Final Report Summary - EPIC-CVD (EPIC-CVD: Individualised CVD risk assessment: tailoring targeted and cost-effective approaches to Europe's diverse populations)
Executive Summary:
1. Executive summary
EPIC-CVD was an ambitious project involving 28 partner organisations across 11 European countries with the overall aim to provide clinicians and policy-makers across Europe with an evidence-based menu of options for targeted and cost-effective CVD risk approaches tailored to the needs of Europe's diverse populations. To achieve this broad objective, EPIC-CVD intended to engage and cross-link 4 broad areas of scientific endeavour:
(1) population health science (epidemiology, public health, biostatistics)
(2) laboratory science (biochemistry, molecular genetics, nutrition)
(3) translational science (medical anthropology, primary care and cardiovascular medicine, health economics)
(4) implementation science (risk communication tools, comparative analysis, policy development).
Our research was divided into 4 main Research Lines of work, namely:
RL1 - Creation of a framework to develop new risk assessment approaches
RL2 - Analysis of factors related to nutrition in a large prospective study
RL3 - Analysis of intermediate biomarkers and genetic make-up in incident CVD
RL4 - Testing of clinical and public health utility
Major successes of the project to date include:
- creation of the world’s largest harmonised case-cohort study of incident coronary heart disease and stroke with >25,000 incident outcomes;
- detailed phenotyping of participants including >1million genetic variants, 300 circulating biomarkers (including metabolomics, nutritional biomarkers and clinical biomarkers), extensive dietary and lifestyle information, permitting aetiological and methodological analyses in addition to studies of risk prediction;
- development of novel risk scores that replace invasive measures with non-invasive modifiable lifestyle factors without greatly sacrificing predictive accuracy;
- development of novel risk scores that increase predictive accuracy through inclusion of non-traditional biomarkers/genetic variants;
- conducting the largest trial of behaviour change in relation to provision of CVD risk estimates (phenotypic and genotypic) with objective outcome measures (eg, physical activity using accelerometry);
- comparison of the regulatory and societal issues associated with CVD risk screening across European countries, particularly with respect to inclusion of biomarkers and genetic information;
- establishment of a framework for capturing information on the costs and benefits of CVD interventions (eg, statins, lifestyle modifications) and utilising this information to assess cost-effectiveness of novel risk scores;
- creation of novel methods to analyse multi-centre case-cohort studies, particularly with respect to risk prediction.
By March 2016, work in EPIC-CVD has led to 15 peer-reviewed scientific publications, with a further 6 already submitted, and another ~40 to be submitted in 2016/2017.
Project Context and Objectives:
2. Summary of project context and main objectives
Coronary heart disease (CHD) and stroke are Europe's first and second leading causes of death, respectively, as well as major causes of disability. Cardiovascular disease (CVD) is currently responsible for more than €170 billion annually in economic costs in the increasingly resource-constrained countries of the EU. Despite reductions in age-specific CVD death rates in many European countries in recent decades, the overall future CVD burden in the EU is predicted to increase as a result of the ageing population and increasing prevalence of obesity and diabetes. It has been estimated that tens of thousands of additional CVD outcomes could be prevented in the EU per year by better screening and management of risk. These potential gains are more likely to be realised if more compelling evidence and policies emerge to support future approaches to primary prevention of CVD. Furthermore, we believe that it is unlikely that a single approach will be optimum or acceptable across the diverse healthcare systems, budgets, cultures, and lifestyles of Europe.
The overall aim of the EPIC-CVD project was therefore to provide clinicians and policy-makers across Europe with an evidence-based menu of options for targeted and cost-effective CVD risk approaches tailored to the needs of Europe's diverse populations.
To achieve this overall aim we targeted 5 key areas:
1) Creating synergy between population-wide and personalised prevention
• EPIC-CVD aimed to create Europe’s first risk assessment approaches focusing on “modifiable” risk scores with behavioural risk factors and objective lifestyle biomarkers
Our hypothesis was that approaches based on modifiable scores should cost-effectively:
(1) enhance accuracy of risk prediction, resulting in improved identification and risk assessment
(2) maximise potential for behaviour change in patients and physicians through personalised lifestyle feedback, leading to better clinical decision making and clinical outcomes
(3) contribute to the prevention of other conditions influenced by lifestyle choices and of major public health importance in Europe, such as obesity, type 2 diabetes and certain cancers, again leading to better clinical outcomes
(4) achieve synergy with population-wide approaches.
Modifiable risk scores should genuinely be complementary to public health activities that aim to reduce the overall population risk of cardiovascular disease by promoting a healthy lifestyle (diet, exercise, avoidance of smoking).
2) Making existing CVD risk tools more relevant to European populations
• EPIC-CVD aimed to provide the first systematic comparison of existing major risk scores in a common set of participants across European populations
By exploiting data and samples in the prospective 10-country EPIC study of 520,000 people in whom extensive information on lifestyle, socioeconomic factors, blood samples, and other characteristics had already been collected in a standardised manner, we aimed to directly address the original EC call’s mandate: “The impact of biomarkers on cardiovascular disease risk prediction will need to be assessed across different European populations as they have different lifestyles (eg, dietary patterns) and varying biomarker levels”.
By achieving this aim, we hoped to:
(1) improve the performance of existing risk scores in Europe
(2) lead to stratified risk scores for subpopulations that “contribute to the development of personalised predictive medicine”
(3) enhance the adoption of these methods by patients and healthcare professionals across Europe because of their greater actual (and perceived) relevance to local circumstances.
Thus, EPIC-CVD aimed to enhance our ability to identify people at high CVD risk and manage them appropriately, leading to improvements in identification, risk assessment, clinical decision making and clinical outcomes.
3) Developing clinically useful new CVD risk scores and tools
• EPIC-CVD aimed to create the first risk score that considered biomarkers of 'nature' (eg, genetic variation), 'nurture' (ie, lifestyle biomarkers), and intermediate biological pathways (eg, lipids, inflammation etc), and to be the first study capable of incorporating the interplay of nature and nurture for CVD risk prediction across Europe.
We hypothesised that detailed risk scores based on parsimonious sets of these markers would be likely to confer a much higher degree of predictive accuracy than existing scores and be especially amenable to tailoring to subpopulations, such as younger people who may be contemplating lifelong treatment, resulting in improvements in identification, risk assessment, clinical decision making and clinical outcomes. In partnership with research intensive SMEs (eg, VITAS in Oslo, an international leader in biomarker assays), we aimed to develop innovative methods for high-throughput blood sample collection and biomarker assays, contributing to research and innovation and yielding approaches and potential products of interest and potential benefit to SME(s).
