Factors Associated With Improved A1C Values Among People With Uncontrolled Type 2 Diabetes

This retrospective case-control study used the electronic health records of patients at a single primary care clinic that provides care to a majority low-income, Hispanic population in Harris County, TX. Patients with type 2 diabetes were eligible for inclusion if they presented for a physician visit with an A1C ≥9.0% between 1 December 2019 and 29 February 2020. The index date was the date of the first physician visit during this period. Each patient was retrospectively followed for 8 months after the index date. An 8-month study period was used to allow collection of up to two subsequent A1C values, assuming that each value would be collected 3–4 months apart. Patients were excluded if they met one or more of the following criteria: <18 years of age, incarcerated, pregnant, diagnosis of type 1 diabetes, or lost to follow-up (defined as no documented clinical encounters or no subsequent A1C values after the index date). The study was approved by Harris Health System’s Human Subjects Institutional Review Board.

Information collected for each study subject included both demographic and clinical data relevant to assess study outcomes. Baseline information for each subject included age, sex, race/ethnicity, health insurance or financial assistance status, A1C value, BMI, duration of type 2 diabetes, use of diabetes medications according to drug class, and high-dose insulin use (a potential signifier of overt insulin resistance, defined as being prescribed insulin ≥1 unit/kg/day [8,9]). Eight months after their index date, the patients’ electronic medical records were reexamined to determine their ending A1C values, change in use of diabetes medications according to drug class, and verified adherence to diabetes medications during the study period (defined by a verified medication possession ratio of ≥80% during the 8-month follow-up period). The number of appointments attended by each subject during the study period was also recorded, including the total number of physician visits, the number of physician visits for a chief complaint of diabetes, and visits to any of the following ancillary services: clinical pharmacy specialist for diabetes medication management, nutritionist for dietary counseling, patient education for diabetes education, or home health nurse.

The primary outcome of the study was the odds ratio (OR; based on nominal logistic regression) of factors that were associated with an improvement in A1C from ≥9.0 to <9.0%. Secondary outcomes included the ORs of factors that were associated with an improvement in A1C from ≥9.0 to <8.0%.

Demographics, clinical characteristics, and outcomes of the subject cohort were summarized with descriptive statistics. For any combination drug product containing two or more active therapeutic agents, each therapeutic agent was regarded as a separate medication for statistical analysis. Dichotomous variables were analyzed using χ² or Fisher exact tests, as appropriate. All continuous variables were tested for normality using the D’Agostino-Pearson normality test. Normally distributed continuous variables were assessed using the Student t test. Non–normally distributed continuous variables were assessed using the Mann-Whitney U test. Nominal logistic regression analysis was performed to assess the effect of significant variables or other variables of interest (including clinical pharmacy specialist visits) on A1C values. A significant variable was defined as any variable that differed significantly between patients with an ending A1C ≥9.0% versus patients with an ending A1C <9.0 or <8.0% (for the primary and secondary end points, respectively).

Based on conservative estimates of each categorical variable being present in 50% of the population, a minimum sample size of 388 was required to detect a 10% difference between proportions in the comparator groups for the purpose of identifying significant variables for inclusion in the logistic regression analysis (using 80% power with an α of 0.05). GraphPad Prism 9.0.0 statistical software (La Jolla, CA) was used for data analysis.

During the eligibility period of 1 December 2019 to 29 February 2020, 365 patients presented to a physician visit with an A1C ≥9.0% at the study site. Of these patients, 156 (42.7%) were excluded based on predetermined criteria, and 209 (57.3%) were included in the study cohort (Figure 1). The most common reason for exclusion was being lost to follow-up (n = 132 [36.2%]). Patients included in the study were primarily Hispanic (n = 186 [89.0%]) and often used the county’s financial assistance program to receive medical care (n = 86 [41.2%]), and the median age was 55 years (interquartile range [IQR] 45–61 years) (Table 2). A minority of the patients were male (n = 64 [30.6%]).

At the time of enrollment, diabetes medication use was widely distributed across the available medication classes (Table 2). Biguanides were the most commonly used agents (n = 151 [72.3%]), followed by basal insulins (n = 126 [60.3%]), dipeptidyl peptidase 4 (DPP-4) inhibitors (n = 64 [30.6%]), insulin secretagogues (n = 61 [29.2%]), sodium–glucose cotransporter 2 (SGLT2) inhibitors (n = 36 [17.2%]), prandial insulins (n = 19 [9.1%]), thiazolidinediones (n = 17 [8.1%]), glucagon-like peptide 1 (GLP-1) receptor agonists (n = 7 [3.3%]), and α-glucosidase inhibitors (n = 6 [2.9%]).

