Early Glycemic Control and Magnitude of HbA_1c Reduction Predict Cardiovascular Events and Mortality: Population-Based Cohort Study of 24,752 Metformin Initiators

Lowering glycated hemoglobin (HbA_1c) levels to <7% (<53 mmol/mol) in most adults with type 2 diabetes has been a recommended target in treatment guidelines for more than a decade (1–4) because of the documented effect in reducing microvascular complications (5). In contrast, it remains debated whether even tighter glucose control (such as HbA_1c <6.5%) may be more beneficial or harmful (6) and what the true effect of tight early glucose control is on subsequent cardiovascular disease (7). Although the large clinical randomized controlled trials have failed to show a clear beneficial effect of early intensive glycemic control on cardiovascular events in type 2 diabetes (5), long-term follow-up studies from the UK Prospective Diabetes Study (UKPDS) (8) and Veterans study (9) have suggested a beneficial cardiovascular effect, termed “metabolic memory” or “legacy effect.” (10) In observational research, most studies, but not all, suggest that having a low glycemic level measured at some point of time is associated with fewer cardiovascular events and mortality in type 2 diabetes (11–18). Some studies found a linear relationship between successively lower glycemic levels and fewer cardiovascular events (11,12,18), whereas others reported a J- or U-shaped curve (14–16). Comparison of these studies is hampered by inclusion of case patients with prevalent diabetes with different time of diabetes duration and by different ways of measuring glycemic control. Only two of the observational studies on cardiovascular risk began at the diagnosis of diabetes (15,18). Olsson et al. (18) found an overall increased risk of acute myocardial infarction with a time-updated HbA_1c <6.0% (42 mmol/mol) compared with 6–7%, whereas Östgren et al. (15) reported the lowest cardiovascular risk was seen with an early achieved HbA_1c level of 6.8% (51 mmol/mol). Whether the magnitude of early HbA_1c reduction predicts subsequent prognosis remains unknown.

Because current guidelines emphasize the importance of tight glycemic control early after diabetes diagnosis, before complications have occurred (4), further research is warranted on the association of early glycemic control with cardiovascular events, taking into account HbA_1c at the initiation of metformin use. Thus, using real-life data from Danish medical registries, we investigated the association of achieved HbA_1c level and magnitude of HbA_1c reduction with subsequent risk of myocardial infarction, stroke, and death in a population-based cohort of patients with incident type 2 diabetes initiating metformin.

We conducted this cohort study using data from existing population-based medical registries covering Northern Denmark region’s 1.8 million residents (30% of Denmark’s population) during 2000–2012. Linkage of all registries was made possible through the unique Civil Personal Registration number (19). In Denmark, most cases of type 2 diabetes are diagnosed by general practitioners (GPs), and an estimated 80% are also treated and monitored there. The remaining 20% are referred by their GP to diabetes outpatient specialist clinics at public hospitals (20). All prescribed medications can only be redeemed at community pharmacies (21). This study was approved by the Danish Data Protection Agency (Record number 1-16-02-1-08). Because this registry-based study did not involve patient contact, no separate permission from the Danish Scientific Ethical Committee was required according to Danish Legislation.

Overall methods are described in more detail in previous reports (22,23). In brief, we identified all patients with type 2 diabetes aged 30 or older living in Northern Denmark who initiated first-time ever glucose-lowering drug treatment between 1 January 2000 and 31 December 2012. Patients had to have at least one available HbA_1c measurement within 12 months before and again ∼6 months after treatment initiation (if several HbA_1c measurements occurred during day 60–180, the value closest to day 180 was chosen) (n = 38,418) (22). Among these individuals, we selected all patients with incident metformin monotherapy use (n = 24,752). Information on prescriptions was obtained via the Aarhus University Prescription Database, which has held complete data coverage in Northern Denmark since 1998 (21). Information on HbA_1c measurements was obtained via the clinical laboratory information system research database (LABKA), which has held complete data on biochemistry test results from all GPs and hospitals in the region since 2000 (24).

We also identified all metformin monotherapy initiators during 2000–2012 who did not have HbA_1c values measured both within 12 months before and ∼6 months after metformin initiation, and these patients were the nonmeasurement cohort (n = 17,142).

