Trends in Racial and Ethnic Disparities in Early Glycemic Control Among Veterans Receiving Care in the Veterans Health Administration, 2008–2019

Type 2 diabetes (T2D) is a chronic disease affecting >37 million people in the U.S., and ∼10% of Americans had a diagnosis of diabetes in 2023 (1). Maintaining glycemic control from an early stage after diagnosis is crucial to slow down disease progression, as well as to prevent complications and hospitalizations, which are key drivers of medical costs (2,3). Randomized controlled trials, including the Diabetes Control and Complications Trial (DCCT) and the UK Prospective Diabetes Study (UKPDS), have shown that optimal glycemic control reduces microvascular events and neurological and eye complications (4,5).

Currently, American Diabetes Association guidelines recommend hemoglobin A_1c (A1C) <7%, without significant hypoglycemia, as a target goal for glycemic control among patients with diabetes, and a less stringent criterion of A1C <8% is considered in situations where harms outweigh benefits, such as limited life expectancy and cardiovascular comorbidities (6). If uncontrolled A1C reaches ≥9%, it is relatively far from target goal and warrants the use of insulin or a glucagon-like peptide 1 receptor agonist (7).

From 1999 to 2010, the proportion of U.S. adults with diagnosed diabetes who had achieved glycemic control (A1C <7%) increased from 44% to 57%, but control decreased to 51% by 2018 (8). Estimates from a national study suggested that while glycemic control improved among non-Hispanic Black (NHB), Mexican American, and non-Hispanic White (NHW) patients from 1999 to 2016, racial and ethnic disparities in reaching A1C <8% persisted for NHB and widened for Mexican American versus NHW patients (9). Other studies have found that NHB, Hispanic, and Asian American patients are not as likely to meet A1C goals as NHW patients and that American Indian patients have extremely low levels of controlled A1C (10–14).

In the U.S. veteran population, the burden of diabetes is high, with prevalence estimated to be 20–25% and rates of diabetes-related complications also elevated (15,16). The Veterans Health Administration is a centralized national health system that provides affordable and highly protocolized health care for veterans and family members who are eligible for Department of Veterans Affairs (VA) benefits. This affords opportunities to reduce racial and ethnic disparities in diabetes care. Prior studies comparing quality of care between VA and non-VA facilities have generally shown favorable performance in VA facilities, particularly regarding provider adherence to recommended care processes (16,17). Indeed, VA facilities tend to provide annual A1C tests, education, and foot examinations more frequently than non-VA facilities (17).

Despite evidence of high performance, prior studies have identified racial and ethnic disparities in glycemic control among veterans, with lower levels of A1C control observed among NHB and Hispanic veterans compared with NHW veterans (18–20). The first 5 years after diabetes diagnosis may be a key period to stabilize glucose in the long term (21); however, no study has examined longitudinal trends of early glycemic control among veterans to assess whether racial and ethnic disparities have improved. While two VA studies previously estimated average disparities in A1C levels over time by leveraging longitudinal A1C data from electronic health records (EHRs), neither assessed trends in disparities over time (18,19). Thus, the current study examined trends of glycemic control in the first 5 years after diabetes diagnosis across cohorts of veterans with newly identified diabetes between 2008 and 2018 to assess whether disparities in early glycemic control improved or worsened across racial and ethnic groups.

The data come from the Veterans Administration Diabetes Risk cohort, a national cohort of all U.S. veterans receiving care at VA facilities between 1 January 2008 and 31 December 2016 who were free of diabetes at cohort entry and had at least two primary care visits at least 30 days apart in the 5-year period prior to cohort entry. The cohort included 5,987,292 veterans (92% male) at risk for T2D followed through 31 December 2018 to identify incident T2D diagnoses. We excluded patients considered to have T2D if they met any of the following criteria prior to cohort entry: having at least two encounters with a T2D ICD-9/10 code, a documented prescription for T2D medication other than metformin or acarbose alone, or one inpatient or outpatient encounter with a T2D ICD-9/10 code and two elevated A1C measures (≥6.5%). The cohort was constructed by the New York University Grossman School of Medicine and George Mason University (22). The New York University and Veterans Administration institutional review boards approved the study. Given the retrospective nature of the study and use of deidentified data, informed consent was waived in accordance with 45 CFR §46. A detailed description of the cohort has been published elsewhere (22).

