Author + information
- Received December 12, 2012
- Revision received March 26, 2013
- Accepted April 7, 2013
- Published online July 9, 2013.
- ∗Kaiser Permanente Center for Health Research, Portland, Oregon
- †AstraZeneca LP, Wilmington, Delaware
- ↵∗Reprint requests and correspondence:
Dr. Gregory A. Nichols, Center for Health Research, 3800 North Interstate Avenue, Portland, Oregon 97227-1098.
Objectives This study sought to evaluate the relationship between glycemic control and cardiovascular disease (CVD) hospitalizations and all-cause mortality among patients with type 2 diabetes in a real-world setting.
Background Clinical trials have not established that tight glycemic control reduces CVD events and may be associated with increased mortality. Observational studies of specific cohorts have reported increased risk of those outcomes at both high and low glycosylated hemoglobin (HbA1c) levels.
Methods Using the mean of all HbA1c measures over a mean follow-up of 6 years, we created categories of HbA1c (<6.0%, 6.0% to 6.4%, 6.5% to 6.9%, 7.0% to 7.4%, 7.5% to 7.9%, 8.0% to 8.4%, 8.5% to 8.9%, and ≥9.0%) to estimate the risk of CVD hospitalization and all-cause mortality associated with glycemic control, adjusting for demographic and clinical characteristics among 26,673 members of Kaiser Permanente Northwest with type 2 diabetes.
Results Compared with patients with mean HbA1c levels 7.0% to 7.4%, those with mean HbA1c levels <6.0% had a 68% increased risk of CVD hospitalization (hazard ratio [HR]: 1.68 [95% confidence interval (CI): 1.39 to 2.04], p < 0.001) after adjustment for demographic and clinical characteristics. Those with HbA1c levels 6.0% to 6.4% (HR: 1.18 [95% CI: 1.00 to 1.40], p = 0.048) and 6.5% to 6.9% (HR: 1.18 [95% CI: 1.02 to 1.37], p = 0.031) also had significantly higher risk relative to the reference group of 7.0% to 7.4%, as did patients with HbA1c levels 8.5% to 8.9% (HR: 1.55 [95% CI: 1.24 to 1.94], p < 0.001) and ≥9.0% (HR: 1.83 [95% CI: 1.50 to 2.22], p < 0.001). Risk of all-cause mortality was significantly greater than the reference group among HbA1c categories <6.0%, 6.0% to 6.4%, 6.5% to 6.9%, and ≥9.0%.
Conclusions The relationship between mean HbA1c and CVD hospitalizations and all-cause mortality was U-shaped, with greater risk at both higher and lower HbA1c levels.
The American Diabetes Association has long recommended that patients with type 2 diabetes strive for glycosylated hemoglobin (HbA1c) levels <7.0%. Historically, this recommendation was based on the findings of the UKPDS (United Kingdom Prospective Diabetes Study) (1), which showed that intensive therapy that achieved median HbA1c levels of 7.0%, compared with 7.9% in the conventional therapy group, reduced the rate of microvascular complications by 25% (2). Although reduction of cardiovascular disease (CVD) did not reach statistical significance (p = 0.052), subsequent epidemiological evidence revealed that the association between HbA1c and CVD was continuous, with no threshold below which further risk reduction was not achieved (3), thus solidifying the HbA1c recommendation.
The glycemic goal of HbA1c levels <7.0% remained largely unquestioned until the ACCORD (Action to Control Cardiovascular Risk in Diabetes), ADVANCE (Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified-Release Controlled Evaluation), and VADT (Veterans Affairs Diabetes Trial) studies were unable to establish that tight glycemic control (HbA1c targets ≤6.5%) reduced CVD events, and the ACCORD study showed an association with increased mortality (4–6). Subsequent studies conducted among observational cohorts that intensified therapy (7), in the elderly (8,9), and in veterans with heart failure (10) reported an increased risk of mortality and CVD events at both high and low HbA1c levels. Therefore, the objectives of the current study were to evaluate the relationship between glycemic control and CVD hospitalizations and all-cause mortality among a more general diabetes population in a real-world clinical setting followed up for up to 10 years.
This was a retrospective observational cohort study conducted by using the electronic medical records of patients with type 2 diabetes enrolled by Kaiser Permanente Northwest, an integrated healthcare delivery system that provides comprehensive medical services to approximately 480,000 individuals in a 75-mile radius around Portland, Oregon. We identified 26,673 individuals age ≥18 years who were diagnosed with diabetes from 1997 to 2007, had a health plan enrollment period between 2002 and 2011, and did not receive insulin within the first year of diabetes diagnosis. All patients were required to have HbA1c, systolic blood pressure, and low-density lipoprotein cholesterol measured after diabetes diagnosis but no more than 6 months apart between 2002 and 2008. The date of the HbA1c test during the first occurrence of the 3-test combination became the index date. Thus, the index date was not equivalent to the diagnosis date. We excluded patients with a CVD hospitalization before the index date.