We defined two ways to measure early glycemic control:

1. HbA_1c level (%) achieved 6 months after the index prescription of metformin: <6.5, 6.5–6.9, 7.0–7.4, 7.5–7.9, ≥8.

2. Magnitude of HbA_1c reduction, defined as the absolute change in HbA_1c (%) from baseline before metformin initiation to the level achieved at 6 months after prescription: Δ −4, −3, −2, −1, 0, +1, or +2 and above. In more detail, Δ −4 included reductions of −3.5% or more, Δ −3 included reductions from −2.5% to −3.4%, Δ −2 from −1.5% to −2.4%, Δ −1 from −0.5% to −1.4%, Δ 0 from −0.4% to +0.4%, Δ +1 from +0.5% to +1.4%, and Δ +2 included HbA_1c increases of 1.5% or more.

Using the Danish National Patient Registry (DNPR) (25), we identified all patients hospitalized with myocardial infarction and stroke. Diagnoses of myocardial infarction and stroke have been previously validated with predictive values of 90% and 79%, respectively (26,27). We used the Civil Registration System to identify all deaths. We also constructed a combined outcome (death and/or myocardial infarction and/or stroke). All diagnostic codes used in the study are assembled in Supplementary Table 1. We monitored all patients from 180 days after metformin initiation until a cardiovascular event, death, emigration, or 31 December 2012.

From the medical databases, we obtained data on patient characteristics at 180 days after metformin initiation, potentially associated with both the exposure (early glycemic control) and outcome (cardiovascular events and death). We obtained data on 19 major disease categories included in the Charlson Comorbidity Index based on patients’ entire hospital contact history within 5 years before follow-up start. We separately ascertained contacts for obesity, alcoholism-related disorders, and for any macrovascular or microvascular diabetes complication, including prior clinical biochemical indication of renal disease from the DNPR and LABKA (for definitions, see Thomsen et al. [23] and Supplementary Table 1).

From the Prescription Database, we obtained information on the redemption of any other glucose-lowering drug between 14 days after the first metformin monotherapy prescription and until follow-up start. We also obtained information on any antihypertensive treatment, statins, antiplatelet drugs, and treatment with psychiatric medications before follow-up start.

From LABKA we obtained information on the latest baseline HbA_1c and on the latest LDL cholesterol and total cholesterol measurement taken within 12 months before follow-up start. We assessed whether recommended cholesterol targets for persons with diabetes were met or not (<2.5 mmol/L for LDL cholesterol and <4.5 mmol/L for total cholesterol) (28).

We used contingency tables to first describe characteristics for all patients at day 180 after metformin initiation, by achieved HbA_1c (%) level (<6.5, 6.5–6.9, 7.0–7.4, 7.5–7.9, ≥8), and by the magnitude of HbA_1c reduction in % (Δ −4 or more, −3, −2, −1, 0, +1, +2 and above). We created a stacked-bar graph figure to illustrate the distribution of achieved HbA_1c levels for each baseline HbA_1c group (Supplementary Fig. 1).

We used Cox regression analyses to compute crude and adjusted hazard ratios (HRs) with 95% CIs to examine the association between achieved HbA_1c level and absolute change in HbA_1c from baseline (%) at 6 months, respectively, and subsequent risk of cardiovascular events and death after 6 months. We adjusted all analyses for the following covariates (see Table 1 for categorization): age, sex, baseline HbA_1c before metformin start, year of follow-up start, micro- and macrovascular complications, obesity, alcoholism, antiplatelet drugs, statins, antihypertensive drugs, psychiatric medications, reached cholesterol target, and other glucose-lowering therapy than metformin. We repeated the adjusted analysis excluding baseline HbA_1c.

We repeated all analyses, stratified by age-groups and presence of comorbidity at baseline (healthy vs. comorbidity). Adjusted HRs were also stratified by baseline HbA_1c (%) level (<7.5, 7.5–8.9, and ≥9) (Figs. 1 and 2). As a sensitivity analyses, we repeated our analysis while censoring all end points occurring within the first 2 years (as potentially too soon for glycemic control to have had an effect) and another including only individuals with at least 5 years of full follow-up.

To adjust for the potential residual or unmeasured effect of socioeconomic status (using low education as a proxy), we also did a sensitivity analysis with external adjustment (29), where we assumed that the relative risk of cardiovascular events in those with a low educational level was 1.33 (30) and that the proportion of people with low education was 25% among patients with good glycemic control (<7%) and 40% in those with higher HbA_1c levels. We used SAS 9.2 software for all analyses.

