Biomarkers of cardiovascular disease: contributions to risk prediction in individuals with diabetes
Cardiovascular disease is a leading cause of death, especially in individuals with diabetes mellitus, whose risk of morbidity and mortality due to cardiovascular disease is markedly increased compared with the general population. There has been growing interest in the identification of biomarkers of cardiovascular disease in people with diabetes. The present review focuses on the current and potential contributions of these biomarkers to predicting cardiovascular risk in individuals with diabetes. At present, certain biomarkers and biomarker combinations can lead to modest improvements in the prediction of cardiovascular disease in diabetes beyond traditional cardiovascular risk factors. Emerging technologies may enable the discovery of novel biomarkers and generate new information about known biomarkers (such as new combinations of biomarkers), which could lead to significant improvements in cardiovascular disease risk prediction. A critical question, however, is whether improvements in risk prediction will affect processes of care and decision making in clinical practice, as this will be required to achieve the ultimate goal of improving clinical outcomes in diabetes.
KeywordsCardiac complications Clinical diabetes Clinical science Epidemiology Human Macrovascular disease Review
American College of Cardiology
American Heart Association
European Society of Cardiology
High-sensitivity C-reactive protein
High-sensitivity troponin T
Net reclassification improvement
N-terminal prohormone of b-type natriuretic peptide
UK Prospective Diabetes Study
Current approaches to cardiovascular disease risk prediction in diabetes
There are three prevailing approaches to cardiovascular disease risk assessment in diabetes (as summarised in the Text box). The first approach considers diabetes as a cardiovascular disease risk equivalent and treats all people with diabetes as if they have cardiovascular disease . The second and third approaches both assess the global risk of cardiovascular disease in individuals with diabetes using multivariable prediction models. The second approach uses prediction models developed in the general population and applies them to people with diabetes . Thus, the same model is used for individuals with and without diabetes. These multivariable models treat diabetes status as a risk factor, much like conventional cardiovascular disease risk factors such as smoking and age. The third approach uses prediction models developed in populations with diabetes, intended for use specifically in people with diabetes . Implicit in this approach is the belief that separate models should be used for individuals with diabetes and for individuals without diabetes.
With regard to the first approach, diabetes has often been regarded as a cardiovascular disease risk equivalent because individuals with diabetes experience a markedly elevated incidence of cardiovascular disease . However, classifying all individuals with diabetes as having a cardiovascular disease risk equivalent may be an oversimplification because substantial heterogeneity in cardiovascular risk can exist among individuals with diabetes, based on age, duration of diabetes and other comorbidities [4, 8]. This supports the argument for more nuanced approaches (i.e. the second and third approaches above; Text box) to estimate absolute risk of cardiovascular disease in individuals with diabetes using multivariable prediction models that account for multiple cardiovascular disease risk factors.
A challenge is posed by the limitations of the prediction models currently used for assessing cardiovascular risk in people with diabetes. In 2012, van Dieren et al observed that of the 45 models used for predicting cardiovascular disease in individuals with diabetes reported in the literature, 33 were developed in the general population (i.e. the second approach above; Text box) and 12 were specifically developed using diabetic populations (i.e. the third approach above; Text box) . The most commonly included predictors were clinical characteristics (age, sex, tobacco use) and lipid biomarkers (e.g. total cholesterol, LDL). Although many predictive models exist, only a minority have been externally validated in new diabetic populations and tested for predictive accuracy. Even the models that have been externally validated, such as the Framingham risk score  and the United Kingdom Prospective Diabetes Study (UKPDS) model , have limited ability to accurately estimate cardiovascular risk in type 2 diabetes [12, 13].
Improving prediction of cardiovascular disease in individuals with diabetes
Risk models for use in individuals with diabetes typically do not include information from biomarkers other than cholesterol and, occasionally, HbA1c and albuminuria. This raises the question of whether adding novel biomarkers would improve cardiovascular risk prediction in individuals with diabetes.
What are the strategies for determining whether a novel biomarker improves the ability to predict cardiovascular disease risk? Risk prediction can only be improved by a novel biomarker if it provides additional information that is not explained by established cardiovascular disease risk factors such as age, tobacco use and lipid levels. Documenting a statistically significant association between a novel biomarker and cardiovascular disease in models adjusting for conventional risk factors is an important first step. Nonetheless, showing an association is not sufficient to establish whether the biomarker improves risk prediction [14, 16]. Several additional metrics exist to assess whether a novel biomarker improves risk prediction: discrimination, calibration, and reclassification.