4) Designing acceptable and cost-effective approaches for a diverse Europe
• EPIC-CVD aimed to provide the first systematic evaluation of clinical and public health utility of modifiable and detailed risk scores across Europe
We aimed to exploit our expertise in devising guidelines for the evaluation of risk scores to ensure that our development and validation of new risk scores would be robust. By capitalising on our existing multidisciplinary framework, we aimed to conduct qualitative fieldwork and a randomised trial of the impact of different risk scores on "patient-centred" outcomes in 1000 people across Europe. We will build on our previous experience with sequential or "stepwise" screening strategies in which periodic invitations are made to individuals for a preventive visit using already recorded determinants (“pre-stratification”). These approaches should enhance cost effectiveness and organisational efficiency of targeting of preventive resources and reduce anxiety in the population by focusing efforts on subsets of the population most likely to benefit from detailed screening. We intended to tailor these approaches to the diverse parts of Europe. We aimed to capitalise on health economics expertise we previously used to shape national screening policies for vascular conditions in the UK and to conduct detailed health economic modelling that considered alternative approaches taking into account long-term or even lifetime risk to determine the overall costs and benefits of using various different risk markers, risk scores, and screening strategies for CVD, leading to the emergence of operational approaches that show cost effectiveness, safety, validity and incremental benefit over existing prediction methods. EPIC-CVD hoped to benefit from its partnership with RAND-Europe, an independent not-for-profit policy research organisation with specific expertise in comparisons of European health systems, to develop state-of-the-art policy for CVD risk assessment tailored to Europe’s diverse needs.
5) Addressing broader societal concerns
• EPIC-CVD will provide consideration of the broader societal implications of modifiable and detailed CVD risk scores across Europe
In partnership with the PHG Foundation, a genetics policy think-tank, EPIC-CVD aimed to explore potential ethical, legal, and social issues related to targeted CVD prevention testing methods. This work was building on a major work-package of research the PHG Foundation previously carried out as part of the FP7-funded COGS (Collaborative Oncological Gene-environment Study) project. In EPIC-CVD, we intended to apply similar considerations and to extend research in new directions to address considerations distinctive to CVD screening.
Across these areas, EPIC-CVD listed the following objectives:
● To develop and evaluate modifiable risk scores that promote behaviour change and personalised lifestyle feedback
● To compare and calibrate major existing risk scores across diverse European populations
● To develop and evaluate detailed risk scores that maximise predictive accuracy and target subpopulations
● To develop and evaluate user-friendly clinical tools to support risk assessment
● To develop and evaluate cost-effective strategies for the organisation of screening
● To compare the costs and benefits of different risk assessment and screening approaches
● To develop evidence-based policies tailored to diverse European settings
● To disseminate these tools, strategies, and policies widely to relevant audiences across Europe to maximise clinical and public health impact.
To achieve these objectives, EPIC-CVD intended to engage and cross-link 4 broad areas of scientific endeavour:
(1) population health science (epidemiology, public health, biostatistics)
(2) laboratory science (biochemistry, molecular genetics, nutrition)
(3) translational science (medical anthropology, primary care and cardiovascular medicine, health economics)
(4) implementation science (risk communication tools, comparative analysis, policy development).
The EPIC-CVD consortium consists of 28 partner organisations from 11 European countries, including 2 SMEs, a genetics policy think-tank, and an independent not-for-profit policy research organisation.
Project Results:
3. Main findings
The EPIC-CVD project is divided into 4 main interlinking Research Lines:
RL1 - Creation of a framework to develop new risk assessment approaches
RL2 - Analysis of factors related to nutrition in a large prospective study
RL3 - Analysis of intermediate biomarkers and genetic make-up in incident CVD
RL4 - Testing of clinical and public health utility
The 4 Research Lines were further divided into 16 scientific Work Packages.
a) Results from RL1 - Creation of a framework to develop new risk assessment approaches
Research Line 1 had the following objectives:
• To identify and verify diagnoses of incident stroke within the EPIC cohort
• To retrieve and prepare biological samples from stroke cases for assays
• To assess the impact of 22 routinely measured clinical biomarkers on the risk of incident CVD
• To coordinate the activities for designing the analysis of data collected in the project and to supervise the preparation of all analytical plans.
The EPIC-CVD project is underpinned by the European Prospective Investigation into Cancer and Nutrition (EPIC) study. EPIC involves 366,521 women and 153,457 men, mostly aged 35–70 years, recruited by 23 centres in 10 European countries (Denmark, France, Germany, Greece, Italy, the Netherlands, Norway, Spain, Sweden, and the United Kingdom) between 1991 and 1999. Within this large pan-European prospective cohort, a nested case-cohort study of type 2 diabetes (EPIC-InterAct) was created by identification of >12,000 incident type 2 diabetes cases and selection of a random subcohort of >15,000 referents. EPIC-InterAct was funded by a EC Framework Progamme 6 award. Taking advantage of the use of the case-cohort design and the selection of the random subcohort for comparison with disease cases, more than 15,000 incident coronary heart disease cases were subsequently identified through funding from the British Heart Foundation and the UK Medical Research Council (“EPIC-CVD”). We aimed to extend EPIC-CVD to include incident stroke cases and then to conduct assays in the CHD cases and stroke cases, creating a framework to allow study of cardiovascular disease.
WP1 - Stroke case verification
The overall objective of this Work Package was to identify and verify diagnoses of incident stroke within the EPIC cohort.
A group at UCAM, including stroke physicians and epidemiologists, devised a protocol for the standardised verification of stroke cases. The protocol outlined the methods of stroke case verification from medical record review and was circulated to all EPIC centres. All 23 EPIC administrative centres, which were all partner organisations in EPIC-CVD, were then asked to identify potential stroke cases that had occurred after study baseline, using whichever methods they had available. Stroke cases were identified through a range of means, including morbidity and mortality registers, participant surveys, and active follow-up using medical records. Identification numbers for all >10,000 suspected incident stroke cases were then provided to the EPIC-CVD Coordination Centre, along with additional information including family history of cardiovascular disease and medication usage. Other details of baseline measurements (eg, anthropometry and blood pressure), lifestyle information (eg, smoking behaviours and alcohol consumption), and dietary information were then retrieved for these participants from the central EPIC database at IARC (Lyon, France).
EPIC centres then retrieved relevant information to verify the stroke diagnoses, typically by retrieving further information on the suspected stroke cases or by conducting substudies to verify the validity of the stroke case ascertainment approach. Information was then curated, collated and harmonised across the 23 centres at the EPIC-CVD Coordinating Centre to develop a common framework for definition of stroke cases. After exclusion of prevalent cases (since our study of risk prediction was designed to assess first-ever events rather than recurrent disease), case events that fell outside of our stroke definition (eg, transient ischaemic attack), and events that turned out not to be valid or only based on self-report, we successfully identified and validated more than 8000 incident cerebrovascular events, creating one of the largest studies of incident stroke cases to date. Stroke cases were subclassified, according to ICD codes into ischaemic stroke events, haemorrhagic stroke events, subarachnoid haemorrhage events, other stroke events or unclassified stroke events. We also described the certainty of the validated stroke case events according to the sources of information used and depth of information provided.
WP2 - Retrieval and preparation of samples
The overall objective of this Work Pacakge was to retrieve and prepare biological samples from stroke cases for assays.
Once lists of suspected stroke cases had been identified by EPIC centres (see WP1), a complete list of stroke cases with stored blood samples was sent to IARC and local biorepositories in Scandinavia by the EPIC-CVD Coordinating Centre for sample retrieval. The retrieval process at IARC consisted of retrieving 0.5ml straws of red blood cells, plasma (x2), serum, and buffy coat (x2), before extracting DNA from buffy coat samples using a high-throughput automated DNA extraction procedure. The processes used were the same as those used previously to retrieve samples for incident type 2 diabetes, incident CHD cases and subcohort members. Once samples from all participants had been successfully retrieved and had DNA extracted for them, sample straws and DNA aliquots were delivered to the MRC Epidemiology Unit in Cambridge. The overall yields of DNA were high and >90% of samples had DNA above the concentration required for downstream assays (50ng/ul).