Table 1 presents the 24,752 patients with first-time metformin treatment according to achieved HbA_1c level. Compared with those who had an HbA_1c of ≥8% at 6 months, patients who achieved HbA_1c <6.5% were to a larger extent older (≥70 years: 27% vs. 17%), female (47% vs. 38%), and more likely to have initiated metformin in the most recent study years (started in 2010–2012: 60% vs. 35%). They also had slightly more macrovascular (15% vs. 12%) and microvascular (24% vs. 21%) complications at baseline, received more preventive medications, and had less medical obesity (8% vs. 11%). Patients with HbA_1c <6.5% had a much lower baseline HbA_1c compared with those who attained a value ≥8%, and more of them had received no further glucose-lowering add-on therapy within the first 180 days (Table 1). Supplementary Fig. 1 shows the distribution of HbA_1c achieved for each baseline HbA_1c group.

Table 1 also reports patient characteristics by magnitude of HbA_1c change. On the one hand, compared with patients with small HbA_1c reductions, those with a large reduction tended to be younger, had a lower prevalence of macrovascular complications, had less comorbidity, and were prescribed less preventive medication. On the other hand, patients with large reductions in HbA_1c had high HbA_1c at baseline (e.g., 93% of those with HbA_1c reduction of Δ −4% had a baseline HbA_1c >10%) and received add-on glucose-lowering therapy to a greater extent. In contrast, increasing HbA_1c after metformin start was unusual and indicated young age, use of psychiatric medications, and use of drugs other than metformin rather than having a specific level of baseline HbA_1c (Table 1).

Median follow-up for our cohort was 2.6 years (interquartile range [IQR] 1.2–4.7). We observed 439 incident myocardial infarctions, 594 strokes, and 1,845 deaths. Figure 1 shows the unadjusted cumulative incidence of a combined outcome event by achieved early glycemic level at 180 days. The incidence was consistently lower in patients who had achieved HbA_1c <6.5% and was highest in patients who had an HbA_1c ≥8% at 180 days after metformin. As a point of comparison, the outcome incidence in individuals with no available HbA_1c test both before and after metformin (Fig. 1, black stippled line) was similar to that in patients who achieved a low HbA_1c.

Figure 2 shows that after adjustment for confounders, the risk of the composite end point increased with rising levels of early achieved HbA_1c, compared with achievement of HbA_1c <6.5%. The adjusted HR for the combined outcome was 1.18 (95% CI 1.07–1.30) for an HbA_1c level of 6.5–6.99%, HR 1.23 (1.09–1.40) for 7.0–7.49%, HR 1.34 (1.14–1.57) for 7.5–7.99%, and HR 1.59 (1.37–1.84) for ≥8% (Fig. 2 and Supplementary Table 2). Differences between crude and adjusted HRs increased in the highest HbA_1c groups; that is, because patients with high HbA_1c were younger and had less comorbidity at baseline, their high cardiovascular risk further increased after adjustment for these differences in prognostic factors. The adjusted model not including baseline HbA_1c showed consistent results for myocardial infarction, stroke, and mortality but with less precise risk estimates (Supplementary Table 2). Results were also consistent within different strata of age and presence or absence of comorbidity at baseline (Supplementary Fig. 2). The clearest association between higher HbA_1c and worse outcomes was observed in patients aged 70 years or older.

Figure 3 shows that the magnitude of HbA_1c change predicted outcome as well. Large HbA_1c reductions were associated with the greatest outcome risk reductions among patients with a high baseline HbA_1c (i.e., >9%), although statistical precision was limited. In patients with a low baseline HbA_1c (i.e., <7.5%), the pattern tended to be U-shaped, with HbA_1c reductions corresponding to Δ = −2% (from −1.5% to −2.4%) predicting increased outcome risk.

The overall adjusted HR for a combined outcome was 0.80 (95% CI 0.65–0.97) for Δ = −4, HR 0.98 (0.80–1.20) for Δ = −3, HR 0.92 (0.78–1.08) for Δ = −2, and HR 0.99 (0.89–1.10) for Δ = −1 compared with the reference group with no HbA_1c change (Δ = 0) (Fig. 3 and Supplementary Table 3). An increased outcome risk was seen in patients with increasing HbA_1c despite metformin initiation.

A sensitivity analysis that censored all end points occurring within 2 years and was restricted to patients with at least 5 years of full follow-up (Supplementary Tables 4 and 5 and Supplementary Figs. 3 and 4) showed consistent results for both HbA_1c reduction and HbA_1c achieved, albeit statistical precision was poorer.

