Deep learning approach for cardiovascular disease risk stratification and survival analysis on a Canadian cohort

The quantification of carotid plaque has been routinely used to predict cardiovascular risk in cardiovascular disease (CVD) and coronary artery disease (CAD). To determine how well carotid plaque features predict the likelihood of CAD and cardiovascular (CV) events using deep learning (DL) and compare against the machine learning (ML) paradigm. The participants in this study consisted of 459 individuals who had undergone coronary angiography, contrast-enhanced ultrasonography, and focused carotid B-mode ultrasound. Each patient was tracked for thirty days. The measurements on these patients consisted of maximum plaque height (MPH), total plaque area (TPA), carotid intima-media thickness (cIMT), and intraplaque neovascularization (IPN). CAD risk and CV event stratification were performed by applying eight types of DL-based models. Univariate and multivariate analysis was also conducted to predict the most significant risk predictors. The DL's model effectiveness was evaluated by the area-under-the-curve measurement while the CV event prediction was evaluated using the Cox proportional hazard model (CPHM) and compared against the DL-based concordance index (c-index). IPN showed a substantial ability to predict CV events (p < 0.0001). The best DL system improved by 21% (0.929 vs. 0.762) over the best ML system. DL-based CV event prediction showed a ~ 17% increase in DL-based c-index compared to the CPHM (0.86 vs. 0.73). CAD and CV incidents were linked to IPN and carotid imaging characteristics. For survival analysis and CAD prediction, the DL-based system performs superior to ML-based models.

Integrating Phenotypic Information of Obstructive Sleep Apnea and Deep Representation of Sleep-Event Sequences for Cardiovascular Risk Prediction

Background

Advances in mobile, wearable and machine learning (ML) technologies for gathering and analyzing long-term health data have opened up new possibilities for predicting and preventing cardiovascular diseases (CVDs). Meanwhile, the association between obstructive sleep apnea (OSA) and CV risk has been well-recognized. This study seeks to explore effective strategies of incorporating OSA phenotypic information and overnight physiological information for precise CV risk prediction in the general population.

Methods

1,874 participants without a history of CVDs from the MESA dataset were included for the 5-year CV risk prediction. Four OSA phenotypes were first identified by the K-mean clustering based on static polysomnographic (PSG) features. Then several phenotype-agnostic and phenotype-specific ML models, along with deep learning (DL) models that integrate deep representations of overnight sleep-event feature sequences, were built for CV risk prediction. Finally, feature importance analysis was conducted by calculating SHapley Additive exPlanations (SHAP) values for all features across the four phenotypes to provide model interpretability.

Results

All ML models showed improved performance after incorporating the OSA phenotypic information. The DL model trained with the proposed phenotype-contrastive training strategy performed the best, achieving an area under the Receiver Operating Characteristic (ROC) curve of 0.877. Moreover, PSG and FOOD FREQUENCY features were recognized as significant CV risk factors across all phenotypes, with each phenotype emphasizing unique features.

Conclusion

Models that are aware of OSA phenotypes are preferred, and lifestyle factors should be a greater focus for precise CV prevention and risk management in the general population.

Ceramides improve cardiovascular risk prediction beyond low-density lipoprotein cholesterol

Aims

Low-density lipoprotein cholesterol (LDL-C) is the best documented cardiovascular risk predictor and at the same time serves as a target for lipid-lowering therapy. However, the power of LDL-C to predict risk is biased by advanced age, comorbidities, and medical treatment, all known to impact cholesterol levels. Consequently, such biased patient cohorts often feature a U-shaped or inverse association between LDL-C and cardiovascular or overall mortality. It is not clear whether these constraints for risk prediction may likewise apply to other lipid risk markers in particular to ceramides and phosphatidylcholines.

Methods and results

In this observational cohort study, we recorded cardiovascular mortality in 1195 patients over a period of up to 16 years, comprising a total of 12 262 patient-years. The median age of patients at baseline was 67 years. All participants were either consecutively referred to elective coronary angiography or diagnosed with peripheral artery disease, indicating a high cardiovascular risk. At baseline, 51% of the patients were under statin therapy. We found a U-shaped association between LDL-C and cardiovascular mortality with a trough level of around 150 mg/dL of LDL-C. Cox regression analyses revealed that LDL-C and other cholesterol species failed to predict cardiovascular risk. In contrast, no U-shaped but linear association was found for ceramide- and phosphatidylcholine-containing markers and these markers were able to significantly predict the cardiovascular risk even after multivariate adjustment.