Discrimination addresses the model’s ability to distinguish between individuals who will develop the outcome and those who will not. The most commonly used discrimination measure is the area under the receiver operating characteristic curve, also known as the C-statistic. Values range from 0.5 (no discrimination ability) to 1.0 (perfect discrimination). Thus, a possible approach to demonstrating the utility of a novel biomarker is to evaluate whether the C-statistic increases when the biomarker is added to conventional risk factors .
Calibration measures, such as the Hosmer–Lemeshow statistic, address the agreement between the risk predicted by the model compared with actual event rates in the population. Well-calibrated models yield accurate absolute risk estimates, which facilitate the assessment of the risks and benefits of various therapeutic approaches [14, 15]. For instance, an intervention associated with a 20% reduction in cardiovascular disease risk but also with a high incidence of adverse effects might be more reasonable to consider in individuals at high absolute risk of cardiovascular disease.
Reclassification addresses the question of whether the addition of a biomarker leads to an appropriate change in the risk categorisation of individuals across a risk threshold. The most widely used measure of reclassification is the net reclassification improvement (NRI). An appropriate reclassification is one in which individuals who ultimately experience an event are reclassified into a higher risk group and individuals who do not experience an event are reclassified into a lower risk group . The NRI is most useful when clearly defined risk thresholds exist for a condition and when these are tied to treatment decisions.
Examples of cardiovascular disease biomarkers in individuals with diabetes
Most investigations of cardiovascular disease biomarkers in diabetes have been hypothesis-driven, candidate biomarker studies focusing on known pathophysiological pathways, including those related to cardiac stress (N-terminal prohormone of b-type natriuretic peptide [NT-proBNP] [8, 17, 18, 19, 20, 21, 22, 23] and high-sensitivity troponin T [hsTnT] [8, 17, 24]), inflammation (high-sensitivity C-reactive protein [hsCRP] [8, 25], IL-6 , IL-15 ), matrix remodelling (matrix metalloproteinases and osteopontin [17, 23]), lipids , endothelial dysfunction (albuminuria ) and diabetes pathophysiology (AGEs) [17, 28]). As in the general population, there is good evidence in diabetic populations that NT-proBNP [8, 17, 18, 19, 20, 21, 22, 23], hsTnT [8, 17, 24] and hsCRP [8, 25] are predictive of cardiovascular disease.
Summary of selected studies of cardiovascular disease biomarkers in populations with diabetes
Sample size and patient population
Prior CVD (%)
No. of events
No. of biomarkers evaluated originally
Biomarkers in final model
Improvements in discrimination
Was the NRI statistically significant?
Looker et al 
CVD incidence (acute CHD, stroke)
N = 2310 T2DM
• Apolipoprotein C-III
• Soluble receptor for AGE
0.66 → 0.72
van der Leeuw et al 
Composite CV outcome (MI, stroke, CV death)
N = 1002
• Matrix metalloproteinase 3
0.70 → 0.73
Composite CV outcome (MI, stroke, CV death)
N = 288
• Matrix metalloproteinase 3
0.69 → 0.72
Gerstein et al 
Composite CV outcome (MI, stroke, CV death)
N = 8401
Dysglycaemia (T2DM, impaired fasting glucose, impaired glucose tolerance)
• Trefoil factor 3
• Apolipoprotein B
• Glutathione S transferase α
• Growth/differentiation factor 15
• Hepatocyte growth factor receptor
• Chromogranin A
0.64 → 0.71
These data suggest that certain biomarkers and biomarker combinations can lead to modest improvements in risk discrimination, beyond traditional cardiovascular risk factors. Estimates of NRI vary, though such estimates can depend on the selection of risk thresholds which do not always conform to clinical guidelines. ‘Category-free’ NRI measures have been developed that do not depend on the selection of thresholds, although such measures may provide less insight regarding the clinical utility of biomarker data. These observations lead to the question of whether novel biomarker information will have a clinical impact (discussed below).
