Article Text


Original article
Contemporary risk estimates of three HbA1c variables in relation to heart failure following diagnosis of type 2 diabetes
  1. Stanko Skrtic1,2,
  2. Claudia Cabrera1,
  3. Marita Olsson1,3,
  4. Volker Schnecke1,
  5. Marcus Lind2,4
  1. 1 AstraZeneca R&D, Mölndal, Sweden
  2. 2 Institute of Medicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden
  3. 3 Mathematical Sciences, Chalmers University of Technology, Gothenburg, Sweden
  4. 4 Department of Medicine, NU-Hospital Group, Uddevalla Hospital, Sweden
  1. Correspondence to Dr Stanko Skrtic, AstraZeneca R&D Gothenburg, Mölndal SE-431 83, Sweden; stanko.skrtic{at}


Background We evaluated the association between glycaemic control and the risk of heart failure (HF) in a contemporary cohort of persons followed after diagnosis of type 2 diabetes (T2D).

Methods and results Persons with T2D diagnosed between 1998 and 2012 were retrieved from the Clinical Practice Research Data Link in the UK and followed from diagnosis until the event of HF, mortality, drop out from the database due to any other reason, or the end of the study on 1 July 2015. The association between each of three different haemoglobin A1C (HbA1c) metrics and HF was estimated using adjusted proportional hazard models. In the overall cohort (n=94 332), the increased risk for HF per 1% (10 mmol/mol) increase in HbA1c was 1.15 (95% CI 1.13 to 1.18) for updated mean HbA1c, and 1.06 (1.04 to 1.07) and 1.06 (1.04 to 1.08) for baseline HbA1c and updated latest HbA1c, respectively. When categorised, the hazard risk (HR) for the updated mean HbA1c in relation to HF became higher than for baseline and updated latest HbA1c above HbA1c levels of 9%, but did not differ at lower HbA1c levels. The updated latest variable showed an increased risk for HbA1c <6% (42 mmol/mol) of 1.16 (1.07 to 1.25), relative category 6–7%, while the HRs for updated mean and baseline HbA1c showed no such J-shaped pattern.

Conclusions Hyperglycaemia is still a risk factor for HF in persons with T2D of similar magnitude as in earlier cohorts. Such a relationship exists for current glycaemic levels, at diagnosis and the overall level but the pattern differs for these variables.

Statistics from


The global burden of diabetes has risen dramatically over the last two decades, and it is expected to affect over 500 million adults worldwide by 2030, with the majority having type 2 diabetes (T2D).1 Persons with T2D have a shorter life expectancy, and heart failure (HF) is one of the most common causes of the excess risk of death in these patients.2 ,3

Whether T2D should be considered a causal factor or a comorbidity in HF is unclear.4 ,5 In addition, studies of intensive glycaemic control in preventing cardiovascular (CV) events in persons with T2D have shown somewhat differing results. Three large clinical trials, conducted over a period of 3–5 years, failed to demonstrate clearly beneficial effects of intensive glycaemic control on CV outcomes.6 However, the longer follow-up of the United Kingdom Prospective Diabetes Study (UKPDS) showed an association between intensive glucose control and reduced CV risk,7 and the recent empagliflozin CV outcomes trial (CVOT) showed a reduction in the overall CV death (38%) as well as a markedly preventive effect on HF-related events (35%) by this glucose-lowering agent.8

Observational studies have generally shown a lesser risk of HF at lower glycaemic levels.9–13 However, few population-based real-world studies have evaluated the importance of glycaemic control on the development of HF beginning at diagnosis of T2D, and contemporary estimates are sparse.14 Recently, we found that the estimates of glycaemic control in relation to myocardial infarction varied over time with less strong associations during more recent time periods.15

The most commonly used measure of glycaemia is haemoglobin A1c (HbA1c).16 However, a deeper understanding of the statistical application of repeated measures of HbA1c is needed. When evaluating risk factors for cardiovascular disease (CVD) events in a statistical model, the most appropriate method to account for repeated measurements is not obvious. Consequently, various metrics of HbA1c have been used in studies of diabetic complications.15 ,17 Most commonly used have been the baseline HbA1c and the updated mean HbA1c (which at the time point for each new registration is the mean of all measurements taken thus far).

Therefore, following a similar comparative HbA1c metric approach,15 we sought in this study to evaluate HbA1c in relation to HF in a large contemporary population of persons with T2D and compared three distinct methods using HbA1c measurements from diabetes diagnosis and onwards.