Veterans with incident T2D were identified based on the same criteria for T2D as stated above. After excluding patients with gestational diabetes mellitus (n = 24) or missing age and sex information (n = 15 and 3, respectively), 953,089 veterans with incident T2D were identified during 2008–2018. Approximately 88% of patients with incident diabetes (n = 837,023) had A1C measurements at least 7 days after the earliest date of T2D diagnosis (index date) through 5 years after the index date or 31 December 2019 when all veterans were administratively censored in this study.

Because this is a dynamic cohort, the average duration of diabetes among patients increased over time. To assess trends in quality of early diabetes management over time, taking into account this unique feature, we generated four nonoverlapping diagnosis cohorts based on year of diagnosis (2008–2010, 2011–2013, 2014–2016, and 2017–2018) to examine early glycemic control in the VA health system over the four time periods.

In the primary analysis, early glycemic control for each participant during the first 5 years after diagnosis was defined as having an average annual A1C <7% in accordance with American Diabetes Association guidelines (6,8). In secondary analyses, we examined 1) average annual A1C <8% for all veterans (18,20) and 2) a relaxed A1C criterion for veterans >80 years old to ≤8.5% (per VA clinical practice). To facilitate understanding A1C profiles, we also included mean A1C and A1C ≥9% as secondary outcomes based on previous literature (18,19). We defined a patient as lost to follow-up if after missing A1C measurements for 1 year due to any reason, including death, no more A1C measurements were available through the end of follow-up.

Race and ethnicity were self-reported and categorized into Hispanic, non-Hispanic Asian (NHA), non-Hispanic American Indian or Alaska Native (NHAIAN), non-Hispanic Native Hawaiian/Pacific Islander (NHNHPI), NHB, NHW, and unknown. Other covariates included age at T2D diagnosis, sex (male, female), geographic region (Northeast, South, West, Midwest), distance to VA primary care (≥40 miles, <40 miles), and visit pattern (annual number of A1C measurements). A low-income/disability flag was used as a proxy for individual socioeconomic status and was created using VA priority groups and Medicaid eligibility. Veterans were categorized as having a disability, low income, no disability, or none of the above. BMI within 5 years of T2D diagnosis was categorized into underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²), mildly to moderately obese (30–39.9 kg/m²), and severely obese (≥40.0 kg/m²). Elixhauser comorbidity indices were generated by summing the total number of comorbidities among 31 categories, each measured by having at least one ICD-9/10 code at the time of T2D diagnosis (23).

We summarized means with SDs for continuous variables and percent for categorical variables. In descriptive analysis, for each diagnosis cohort, we calculated the average annual A1C value per patient and estimated the annual percentage of patients with average A1C <7% by year since diagnosis. For the entire follow-up, we calculated the weighted average percentage of patients achieving early glycemic control.

To account for irregular visit patterns and repeated binary outcomes, we used mixed-effects modeling with a patient-specific random intercept and estimated population average odds ratios of early glycemic control and corresponding 95% CIs, comparing racial and ethnic groups with NHW. To evaluate racial and ethnic disparities in early glycemic control, we conducted a stratified analysis across the four diagnosis cohorts. In the regression models, we included race and ethnicity, time since diagnosis (centered at 2 years), age at diagnosis (19–34, 35–49, 50–64, 65–79, ≥80 years), sex, annual number of A1C measurements (centered at 1), and an interaction term between age-groups and time since diagnosis. Similar analyses were repeated for secondary outcomes, including early glycemic control, using A1C ≥9% and mean A1C. In addition, we conducted a sensitivity analysis restricting follow-up time to 3 years after diagnosis for all cohorts.

To assess potential selection bias due to missing A1C values, we compared the characteristics (age, sex, race and ethnicity, socioeconomic status, and comorbidities) of veterans lost to follow-up and veterans who had a missing A1C value (intermittent missing A1C) during follow-up with those with complete follow-up in each diagnosis cohort. A two-sided P < 0.05 was used as the cutoff for statistical significance. All analyses were conducted using R 4.0.3 (The R Foundation, Vienna, Austria), GLMMadaptive package version 0.9–1, and lme4 package version 1.1-32.