Similar procedures were followed in Denmark, Malmo and Umea, where samples from Scandinavian participants are stored. From these centres, aliquots in tubes/plates were provided rather than straws, and were shipped to Cambridge or directly to the assay laboratory.
Once samples had been received in Cambridge, the laboratory team at the MRC-Epidemiology Unit then sub-aliquoted and arrayed the non-DNA samples for the assays of clinical biomarkers, fatty acids, and vitamins/carotenoids. For the vitamin C assays, this involved addition of a preservative to the sample once it had thawed (see WP5). Regular shipments of samples for biomarker assays were made from the MRC-Epidemiology Unit to the MRC Human Nutrition Research Unit (MRC-HNR) for fatty acids assays, to VITAS for vitamins/carotenoids assays and to Stiching Ingehousz (SHL, Netherlands) for clinical biomarker assays.
For DNA samples received at the MRC Epidemiology Unit in Cambridge, the concentration was initially tested using a Picogreen method. Once quantified, samples were then normalised to a standard concentration of 50ng/ul for assay, and were then stored in a robotic freezer. Any samples received at a concentration below 50ng/ul were dried down and resuspended to the appropriate concentration. Samples were further tested using a Taqman assay of variants on the sex chromosomes to check that they were genetically the same sex as expected according to the EPIC database. In total, >97% of samples passed DNA QC.
DNA samples were shipped to the labs at Cambridge Genomics Services for variant microarray assays and were subaliquoted for the in-house telomere length assay at the MRC Epidemiology Unit.
WP3 - 22 clinical chemistry biomarkers
The overall objective of this Work Package was to assess the impact of 22 routinely measured clinical biomarkers on the risk of incident CVD.
The 22 clinical chemistry biomarkers investigated included lipids (total cholesterol, HDL-cholesterol, triglycerides), lipoproteins (apolipoprotein AI, apolipoprotein B, lipoprotein[a]), glycaemic markers (HbA1c, glucose), inflammatory markers (C-reactive protein, albumin), iron markers (serum iron, ferritin, transferrin), electrolytes (calcium, magnesium), liver enzymes (ALT, AST, ALP, GGT, bilirubin), and renal markers (creatinine, uric acid). Following an options appraisal to assess the optimal laboratory to conduct these assays, it was decided to use SHL, the laboratory in the Netherlands that has previously conducted the same assays on the 43,000 serum samples comprising the CHD case samples, the type 2 diabetes samples and the subcohort samples.
In total, more than 8000 validated incident stroke case samples were measured and passed quality control steps. Results appeared to be consistent over time with no strong evidence of drift. Distributions of biomarkers were consistent with expectations based on previous studies. For example, triglycerides, C-reactive protein and lipoprotein(a) were all heavily skewed, whilst total cholesterol was normally distributed. Correlations between biomarkers were also of the expected magnitudes and directions. For example, HDL-cholesterol and apolipoprotein AI were strongly correlated (r=0.83) ALT and AST were strongly correlated (r=0.71) and glucose and HbA1c were strongly correlated (r=0.70). There were also inverse correlations, such as between HDL-cholesterol and triglyceride levels (r=-0.50).
An analysis plan was developed for the analysis of the data from the assays of 22 clinical biomarkers with incident CVD. This plan covered methods to assess cross-sectional correlates, assess the shape of dose-response relationships between the 22 clinical biomarkers and CVD outcomes, consistently adjust for potential confounding factors, assess the relevance of pre-specified subgroups, and finally to assess the incremental predictive value of the 22 clinical biomarkers beyond conventional CVD risk factors.
As expected from previous literature and large-scale meta-analyses, the strongest positive associations with disease risk were for pro-atherogenic lipids and lipoproteins, with glycaemic markers also showing positive associations. The strongest inverse associations were with HDL-cholesterol and apolipoprotein AI. C-reactive protein was also positively associated, although its ability to add to existing CVD risk scores has already been well documented. Other notable associations were inverse associations with bilirubin, iron and albumin, and positive associations with liver enzymes (GGT, ALT, ALP), ferritin and uric acid. These associations mostly remained significant after adjustment for traditional cardiovascular risk factors.
Although total cholesterol and HDL-cholesterol are already routinely included in many existing CVD risk scores, there is potential that other emerging biomarkers could usefully add to risk prediction. Whilst some, such as C-reactive protein and glycaemic markers, have been well studied in this context, other biomarkers have not. On the basis of the associations seen with cardiovascular events, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that it would be beneficial to investigate the predictive ability of other associated biomarkers (i.e. not those already in risk scores or previously well covered) given that their associations with disease risk appear to be largely independent of conventional risk factors.
WP4 - Analysis management
The overall objective of this Work Package was to coordinate the activities for designing the analysis of data collected in the project and to supervise the preparation of all analytical plans.
Given the number and range of different analyses to be performed as part of the EPIC-CVD project, and the challenges of analysing a multi-centre case-cohort study, we established a Central Analytical Team, led by Professor Simon Thompson, Dr Angela Wood and Dr Michael Sweeting at the University of Cambridge. This team worked closely with epidemiologists and data managers at the EPIC-CVD Coordinating Centre to produce documents describing recommended approaches to understanding and interpreting the EPIC-CVD data, conducting aetiological analyses in the EPIC-CVD dataset, and conducting risk prediction analyses in the EPIC-CVD dataset. Since not all methods were already clearly established, a number of pieces of methodological work were conducted, overseen by Dr Angela Wood. For example, methods to estimate risk prediction metrics in multi-centre case-cohort studies were developed and published (Sanderson et al., BMC Med Res Methods, 2013). Other publications already produced cover methods for stratified case-cohort studies (Jones et al., J Clin Epi, 2015), recommendations for reporting results from case-cohort studies (Sharp et al., PLoS One, 2014) and methods to assess risk prediction models using individual participant data from multiple studies (Pennells et al., Am J Epid, 2014). A number of other methodological papers will emerge in 2016 and 2017, such as a recently submitted paper looking at analysing genetic data in case-cohort studies (Staley et al., submitted).
The Central Analyatical Team was also responsible for ensuring that appropriate statistical analysis plans were produced prior to analysis for all risk prediction analyses, as well as the aetiological and methodological analyses enabled by the creation of the EPIC-CVD dataset. Hence, analytical plans were produced for the analysis of data from the assays of the 22 clinical biomarkers, the fatty acids panel, the vitamins and carotenoids, the customised SNP array and the telomere length.
b)_Results from RL2 - Analysis of factors related to nutrition in a large prospective study
Research Line 2 had the following objectives:
• to further optimise existing high-throughput assays for nutritional biomarkers
• to assess the potential for inclusion of vitamin C and 6 carotenoids in CVD risk scores.