Data were obtained from the Clinical Practice Research Data Link (CPRD), where primary healthcare practitioners in the UK record patient information captured through Electronic Health Record IT systems and updated on regular intervals. CPRD has compiled patients' electronic health records since 1987 and currently collects data for approximately 8% of the UK population. CPRD provides researchers with access to high-quality anonymous healthcare data that include demographic, laboratory, prescribed drug and diagnosis.18 The CPRD also provides linkage to external data sources which form part of the UK Health System such as the Hospital Episode Statistic (HES) data collected on a subset of patients from England and for whom GP consent for the data linkage has been obtained. Ethical approval was granted by the CPRD scientific committee and the National Information Governance Board of Ethics and Confidentiality Committee.

We identified 102 747 patients with T2D in the CPRD diagnosed between 1 January 1998 and 30 June 2012. Index date was defined as the first recorded diagnosis of T2D. Patients aged 18 years or older were included if they had a record in the CPRD at least 3 years prior to diagnosis, and information on gender, age, blood pressure, CV drug use and at least one recorded baseline HbA1c measurement.

Patients below 40 years of age using insulin at diagnosis and continuing with insulin as the only glucose-lowering medication were excluded due to potential misclassification of type 1 diabetes. Follow-up time was defined as the time from T2D diagnosis until the date of HF, death or dropout from the electronic health records for any other reason, or the end of the study on 1 July 2015, whichever came first. HF was identified using the earliest record from the CPRD or HES databases. From these, the following exclusions were implemented: unknown sex (3 subjects), date of death before index date (69 subjects), HF event registered within a 3-year time period prior to their index date (2957 subjects), 1 registration of HbA1c which coincided with the date of death or HF date (47 subjects), only 1 registration of HbA1c with a follow-up time longer than 2 years (364 subjects), no baseline information on blood pressure (1037 subjects) and no baseline information on body mass index (BMI) (3944 subjects). The remaining selected cohort consisted of 94 332 patients of whom 6068 (6.4%) experienced HF during follow-up. Medcodes in CPRD and International Classification of Disease version 10 (ICD-10) codes in HES were used to define HF events as listed in online supplementary table S1.

Three different HbA1c variables were constructed: baseline, updated latest and updated mean. Baseline HbA1c is the value recorded closest to the date of diagnosis within 90 days before and 30 days after diagnosis. Updated latest HbA1c and updated mean HbA1c are time-varying variables, which are recalculated each time a new HbA1c measurement is recorded during the patient's follow-up. Updated latest HbA1c is set to the most recently recorded value, which then represents the patient's HbA1c until a new measurement is taken. Similarly, updated mean HbA1c is the mean of all available HbA1c measurements.

Baseline values for other risk factors were determined by taking the value closest to the T2D diagnosis date, within a 2-year interval consisting of 1 year before and 1 year after diabetes diagnosis. Smoking status was assigned ‘yes’ if the patient had at least once been recorded as smoker or ex-smoker, ‘no’ if all records indicated non-smoker and ‘unknown’ if no information was available. The use of statins, β blockers, ACE inhibitors (ACEi), angiotensin II receptor blockade (ARBs) and/or acetylsalicylic acid (ASA) was defined as an indicator of any CV drug prescription during the 2-year baseline interval.

Statistical analysis

Proportional hazards models were constructed to assess and compare the association between each HbA1c variable and HF. Overall comparisons of the HbA1c variables were based on the estimated linear effect HRs. To further investigate the shape of the risk curves associated with HF, models were fitted with each HbA1c variable categorised as follows: <6% (42 mmol/mol), 6 to <7% (42–53) used as the reference category, 7 to <8% (53–64), 8 to <9% (64–75), 9 to <10% (75–86) and ≥10% (≥86). Each model (one for each HbA1c variable) was stratified for time period (before and after 1 January 2004) in order to allow for different baseline hazard functions in the two time periods, where the incentives for registration of HbA1c differed. Furthermore, all models were adjusted for sex, age, BMI, smoking, prior MI and prior stroke (counting events occurring 3 years prior to index date), systolic and diastolic blood pressure categorised into five classes each (for systolic <126, 126 to <135, 135 to <142, 142 to <155, ≥155 mm Hg, and for diastolic <72, 72 to <80, 80 to <83, 83 to <90, ≥90 mm Hg) and use of statins, β blockers, ACEi, ARBs and ASA at baseline. As the adjusting covariates changed very little between the three different HbA1c models, only data from the model where HbA1c is included as updated mean HbA1c are presented in online supplementary table S2. All adjusting covariates are baseline measurements (ie, using the registration closest in time to the index date, but no more than ±1 year from index date). Potential deviations from model assumptions were evaluated based on the scaled Schoenfeld residuals, and penalised spline functions were used to check the functional form of continuous covariates.19 Incidence rates of HF were estimated using a Poisson regression model allowing for overdispersion and with follow-up time included as an offset.