When we externally adjusted for unmeasured confounding resulting from socioeconomic status (proxy education) for the association between glycemic control and macrovascular events, the analyses yielded results consistent with the overall findings.

In this population-based study of 24,752 metformin initiators, attaining a stringent HbA_1c goal of <6.5% within 6 months was associated with lower risk of cardiovascular events and mortality, with the risk gradually increasing at higher HbA_1c levels. A large magnitude of HbA_1c reduction similarly was associated with a lower subsequent risk of adverse outcomes.

Our results corroborate findings from a few previous observational studies on early glycemic control showing lowered risk of macrovascular events (15,18), and also in the UKPDS trial (31), all in patients with newly diagnosed type 2 diabetes. Olsson et al. (18), monitoring 101,799 patients from the Clinical Practice Research Data Link in the U.K. between 1995 and 2011 from start of type 2 diabetes diagnoses, found an increased risk for myocardial infarction of ∼60% (HR 1.6) at an HbA_1c of 7–8% vs. 6–7%, with a median follow-up of 5.4 years (18). Our corresponding finding was slightly lower (HR 1.5 [95% CI 1.0–2.1]) comparing HbA_1c of ≥8% vs. <6.5%, with a median follow-up of 2.6 years. The similar findings, despite different categorization of HbA_1c and use of different HbA_1c measure (updated means vs. early glycemic control), strengthens the validity of our results.

Our data show that achievement of stringent glycemic levels is possible in real life in at least some elderly patients with comorbidities and that reaching such levels predicts lower risk of vascular events and death. Of note, elderly patients with complications who attained stringent HbA_1c levels in our observational study may have been selected by caregivers through criteria that are not well described in our data, for example, by having a low risk of hypoglycemia or high patient motivation and self-care (3). In accordance, the updated statement from the American Heart Association and the American Diabetes Association suggests stringent targets for selected individual patients, including patients with a short disease duration (7).

A novel finding in our study was that the magnitude of early HbA_1c reduction is an independent predictor of lower cardiovascular risk and death, also after taking pretreatment HbA_1c level into account. Nonetheless, in the subgroup of patients who had a low baseline HbA_1c (<7.5%) before metformin initiation, a reduction in the order of −2% tended to be associated with worse outcomes, consistent with findings from randomized trials (5). As a result of the statistical variation and imprecision of outcome HRs associated with limited size of the change in HbA_1c subgroups, our findings should be interpreted with caution.

Overall, our findings suggest that rapid glycemic response may be used as a possible source of identification of a subgroup of patients with a lower risk of adverse outcomes. It is possible that rapid glycemic responders to therapy initiation (i.e., more easy-to-treat patients) may have a different pathological trajectory and a milder variant of type 2 diabetes than patients who are poor responders. It is also possible that it is the young patients with type 2 diabetes without complications that clinicians dare to treat intensively; for example , reduce HbA_1c >10% to <7%, supported by a recent study from Denmark (32). However, we observed a clear association between early control and improved outcomes also when restricted to people older than 70 years in our study. Moreover, we have previously observed that young patients, to a lesser extent than older patients, reach an early glycemic control <7% within 6 months, possibly related to the fact that the pretreatment HbA_1c with young debut of diabetes often is high (33).

The strengths of this study include a population-based design within the comprehensive Danish public health care system, and accordingly, our data reflect actual clinical practice in diabetes care. Carstensen et al. (34) showed a high sensitivity and positive predictive value (>95%) for identifying patients with type 2 diabetes using Danish registries, with GP registration as the gold standard. Positive predictive values for important comorbidities are also documented as being high in the DNPR (25). Furthermore, we have a comprehensive assessment of cardiovascular events and death, and validity of these codes are also high (26,27). We only looked at myocardial infarction, stroke, and death, yet other studies have found similar associations with glycemia for heart failure, as summed in the meta-analysis by Erqou et al. (35) Finally, we only included metformin initiators, increasing the homogeneity of the population studied and representing the clinical practice today with metformin as the preferred initial pharmacological agent for type 2 diabetes (4). Whether early glycemic control by other oral glucose-lowering drugs is associated with similar benefits remains to be proven.