• to assess the potential for inclusion of vitamin D and its epimers in CVD risk scores
• to assess the potential for inclusion of of 37 fatty acids in CVD risk scores.
WP5 - Optimising high-throughput assays
The overall objective of this Work Package was to further optimise existing high-throughput assays for nutritional biomarkers.
Dr Thomas Gundersen, a world authority on nutritional biomarkers, at VITAS conducted several optimisation procedures for assay of vitamin C, vitamin D (including its epimer) and a panel of carotenoids.
a) Carotenoids
The gold standard method for assay of carotenoids is a chromatographic approach known as HPLC-UV. This approach has the ability to separate out the six carotenoids (alpha-carotene, beta-carotene, beta-cryptoxanthin, lutein, lycopene and zeaxanthin). However, doing so required a lengthy run-time (28 minutes at the start of the project) which would not allow the assay of the requisite number of samples in the project timescale. VITAS therefore tested a number of column chemistries and particle technologies to
speed up the cycle time without sacrificing the resolution of the carotenoid isomers. On each column type numerous variations in conditions were investigated. These included mobile phase composition, temperature, flow, column length, particle size, pH and various modifiers. The C30 stationary phase was then looked at again to attempt to optimize condition on this chemistry. By using an YMC C30 column, optimizing the above mentioned parameters VITAS were able to reduce the cycle time of the method from 28 minutes to 12 minutes, a 64% reduction, still keeping the required resolution between the critical pairs of carotenoids. As plasma sample was also limited in the EPIC-CVD study, VITAS also optimized several parameters that might influence the signal strength and the signal to noise ratio. Altogether VITAS achieved a 6.5 time increase in sensitivity allowing analysis from as little as 30µL of sample.
b) Vitamin C
Vitamin C, also known as ascorbic acid, is very unstable in plasma and it is expected that significant degradation would have occurred in the EPIC-CVD samples, which are typically more than 10 years old. However, the EPIC samples have been stored in liquid nitrogen and the stability of vitamin C in plasma stored for this length of time under these conditions had not been previously examined. To investigate if detectable levels of vitamin C remain in the sample and to what degree these levels correlate with levels measured fresh at baseline, a subset of 96 samples from EPIC-Norfolk were analysed. For these samples baseline data for vitamin C were available. A protocol for this pilot investigation was designed and written. An SOP for the transfer of serum/plasma from the EPIC plasma straws to vials and subsequently to microtiter plates, addition of stabilizers and shipment on dry ice from Cambridge to Oslo was also written. The results from the pilot showed a surprisingly strong correlation of r=0.94. This indicates that vitamin C has been very well preserved when stored in liquid nitrogen as both labs providing data for this comparison obtained very similar results for all of these 96 samples.
c) Vitamin D and its epimer
It has previously been shown that 25-OH-Vitamin D3 also exists in its epimerised form in human plasma and serum. Separating isoforms of molecules is a challenging task and often requires prolonged analysis time. The presence of this isoform has only been studied in smaller sample cohorts. To include this isoform in the large EPIC-CVD project required development of a method with the isoforms resolved, but still retaining the high throughput properties of the method due to the large numbers of samples to be assayed. By applying modern HPLC columns with sun 3µm core shell particles and pentafluorophenylpropyl chemistry, VITAS were able to separate these and keep the cycle time of the method within five minutes.
WP6 - Vitamin C and carotenoids
The overall objective was to assess the potential for inclusion of vitamin C and 6 carotenoids in CVD risk scores. To do this required assay of this panel of 7 nutritional biomarkers in the EPIC-CVD dataset, including the incident CHD cases, incident stroke cases and subcohort members.
Regular shipments of samples were made to Dr Gundersen’s laboratory at VITAS for vitamin C assays to be performed. Dr Gundersen’s team regularly returned data, including quality control (QC) data, for the assays as they passed through the pipeline. In total, ~45,000 samples were assayed on the vitamin C and carotenoids platform, including ~12,000 incident CHD cases, ~8,000 incident stroke cases, ~15,000 subcohort members (as well as ~10,000 incident type 2 diabetes cases not included in EPIC-CVD). We additionally completed the measurement of around 2,100 samples from the EPIC-Norfolk study for a repeat measurements validation study. This included samples from ~700 participants with measurements available at each of three time points of the three health-check visits in EPIC-Norfolk study (http://www.srl.cam.ac.uk/epic/). Final analyses of these data are in progress, but initial impressions suggest that the measurements are highly repeatable, but have a reasonable degree of within-person change over time.
The consistency of results over the long time period of the assay was assessed to ensure that results were not susceptible to drift or batch effects. QC standard samples were plotted and continuously monitored throughout the project, with samples and plates showing strong deviations excluded. Overall consistency of measurements over time was high. Vitamin C and carotenoid levels were typically higher in women compared to men and varied by country. For example, lutein levels were considerably higher in Greece, Italy and France, than in central or northern European countries.
Plasma vitamin C levels showed a linear inverse relationship with risk of CHD, which attenuated somewhat after adjustment for conventional cardiovascular risk factors (age, energy intake, smoking, education, physical activity, alcohol, history of diabetes, hypertension, BMI) but remained statistically significant. Results varied for carotenoids but typically showed inverse associations, albeit generally weaker than for vitamin C. As these associations appear to be largely independent of risk factors in existing CVD risk scores, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that there is potential for vitamin C/carotenoid levels to improve risk prediction if added to risk scores.
WP7 - Vitamin D and its epimer
The overall objective was to assess the potential for inclusion of 25(OH)D3 and its epimer, and 25(OH)D2 in CVD risk scores. To do this required assay of this panel of these 3 biomarkers in the EPIC-CVD dataset, including the incident CHD cases, incident stroke cases and subcohort members.
Regular shipments of samples were made to Dr Gundersen’s laboratory at VITAS for assays of vitamin D and its epimers to be performed. Dr Gundersen’s team regularly returned data, including QC data, for the assays they completed as they passed through the pipeline.
In total, ~45,000 samples were assayed on the vitamin D platform, including ~12,000 incident CHD cases, ~8,000 incident stroke cases, ~15,000 subcohort members (as well as ~10,000 incident type 2 diabetes cases not included in EPIC-CVD). We additionally completed the measurement of around 2,100 samples from the EPIC-Norfolk study for a repeat measurements validation study. This included samples from ~700 participants with measurements available at each of three time points of the three health-check visits in EPIC-Norfolk study (http://www.srl.cam.ac.uk/epic/). Final analyses of these data are in progress, but initial impressions suggest that the measurements are highly repeatable, but have a reasonable degree of within-person change over time.
The consistency of results over the long time period of the assay was assessed to ensure that results were not susceptible to drift or batch effects. QC standard samples were plotted and continuously monitored throughout the project, with samples and plates showing strong deviations excluded. Overall consistency of measurements over time was high. Vitamin D levels in subcohort members varied by country and showed a strong expected correlation with the time of year in which the blood sample was taken, reflecting recent sunlight exposure at baseline in participants.
Similarly to previous smaller studies, a non-linear inverse association with vitamin D levels was seen with cardiovascular events, suggesting that only participants with vitamin D insufficiency are at increased cardiovascular risk. The strength of the association attenuated only slightly upon adjustment for conventional cardiovascular risk factors (i.e. age, sex, smoking status, systolic blood pressure, history of diabetes, BMI, total cholesterol, HDL cholesterol). As the association appeared to be largely independent of risk factors in existing CVD risk scores, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that there is potential for vitamin D levels to improve risk prediction if added to risk scores.