Conclusion

We thus suggest that ceramides- and phosphatidylcholine-based predictors rather than LDL-C may be used for a more accurate cardiovascular risk prediction in high-risk patients.

Individual lifetime benefit from low-dose colchicine in patients with chronic coronary artery disease

AIMS

Low-dose colchicine reduces cardiovascular risk in patients with coronary artery disease (CAD), but absolute benefits may vary between individuals. This study aimed to assess the range of individual absolute benefits from low-dose colchicine according to patient risk profile.

METHODS AND RESULTS

The European Society of Cardiology (ESC) guideline-recommended SMART-REACH model was combined with the relative treatment effect of low-dose colchicine and applied to patients with CAD from the Low-Dose Colchicine 2 (LoDoCo2) trial and the Utrecht Cardiovascular Cohort-Second Manifestations of ARTerial disease (UCC-SMART) study (n = 10 830). Individual treatment benefits were expressed as 10-year absolute risk reductions (ARRs) for myocardial infarction, stroke, or cardiovascular death (MACE), and MACE-free life-years gained. Predictions were also performed for MACE plus coronary revascularization (MACE+), using a new lifetime model derived in the REduction of Atherothrombosis for Continued Health (REACH) registry. Colchicine was compared with other ESC guideline-recommended intensified (Step 2) prevention strategies, i.e. LDL cholesterol (LDL-c) reduction to 1.4 mmol/L and systolic blood pressure (SBP) reduction to 130 mmHg. The generalizability to other populations was assessed in patients with CAD from REACH North America and Western Europe (n = 25 812). The median 10-year ARR from low-dose colchicine was 4.6% [interquartile range (IQR) 3.6-6.0%] for MACE and 8.6% (IQR 7.6-9.8%) for MACE+. Lifetime benefit was 2.0 (IQR 1.6-2.5) MACE-free years, and 3.4 (IQR 2.6-4.2) MACE+-free life-years gained. For LDL-c and SBP reduction, respectively, the median 10-year ARR for MACE was 3.0% (IQR 1.5-5.1%) and 1.7% (IQR 0.0-5.7%), and the lifetime benefit was 1.2 (IQR 0.6-2.1) and 0.7 (IQR 0.0-2.3) MACE-free life-years gained. Similar results were obtained for MACE+ and in American and European patients from REACH.

CONCLUSION

The absolute benefits of low-dose colchicine vary between individual patients with chronic CAD. They may be expected to be of at least similar magnitude to those of intensified LDL-c and SBP reduction in a majority of patients already on conventional lipid-lowering and blood pressure-lowering therapy.

A new score for improving cardiovascular risk prediction and prevention

BACKGROUND AND AIMS

The ultrasonographic detection of subclinical atherosclerosis (scATS) at carotid and femoral vascular sites using the atherosclerosis burden score (ABS) improves the risk stratification for atherosclerotic cardiovascular disease beyond traditional cardiovascular (CV) risk factors. However, its predictive value should be further enhanced. We hypothesize that combining the ABS and the Framingham risk score (FHRS) to create a new score called the FHRABS will improve CV risk prediction and prevention. We aim to investigate if incorporating the ABS into the FHRS improved CV risk prediction in a primary prevention setting.

METHODS AND RESULTS

1024 patients were included in this prospective observational cohort study. Carotid and femoral plaques were ultra-sonographic detected. Major incident cardiovascular events (MACEs) were collected. The receiver operating characteristic curve (ROC-AUC) and Youden's index (Ysi) were used to compare the incremental contributions of each marker to predict MACEs. After a median follow-up of 6.0 ± 3.3 years, 60 primary MACEs (5.8%) occurred. The ROC-AUC for MACEs prediction was significantly higher for the FHRABS (0.74, p < 0.024) and for the ABS (0.71, p < 0.013) compared to the FHRS alone (0.71, p < 0.46). Ysi or the FHRABS (42%, p < 0.001) and ABS (37%, p < 0.001) than for the FHRS (31%). Cox proportional-hazard models showed that the CV predictive performance of FHRS was significantly enhanced by the ABS (10.8 vs. 5.5, p < 0.001) and FHRABS (HR 23.30 vs. 5.50, p < 0.001).