Outcomes and observation periods
We examined 2 outcomes independently and as a composite. The first outcome was a first observed hospitalization with a primary diagnosis of CVD (International Classification of Diseases-Ninth Revision-Clinical Modification, codes 410.x, 411.x, 413.x, 414.x) or stroke (codes 430.x, 431.x, 432.x, 434.x, 435.x, 436.x, 437.1). The observation period was from the index date until patients first experienced the outcome, with censoring if they died or left the health plan, or on December 31, 2011. The second outcome was all-cause mortality, with an observation period from the index date until death, with censoring if they left the health plan or on December 31, 2011. The observation period for the composite outcome was from the index date until the earliest occurrence of death or a CVD hospitalization, with censoring if the patient left the health plan or on December 31, 2011.
The exposure variable was the mean of all HbA1c measures during the observation periods. Based on mean values, we used 0.5% increments of HbA1c to categorize patients into 8 levels of glycemic control: <6.0%, 6.0% to 6.4%, 6.5% to 6.9%, 7.0% to 7.4%, 7.5% to 7.9%, 8.0% to 8.4%, 8.5% to 8.9%, and ≥9.0%. Because the observation periods for the 2 outcomes were of different length, patients could be in a different category for each outcome, but 87% were in the same category for all analyses.
We calculated incidence rates for CVD hospitalizations and all-cause mortality per 1,000 person-years for each HbA1c category by using generalized linear regression with Poisson errors and the natural log of person-years as an adjustment for unequal follow-up. PROC GENMOD in SAS version 9.2 (SAS Institute, Cary, North Carolina) was used, controlling for baseline age, sex, and duration of diabetes. A Cox proportional hazards regression analysis was used to more fully adjust the comparisons of HbA1c categories and to evaluate the continuous relationship between HbA1c and the outcomes. The models adjusted for age, sex, diabetes duration, African-American race, smoking status, other clinical CVD risk factors (systolic blood pressure, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, triglycerides, and body mass index) recorded in the outpatient setting within 6 months before or after the index date. We also included variables for presence of macrovascular complications (a composite of ischemic heart disease, stroke, and heart failure) and presence of microvascular complications (retinopathy, neuropathy, or chronic kidney disease defined by a glomerular filtration rate <60 ml/min/1.73 m2) recorded in the outpatient setting any time before the index date. We also adjusted for use of antihypertensive medications (angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, diuretics, beta-blockers), oral antihyperglycemic agents (metformin, sulphonylureas, thiazolidinediones, alpha-glucosidase inhibitors, meglitinides, and dipeptidyl peptidase-4 inhibitors), insulin, and statins within 90 days of the index date. To account for the possibility that HbA1c testing was related to the outcomes, we controlled for the number of tests per year of follow-up. We tested the proportional hazards assumption by creating interaction terms between survival time and the HbA1c categories; none of the interaction terms was significant, indicating each was proportional. The proportionality test of all time-dependent covariates at once produced a p value of 0.10, indicating the proportionality assumption was met.
Patient characteristics for each category of mean HbA1c analyzed for the composite outcome are displayed in Table 1. Patients in lower HbA1c categories were older, with slightly shorter diabetes duration compared with patients in higher HbA1c categories. Those in lower HbA1c categories were more likely to have macrovascular complications and less likely to be using insulin. Nearly identical results were observed when examining patient characteristics across HbA1c categories analyzed for CVD hospitalizations or all-cause mortality (data not shown to avoid redundancy).
Variability of HbA1c values among patients in each HbA1c category is shown in Table 2. SDs and coefficients of variation were low in all HbA1c categories. The proportion of all HbA1c measurements that occurred within the category range was greater among lower HbA1c categories except for HbA1c ≥9.0%, as was the mean difference between each individual HbA1c measurement and the patients' mean values. The number of tests per year of observation was consistent across categories.