Study limitations included that only approximately half of all potentially eligible patients had HbA_1c measurements within the right time frame around therapy start. The fact that patients with no HbA_1c measurement available had low outcome risks suggests that this subgroup might have been less severe at start. Over time, the frequency of HbA_1c measurements has increased, and because treatment guidelines have changed over time from initial lifestyle modification to emphasizing early drug treatment, it is likely that patients had high HbA_1c levels in the beginning of our study period primarily as a result of late initiation of medical therapy and that HbA_1c measurements from recent years more reliably reflect baseline glycemic control in newly diagnosed type 2 diabetes. Nonetheless, analyses stratified by period of metformin initiation showed consistent results. Moreover, prescription redemption is only a marker of actual drug consumption. We also have to bear in mind that the guidelines have changed over time to encompass individualized therapy (3,4). Pretreatment HbA_1c levels have decreased substantially over time in Denmark and in other countries, and achievement of early glycemic control has improved (22).

By regression analyses and stratification, we were able to evaluate the effect of a range of possible confounders of the association between early glycemic control and cardiovascular events, and interestingly, few differences were seen comparing crude and adjusted results and in stratified analyses. Also, the analysis adjusted for potential unmeasured confounding by educational level showed consistent results. We had no data on tobacco smoking but were able to adjust for a number of smoking-related diseases. Still, imperfectly measured, unmeasured (e.g., BMI, diet, physical activity, social support, motivation, and self-care), or unknown factors may have affected our risk estimates.

In conclusion, these real-world data provide evidence that not only achievement of early glycemic control but also the magnitude of HbA_1c reduction predicts decreased risk of cardiovascular outcomes and mortality in metformin initiators, independent of baseline HbA_1c levels at treatment initiation. Whereas causality is difficult to prove in our observational study design, these results provide an early prediction tool for identification of patient subgroups with type 2 diabetes that have increased risk for cardiovascular complications and death.

Acknowledgments. The authors thank biostatisticians Johnny A. Kahlert and Nickolaj Risbo Kristensen, both from the Department of Clinical Epidemiology, Aarhus University Hospital, for their kind help and advice on data analyses and preparation of figures.

Funding. The Department of Clinical Epidemiology is a member of the Danish Centre for Strategic Research in Type 2 Diabetes (DD2), supported by the Danish Agency for Science (grant no. 09-067009 and 09-075724), the Danish Health and Medicines Authority, and the Danish Diabetes Association. Project partners are listed on the website at www.DD2.nu.

Duality of Interest. E.S.B. has been an employee of Novo Nordisk Scandinavia AB but has taken a position as a general practice physician in Søndre Land Kommune, Norway as of 1 February 2016. C.L.H. is an employee of Novo Nordisk Scandinavia AB. This study was partly supported by a research grant from Novo Nordisk A/S to Aarhus University. The Department of Clinical Epidemiology is involved in studies with funding from various companies as research grants to (and administered by) Aarhus University, including the current study. No other potential conflicts of interest relevant to this article were reported.

The Department of Clinical Epidemiology at Aarhus University had control of the data and retained final authority over design, content, and interpretation of the analyses, as well as the decision to submit the manuscript for publication.

Author Contributions. E.S., L.M.B., S.P.J., E.S.B., C.L.H., and R.W.T. contributed to the study concept and design. E.S., L.M.B., S.P.J., L.P., H.N., E.S.B., C.L.H., and R.W.T. analyzed and interpreted data. E.S., S.P.J., and R.W.T. drafted the manuscript. L.M.B., S.P.J., L.P., and R.W.T. contributed to data acquisition. L.M.B. and L.P. contributed to statistical analysis. L.M.B., L.P., H.N., E.S.B., and C.L.H. critically revised the manuscript for important intellectual content. S.P.J., L.P., E.S.B., C.L.H., and R.W.T. supervised the study. S.P.J., E.S.B., C.L.H., and R.W.T. obtained funding. S.P.J. and R.W.T. provided administrative, technical, or material support. R.W.T. 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. Part of the results were presented as poster number 1092-P (General Poster Session 11 June and selected for showcase in a Moderated Poster Discussion e-Poster Theater 12 June) at the 76th Scientific Sessions of the American Diabetes Association, New Orleans, LA, 10–14 June 2016. Part of the results were presented as oral presentation number 261 at the 32nd International Conference on Pharmacoepidemiology & Therapeutic Risk Management of the International Society for Pharmacoepidemiology, Dublin, Ireland, 25–28 August 2016.