CONCLUSIONS

FHRABS is a useful score for improving CV risk stratification and detecting patients at high risk of future MACEs. FHRABS offers a simple-to-use, and radiation-free score with which to detect scATS in order to promote personalized CV prevention.

Cardiovascular Risk Prediction Using Machine Learning in a Large Japanese Cohort

Background: Cardiovascular disease (CVD) screening entails precise event prediction to orient risk stratification, resource allocation, and insurance policy. We used random survival forests (RSF) to identify markers of incident CVD among Japanese adults enrolled in an employer-mandated screening program. Methods and Results: We examined biomarker, health history, medication use, and lifestyle data from 155,108 adults aged ≥40 years. The occurrence of coronary artery disease (CAD) or atherosclerotic CVD (ASCVD) events was examined over 6 years of follow-up. The analysis used RSF to identify predictors, then investigated simplified RSF models with fewer predictors for individual-level risk prediction. Data were split into training (70%) and test (30%) datasets. At baseline, the median patient age was 47 years (interquartile range 41-56 years), with 65% males. In all, 1,642 CAD and 2,164 ASCVD events were observed. RSF identified history of heart disease, age, self-reported blood pressure medication, HbA1c, fasting blood sugar, and high-density lipoprotein as important markers of both endpoints. RSF analyses with only the top 20 predictors demonstrated good performance, with areas under the curve of >84% for CAD and >82% for ASCVD in test data across 6 years. Conclusions: We present a machine learning technique for accurate assessment of cardiovascular risk using employer-mandated annual health checkup information. The algorithm produces individual-level risk curves over time, empowering clinicians to efficiently implement prevention strategies in a low-risk population.

Changes in the prognostic values of modern cardiovascular biomarkers in relation to duration of diabetes mellitus

BACKGROUND

Based on the complex pathophysiology of type 2 diabetes and atherosclerosis we hypothesized a dynamic change in prognostic value of cardiovascular biomarkers over time.

METHODS

In this prospective study 746 patients with type 2 diabetes mellitus, being followed up for 60 months were analysed. The primary endpoint was defined as unplanned hospitalization for cardiovascular disease or death. Beside others, especially the prognostic performance of the biomarkers of interest (GDF-15, NT-proBNP, hs-TnT) was evaluated in relation to quartiles of diabetes duration.

RESULTS

In patients having a diabetes duration below 7 years lnGDF-15 (HR 2.84; p < 0.01) and lnhs-TnT (HR 2.96; p < 0.01) were significant predictors of the primary endpoint. LnAge (HR 40.01; p < 0.01) and lnNT-proBNP (HR 1.56; p = 0.03) were significant predictors in patients with a diabetes duration between 7 and 12 years. In the third quartile (diabetes duration 12-22 years) lnurinary albumin to creatinine ratio (HR 1.25; p = 0.005) and lnNT-proBNP (HR 2.13, p < 0.001) predicted the endpoint. In patients with a diabetes duration above 22 years, lnAge (HR 75.35; p = 0.001) and lnNT-proBNP (HR 2.0; p < 0.01) were the only significant predictors of the endpoint.

CONCLUSION

Prognostic power of cardiovascular biomarkers changes dynamically in relation to duration of type 2 diabetes mellitus. In patients with shorter duration of the disease markers of subclinical cardiovascular dysfunction and inflammation perform better than markers of systemic advanced organ dysfunction and cardiovascular disease.

Comparison of recent ceramide-based coronary risk prediction scores in cardiovascular disease patients

AIM

Cholesterol-based risk prediction is often insufficient in cardiovascular disease (CVD) patients. Ceramides are a new kind of biomarkers for CVD. The Coronary Event Risk Test (CERT) is a validated cardiovascular risk predictor that uses only circulating ceramide levels, determined by coupled liquid chromatography-mass spectrometry, to allocate patients into one of four risk categories. This test has recently been modified (CERT2) by additionally including phosphatidylcholine levels.

METHODS AND RESULTS

In this observational cohort study, we have recruited 999 Austrian patients with CVD and followed them for up to 13 years. We found that CERT and CERT2 both predicted cardiovascular events, cardiovascular mortality, and overall mortality. CERT2 had the higher performance compared to CERT and also to the recent cardiovascular risk score of the ESC/EAS guidelines (Systematic COronary Risk Evaluation (SCORE)) for low-risk European countries. Combining CERT2 with the ESC/EAS-SCORE, predictive capacity was further increased leading to a hazard ratio of 3.58 (2.02-6.36; P < 0.001) for cardiovascular events, 11.60 (2.72-49.56; P = 0.001) for cardiovascular mortality, and 9.86 (4.23-22.99; P < 0.001) for overall mortality when patients with very high risk (category 4) were compared to those with low risk (category 1). The use of the combined score instead of the ESC/EAS-SCORE significantly improved the predictive power according to the integrated discrimination improvement index (P = 0.004).