Among the 26,673 study subjects, 2,176 (8.2%) experienced a CVD hospitalization, 3,360 (12.6%) died, and 4,853 (18.2%) experienced either outcome (Table 3). The proportion of patients with a CVD hospitalization according to HbA1c category ranged from 7.3% to 9.3%. All-cause mortality seemed to be inversely related to HbA1c category, ranging from 20.4% to 7.4%. CVD hospitalization incidence was greatest among patients with mean HbA1c levels ≥9.0% (18.2 per 1,000 person-years [95% confidence interval (CI): 15.9 to 21.0]), followed by those with HbA1c levels 8.5% to 8.9% (14.6 per 1,000 person-years [95% CI: 12.3 to 17.4]) and <6.0% (13.4 per 1,000 person-years [95% CI: 11.6 to 15.3]). Incidence rates for all other categories were significantly lower and statistically similar to one another. Mortality was highest among those with mean HbA1c levels <6.0% (17.5 per 1,000 person-years [95% CI: 15.8 to 19.4]) and was also statistically higher among those with mean HbA1c levels 6.0% to 6.4% (13.1 per 1,000 person-years [95% CI: 12.0 to 14.3]) and ≥9.0% (15.3 per 1,000 person-years [95% CI: 13.2 to 17.7]) compared with other HbA1c categories. Adjusted incidence rates for the composite outcome were also statistically significantly greater among the two highest and two lowest HbA1c categories.
Results of adjustment using Cox regression models and the covariates displayed in Table 1 are portrayed in Figure 1. Compared with patients with mean HbA1c levels 7.0% to 7.4%, those with mean HbA1c levels <6.0% had a 68% increased risk of CVD hospitalization (hazard ratio [HR]: 1.68 [95% CI: 1.39 to 2.04], p < 0.001). Risk was also statistically higher among those in HbA1c categories 6.0% to 6.4% (HR: 1.18 [95% CI: 1.00 to 1.40], p = 0.048) and 6.5% to 6.9% (HR: 1.18 [95% CI: 1.02 to 1.37], p = 0.031). Patients with HbA1c levels 8.5% to 8.9% (HR: 1.58 [95% CI: 1.26 to 1.99], p < 0.001) and ≥9.0% (HR: 1.98 [95% CI: 1.62 to 2.41], p < 0.001) also had significantly greater risk of CVD hospitalization relative to the reference group. Risk of all-cause mortality was significantly greater than the reference group among the HbA1c categories <6.0%, 6.0% to 6.4%, 6.5% to 6.9%, and ≥9.0%. Results for the composite outcome showed greater risk among patients with HbA1c in categories <7.0% and ≥8.5%. The full Cox models are displayed in Online Table 1.
In a general cohort of 26,673 diabetes patients followed up for a mean of 6.0 years, the relationship between mean HbA1c and CVD hospitalizations and all-cause mortality was approximately U-shaped. Age-, sex-, and duration-adjusted incidence of CVD hospitalizations and mortality was greatest among patients with relatively high mean HbA1c levels (≥8.5%) and also among those with low mean HbA1c levels (<7.0%). These results were robust after further adjustment for known CVD risk factors, complications, and medications typical of patients with diabetes.
Ours is not the first study to observe an association between both high and low HbA1c levels and increased risk of CVD events or mortality. Aguilar et al. (10) reported that in U.S. veteran patients with diabetes and heart failure, the lowest risk of mortality occurred among those with baseline HbA1c levels 7.1% to 7.8%. An analysis of the UK General Practice Research Database found a U-shaped relationship between mean HbA1c levels and mortality and large-vessel disease over 4.5 years after diabetes treatment escalation (7). Two other studies conducted among elderly subjects reported increased risk of CVD or mortality at high and low HbA1c levels (8,9). The current study confirms the U-shaped relationship between HbA1c and CVD hospitalizations and mortality in a general population suggested by these more specific cohort studies.
An observational analysis of UKPDS data found that each 1% reduction of updated mean HbA1c level was associated with 14% reductions in risk of all-cause mortality and myocardial infarction (3). The UKPDS was a trial of newly diagnosed type 2 diabetes participants in whom vascular damage from long-term exposure to hyperglycemia may not yet have occurred. Indeed, studies of patients without diabetes have shown a continuous relationship between HbA1c levels and CVD and mortality (11–13), although the Strong Heart Study found no such relationship (14). It may be that established diabetes patients with more cumulative exposure to hyperglycemia exhibit a different relationship between HbA1c levels and CVD and mortality than nondiabetic or newly diagnosed patients. However, at least 1 cohort study of general diabetes patients in the Swedish National Diabetes Register found progressive risk of CVD and mortality at higher but not lower HbA1c levels, even among registrants with longer duration of disease (15). Other studies suggest a threshold effect such that risk of CVD events and mortality is increased at higher levels of HbA1c but neither increased nor decreased below 7.0% (16,17). Although the observational design precludes the establishment of a causal relationship, our data support the concept of a “sweet spot” for controlling glycemia in type 2 diabetes (18) and also support the recent position statement from the American Diabetes Association and the European Association for the Study of Diabetes that argues for a patient-centered approach to glycemic control (19).