Median follow-up of the 94 332 patients was 5.8 years, men comprised 56% of the cohort, mean age was 62 years at diabetes diagnosis, mean systolic blood pressure was 141 mm Hg, 64% were on statins, 39% were on ACEi and 54% were smokers or ex-smokers at diagnosis (table 1). In total, there were 6068 HF events registered resulting in a cumulative incidence of 6.4% persons with T2D (table 1). The incidence rate of HF was significantly higher in men throughout all age intervals (tables 2 and 3). Figure 1 shows the estimated incidence rates per age quintile for men and women separately.

Table 1

Patient characteristics at baseline by HbA1c categories

Table 2

Incidence rates (IR) of heart failure (HF) per 1000 patient years in a population with T2D

Table 3

Incidence rates (IR) of heart failure by sex and age per 1000 patient years in a population with T2D

Figure 1

Incidence heart failure rates per 1000 patient years with 95% CIs, for men (squares) and women (circles) across age categories.

Relationship between HF and HbA1c

Regardless of HbA1c modelling, there was a significant association between HbA1c and HF. The estimated overall risk increase per 1% (10 mmol/mol) increase in HbA1c ranged from 6% for baseline HbA1c to 15% for the updated mean HbA1c (table 4). When categorised by HbA1c, the latest variable showed a J-shaped increased risk for HbA1c <6% (42 mmol/mol) of 1.16 (1.07 to 1.25), relative category 6–7%, which was not observed for the updated mean and baseline HbA1c (table 4).

Table 4

HR estimates (95% CI) of the association between heart failure and HbA1c, based on the linear effect model (increase per 1% HbA1c) and categorised HbA1c variables

Comparisons of the three HbA1c variables

According to the estimated linear effect HRs, baseline HbA1c showed the lowest HR for HF, followed by updated latest and updated mean HbA1c (table 4). By HbA1c categories, there were discernible differences in the shape of the risk curves across HbA1c levels (figure 2). Most notable was a significantly increased risk of 16% for the updated latest variable in HbA1c <6%, relative to the reference category 6–7%, where the corresponding estimates of the baseline and updated mean HbA1c showed no risk increase. The updated mean HbA1c variable also notably indicated higher HRs at the upper end of HbA1c categories versus baseline and latest. For the baseline HbA1c, the HRs levelled out above the 8–9% HbA1c category, while for the updated mean HbA1c variable the HRs showed a monotonic increase with increasing HbA1c category.

Figure 2

Estimated HRs with 95% CI for each of the three HbA1c variables, across categories of HbA1c level for heart failure events. Reference category is 6%–<7% (42–53 mmol/mol). The dashed line indicates HR=1.


In this contemporary population study of 94 332 persons followed from diagnosis of T2D, we found that hyperglycaemia remains as an essential risk factor for HF events of similar magnitude as that demonstrated in earlier studies. An association between hyperglycaemia and HF was apparent when measures of glycaemic control were taken at the time of diagnosis of T2D, for the current levels of glycaemic control as well as for the overall control since the diagnosis of T2D. The risk increase for HF per 1% (10 mmol/mol) increase in HbA1c was 15% for the updated mean HbA1c, whereas it was only 6% for baseline HbA1c and the latest HbA1c variables. The risk pattern of HF in relation to the updated mean HbA1c diverged when compared with the other variables mainly at HbA1c levels above 9%, whereas no major differences were found below this level. Also noteworthy, at HbA1c lower than 6% (42 mmol/mol) the updated latest HbA1c variable showed an increased risk of HF compared with the referent HbA1c 6–7% (42–52 mmol/mol), whereas the two other variables showed no such J-shaped association.

The risk increase of HF of 15% by 1% (10 mmol/mol) higher updated mean HbA1c is in line with earlier studies.7 ,10–13 To our knowledge, risk associations between the updated latest HbA1c value and HF have not been evaluated previously in large population-based studies although recently a smaller population-based study found a U-shaped association between mean HbA1c and mortality in chronic HF subjects with T2D.20 In a meta-analysis of four randomised trials of intensive glycaemic control, no preventive effect on HF could be shown.21 It is noteworthy that three of these studies were relatively short and that the effects of intensive therapy on HF may act over longer time periods. Furthermore, in three of these studies the intervention of intensive CV therapy was generally initiated many years after the diagnosis of T2D and, therefore, the effects of CV treatment initiation may differ between early and later stages of the disease. However, the recent empagliflozin CVOT showed a clear preventive effect on HF in persons with T2D and CV disease, which may be due to contributing effects of the medication beyond its glucose-lowering factors.8

In the light of few existing clinical trials of patients with T2D where intensive glycaemic control was initiated at diagnosis and preventive effects on HF were investigated, it is essential that we confirm that a strong association exists between hyperglycaemia and HF in this contemporary population-based study following persons from diagnosis. This finding implies that good glycaemic control is likely essential in preventing HF in persons with T2D and that early as well as overall control is essential. Based on preventive effects by metformin in the UKPDS study,7 metformin is still a first-line option at diagnosis of T2D. The sodium glucose cotransporter 2 inhibitors may from the recent findings in the empagliflozin CVOT be especially efficient in preventing HF, but it will be valuable to further confirm that this is the case also in a more general population of persons with T2D, without known CVD morbidity. The clinical trials of incretin-based therapies have not so far shown any preventive effect on CVD but were generally designed to show non-inferiority and may have preventive effects over longer time periods.22–24