CONCLUSION

We conclude that CERT and CERT2 are powerful predictors of cardiovascular events, cardiovascular mortality, and overall mortality in CVD patients. Including phosphatidylcholine to a ceramide-based score increases the predictive performance and is best in combination with classical risk factors as used in the ESC/EAS-SCORE.

Coronary artery disease and the risk-associated LPA variants, rs3798220 and rs10455872, in patients with suspected familial hypercholesterolaemia

BACKGROUND

The rs3798220 and rs10455872 single nucleotide polymorphisms (SNPs) in LPA are associated with increased plasma concentrations of lipoprotein(a) [Lp(a)] and coronary artery disease (CAD).

METHODS

We investigated the association between rs3798220 and rs10455872 and prevalent CAD in 763 patients with suspected familial hypercholesterolaemia (FH). The rs3798220 and rs10455872 SNPs in LPA were detected using a SEQUENOM platform.

RESULTS

Both LPA SNPs were significantly associated with CAD, but only rs3798220 after adjustment for other risk factors (odds ratio [OR] 2.05; 95% confidence interval [CI] 1.02-4.12; p = 0.045), and neither after adjustment for Lp(a) concentrations. Both SNPs were positively and independently associated with increased Lp(a) (rs3798220: OR 1.27; 95% CI 0.96-1.57; p < 0.001. rs10455872: OR 1.41; 95% CI 1.24-1.58; p < 0.001). Plasma concentrations of Lp(a) were independently associated with prevalent CAD (OR 1.28; 95% CI 1.08-1.52, p = 0.005) after adjustment for LPA SNPs and other cardiovascular risk factors. While both the rs3798220 and rs10455872 SNPs were associated with Lp(a) concentrations and prevalent CAD in patients with suspected FH, this was not independent of Lp(a) concentration.

CONCLUSIONS

Quantification of Lp(a) is more likely to be useful than assessment of these Lp(a)-associated SNPs to augment CAD risk prediction.

Cardiovascular risk prediction in the elderly

Heart failure (HF) in the elderly, besides being a leading cause of mortality and morbidity, is rapidly increasing in prevalence with patients aged 65 and older accounting for more than 75% of heart failure hospitalizations. Elderly patients have historically been unrepresented in clinical HF trials and often present with multiple comorbidities, including frailty, depression, nutritional, functional and cognitive impairments. Additionally, pharmacologic challenges such as adherence to therapy, polypharmacy, altered drug pharmacokinetics and/or renal derangements make them less likely to receive guideline-directed medical therapies for HF. Recognition of these various interrelated domains is key and should prompt a multidisciplinary, holistic management approach so as to optimize prognosis in this vulnerable subset of the population.

Segment-specific association of carotid-intima-media thickness with cardiovascular risk factors - findings from the STAAB cohort study

Background

The guideline recommendation to not measure carotid intima-media thickness (CIMT) for cardiovascular risk prediction is based on the assessment of just one single carotid segment. We evaluated whether there is a segment-specific association between different measurement locations of CIMT and cardiovascular risk factors.

Methods

Subjects from the population-based STAAB cohort study comprising subjects aged 30 to 79 years of the general population from Würzburg, Germany, were investigated. CIMT was measured on the far wall of both sides in three different predefined locations: common carotid artery (CCA), bulb, and internal carotid artery (ICA). Diabetes, dyslipidemia, hypertension, smoking, and obesity were considered as risk factors. In multivariable logistic regression analysis, odds ratios of risk factors per location were estimated for the endpoint of individual age- and sex-adjusted 75th percentile of CIMT.

Results

2492 subjects were included in the analysis. Segment-specific CIMT was highest in the bulb, followed by CCA, and lowest in the ICA. Dyslipidemia, hypertension, and smoking were associated with CIMT, but not diabetes and obesity. We observed no relevant segment-specific association between the three different locations and risk factors, except for a possible interaction between smoking and ICA.

