Differences in Persistently Elevated Hemoglobin A1c for African American and White Patients and the Role of Provider-Level Variation

Introduction

For individuals with diabetes, hemoglobin A1c (A1c) control is important for prevention of microvascular complications. ¹ However, a significant portion of individuals with diabetes have persistently elevated A1c. ² Furthermore, there are well-established racial disparities in A1c levels: On average, African-American patients have A1c levels 0.65% higher than White patients. ³ These disparities remain even after adjusting for socioeconomic status and health care access. ^{4 –6}

Patients with diabetes should have A1c measurement at least every 6 months, according to American Diabetes Association guidelines. In patients with elevated A1c, measurement should occur every 3 months until A1c goals are achieved through medication and/or lifestyle changes. ¹ Primary care providers deliver the vast majority of type 2 diabetes (T2D) management, including HbA1c monitoring and responses to elevated HbA1c, such as T2D medication initiation or intensification; diet, physical activity, and weight loss counseling; performance or ordering of foot and retinal exams; and referral to an endocrinologist, dietitian, or other relevant specialist. ⁷ A lack of treatment responses to persistently elevated A1c is sometimes referred to as clinical or therapeutic inertia, and is thought to be related to a combination of provider-, patient-, and system-level factors. ⁸

However, few studies have examined racial disparities in T2D management in primary care, particularly the contribution of provider-level factors to these disparities. This retrospective study of primary care patients explores differences in persistently elevated A1c between African American and White patients, differences in their primary care visit patterns, and the contribution of primary care providers to A1c variability.

This analysis had 4 goals, mapped to the research questions: (1) to analyze racial differences in patients with any A1c ≥8.5, (2) to analyze racial differences for patients with ≥2 A1c readings ≥8.5, (3) to explore patterns of visits and providers seen in patients with at least 1 elevated A1c, and (4) to explore the contributions of provider variability in patient A1c (because of clustering of patients within providers).

Methods

Data used in this study were obtained from a large urban health system's electronic medical record (EMR; Epic, Epic Systems Corporation, Verona, WI) database. The Thomas Jefferson University Institutional Review Board granted approval for this study. Patient-level data were collected from primary care practices in the health system for all adult patients (ages 18–65 years) with a diagnosis of T2D and ≥2 measured A1cs in the study time frame of January 1, 2017–February 1, 2018. These data included patient demographics (age, sex, race), dates and measured A1cs for the first 4 visits in the time frame, and the provider seen at each visit. Patient-level data also included the medication list at each visit, and the patients' diagnoses that were captured in the Epic EMR problem list.

The majority of patients (80%) had only 2 visits in the specified time frame, 16% had 3 visits, and 4% had 4 visits. Because of the small proportion with more than 3 visits, data were filtered to include only 3 visits. Data were filtered further to include only cases with race coded in the EMR as Black/African American and White/Caucasian. The numbers of participants included in other racial/ethnic categories were too small to conduct any meaningful statistical comparisons. The final number of patients included in this study was 1764.

Descriptive statistics (means, frequencies, and percentages) were calculated for the demographic variables (age, race, sex), as well as for elevated A1c measurements, number of visits, and number of providers seen. Paired t tests were used to evaluate changes in A1c levels across the 2 most recent measurements and a repeated measures analysis of variance was conducted to evaluate the impact of race on these changes. Logistic regression analyses were performed to examine the relationship of the 3 main demographic variables (age, race, sex) with the occurrence of any elevated A1c levels and with persistent A1c levels (at least 2 consecutive A1c measurements ≥8.5). Finally, to investigate associations of elevated A1c by provider, the researchers coded providers who had seen at least 4 patients (necessary to estimate within-provider variability), and used the intraclass correlation coefficient (ICC) to estimate clustering of A1c by provider at all 3 time points. All analyses were conducted using SAS statistical software version 9.4 (SAS Institute Inc., Cary, NC).

Results

The majority of patients included in this study (N = 1764) were female and African American (Table 1). Mean age was 52.6 years (SD = 9.3, range 19–65). Results for patients with at least 1 elevated A1c measurement were similar for sex and age: female (54.1%), mean age 51.2. However, a higher percentage of those with an elevated A1c were African American (Table 1).

A total of 405 patients had at least 1 measured A1c ≥8.5 (23%), and 225 had 2 (or 3) elevated readings (12.8% of the total, 56% of those with 1 elevated reading). The unadjusted relative risk (RR) for any A1c ≥8.5 was 1.36 (odds ratio [OR] = 1.48, P < .001) for African Americans.

A logistic regression predicting any elevated A1c from sex, race, age, and number of visits (2 vs 3), was significant, χ ²  = 72.95, P < .001. All model predictors were significant at P < .01; results for race, sex, and visits are shown in Table 2. The regression accounted for 11.7% of the variability in the outcome using Nagelkerke's R2.

Discussion

A key question asked in this study was whether or not the association of elevated A1c with race was also present among patients receiving care in a large urban primary care system. All of the patients in this study had their A1c levels measured during a 13-month period with this practice. All of the results from the descriptive statistics and the bivariate and multivariate analyses showed an association between race and elevated A1c levels. African Americans were more likely to have elevated and persistently elevated A1c levels than Whites. A positive result from this study was that both White and African American patients with an elevated A1c at time 2 showed significant decreases in A1c levels at the next visit/measurement. The average change was slightly greater for White patients but the difference was not statistically significant.

Among the other demographic characteristics examined, the logistic regressions showed statistically significant associations with age and number of visits. Older patients and patients with more visits (3 versus 2) were more likely to have elevated and persistently elevated A1c levels. However, sex was only statistically significant when analyzing “any elevated A1c.”

The majority of patients (84%) saw the same provider at least twice, and 78% saw the same provider for 3 visits. High percentages in the variability in the occurrence of any elevated A1c and persistently elevated A1c could be accounted for by provider (27% and 32%, respectively). These results suggest a clustering of these elevated measurements among certain providers.

Conclusion

Future research should examine persistently elevated A1c in larger primary care samples to further explore racial disparities in persistently elevated A1c. Future research also should examine the effect of other patient- and provider-level characteristics, including patient demographics and comorbidities and provider activities, on this outcome to identify potential targets for future interventions.