We observed some potentially important differences in the characteristics of patients with low versus high mean HbA1c levels. Specifically, mean age among those with low HbA1c levels (<6.0% and 6.0% to 6.5%) was approximately 62 years, with a mean duration of diabetes of about 2 years. Mean age at diagnosis was therefore about 60 years. In contrast, patients with high mean HbA1c levels (8.5% to 8.9% and ≥9.0%) had a mean age at diagnosis of <50 years. Mean body mass index and systolic blood pressure were somewhat lower among those patients with low versus high HbA1c levels. Taken together, these data suggest that we may be observing different “phenotypes” of type 2 diabetes, with differential contribution of insulin resistance and beta cell failure. If so, how glycemia is lowered may be as important as how much it is lowered. Further exploration of this hypothesis is warranted but is beyond the scope of the current study.
First, as an observational study, we cannot conclude that the observed associations were causal. It is possible that unmeasured confounders contributed to our results. It is also possible that multiple underlying mechanisms are at work, the sum total of which contributes to the associations we report. Thus, our results must be viewed with caution. Second, we did not attempt to adjudicate CVD hospitalization events, relying on the accuracy of coding of inpatient diagnoses. Third, we used statistical models to account for heterogeneity between HbA1c categories, but confounding from unmeasured variables may remain. However, the associations between HbA1c and age, sex, and duration-adjusted incidence rates of the outcomes were robust when we further adjusted for clinical characteristics. The sequential addition of blocks of variables had little to no effect on the size or significance of the HRs for the HbA1c categories. Fourth, we recognize that using the mean of HbA1c values collected during follow-up violates statistical assumptions of survival models. However, re-analysis using time-dependent HbA1c measures did not affect our findings and are more difficult to interpret. Therefore, we report our results in the most easily understood manner possible. Last, the study benefited from the comprehensiveness of the electronic medical records maintained by a fully integrated health system in which patients are all insured and predominantly white; the results may not generalize to other health systems with less information technology support, a more diverse population, or the uninsured.
In this large cohort of patients with type 2 diabetes, we found an increased risk of CVD hospitalizations and all-cause mortality at both higher and lower mean HbA1c levels. Our results confirm that maintaining HbA1c levels <8.0% is important for the prevention of CVD events and death. Lower HbA1c levels are recommended because of well-established microvascular risk reduction (20). Without accounting for how HbA1c is lowered, our results do not support an HbA1c target level of <7.0% for CVD and mortality prevention, a finding consistent with the ACCORD trial and with other epidemiological analyses (4,7–10). The current study supports a patient-centered approach to managing hyperglycemia as recently advocated by the American Diabetes Association and the European Association for the Study of Diabetes that allows less-stringent HbA1c targets for specific patients (19).
Funding for this research was provided by AstraZeneca LP and Bristol-Myers Squibb. Dr. Nichols is an employee of Kaiser Permanente; and has received research funding from AstraZeneca LP, Bristol-Myers Squibb, Merck Research Laboratories, GlaxoSmithKline, Takeda Pharmaceuticals, Novartis Pharmaceuticals, and Tethys Bioscience. Ms. Joshua-Gotlib is a former employee of and a stockholder in AstraZeneca. Dr. Parasuraman is an employee and stockholder of AstraZeneca.
- Abbreviations and Acronyms
- confidence interval
- cardiovascular disease
- glycosylated hemoglobin
- hazard ratio
- Received December 12, 2012.
- Revision received March 26, 2013.
- Accepted April 7, 2013.
- 2013 American College of Cardiology Foundation
- Stratton I.M.,
- Alder A.I.,
- Neil H.A.,
- et al.
- Colayco D.C.,
- Niu F.,
- McCombs J.S.,
- Cheetham T.C.
- Huang E.S.,
- Liu J.Y.,
- Moffet H.H.,
- John P.M.,
- Karter A.J.
- Aguilar D.,
- Bozkurt B.,
- Ramasubbu K.,
- Deswal A.
- Khaw K.T.,
- Wareham N.,
- Bingham S.,
- Luben R.,
- Welch A.,
- Day N.
- Wang H.,
- Shara N.M.,
- Lee E.T.,
- et al.
- Weinrauch L.A.,
- Lewis E.F.
- Inzucchi S.E.,
- Bergenstal R.M.,
- Buse J.B.,
- et al.
- Skyler J.S.,
- Bergenstal R.,
- Bonow R.O.,
- et al.