Applications of cardiovascular disease risk prediction models in clinical practice
As already discussed, risk prediction models (the second and third approaches above; Text box) have been developed to estimate absolute risk of cardiovascular disease in individuals with diabetes. Risk prediction models have a clinical impact only if they are adopted into clinical practice and if their use in the clinical setting changes providers’ management decisions (with regard to therapy, screening and education) and/or patients’ behaviour [29, 30, 31]. To what extent are cardiovascular disease risk prediction models incorporated into current clinical guidelines issued by major medical societies? The European Society of Cardiology (ESC) and EASD 2013 guidelines  mention several available cardiovascular disease risk calculators for individuals with diabetes and also mention several cardiovascular disease biomarkers in diabetes. However, they do not explicitly advocate the use of risk calculators or biomarkers to guide clinical management. The ESC/EASD 2013 guidelines recommend classifying all individuals with diabetes as being at ‘high risk’ of cardiovascular disease (based solely on the presence of diabetes, without any other cardiovascular risk factor or evidence of target organ damage) or at ‘very high risk’ (documented cardiovascular disease, severe chronic kidney disease or presence of at least one cardiovascular risk factor or target organ damage). Thus, these guidelines recommend statin therapy for most individuals with diabetes.
The American Heart Association (AHA)/American College of Cardiology (ACC) 2013 guidelines recommend statin therapy for all individuals with diabetes aged 40–75 years, based solely on the presence of diabetes, and recommend that statin therapy be considered for patients with diabetes outside that age range based on the individuals’ potential for cardiovascular disease risk reduction and personal preferences . The AHA/ACC 2013 guidelines encourage healthcare providers to calculate the absolute risk using the AHA/ACC Pooled Cohort Risk Assessment Equation to guide the intensity of statin treatment in diabetic individuals who are 40–75 years of age.
In summary, the current lipid management guidelines from major European and American medical societies recommend statin therapy for most individuals with diabetes, based solely on the presence of diabetes, reflecting the first approach discussed above (Text box) which considers diabetes as a cardiovascular disease risk equivalent. However, even with this approach, risk prediction models and biomarker data may be used to inform the intensity of statin treatment or to help communicate risk to patients.
It is worth repeating that substantial heterogeneity in the risk of cardiovascular disease exists among individuals with diabetes [4, 8]. This raises the question whether a more nuanced approach to the treatment of cardiovascular risk in individuals with diabetes should be considered. In one study, individuals with type 2 diabetes and multiple risk factors for cardiovascular disease had a 2 year rate of cardiovascular death, myocardial infarction or ischaemic stroke of 20% when levels of NT-proBNP, hsTnT and hsCRP were elevated. In contrast, the risk was only 1.6% for those with no elevation in any of these three biomarkers . The absolute risk would likely be even lower in younger individuals with no clinical risk factors other than diabetes. According to both ESC/EASD and AHA/ACC guidelines, a 40-year-old individual with diabetes without any other clinical risk factors should be given statin therapy. However, if biomarker data revealed that this individual had an extremely low 10 year risk of cardiovascular events, it may be reasonable to defer statin therapy.
Thus, it remains possible, though unproven, that refinements in risk prediction may inform the application of current clinical guidelines, the aggressiveness of lipid reduction strategies, use of adjunctive medications beyond statins and communication of risk to patients. Very few publications have reported the impact of risk models on clinical management, not to mention clinical outcomes . Well-designed, prospective studies examining the effectiveness of different screening and prevention strategies can be expensive and take many years to complete .
Cardiovascular risk and management in individuals with prediabetes
In contrast to diabetes, other states of hyperglycemia such as prediabetes (including impaired fasting glucose and impaired glucose tolerance) are not currently considered to be cardiovascular disease risk equivalents. Cardiovascular risk is elevated in individuals with prediabetes and hyperglycaemic conditions other than diabetes  but risk models such as the Framingham risk score and the AHA/ACC risk calculator do not account for these conditions. Moreover, there are no specific recommendations regarding the management of cardiovascular risk in individuals with prediabetes. Individuals with prediabetes are advised to institute healthy lifestyle changes (diet and exercise) but otherwise are treated the same as non-diabetic individuals regarding cardiovascular risk management. For instance, neither the ESC/EASD nor the AHA/ACC guidelines for lipid management currently distinguish between individuals with pre-diabetes and those without diabetes.