Conclusions

As no segment-specific association between cardiovascular risk factors and CIMT became evident, one simple measurement of one location may suffice to assess the cardiovascular risk of an individual.

Cost-effectiveness of the non-laboratory based Framingham algorithm in primary prevention of cardiovascular disease: A simulated analysis of a cohort of African American adults

The non-lab Framingham algorithm, which substitute body mass index for lipids in the laboratory based (lab-based) Framingham algorithm, has been validated among African Americans (AAs). However, its cost-effectiveness and economic tradeoffs have not been evaluated. This study examines the incremental cost-effectiveness ratio (ICER) of two cardiovascular disease (CVD) prevention programs guided by the non-lab versus lab-based Framingham algorithm. We simulated the World Health Organization CVD prevention guidelines on a cohort of 2690 AA participants in the Atherosclerosis Risk in Communities (ARIC) cohort. Costs were estimated using Medicare fee schedules (diagnostic tests, drugs & visits), Bureau of Labor Statistics (RN wages), and estimates for managing incident CVD events. Outcomes were assumed to be true positive cases detected at a data driven treatment threshold. Both algorithms had the best balance of sensitivity/specificity at the moderate risk threshold (>10% risk). Over 12years, 82% and 77% of 401 incident CVD events were accurately predicted via the non-lab and lab-based Framingham algorithms, respectively. There were 20 fewer false negative cases in the non-lab approach translating into over $900,000 in savings over 12years. The ICER was -$57,153 for every extra CVD event prevented when using the non-lab algorithm. The approach guided by the non-lab Framingham strategy dominated the lab-based approach with respect to both costs and predictive ability. Consequently, the non-lab Framingham algorithm could potentially provide a highly effective screening tool at lower cost to address the high burden of CVD especially among AA and in resource-constrained settings where lab tests are unavailable.

Association between traditional cardiovascular risk factors and mortality in the oldest old: untangling the role of frailty

Background

To date, there is no consensus regarding cardiovascular risk management in the very old. Studies have shown that the relationship between traditional cardiovascular risk factors and mortality is null or even inverted within this age group. This relationship could be modified by the presence of frailty. This study was performed to examine the effect of frailty on the association between cardiovascular risk factors and mortality in the oldest old.

Methods

The BELFRAIL study is a prospective, observational, population-based cohort study of 567 subjects aged 80 years and older. Data on cardiovascular risk factors were recorded. Frailty was assessed using three different models: the Groningen Frailty Indicator, Fried and Puts models. Participants were considered robust if they were ‘not frail’ according to all three models, and frail if they met the frailty criteria for one of the three models. The follow-up data on mortality and cause of death were registered.

Results

No cardiovascular risk factor was associated with mortality in subjects with and without cardiovascular disease. The presence of frailty was a strong risk factor for mortality [HR: 2.5, 95%CI: (1.9–3.2) for all-cause mortality; HR: 2.2, 95%CI: (1.4–3.4) for cardiovascular mortality]. In robust patients, a history of cardiovascular disease increased the risk for mortality [HR: 1.7, 95%CI: (1.1–2.5) for all-cause mortality; HR: 2.2, 95%CI: (1.2–3.9) for cardiovascular mortality]. In frail patients, there was no association between any of the traditional risk factors and mortality.

Conclusions

Traditional cardiovascular risk factors were not associated with mortality in very old subjects. Frailty was shown to be a strong risk factor for mortality in this age group. However, frailty could not be used to identify additional subjects who might benefit more from cardiovascular risk management.

Electronic supplementary material

The online version of this article (10.1186/s12877-017-0626-x) contains supplementary material, which is available to authorized users.

Risk score overestimation: the impact of individual cardiovascular risk factors and preventive therapies on the performance of the American Heart Association-American College of Cardiology-Atherosclerotic Cardiovascular Disease risk score in a modern multi-ethnic cohort

AIMS

To evaluate the 2013 American Heart Association (AHA)-American College of Cardiology (ACC)-Atherosclerotic Cardiovascular Disease (ASCVD) risk score among four different race/ethnic groups and to ascertain which factors are most associated with risk overestimation by the AHA-ACC-ASCVD score.