WP8 –Fatty acids
The overall objective was to assess the potential for inclusion of a panel of 37 plasma phospholipid fatty acids in CVD risk scores. To do this required assay of this panel of the 37 fatty acids in the incident stroke cases from EPIC-CVD, since assays in the incident CHD cases had been previously funded (by a European Research Council award), and assays in the subcohort members had been previously funded by the EPIC-InterAct Framework Programme 6 award.
An options appraisal was undertaken to select a laboratory to conduct the fatty acids assays and MRC Human Nutrition Research (MRC HNR) was selected as they have world leading expertise in the field with Dr Jules Griffin and Dr Albert Koulman. MRC HNR had already performed assays for T2D, CHD cases and common controls, so there was a strong scientific argument for undertaking the assays at the same laboratory, to reduce batch effects.
The measurement of plasma phospholipid fatty acids was successfully conducted by the MRC-HNR lab, using gas chromatography. Specifically, the method employed robotics to maintain a high throughput and generated results in relative concentrations of fatty acids in mol% units. Individual fatty acids within the major classes of fatty acids (saturated fatty acids, omega-3 and omega-6 polyunsaturated fatty acids, monounsaturated fatty acids and trans- fatty acids) were measured.
Quality control (QC) samples were included into every batch of samples analysed across the projects [InterAct (subcohort participants) and EPIC-CVD cases (CHD, stroke)]. This QC material was used to monitor assay precision and to assess if any drift occurred across the study or any significant differences were observed between the measurements of the study samples and the QC samples with known concentrations of fatty acids. We observed high QC stability of the fatty acids across the duration of the fatty acid measurements. In addition, no QC differences between measurement of InterAct (subcohort) and EPIC-CVD cases (CHD, Stroke) were observed across all the fatty acids. This is an essential prerequisite to be able to use the data on both cases and subcohort with confidence.
In addition to the samples measured in EPIC-CVD, we additionally completed the measurement of around 2,100 samples from the EPIC-Norfolk study for a repeat measurements validation study. This included samples from ~700 participants with measurements available at each of three time points of the three health-check visits in EPIC-Norfolk study (http://www.srl.cam.ac.uk/epic/). The analyses of repeat measures data are in progress and will help with understanding the determinants of change in fatty acid concentrations over time, and also with repeatability of measurements.
Overall, 92.5% of the EPIC-CVD participants (i.e. subcohort members, incident CHD cases and incident stroke cases) had measurements for all 37 fatty acids (FAs). FAs with a relative mean<0.05% (based on the subcohort) were excluded (C8:0, C10:0, C11:0, C12:0, C13:0, C21:0, C22:2, C14:1, C15:1 and C22:1-n9) due to their low levels leaving 27 FAs for analysis. 94 participants were excluded due to having outlying measurements. Levels of FAs varied across European countries – for example, trans fatty acid levels, although low overall, were substantially higher in the UK and the Netherlands compared to Mediterranean countries.
As expected, fatty acid levels were correlated within subgroups and clustered closely in hierarchical clustering models. For example, n-3 PUFAs showed strong positive correlations. Fatty acids were only weakly correlated with other cardiovascular risk factors, suggesting that any association with cardiovascular events is likely to be largely independent.
Associations of each FA with incident cardiovascular events were assessed using Prentice-weighted Cox proportional hazards models with robust standard errors to account for the case-cohort design. Shapes of association were assessed by classifying participants into quintiles of FA levels. Analyses were run within country and then combined using random-effects meta-analysis. Results were largely consistent with previous literature. For example, even chain saturated fatty acids were positively associated with CHD risk, whilst odd chain saturated fatty acids were inversely associated. These associations persisted after adjustment for conventional cardiovascular risk factors (i.e. age, sex, smoking, systolic blood pressure, BMI, history of diabetes, total cholesterol and HDL-cholesterol) suggesting that the associations with risk are independent. In conclusion, specific plasma phospholipid fatty acids are associated with cardiovascular events, independently from traditional CVD risk factors, hence the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that the inclusion of some plasma phospholipid fatty acids in CVD risk scores may provide predictive benefit.
c)_Results from RL3 - Analysis of intermediate biomarkers and genetic make-up in incident CVD
Research Line 3 had the following objectives:
• to assess the potential for inclusion of 225 metabolites in CVD risk scores
• to assess the potential for inclusion of common SNPs in CVD risk scores
• to assess the potential for inclusion of telomere length in CVD risk scores
WP9 –225 serum metabolites
The overall objective was to assess the potential for inclusion of a panel of 225 serum metabolites in CVD risk scores. To achieve this, an NMR metabolomics pipeline providing molar serum concentrations of 225 analytes (including lipids, lipoproteins, and low molecular weight metabolites) was implemented at an SME in Finland called “Brainshake”, a world-leader in developing and applying this technology. A pilot study of 930 samples showed reliable measurements could be obtained and that levels of analytes were mostly very similar to other population studies.
Samples were prepared and regular shipments of samples are being sent to Professor Ala-Korpela’s laboratory at Brainshake for metabolomics assays to be performed. Professor Ala-Korpela’s team regularly returned data, including QC data, for the assays they completed as they passed through the pipeline. The assay was successfully completed on ~46,000 samples.
Distributions of analytes looked broadly as expected. For example, total cholesterol was approximately normally distributed, triglycerides were positively skewed, whilst levels of XXL-VLDL-c were positively skewed with some samples having zero levels. Time-ordered measurements were plotted to assess drift over the timecourse of the project and coefficients of variation (CVs) were used to assess reliability of the assays. There was no evidence of drift in any of the analytes and CVs were below 10% for almost all analytes. For example, the CV for remant cholesterol was 5.9%, the CV for LDL-cholesterol was 3.8%, and the CV for alanine was 2.8%. Expected relationships were seen between lipoprotein fractions and lipid species, such as the strong relationship between triglyceride particles and lower density lipoproteins.
To assess the potential for inclusion in CVD risk scores, we investigated the correlation between NMR metabolites and cardiovascular risk, as predicted by a standard Framingham risk score (i.e. a model including age, sex, smoking status, history of diabetes, total cholesterol and HDL-cholesterol). Many metabolites were positively correlated with CVD risk, albeit weakly. For example, a range of types of cholesterol were all similarly positively correlated, suggesting that that, since they are themselves strongly positively correlated, they may not individually add to existing CVD risk scores. Other notable findings were: differing directions of association between individual metabolites and their ratios, and weak or null associations with low molecular weight metabolites.
In conclusion, we reliably measured 225 metabolite species in >45,000 serum samples. There were strong correlations between metabolites. Many metabolites, particularly those related to lipids and lipoproteins, were correlated with cardiovascular risk, usually in a positive direction. Hence the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that metabolites could usefully add to existing CVD risk scores, although only selected individual metabolites (or metabolite ratios) will be needed to capture the relevant predictive information.
