were moderately attenuated and changes over time were altered, but the pattern remained similar. Adding medication type, MPR, and comorbidities to the final model altered racial/ethnic group HbA_1c differences only slightly. A small, but statistically significant difference in HbA_1c with respect to rural/urban residence was found with rural residence being associated with slightly higher HbA_1c than urban residence. Small, but statistically significant regional differences were also noted with HbA_1c being lowest in the South and highest in the Mid-Atlantic. Adding medication type (i.e., insulin combined with oral hypoglycemic medication, insulin only, and hypoglycemic medication only), MPR, and comorbidities to the final model altered regional differences in HbA_1c levels moderately, but HbA_1c levels remained lowest in the South and highest in the Mid-Atlantic. Rural/urban differences were attenuated. Medication type was associated with HbA_1c levels. Relative to individuals using only oral hypoglycemic medications, HbA_1c was 0.53% higher in individuals using insulin only and 0.81% higher in individuals using insulin combined with oral hypoglycemic medication. MPR was not significantly associated with HbA_1c levels.

Relative to NHWs, the odds of having uncontrolled versus controlled HbA_1c (defined as HbA_1c >8.0%) was 1.71-fold (95% CI 1.68–1.73) higher in NHBs, 1.63-fold (1.60–1.66) higher in Hispanics, and 1.91-fold (1.87–1.94) higher in individuals with other/unknown/missing racial/ethnic group after adjusting for trends over time (Table 3). The odds of having uncontrolled versus controlled HbA_1c initially decreased, but then increased over time. After further adjusting for demographic factors (including location of residence and geographic region), changes over time were altered, but the pattern remained similar. Odds ratios relative to NHWs for uncontrolled HbA1c were 1.35 (1.33–1.37), 1.53 (1.50–1.57), and 1.14 (1.11–1.16) for NHBs, Hispanics, and individuals with other/unknown/missing racial/ethnic group over time, respectively (Fig. 1). The odds of having uncontrolled HbA_1c levels decreased with increasing age, were lower in women than men, and were higher in single versus married individuals, but were not associated with service-connected disability. Small, but statistically significant differences in the odds of having uncontrolled HbA_1c with respect to rural/urban residence were found; rural residence was associated with a slightly higher odds of having uncontrolled HbA_1c compared with urban residence (Fig. 1). Small, but statistically significant regional differences were also noted with the odds of having uncontrolled HbA_1c lowest in the Midwest and highest in the Mid-Atlantic (Fig. 1). Adding medication type, MPR, and comorbidities to a final model altered changes in HbA_1c control over time, racial/ethnic group differences in HbA_1c control, and regional differences in HbA_1c control only slightly. Medication type was associated with HbA_1c control relative to individuals using only oral hypoglycemic medications. The odds of having uncontrolled HbA_1c were 2.35-fold (2.32–2.38) higher in individuals using insulin only and 3.53-fold (3.49–3.57) higher in individuals using insulin combined with oral hypoglycemic medication. In addition, having an MPR less than 80% was associated with a 5% increase in the odds of having uncontrolled HbA_1c (odds ratio 1.05 [1.04–1.06]).

In the final models, HbA_1c levels and control (i.e., defined as HbA_1c ≤8.0%) were also associated with comorbidities; however, the strength and direction of the association depended on the comorbidity (Tables 2 and 3).

The findings of this study are important for the following reasons. First, few prior studies have examined geographic or regional differences in either HbA_1c testing or HbA_1c levels, and prior national studies have not examined geographic or regional differences in HbA_1c levels. Moreover, prior national studies on racial/ethnic differences in HbA_1c levels at the VA have not focused on HbA_1c levels in Hispanics relative to NHWs or adjusted for type of diabetes medication, medication adherence, or comorbidities. These unique features of our study combined with the unique setting of the VA, where racial/ethnic disparities in access to and use of medical care, and provider behavior are limited, are novel aspects of our study that contribute to the literature and to our understanding of factors that influence HbA_1c levels. Finally, understanding geographic and regional differences in HbA_1c levels and control may have important policy implications if we are able to identify factors within each setting that contribute to differences.