METHODS AND RESULTS

The Multi-Ethnic Study of Atherosclerosis (MESA), a prospective community-based cohort, was used to examine calibration and discrimination of the AHA-ACC-ASCVD risk score in 6441 White, Black, Chinese, and Hispanic Americans (aged 45-79 years and free of known ASCVD at baseline). Using univariable and multivariable absolute risk regression, we modelled the impact of individual risk factors on the discordance between observed and predicted 10-year ASCVD risk. Overestimation was observed in all race/ethnic groups in MESA and was highest among Chinese (252% for women and 314% for men) and lowest in White women (72%) and Hispanic men (67%). Higher age, Chinese race/ethnicity (when compared with White), systolic blood pressure (treated and untreated), diabetes, alcohol use, exercise, lipid-lowering medication, and aspirin use were all associated with more risk overestimation, whereas family history was associated with less risk overestimation in a multivariable model (all P < 0.05).

CONCLUSION

The AHA-ACC-ASCVD risk score overestimates ASCVD risk among men, women, and all four race/ethnic groups evaluated in a modern American primary prevention cohort. Clinicians treating patients similar to those in MESA, particularly older individuals and those with factors associated with more risk overestimation, may consider interpreting absolute ASCVD risk estimates with caution.

Differing predictive relationships between baseline LDL-C, systolic blood pressure, and cardiovascular outcomes

BACKGROUND

Traditional cardiovascular risk factors, such as hypertension and dyslipidemia, predispose individuals to cardiovascular disease, particularly patients with diabetes. We investigated the predictive value of baseline systolic blood pressure (SBP) and low-density lipoprotein cholesterol (LDL-C) on the risk of vascular outcomes in a large population of patients at high risk of future cardiovascular events.

METHODS

Data were pooled from the TNT (Treating to New Targets), CARDS (Collaborative Atorvastatin Diabetes Study), and IDEAL (Incremental Decrease in End-Points Through Aggressive Lipid Lowering) trials and included a total of 21,727 patients (TNT: 10,001; CARDS: 2838; IDEAL: 8888). The effect of baseline SBP and LDL-C on cardiovascular events, coronary events, and stroke was evaluated using a multivariate Cox proportional-hazards model.

RESULTS

Overall, risk of cardiovascular events was significantly higher for patients with higher baseline SBP or LDL-C. Higher baseline SBP was significantly predictive of stroke but not coronary events. Conversely, higher baseline LDL-C was significantly predictive of coronary events but not stroke. Results from the subgroup with diabetes (5408 patients; TNT: 1501; CARDS: 2838; IDEAL: 1069) were broadly consistent with those of the total cohort: baseline SBP and LDL-C were significantly predictive of cardiovascular events overall, with the association to LDL-C predominantly related to an effect on coronary events. However, baseline SBP was not predictive of either coronary or stroke events in the pooled diabetic population.

CONCLUSIONS

In this cohort of high-risk patients, baseline SBP and LDL-C were significantly predictive of cardiovascular outcomes, but this effect may differ between the cerebrovascular and coronary systems.

TRIAL REGISTRATION NUMBER

NCT00327691 (TNT); NCT00327418 (CARDS); NCT00159835 (IDEAL).

Vascular risk assessment in older adults without a history of cardiovascular disease

Modern cardiovascular risk prediction tools, which have their genesis in the Framingham Heart Study, have allowed more accurate risk stratification and targeting of treatments worldwide over the last seven decades. Better cardiovascular risk factor control during this time has led to a reduction in cardiovascular mortality and, at least in part, to improved life expectancy. As a result, western societies as a whole have seen a steady increase in the proportion of older persons in their populations. Unfortunately, several studies have shown that the same tools which have contributed to this increase cannot be reliably extrapolated for use in older generations. Recent work has allowed recalibration of existing models for use in older populations but these modified tools still require external validation before they can be confidently applied in clinical practice. Another complication is emerging evidence that aggressive risk factor modification in older adults, particularly more frail individuals, may actually be harmful. This review looks at currently available cardiovascular risk prediction models and the specific challenges faced with their use in older adults, followed by analysis of recent attempts at recalibration for this cohort. We discuss the issue of frailty, looking at our evolving understanding of its constituent features and various tools for its assessment. We also review work to date on the impact of frailty on cardiovascular risk modification and outline its potentially central role in determining the most sensible approach in older patients. We summarise the most promising novel markers of cardiovascular risk which may be of use in improving risk prediction in older adults in the future. These include markers of vascular compliance (such as aortic pulse wave velocity and pulse wave analysis), of endothelial function (such as flow mediated dilation, carotid intima-media thickness and coronary artery calcium scores), and also biochemical and circulating cellular markers.