The base data set generated for the current study is not publicly available, but a curated summary data set can be made available from the corresponding author upon reasonable request.

Mean age of patients at T2D diagnosis were similar in the four cohorts, ranging from 63.7 to 64.0 years. Over time, among patients with incident diagnoses, the proportion of NHW patients decreased from 68.1% to 64.6%, and the proportion of patients from other racial and ethnic groups increased. The proportion of male veterans decreased from 96% to 92%. Differences in other characteristics across diagnosis cohorts were minor, except for a lower accessibility to VA facilities within 40 miles in the 2008–2010 cohort. Elixhauser comorbidity indices were higher in the 2017–2018 cohort than the other cohorts (Table 1).

Patients had a median of one A1C measurement each calendar year (interquartile range 1–2). During the study period, median follow-up time of patients was 4.5, 4.5, 3.7, and 1.5 years in the 2008–2010, 2011–2013, 2014–2016 and 2017–2018 diagnosis cohorts, respectively. Overall, 70% veterans had complete follow-up, while 7% were lost to follow-up and 23% had an intermittent missing A1C during follow-up.

In the 2008–2010, 2011–2013, 2014–2016, and 2017–2018 cohorts, the percentage of veterans with average annual A1C <7% during follow-up was 73%, 72%, 72%, and 76%, respectively. Differences in early glycemic control by diagnosis cohorts remained <5% across year strata since T2D diagnosis (Fig. 1). In each cohort, early glycemic control declined gradually by year since diagnosis (Fig. 1). A sensitivity analysis showed similar downward trends in early glycemic control among veterans with intermittent missing A1C during follow-up; however, the trends decreased less among those who were lost to follow-up (Supplementary Fig. 1).

Adjusting for age and sex, we observed improved early glycemic control for some racial and ethnic groups only relative to NHW across cohorts (Fig. 2A). In the 2008–2010 cohort, veterans of all racial and ethnic groups were less likely to reach A1C <7% compared with NHW veterans (Fig. 2A). In the 2011–2013 and 2014–2016 cohorts, NHB and Hispanic veterans were more likely to reach A1C targets than NHW veterans, while NHA, NHAIAN, and NHNHPI veterans remained less likely to do so (Fig. 2A). In the 2017–2018 cohort, all racial and ethnic groups were equally or more likely to achieve early glycemic control as NHW veterans (Fig. 2A). Results were similar when we further adjusted for visit pattern and when we limited follow-up time to 3 years after diagnosis for all cohorts (Supplementary Table 1 and Supplementary Fig. 2A).

In analyses of secondary outcomes, we observed similar patterns of better early glycemic control (<8%) or lower mean A1C between NHB and NHW veterans in later cohorts and worse early glycemic control between NHAIAN and NHW veterans across cohorts (Fig. 2B and Supplementary Table 2). We also observed that Hispanic and NHAIAN veterans were more likely to have A1C ≥9% than NHW veterans in all cohorts, including the 2017–2018 cohort, which had a shorter follow-up than the other cohorts (Fig. 2C). The same was observed when follow-up time was restricted to 3 years after diagnosis (Supplementary Fig. 2C).

We found that participants lost to follow-up or with intermittent missing A1C during follow-up may have had better health status, including lower average A1C at diagnosis and at the last visit before censoring, suggesting potential differential retention of patients doing worse in managing their diabetes (Supplementary Table 3). This pattern did not differ by racial and ethnic groups across cohorts. Veterans aged <65 years at diagnosis were more likely to drop out during follow-up (Supplementary Table 3). In addition, there were slightly more NHB veterans and veterans with BMI <30 kg/m² at diagnosis among those lost to follow-up or with intermittent missing A1C during follow-up (Supplementary Table 3).