Despite equal access in the VHA, we report longitudinal racial/ethnic differences in HbA_1c. A major contribution of this study is the ability to control for demographic factors, type of medication used (as a surrogate for disease severity), medication adherence, and a variety of comorbidities. In addition, the study included a large sample of Hispanics. Previous VA studies have not examined HbA_1c levels in Hispanics relative to NHWs or NHBs. Notably, our adjusted reported mean difference in HbA_1c levels between NHBs and NHWs is smaller (0.25%) than that reported in a recent cross-sectional study (0.47%), which adjusted for plasma glucose levels, demographics, and a few other variables, not including type of medication used (9). The magnitude of the racial/ethnic disparity was similar in our study as in a smaller VA study (n = 4,080) that reported glucose-independent mean differences in HbA_1c levels between NHBs and NHWs of ∼0.2% among individuals receiving consistent primary care at the VA (22).

Studies outside the VHA have consistently reported racial/ethnic disparities (7). A meta-analysis of disparities in HbA_1c levels between NHWs and NHBs with diabetes, which included 11 articles published between 1993 and 2005, reported crude unadjusted HbA_1c differences between NHWs and NHBs of ∼0.65% (7). Our adjusted estimate of mean difference in HbA_1c between NHBs and NHWs (0.25%) is substantially lower (7). Studies conducted outside a managed care setting generally report wider disparities. One smaller (n = 1,034) study conducted in a nationally representative sample reported mean differences in HbA_1c levels of 0.73% between NHBs and NHWs and 0.74% between Latinos and NHWs after adjusting for demographics, clinical characteristics, access and quality of care, health behaviors, medication adherence, and self-management attitudes (23).

HbA_1c levels were highly correlated with average glucose levels over the proceeding 3-month period (24); however, glucose is not the only factor that affects HbA_1c levels. Current proposed mechanisms for racial/ethnic disparities in HbA_1c levels are physiologic, genetic, socioeconomic, and cultural in nature. The social determinants of poor adherence need to be better elucidated to adequately understand why Hispanics and NHBs have more problems with poor control of their diabetes even when access to care is not a factor. However, in considering mechanisms in racial/ethnic disparities in HbA_1c levels, it is important to remember that glucose-adjusted racial and ethnic disparities in HbA_1c are estimated to be less than 0.5% in individuals with diabetes (5,6,9).

Few studies have examined geographic or regional differences in HbA_1c testing or control (13–15). Results from these studies indicate that HbA_1c testing is lower in rural than urban areas, especially in the rural south (13–15). In one study conducted in Oregon, HbA_1c testing rates in areas with rural health clinics were higher than in areas without rural health clinics and similar to levels in urban areas (15). Thus, there may be important policy implications given the geographic differences that we found.

Strengths of our study include the study population, all veterans with type 2 diabetes receiving care at the VHA had prescriptions for diabetes medication in 2002, the longitudinal design with 5 years of follow-up data, the extensive data available on comorbidities, the ability to identify racial/ethnic group in >90% of the cohort, the longitudinal information on the type of medication used to treat diabetes, and the information on medication nonadherence as measured by an MPR <80%. There are also limitations of our study. First, we do not have information on plasma glucose levels and are therefore unable to examine glucose-independent variability in HbA_1c levels. Second, MPR is really only a proxy for medication adherence, and because of wastage the reliability of MPR for insulin needs further validation; however, prior studies of insulin MPR have been conducted using VA data (25). Third, the VA medical record does not include information on socioeconomic status, which is likely to be an important factor. Fourth, women make up a small proportion, and women veterans may not be representative of women in the general population.

In summary, in a national longitudinal cohort of veterans with diabetes, we found racial/ethnic disparities in HbA_1c levels and HbA_1c control after adjustment for demographic characteristics, medication adherence, type of medication used to treat diabetes, and comorbidities. As efforts are made to better elucidate the mechanisms of racial/ethnic disparities in HbA_1c levels, it will be important to determine whether similar HbA_1c levels across racial/ethnic groups have the same impact on macrovascular outcomes and mortality risk. Moreover, understanding geographic differences in HbA_1c levels and control may have important policy implications if the findings are replicated in non-VA samples and we are able to identify factors within each setting that contribute to differences.

This study was supported by Grant IIR-06-219 funded by the VHA Health Services Research and Development program. The funding agency did not participate in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript. The article represents the views of the authors and not those of the VA or Health Services Research and Development.

No potential conflicts of interest relevant to this article were reported.

L.E.E. conceived and designed the study, researched data, analyzed and interpreted data, reviewed and edited the manuscript, and supervised the study. M.G. and K.J.H. analyzed and interpreted data, wrote the manuscript, and reviewed and edited the manuscript. R.N.A., C.E., and G.E.G. analyzed and interpreted data and wrote the manuscript. P.D.M. conceived and designed the study, researched data, analyzed and interpreted data, wrote the manuscript, reviewed and edited the manuscript, and supervised the study.