WP10 – Common SNPs
The main objective was to assess the potential for inclusion of common SNPs in CVD risk scores. Cambridge Genomic Services (CGS, part of UCAM) was selected as the laboratory to perform the genetic assays using the 215,000 SNP CardioMetabochip+ array in the coronary disease cases and subcohort members due to their proven track record in providing high quality data from high-throughput projects using Illumina platforms, and their low costs. CGS was also used to perform the genetic assays using the 550,000 SNP HumanCoreExome array in the coronary and stroke cases, and subcohort members.
Genotyping of the 215,000 SNPs on the CardioMetabochip+ was been completed and the data delivered. The 550,000 SNP HumanCoreExome array was also assayed in all coronary disease cases, stroke cases and subcohort samples. Stringent quality control of both arrays was performed to ensure that only high-quality variants were included in analyses.
An initial report describing the possible genetic content for CVD risk prediction scores was produced and provided recommendations to leaders of other WPs in RL4 tasked with risk prediction analyses within EPIC-CVD. In summary, four methods of selecting common genetic variants for inclusion in risk prediction models were assessed. The method using variants known to be associated with CHD risk has the most immediate application and can be tested most simply. The unbiased selection of variants from across the genome shows great methodological promise but potentially at the cost of clinical translation. Using data from the CardioMetabochip, these methods have since been tested in relation to improvements in CVD risk prediction algorithms.
The ~50 known common variants associated with risk of coronary heart disease nearly all had associations in the same direction as the previous literature. Participants who had incident CHD events had slightly higher genetic risk scores than those who didn’t, regardless of the method used to weight the variants in the score. Participants with higher genetic risk scores had a higher risk of incident CHD. For example, there was a hazard ratio of 1.8 comparing participants in the top quintile of genetic risk score with those in the bottom quintile. Including the genetic risk score in a score based on conventional risk factors (age, sex, smoking, diabetes, blood pressure, total cholesterol and HDL-cholesterol) increased the C-index slightly, but did not statistically significantly improve reclassification. The change in C-index was smaller than that seen for other clinical biomarkers e.g. C-reactive protein. Further analyses will investigate inclusion of a broader panel of genetic variants.
WP11 – Relative telomere length
The main objective of this Work Package was to assess the potential for inclusion of telomere length in CVD risk scores. To do this, we aimed to measure leukocyte relative telomere length in the incident stroke cases, incident CHD cases and subcohort members in EPIC-CVD (additional funding was identified to allow concomitant assay of the incident type 2 diabetes cases also).
An options appraisal to identify suitable laboratories capable of measuring leukocyte telomere length (LTL) was carried out. The MRC Epidemiology Unit (part of the University of Cambridge) in Cambridge was identified as the best laboratory to carry out the work and this was based on technical capability, experience of high-throughput DNA applications and capacity to analyse >40,000 samples in the necessary timeframe. Following a thorough appraisal of the methods currently available for telomere length assessment, it was decided to use a quantitative PCR method (qPCR) since this is the only assay that allowed the high throughput necessary to conduct assays on >40,000 samples in the required timeframe. Relative telomere lengths were measured using the ratio between the relative quantities of the telomeric TTAGGGn repeat (T) and a single copy region of the Albumin gene (S), using a previously described monochrome multiplex qPCR method.
The initial optimisation and validation work for the assay was carried at the MRC Epidemiology Unit (EPID) in Cambridge (UCAM). Matullo’s laboratory at HUGEF provided 30 DNA samples, extracted from peripheral blood mononuclear cells, where telomere length had been previously measured by Southern Blot (considered the gold standard method for telomere length measurement). A blind validation assessment was carried out by EPID and a very good agreement between the two methods was observed (correlation coefficient r2=0.69). Excellent inter-plate agreement was obtained (correlation coefficient r2=0.99) and the intra-sample coefficient of variation was also very low (4.5% on average). Finally, an inverse and significant correlation between telomere length and age was found, as expected (r2=-0.61 p<0.0001) suggesting that the qPCR method was providing valid estimates of RTL.
Each sample was further diluted from 50ng/µl to 2ng/µl and a total of 10ng of sample was pipetted into each reaction well. The qPCR reagents used were SYBR® GreenER™ (ThermoFisher Scientific) and primers (IDT, Leuven, Belgium) in a total reaction volume of 15 µl. A ViiA™ 7 Real-Time PCR System (ThermoFisher Scientific) was used to run and measure all of the qPCR reactions. The main experiment was run in four separate batches that had either used different reagent lots or machine calibration settings. Each 384 well plate contained 358 unique samples, four non-template control wells, five intra-plate duplicates, five inter-plate duplicates, two global control duplicate samples and two identical 5 point standard curves. An average standard curve for T and S was calculated using the standard parameters within the ViiA™ 7 software.
Each sample was measured independently three times within the same batch or six times for duplicate samples. For a sample to pass the Coefficient of Variation (CV) needed to be <10%. Individual results were excluded if the T or S value was similar to the non-template control, if any result was measured in a well that had a systematic problem or if there were <3 T/S results. Full plates were excluded and (where possible) repeated if there were <4 passed intra-plate duplicates or they were identified as an outlier by the T/S ratio of the global duplicate. A duplicate was failed if there were less than 5 out of 6 measurements available or if the CV between them was <10%.
To adjust for any variation in assay batch we carried out a bridging experiment repeating the measurement of samples in batch 4 from samples that were run in batch 1, 2 and 3. Only samples that passed the QC criteria above were used for bridging comparisons. The mean value of the RTL for each batch was calculated and compared to the mean value of the same samples in batch 4 to calculate an adjustment factor of RTL for batch 1, 2 and 3. The correlation between T/S ratios in each bridging comparison was statistically significant (p<0.00005). The calculated adjustment factor was applied to every result for batches 1-3.
After exclusion of outlying plates and the bridging normalisation step, 37,281 samples passed quality control and were considered for statistical analysis. Cross-sectional associations for males and females of RTL with baseline variables were tested: an inverse and significant correlation between RTL and age was found, (r=-0.16) while females showed longer RTL than males (0.009; 95% CI,0.005-0.013 p<0.001) consistent with previous literature and therefore validating the reliability of the measurements in this study.
No other cross-sectional associations were observed in males or females with other baseline variables tested (Total cholesterol, HDL-C, ln(triglycerides), apolipoprotein A1, apolipoprotein B, ln(Lipoprotein[a]), ln(C-reactive protein), albumin, creatinine, uric acid, glucose, HbA1c, ferritin, total energy intake, smoking status, history of diabetes, physical activity index).
Associations between genetic variants in several genes and RTL have been previously discovered (Pooley et al., 2013). In addition to investigating the interindividual variability in RTL measurements due to different genetic background, we can use RTL-associated variants as instrumental variables in order to verify if our RTL measurements go in the expected direction. Within this very large study, our results confirmed the association between RTL measured and 5 variants previously reported in the literature to be associated with RTL, further supporting the reliability of the measurements.