To our knowledge, this study is the first to examine trends in racial and ethnic disparities in early diabetes management among veterans recently diagnosed with T2D. We identified consistently high and stable levels of glycemic control (>70%) in early postdiagnosis years among four cohorts of veterans recently diagnosed with T2D and treated at the VA over the course of 12 years since 2008. Racial and ethnic disparities in early glycemic control existed during these early stages of diabetes management but by smaller magnitudes than previously reported, (18) and trends in early glycemic control differed among racial and ethnic groups. Specifically, disparities in early glycemic control persisted longer, until 2017–2018, for NHAIAN and NHNHPI veterans relative to NHW veterans, while for NHB and Hispanic veterans, disparities largely attenuated over time but still existed among patients with very poor glycemic control (A1C >9%).

Trends in glycemic control at the population level in the U.S. have been reported using the National Health and Nutrition Examination Survey, which represents individuals with both recent and longstanding diabetes (8,9,24,25). One study found a declining trend in A1C <7% at the national level from 2007–2010 (57%) to 2015–2018 (51%), while another study found an improving trend in A1C <8% from 2005–2010 (74%) to 2011–2016 (80%) potentially related to a shift toward more conservative treatments due to side effects of intensive treatments in trials (8,9). In this study of veterans with recently diagnosed T2D, we saw consistently high levels of glycemic control across the various cohorts, with >70% having an A1C <7% and >85% having an A1C <8%. Observed levels of glycemic control using EHR-recorded A1C laboratory results were substantially higher than estimated glycemic control among U.S. adults with diagnosed diabetes during the same period. Although patients in each diagnosis cohort experienced declines in A1C control during follow-up, partially attributed to loss to follow-up of patients in better condition, the overall trajectory of A1C control remained similar across cohorts. High levels of diabetes control may be explained by the restriction of the study period to early postdiagnosis years, a high percentage of veterans (70%) receiving continuous care in the VA health system after T2D diagnosis, and quality improvement strategies implemented in VA hospitals since early 2000, including interventions to promote guideline-adherent practice, standardization of care protocols, and monitoring quality standards (19,26).

Racial and ethnic disparities in glycemic control between NHW and NHB veterans have been previously reported in the VA as well as in other health systems (18–20,25,27,28). In these studies, including the VA, researchers have found that NHB patients had a higher mean A1C and were more likely to have uncontrolled A1C compared with NHW patients (18–20,25,27). More recent studies have also elucidated lower glycemic control among Hispanic and Asian American patients relative to NHW patients (14,28). Another study among American Indian patients in 2002–2004 documented a low level of glycemic control, with 33% of participants having an A1C <7% (13). Many potential causes for these persistent disparities in glycemic control have been identified, including medication adherence, concerns over medications, self-care behaviors, emotional distress, depression, acculturation, language barriers, health insurance coverage, access to care, and system-level quality of care (11,14,28,29). In this study of disparities in early diabetes management, we found relatively small differences (<6%) in glycemic control between the various racial and ethnic groups and NHW group during 2008–2019, which may reflect quality improvement initiatives in the VA since 2000 as well as our focus on the early years after diagnosis (19,26). Nevertheless, certain racial and ethnic groups, including NHNHPI and NHANAI veterans, were still behind in achieving target glucose levels in most diagnosis cohorts in our study, although statistical power may be limited. Trends in diabetes control may be more easily seen when evaluating health outcomes among patients with a new diagnosis, which directly reflect care quality, as opposed to prevalent cases, which may carry influences from prior care, lifestyle, and other health conditions. Findings from this study suggest that differential early glycemic control by racial and ethnic group may have improved in 2017–2019 among veterans during early management of newly diagnosed T2D. Continuous monitoring of quality of care among patients with a new T2D diagnosis may provide valuable information and timely feedback to support additional strategies aimed at improving glycemic control and minimizing disparities.