At the end of the study period, 125 patients (59.8%) had an A1C that remained ≥9.0%, 84 (40.2%) had an A1C that decreased to <9.0%, and 54 (25.8%) had an A1C that decreased to <8.0%. When comparing patients who achieved an ending A1C <9.0 or <8.0% versus those with an ending A1C that remained ≥9.0%, there were differences in baseline characteristics between groups (Table 2). Compared with patients whose A1C remained ≥9.0%, those who achieved an A1C <9.0% had a lower median baseline BMI (30.0 vs. 32.6 kg/m², P = 0.0437) and a shorter mean duration of diabetes (8.5 vs. 10.1 years, P = 0.0428). In addition, patients who achieved an A1C <8.0% had a slightly lower median baseline A1C (10.2 vs. 10.6%, P = 0.0344). Compared with patients whose glycemia remained uncontrolled, both groups of patients with improved A1C values had higher baseline use of α-glucosidase inhibitors, lower baseline use of insulin products, and a lower prevalence of high-dose insulin use at baseline.

At the end of the study period, several differences were found between patients who achieved an ending A1C <9.0% or <8.0% and those with an ending A1C that remained ≥9.0% (Table 3). Both groups of patients with improved A1C values demonstrated higher rates of medication adherence. Inversely, use of insulin products was lower among both groups of patients with improved A1C values. Although there was no statistically significant difference in the use of clinical pharmacy services between groups, there was a trend of greater utilization of clinical pharmacy services in patients with improved A1C values (61.9%) versus patients whose A1C remained ≥9.0% (49.6%). Patients whose A1C decreased to <8.0% also had a slightly lower number of medications at the end of the study. There were no differences between groups regarding the addition of various classes of diabetes medications during the 8-month follow-up period.

A multivariable logistic regression model was performed to measure the predictive value of key variables for improvement in A1C values. Variables incorporated into the logistic regression model included baseline A1C, baseline BMI, baseline duration of diabetes, verified adherence to diabetes medications, total number of diabetes medications, any visit to a clinical pharmacy specialist, and use of high-dose insulin at baseline. Use of α-glucosidase inhibitors was not included in the logistic regression analysis because of marginal event rates across all groups. High-dose insulin use was chosen for inclusion in the logistic regression in lieu of basal or prandial insulin use, as this value was considered by the authors to better indicate the use of insulin across multiple formulations and has been previously shown to correlate with objectively measured insulin resistance (9).

The logistic regression model revealed that the following factors were associated with A1C values decreasing from ≥9.0 to <9.0%: high adherence to medications (OR 2.104 [95% CI 1.09–4.115]) and any visit with a clinical pharmacy specialist (OR 1.822 [95% CI 1.004–3.352]) (Table 4). Any visit with a clinical pharmacy specialist was also associated with A1C decreasing from ≥9.0 to <8.0% (OR 2.002 [95% CI 1.026–4.012]). Additionally, the model revealed that high-dose insulin use decreased the odds of achieving an A1C <9.0% (OR 0.360 [95% CI 0.157–0.773]) or an A1C <8.0% (OR 0.253 [95% CI 0.081–0.655]). Other variables including baseline A1C, BMI, diabetes duration, and total number of medications at study end were not predictive of any outcome.

This study illustrates that several avenues remain available for improving A1C values in patient populations with poorly controlled type 2 diabetes. First, high adherence to medications (defined by a medication possession ratio ≥80% of prescribed doses of diabetes medications) was independently associated with an improvement in A1C compared with patients who were nonadherent to their diabetes medications. Although previous studies have discussed both the generally poor medication adherence rates among people with diabetes and the correlation between medication adherence and improvement in diabetes outcomes (10–12), our study demonstrated that, even with low overall rates of adherence (29%), patients who achieve high medication adherence were more than twice as likely to achieve an A1C <9.0% independent of other factors. The strength of this association suggests that more effort should be invested in improving medication adherence for people with poorly controlled type 2 diabetes by implementing local interventions to address population-specific barriers to obtaining and taking prescribed therapies.

Second, our study showed that any visit with a clinical pharmacy specialist nearly doubled the odds of improving A1C values to <9.0 or <8.0%. This effect appeared durable with respect to the high use of clinical pharmacy specialist services (n = 114 [54.6%]). Again, previous studies have shown improvements in clinical outcomes associated with the use of clinical pharmacy specialist services, including decreases in measured A1C, blood pressure, and lipid levels over time (13–15). This study was unique in that it found that pharmacist interventions were effective independent of other critical factors, including choice of medications used, medication adherence, visits with other providers, and other patient characteristics. Clinical pharmacy specialists at the study site established collaborative practice agreements with the physician staff to initiate and modify diabetes medications, order laboratory tests, and complete other health maintenance activities. The beneficial effect of clinical pharmacy specialist visits in this study may be attributed to more frequent follow-up, longer visit times, prioritized focus on diabetes care, and optimal diabetes medication use compared with what patients are able to routinely receive during visits with their primary care provider. The strength of association between clinical pharmacy services and achievement of a meaningful A1C reduction suggests that use of clinical pharmacy specialists should be optimized as a complement to physician services in patients with poorly controlled type 2 diabetes.