Nonetheless, diabetes and prediabetes are not binary outcomes but rather fall along a continuum. Individuals with advanced prediabetes and those with early, mild diabetes may have a similar risk of cardiovascular disease. Thus, using the clinical onset of diabetes as the starting point for cardiovascular disease risk management may be too late. If biomarkers could refine the prediction of risk in prediabetic individuals, then those who would benefit from more aggressive risk management might be identified. An additional consideration is preventing diabetes itself since strategies aimed at preventing diabetes also lower cardiovascular disease risk. Identifying those at highest risk of developing diabetes would help target diabetes prevention efforts. Most of the currently known biomarkers offer only a modest improvement in risk prediction for type 2 diabetes beyond traditional risk factors . However, recent technologies enabling high-throughput profiling of metabolic status (metabolomics) have identified novel biomarkers that improve prediction of diabetes and show promise for the future .
Biomarkers as an approach to increasing the efficiency of clinical trials
The US Food and Drug Administration requires that manufacturers of novel diabetes medications conduct large-scale outcomes trials to evaluate the cardiovascular risks of these drugs. Despite the elevated risk of cardiovascular disease in individuals with diabetes, short-term event rates are still moderate, meaning that very large sample sizes are required for adequately powered clinical trials. Accordingly, recent and ongoing cardiovascular trials of diabetes medications have had sample sizes of 3300–17,000 individuals . Biomarkers that improve the identification of high-risk individuals with diabetes may allow smaller trials to be performed in these selected subgroups.
As noted above, diabetes is a heterogeneous disease. The physiology of diabetes may differ by sex, race and age. Thus, it is conceivable that the predictive ability of cardiovascular disease biomarkers in individuals with diabetes may differ by clinical covariates. In non-diabetic individuals, biomarkers may perform differently according to clinical characteristics (e.g. in healthy postmenopausal women, hsCRP is an even stronger predictor of cardiovascular disease than is LDL ). There are few analogous data in diabetic populations but investigations to determine whether clinical characteristics such as sex, race and age influence the association of cardiovascular disease biomarkers with events are warranted. This information may be particularly useful in subgroups of younger individuals in whom the average absolute risk of events is low despite the presence of diabetes.
A related question is whether certain biomarkers may improve cardiovascular disease risk prediction selectively in diabetic vs non-diabetic individuals. Because the distribution of certain biomarkers may differ by diabetes status, their abilities to predict cardiovascular disease may also differ by diabetes status . This appears to be the case for glucose levels and for markers of glucose-mediated damage. For example, whereas higher glucose levels are positively associated with cardiovascular disease risk in both diabetic and non-diabetic individuals [40, 41], HbA1c levels improve cardiovascular disease risk prediction in individuals with diabetes [42, 43] but not consistently in the general population [44, 45]. In addition, markers of glucose-mediated damage, such as urinary microalbumin and AGEs, may be particularly useful for predicting cardiovascular disease risk in individuals with diabetes [17, 43]. In contrast, the predictive abilities of C-reactive protein , fibrinogen  or γ-glutamyltransferase  do not appear to differ by diabetes status. Whether the predictive abilities of many cardiovascular disease biomarkers are modified by diabetes status is largely unknown. Addressing these questions is valuable not only for improving risk prediction but also for understanding how the pathophysiology of cardiovascular disease differs in individuals with and without diabetes.
Another important area of future research is the discovery of new biomarkers of cardiovascular disease risk in individuals with diabetes. As discussed above, if a biomarker has a relatively low correlation with existing predictors, it is better able to improve prediction of disease (Fig. 1). Because most studies so far have focused on traditional pathophysiological pathways for cardiovascular disease, such as inflammation, cardiac injury and oxidative stress, the potential improvement in prediction afforded by identifying additional biomarkers from the same pathways may be limited as such biomarkers are likely to be correlated.
In contrast to candidate-based designs, hypothesis-free approaches such as proteomics and metabolomics can simultaneously evaluate hundreds or thousands of biomarkers and offer the potential of identifying biomarkers outside the traditionally studied pathophysiological pathways. Proteomics- and metabolomics-based studies in related fields have enabled the discovery of novel biomarkers that improve the prediction of new-onset type 2 diabetes  and incident renal disease in type 2 diabetes . These studies highlight the potential value of analogous studies to identify biomarkers of cardiovascular disease among individuals with diabetes.