Building on prior studies that reported disparities in glycemic control by racial and ethnic group, we observed improvement in early glycemic control among NHB and Hispanic veterans relative to NHW veterans. Triangulating findings from other VA studies, however, we found evidence that disparities between NHB and NHW veterans may have been narrowing gradually prior to our study period (18–20). For example, one study identified that the relative odds of uncontrolled diabetes among NHB compared with NHW veterans was 1.8 during 1997–2006 (18); in another study during 2000–2009, the relative odds were roughly 1.5 (19), and a third study in 2015 observed a relative odds of 1.1 (20). Changes in health care environment, such as reduced out-of-pocket costs of metformin, and efforts to improve health equity, including increased awareness of health disparities among physicians, may have differentially improved outcomes among racial and ethnic minority patients. Moreover, increased knowledge and awareness of diabetes among patient populations with high diabetes burden may have contributed to better diabetes self-management. For example, while national studies such as the National Health and Nutrition Examination Survey (NHANES) identified that awareness of prediabetes increased overall from 2005 to 2020 (30), awareness was higher among NHB and Hispanic patients than NHW patients (31). Future studies are needed to validate current findings among NHB and Hispanic veterans.

While greater equity in early glycemic control among racial and ethnic groups in the 2017–2018 cohort is promising, results need to be interpreted with caution. For example, Hispanic and NHAIAN veterans were still persistently more likely to have an A1C >9%, suggesting that continued efforts are needed to improve diabetes management, particularly among those with very poor glycemic control. We were not able to include data after 2020 due to concerns about disruptions to clinical services during the pandemic. Although the follow-up of this diagnosis cohort was shorter than other diagnosis cohorts by design, we did not find differential loss to follow-up by racial and ethnic groups.

This study leveraged four cohorts of veterans who were newly diagnosed with T2D to provide unique insight into early glycemic control across time in the VA health system. Studying glycemic control using incident cases and longitudinal data has advantages over a serial cross-sectional approach, including controlling for disease duration (32). Use of patients with a new diagnosis also teased out disparities among prevalent cases. Moreover, most published studies focused on African Americans and had limited sample sizes for Hispanic American, Asian American, Native Hawaiian/Pacific Islander, and American Indians/Alaska Native patients. The large national VA database ensured sufficient samples in each racial and ethnic group to allow for valid comparison among these groups.

However, this study also has limitations. First, we did not account for variation in care by facilities. Previous studies have found that most variation in A1C control is from within facilities rather than between facilities (18,19). Second, due to the nature of EHR data, sicker patients might be overrepresented in the health system, while heathier patients might be more likely to have intermittent or discontinued care. To address this issue, we used a random intercept model to account for the underlying heterogeneity in EHR visit patterns, and we obtained consistent findings from a sensitivity analysis adjusting for visit frequency. In addition, we did not find meaningful differences in A1C missing patterns by racial and ethnic group. Third, >90% of veterans were male in this study; thus, the findings may not be generalizable to female veterans. Fourth, the shorter follow-up in the 2017–2018 cohort may bias the racial and ethnic disparities in early glycemic control to be smaller compared with the other cohorts. However, our sensitivity analysis showed minor changes in estimates when restricting follow-up time to 3 years after diagnosis. Fifth, only broad racial and ethnic categories were examined in this study, and future studies may delve into more refined categories of race and ethnicity (i.e., among Asian Americans and Hispanic Americans). Finally, the findings in this study may not be generalizable to U.S. veterans who receive care in other health systems.

In conclusion, since 2008, overall trends in early glycemic control in veterans with a recent diagnosis of T2D have remained stable. Within that, trends differed across racial and ethnic groups. Racial and ethnic disparities in early glycemic control may have improved for all groups and particularly for NHB, though disparities in very poor glycemic control were still observed for Hispanic and NHAIAN veterans. More efforts should be directed to understanding diabetes management in patients with poor glycemic control and minimizing any disparities among racial and ethnic groups.

Funding. S.H. was supported by National Center for Advancing Translational Sciences, National Institutes of Health, grant TL1TR001447.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This study is the result of work supported with resources and patient data from the Veterans Health Administration and VA New York Harbor Healthcare System. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the U.S. government.

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

Author Contributions. S.H. wrote the first draft of the manuscript. S.H., R.K., R.A., A.R.T., and L.E.T. conducted the analysis. S.H. and L.E.T. designed the study. All authors were involved in the interpretation of results and edited, reviewed, and approved the final version of the manuscript. S.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at Translational Science 2024, Las Vegas, NV, 3–5 April 2024.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Meghana D. Gadgil.