Third, patients who failed to achieve improvement in A1C were more likely to use insulin and also had high rates of high-dose insulin use. This study revealed that high-dose insulin use at baseline (defined as being prescribed ≥1 units/kg/day [8,9]) reduced the odds of improving A1C by ∼64–75%. Although there is no unified definition of what is considered to be high-dose insulin, it has been suggested previously that the weight-adjusted insulin dose can serve as a surrogate marker for overt insulin resistance, which correlates with gold-standard laboratory tests for insulin resistance that are expensive, labor-intensive, and infeasible for clinical practice (8,9). The failure of these patients to improve suggests that patients receiving high-dose insulin therapy may need a different approach to care than those who do not require such high doses of insulin. Although insulin initiation or titration may be necessary to overcome severe β-cell dysfunction or insulin resistance in type 2 diabetes, the results of this study suggest that high-dose insulin alone is not a universal solution to poor glycemic control; rather, some patients may benefit from additional noninsulin therapies antecedent to or in addition to insulin therapy (16–18). Further research is warranted to improve therapeutic pathways for patients who fail to respond to high-dose insulin therapy.

This study offers other findings worthy of attention. In the patients whose A1C remained ≥9.0%, the authors expected some improvement in A1C despite failing to reach the 9.0% threshold; however, in these patients, there was only a median 0.2% reduction in A1C by 8 months, compared with a median 2.9% reduction in A1C among patients who achieved an A1C <9.0%. This finding suggests that patients generally improved greatly or not at all and that there was a limited number of patients whose ending A1C was near the 9.0% cutoff. These findings were unexpected and may warrant future examination.

Another unexpected finding was that rates of noninsulin diabetes medication use across therapeutic classes were similar between patients whose A1C improved versus those whose A1C did not improve. This finding surprised the authors, who anticipated that the use of newer therapeutic agents (e.g., SGLT2 inhibitors and GLP-1 receptor agonists) may offer an advantage compared with older medication classes (e.g., biguanides, insulin secretagogues, and thiazolidinediones). The majority of patients were prescribed three to four different diabetes medications simultaneously; however, the unique combinations of various treatment modalities were not examined. In addition, GLP-1 receptor agonist usage was low, likely because of a system-wide focus on the use of multiple oral diabetes medications before initiating injectable medications.

Other patient factors that were hypothesized to influence improvement in A1C (i.e., age, diabetes duration, baseline A1C, health care payer status, and BMI) were not predictive of therapeutic response in this study; however, the differences between groups were relatively small regarding these factors, and a larger sample size would likely be needed to determine the potential effect of these small differences.

An important external factor that may have affected the results is the occurrence of the coronavirus disease 2019 (COVID-19) pandemic during the retrospective chart review from 1 December 2019 through 29 October 2020. At the study site, the onset of the pandemic resulted in a temporary shift to nearly 100% telemedicine visits for physician and clinical pharmacy services in March 2020, as well as a transition to mail-order pharmacy services. In addition, patient education visits, nutrition classes, and home health visits were all limited in accordance with social distancing guidelines, thus decreasing overall use of these services. COVID-19–related disruptions in health care, schools, jobs, and other domains likely contributed to the high number of patients who were lost to follow-up during this period (n = 132 [36.4%]). In addition, in August 2020, the New England Journal of Medicine published an article on the relationship between COVID-19 and diabetes, reporting cases of severe worsening of preexisting diabetes and new-onset diabetes in patients with COVID-19 (19). Unfortunately, we were unable to retrospectively determine which patients in our study were exposed to or infected with COVID-19. The results of our study may have been different in the absence of a global pandemic.

Other limitations of this study include our inability to obtain outside pharmacy records to assess medication adherence. In a post-hoc analysis of a small sample of 105 patients from the original 365 patients, 17 patients (16.2%) filled their medications outside of our health care system and thus were counted as nonadherent because their adherence could not be verified. Our study also used medication possession ratios to evaluate medication adherence; although this approach has been validated for oral formulations, it may not accurately measure adherence to injectable products (20). Additionally, our study was not able to assess other potential factors related to diabetes care such as continuity of health insurance/financial assistance, whether patients self-monitored their blood glucose, and whether patients had metabolic syndrome. The generalizability of this study is also limited to similar populations, as it took place at a single community health center with a predominantly Hispanic, low-income population.

This retrospective case-control study identified multiple factors that may influence A1C-lowering. Factors associated with achieving an A1C <9.0% included medication adherence and any visit with a clinical pharmacy specialist. Conversely, high-dose insulin use decreased the odds of achieving an A1C of <9.0%. Further research should be conducted to determine the generalizability of these findings and to further explore how addressing medication adherence and high-dose insulin use may improve therapeutic outcomes.

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

S.R.W. developed methods, collected data, and wrote the manuscript. R.A.P. developed methods, collected data, contributed to the discussion, and reviewed/edited the manuscript. S.R.W. 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.

The study’s methods were presented in abstract form as a poster at the American Society of Health System Pharmacists’ virtual Midyear Clinical Conference on 10 December 2020. The study’s results were presented in abstract form as a poster at the American College of Clinical Pharmacy Virtual Poster Symposium on 26 May 2021.