Treatment guidelines for the primary and secondary prevention of cardiovascular disease in diabetic populations do not fully account for the heterogeneity in risk among individuals with diabetes. Certain biomarkers and combinations of biomarkers can lead to moderate improvements in cardiovascular disease risk prediction in diabetes. A critical question is whether the resultant improvement in risk stratification would lead to changes in clinical care for individuals with diabetes. The prospects for biomarker-guided management in diabetes would be further enhanced by expanding the number of biomarkers that contribute to risk stratification. Efforts to identify these biomarkers should be aided by recent advances in platforms for biomarker discovery that enable investigations involving broad panels of biomarkers not restricted to selected pathways. Ultimately, evidence from randomised trials is needed to establish whether adding biomarkers to conventional clinical assessment could lead to improved patient outcomes.
Work in the authors’ laboratories is supported by the National Institutes of Health (R01 DK081572, R01 DK108159) and by the Vanderbilt University Medical Center Faculty Research Scholars Program.
Duality of interest
The authors declare that there is no duality of interest associated with this manuscript.
Both authors were responsible for drafting the article and revising it critically for important intellectual content and both approved the final version.
- 6.Berezin AE (2016) Cardiac biomarkers in diabetes mellitus: new dawn for risk stratification? Diabetes Metab Syndr https://doi.org/10.1016/j.dsx.2016.12.032
- 12.Kengne AP, Patel A, Colagiuri S et al (2010) The Framingham and UK prospective diabetes study (UKPDS) risk equations do not reliably estimate the probability of cardiovascular events in a large ethnically diverse sample of patients with diabetes: the action in diabetes and vascular disease: Preterax and Diamicron-MR controlled evaluation (ADVANCE) study. Diabetologia 53:821–831CrossRefPubMedGoogle Scholar
- 15.Cook NR (2007) Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation 115:928–935Google Scholar
- 22.Bruno G, Landi A, Barutta F et al (2013) N-terminal probrain natriuretic peptide is a stronger predictor of cardiovascular mortality than C-reactive protein and albumin excretion rate in elderly patients with type 2 diabetes: the Casale Monferrato population-based study. Diabetes Care 36:2677–2682CrossRefPubMedCentralPubMedGoogle Scholar
- 23.van der Leeuw J, Beulens JW, van Dieren S et al (2016) Novel biomarkers to improve the prediction of cardiovascular event risk in type 2 diabetes mellitus. J Am Heart Assoc 5 https://doi.org/10.1161/JAHA.115.003048
- 30.Moons KG, Altman DG, Vergouwe Y, Royston P (2009) Prognosis and prognostic research: application and impact of prognostic models in clinical practice. BMJ 338:b606Google Scholar
- 31.Reilly BM, Evans AT (2006) Translating clinical research into clinical practice: impact of using prediction rules to make decisions. Ann Intern Med 144:201–209Google Scholar
- 32.Authors/Task Force Members, Ryden L, Grant PJ et al (2013) ESC guidelines on diabetes, pre-diabetes, and cardiovascular diseases developed in collaboration with the EASD: the task force on diabetes, pre-diabetes, and cardiovascular diseases of the European Society of Cardiology (ESC) and developed in collaboration with the European Association for the Study of Diabetes (EASD). Eur Heart J 34:3035–3087CrossRefGoogle Scholar
- 33.Stone NJ, Robinson JG, Lichtenstein AH et al (2014) 2013 ACC/AHA guideline on the treatment of blood cholesterol to reduce atherosclerotic cardiovascular risk in adults: a report of the American College of Cardiology/American Heart Association task force on practice guidelines. Circulation 129:S1–S45CrossRefPubMedGoogle Scholar
- 44.Simmons RK, Sharp S, Boekholdt SM et al (2008) Evaluation of the Framingham risk score in the European prospective investigation of Cancer-Norfolk cohort: does adding glycated hemoglobin improve the prediction of coronary heart disease events? Arch Intern Med 168:1209–1216CrossRefPubMedGoogle Scholar
- 46.Kengne AP, Batty GD, Hamer M, Stamatakis E, Czernichow S (2012) Association of C-reactive protein with cardiovascular disease mortality according to diabetes status: pooled analyses of 25,979 participants from four U.K. prospective cohort studies. Diabetes Care 35:396–403CrossRefPubMedCentralPubMedGoogle Scholar
- 47.Kengne AP, Czernichow S, Stamatakis E, Hamer M, Batty GD (2013) Fibrinogen and future cardiovascular disease in people with diabetes: aetiological associations and risk prediction using individual participant data from nine community-based prospective cohort studies. Diab Vasc Dis Res 10:143–151CrossRefPubMedGoogle Scholar