In conclusion, based on the independence of the telomere length measurement from other cardiovascular risk factors, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that there is the potential for RTL to usefully add to a CVD risk score.
d)_Results from RL4 - Testing of clinical and public health utility
Research Line 4 had the following objectives:
• to identify lifestyle factors independently associated with CVD risk for use in modifiable risk scores
• to compare and calibrate existing risk scores to European populations, and recalibrate, develop and validate new risk scores
• to assess the impact on patients and physicians of new “modifiable” versus “detailed” risk scores
• to estimate the costs and benefits of different risk assessment strategies
• to identify and analyse the societal issues relating to European CVD risk screening.
WP12 - Development of modifiable risk scores
The main objective of this Work Package was to identify lifestyle factors independently associated with CVD risk for use in modifiable risk scores. In EPIC-CVD we aimed to optimize risk prediction for cardiovascular disease for the general population, by incorporating biomarkers and genetic markers on top of known risk factors for cardiovascular disease. Before developing new models, we first identified existing risk models through a systematic review of published scientific literature and validated their performance. The aim of the review was to provide an overview of all prognostic models that predict future risk of developing cardiovascular disease in the general population, assessing both modifiable and detailed risk scores.
Our systematic review of modifiable and detailed risk scores, conducted in conjunction with investigators from WP13, was conducted with 314 articles with risk scores focused on CVD outcomes being included in the literature review. Within these articles, 373 different prognostic models were developed, 519 external validations were performed and 278 incremental values were assessed. Models that are analysed in development, validation, and incremental value studies, and the populations used in these studies, show heterogeneity on many aspects, which need to be taken into account in the comparison of model performance. In the studies on validation of existing models on the EPIC-CVD dataset, and in assessing incremental value of biomarkers, genetic markers, and modifiable risk factors, this heterogeneity was taken into account. The systematic review has been completed and disseminated at conferences, with a scientific manuscript currently under revision at a major scientific journal. The results of the review have served as an input for decision regarding the design of the external validation study and the development of new detailed risk scores and risk scores with modifiable risk factors.
Having identified previous modifiable risk scores through the systematic review, and put in place a statistical analysis plan for validating existing risk scores and developing new modifiable risk scores (including a set of modifiable risk factors for evaluation), data generated within EPIC-CVD were analysed, initially using the CHD cases. The two main aims were:
1) To develop a model to predict cardiovascular events within 10 years that included only non-invasively measured lifestyle factors, including diet and physical activity
2) To assess the added predictive value of lifestyle factors when added to a set of well established cardiovascular risk factors, such as age, sex and smoking.
We performed backwards step-wise model selection based on a Wald t-test (p-value<0.05) to build a reduced prediction model including easy-to-obtain, non-invasive predictors. Criteria for consideration of lifestyle factors and non-invasively measured predictors (e.g. anthropometry, physical activity, diet) included strength and consistency of association with cardiovascular risk in the literature and availability of data within the EPIC-CVD dataset. To assess the added value of dietary information and physical activity over and above conventional predictors, we estimated the change in C-index, Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI).
Consistent with previous studies, known cardiovascular risk factors (e.g. age, smoking, blood pressure, pro-atherogenic lipids) were all higher in cases than non-cases. Consumption of fruit, vegetables and nuts was higher in non-cases, whilst red meat consumption was higher in cases. Our model including only non-invasive predictors, stratified by sex and country, included age, history of hypertension and diabetes, body mass index, waist-to-hip ratio, smoking, physical activity, alcohol consumption and three food groups: fruit, nuts (inverse association) and red meat (positive). Apparent discrimination of the combined model at 10 years was good with a C-index of 0.714 (95% confidence interval: 0.707-0.722). Heterogeneity across countries was large and statistically significant (I2=95%, p<0.0001) with country-specific C-index estimates ranging from 0.602 to 0.800.
Compared to a model including age, blood pressure, history of hypertension, total and HDL cholesterol, diabetes, and smoking, diet (alcohol, fruit, red meat) provided minimal additional predictive value: change in C-index of 0.002 and NRI of 0.72%. Further addition of physical activity did not improve discrimination or reclassification further (NRI=0.75%; change in C-index 0.002). Qualitatively similar results were found with multiple imputation.
In conclusion, there was little evidence of additional predictive value of non-invasive modifiable lifestyle factors, such as dietary factors or physical activity, when added to traditional cardiovascular risk scores in terms of discrimination, classification and calibration. However, the lifestyle-based risk score that we developed, including only non-invasively measured predictors, showed acceptable discrimination and therefore may have potential to be used for self-assessment of risk or inexpensive population screening. Use of such a score may have a range of benefits including greater acceptability to patients, cost and time saving for healthcare services (particularly those with limited resources), and greater potential to motivate behaviour change.
WP13 – Existing and detailed risk scores
This Work Package had two main objectives, namely (i) to compare and calibrate existing risk scores to European populations, and (ii) to recalibrate, develop and validate new risk scores.
Initially, we compared three varieties of the Framingham risk score (the most-often-validated risk score for CVD prediction): the original risk score, the original score but adapted to the incidence of the European population (recalibrated) and an adapted score, in which also coefficients were fit according to the European population (refit). By doing so, we assessed whether using an existing score would be sufficient, or whether adaptation would be necessary for risk prediction in the European population when assessing added value of modifiable risk factors and biomarkers).
We compared predicted probabilities by the original FRS (original) with a risk score in which the baseline hazard was recalibrated to the baseline hazard in our population (recalibrated) and an entirely new risk score using the same predictors (refitted) with the observed outcome, focusing on predictive performance. This was assessed using Harrell’s C-index, similar to the area under the receiver operator characteristic curve (AUC) adapted to survival analysis. This indicator ranges from 0.5 (no discrimination) to 1 (perfect discrimination). Validation was done per country and sex group and results were subsequently pooled by random-effects meta-analysis.
C indices with 95% confidence interval for the original, recalibrated and refit model were 0.698 (0.673-0.724) 0.693 (0.667-0.720) and 0.703 (0.676-0.731) respectively. These differences may seem small on this absolute scale, but as potential added value of biomarkers or modifiable factors may also be small, we consider the refitted model to be superior to the others. Using the existing Framingham risk score shows reasonable discrimination. Recalibration of the intercept did not improve the discriminative performance, in fact it even slightly decreased, probably due to heterogeneity among country and sex groups. The best discriminative performance was obtained by the model in which the predictors were actually refitted to the data. Therefore, in the analyses on modifiable risk scores and detailed risk scores, we used predictors from the Framingham model as a basis, but refit the coefficients of these variables.
For investigation of detailed risk scores, we developed four different models in a stepwise fashion. The first model used the easily obtainable predictors: age, smoking, history of diabetes, history of hypertension, and body mass index. Next we added systolic blood pressure. In model 3 total cholesterol and HDL were added to the model (as these factors are more often used in risk scores). Finally all other biomarkers were added to the model to investigate the maximum predictive performance that could be obtained by these biomarkers. Similarly, sex-specific models were developed for males and females separately. Performance of the four models in predicting 10-year risk was evaluated using discrimination (i.e. whether the model can distinguish between individuals who did and did not have the event) and calibration (i.e. to what extent the predicted probabilities agree with the observed risk) measures.