Studying various HbA1c metrics offers clinicians compound perspectives on this biomarker with potentially differing purposes and utilities. The baseline HbA1c provides the clinician with information on whether the glycaemic control at diagnosis already has a predictive value on complications at later stages. However, when meeting patients in the clinic, the current HbA1c value is the main focus from a treatment perspective and may therefore also be so in prognosis. On the other hand, the overall glycaemic control from diagnosis, measured as a mean value, is likely to be a better prognostic marker. This is also essential to account for when estimating the magnitude of hyperglycaemia as a contributing factor to HF, which is crucial in health-economic models and risk-engines used in clinical practice.25–29 However, being able to use the updated mean HbA1c in risk engines requires that the risk engine is incorporated in the electronic medical record system, since it will be burdensome for the clinician to insert multiple historical HbA1c values. Nonetheless, this is an essential point from our findings since the HRs described by 1% higher HbA1c differ greatly depending on the HbA1c metric used. It is noteworthy that in accordance with our recent analysis of HbA1c in relation to MI (15), there was a J-shaped pattern for the latest HbA1c variable. Although this finding is repeated here for another CV complication and it could be inferred that very tight glycaemic control, for example, HbA1c close to normal levels may be harmful, it should be interpreted with caution since patients with HbA1c <6% (42 mmol/mol) have a glycaemic control lower than general targets. There may be characteristics essential for this patient group, which are difficult to control for in the current analyses.

A strength of the present study is the large population studied, which we believe to be the largest observational study of glycaemic control and HF in patients with T2D. The size is essential for obtaining adequately precise risk estimates to compare different HbA1c variables and whether patterns of the associations differ at different HbA1c levels. We also adjusted for the main risk factors but it should be noted that residual confounding cannot be excluded due to the observational nature of this study and lack of availability of recognised prognostic HF biomarkers such as N-terminal pro b-type natriuretic peptide (NT-pro-BNP). Previous studies that have validated HF diagnoses and HF risk assessment methods have found the CPRD to have good levels of accuracy and completeness.30 We are, therefore, confident in our outcomes.

In conclusion, hyperglycaemia remains a strong risk factor for HF in persons with T2D, of similar magnitude as in earlier cohorts. Such a relationship exists both for current glycaemic levels, at diagnosis and the overall level but the patterns differ for these variables.

Key messages

What is already known on this subject?

  • Observational studies have generally shown a lesser risk of heart failure (HF) at lower glycaemic levels. However, there are few contemporary population-based real-world studies that have evaluated the importance of glycaemic control on the development of HF from the beginning at diagnosis of T2D. Also, it is not known which metric of HbA1c is best suited to estimate the glycaemic hazard risk on HF.

What might this study add?

  • In a large contemporary population-based real-world study, we demonstrate that all studied metrics of HbA1c (baseline, updated latest and updated mean) confer a risk increase for HF in incident T2D and thus that hyperglycaemia continues to be a strong risk factor for HF in persons with T2D. Of the studied HbA1c metrics, the updated mean HbA1c showed higher hazard risk estimates than the others, indicating that the average long-term glycaemic control is of clinical importance to reduce HF outcomes. Also by only estimating risk based on baseline or latest HbA1c, the impact of glycemic control can be underestimated.

How might this impact on clinical practice?

  • Hyperglycaemia continues to be a strong risk factor of HF in persons with T2D, and the higher risk estimates for the updated mean HbA1c indicate that there is clinically significant benefit on reducing HF outcomes by implementing glycaemic control.

View Abstract


  • Contributors All authors took part in the design of the study and interpretation of results. MO conducted the statistical analysis. SS and ML wrote the manuscript. VS retrieved the data from CPRD and reviewed and contributed to writing of the manuscript. CC and MO reviewed and contributed to writing of the manuscript. SS 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.

  • Funding AstraZeneca funded access to the CPRD database.

  • Competing interests MO, VS, CC and SS are employed by AstraZeneca. ML has been consultant or received honoraria from AstraZeneca, Medtronic, Novo Nordisk and Pfizer and received research grants from Abbot Scandinavia, AstraZeneca, DexCom, Novo Nordisk, Pfizer and participated in advisory boards for Novo Nordisk.

  • Ethics approval Ethical approval for this study has been obtained from the ISAC (Independent Scientific Advisory Committee) for MHRA Database Research for the Clinical Practice Research Datalink (CPRD).

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.