Consistent with previous studies, known cardiovascular risk factors (e.g. age, smoking, blood pressure, pro-atherogenic lipids) were all higher in cases than non-cases. Glycaemic markers, liver markers, CRP and lipoprotein(a) also had notably higher levels in cases. The model containing non-invasive easily measured variables only (model 1) achieved an apparent C statistic of 0.71 (95% confidence interval 0.70 - 0.72) which was similar after correction for optimism: 0.71 (0.68-0.73). However, heterogeneity across countries was very large (I2 = 93%, p < 0.0001) in particular among females. Adding systolic blood pressure (model 2) and cholesterol levels (model 3), variables frequently included in currently used risk scores, improved the optimism-corrected C index to 0.71 (0.69-0.74) and 0.73 (0.71-0.76) respectively. Additionally adding the other biomarkers (model4) further increased the optimism-corrected C statistic to 0.75 (0.72-0.77). The most important predictors in the full model which remained statistically significant were age, smoking, history of hypertension, systolic blood pressure, HDL, apolipoprotein B and lipoprotein(a), creatinine and uric acid, glucose and HbA1c, and alkaline phosphatase.
In conclusion, adding a set of routinely measured clinical biomarkers into a cardiovascular risk score including standard risk factors improved the potential of the risk score to identify persons at increased risk of future events. However, the major improvement above the non-invasive well-established risk factors was achieved by adding systolic blood pressure, total cholesterol and HDL-cholesterol to the model. Adding biomarkers related to different areas of biology (lipids and glucose metabolism, inflammation, renal function) improved discrimination further. However, the considerable heterogeneity between countries and between males and females should be taken into account and warrants further investigation. In resource-rich settings, inclusion of additional biomarkers beyond those routinely used (total cholesterol and HDL-cholesterol) may improve our ability to identify individuals at high risk of cardiovascular disease for targeting interventions.
WP14 - Assessment of impact of risk scores with differing content on behaviour
The main objective of this Work Package was to assess the impact of ‘new modifiable’ versus ‘detailed’ risk scores on patients and physicians. As a form of triangulation, and in order to collect data that is rich as possible, we conducted three inter-related studies in this working package:
a) Randomised controlled trial - Information and Risk Modification Trial (INFORM)
b) Systematic review - Impact of provision of cardiovascular disease risk estimates to healthcare professionals and patients
c) Qualitative study - The response to receiving phenotyping and genetic coronary heart disease risk scores and lifestyle advice
a) Randomised controlled trial - Information and Risk Modification Trial (INFORM)
The INFORM randomised trial was designed to contribute toward the understanding of the impact of provision of phenotypic and genetic coronary heart disease (CHD) risk information on health-related behaviour change and other important clinical outcomes. Comparison between randomised trial arms enabled estimation of: 1) the effect of providing web-based lifestyle advice; 2) the effect of providing a genetic risk score in addition to lifestyle advice and a phenotypic risk score; 3) the effect of providing risk score information and lifestyle advice and 4) the effect of providing risk score information in addition to lifestyle advice. The INFORM study will also provide information on potential moderators and mediators of associations between the intervention and health-related behaviour change and other clinical outcomes.
In a parallel-group, open randomised trial, we allocated 932 male and female blood donors with no previous history of CVD aged 40¬84 years in England to either no intervention (control group) or to one of three active intervention groups: i) lifestyle advice only; ii) lifestyle advice plus information on estimated 10-year CHD risk based on phenotypic characteristics; and iii) lifestyle advice plus information on estimated 10-year CHD risk based on phenotypic and genetic characteristics. The primary outcome was change in objectively measured physical activity. Secondary outcomes included: objectively measured dietary behaviours; cardiovascular risk factors; current medication and healthcare usage; perceived risk; cognitive evaluation of provision of CHD risk scores; and psychological outcomes. The follow-up assessment took place 12 weeks after randomisation. The protocol paper for INFORM study was published in September 2015 by BMC Public Health.
Between March 2015 and June 2015, electronic invitations approved by the local ethics committee were sent to participants who met the invitation criteria. Recruitment was completed earlier than planned on 25/06/2015 as we reached our target of 932 participants. We completed follow up of participants on 31/12/2015. 5124 participants of the INTERVAL study (study within which INFORM was nested) were available for invitation, we sent 2908 invitations (57%), 1152 participants responded (39.6%), 166 participants were not eligible/did not consent (14.4%), 986 participants consented (85.6%), 956 participants were randomised (97%).
Initial analyses have been completed and an abstract describing the preliminary results has been submitted to the SAPC (Society for Academic Primary Care) conference 2016. These results suggest that the provision of risk information (either based on phenotype or genotype) is unlikely to enhance preventive strategies but neither does it cause psychological harm in the short-term. Provision of lifestyle advice alone or alongside CHD risk information might improve fruit and vegetable intake, which needs to be confirmed once analyses of carotenoid levels are completed.
b) Systematic review - Impact of provision of cardiovascular disease risk estimates to healthcare professionals and patients
The purpose of this review was to assess whether provision of a CVD risk model estimate to either patients or practitioners, as opposed to other simultaneous or subsequent interventions, such as lifestyle advice or exercise programmes, impacts patient or practitioner behaviour or health outcomes.
An electronic literature search of Medline and PubMed from 01/01/2004 to 01/06/2013 was conducted with no language restriction and manual screening of reference lists of systematic reviews on similar topics and all included papers. Eligibility criteria for selecting studies were: (1) primary research published in a peer reviewed journal; (2) inclusion of participants with no history of CVD; (3) intervention strategy consisted of provision of a CVD risk model estimate to either professionals or patients; and (4) the only difference between the intervention group and control group (or the only intervention in the case of before-after studies) was the provision of a CVD risk model estimate.
After duplicates were removed, the initial electronic search identified 9671 papers. We screened 196 papers at title and abstract level and included 17 studies. The heterogeneity of the studies limited the analysis but together they showed that provision of risk information to patients improved the accuracy of risk perception without decreasing quality of life or increasing anxiety, but had little effect on lifestyle. Providing risk information to physicians increased prescribing of lipid lowering and blood pressure medication, with greatest effects in those with CVD risk >20% (RR for change in prescribing 2.13 (1.02 to 4.63) and 2.38 (1.11 to 5.10) respectively). Overall there was a trend towards reductions in cholesterol and blood pressure and a statistically significant reduction in modelled CVD risk (-0.39% (-0.71 to -0.07) after, on average, 12 months.
In conclusion, our review suggested that providing CVD risk model estimates to professionals and patients improves perceived CVD risk and medical prescribing, with little evidence of harmful effects on psychological well-being. The systematic review was published in October 2015 by BMJ Open.
c) Qualitative study - The response to receiving phenotypic and genetic coronary heart disease risk scores and lifestyle advice
The aim of this study was to use qualitative methods to explore the short term response to receiving different forms of coronary heart disease (CHD) risk information and lifestyle advice for risk reduction on health-related behaviours. Forty two face to face interviews and two focus groups were conducted across England with participants from the INFORM trial who received a combination of individualised phenotypic and genotypic CHD risk score and web-based lifestyle advice. The risk scores were presented in different formats, including absolute 10 year risk as a percentage, ‘heart age’ and colour visualisation of absolute and comparative risk as a thermometer. Interviews and focus groups explored participants’ understanding and reaction to the risk scores, the usefulness of the lifestyle advice and attempts to change lifestyle during the intervention. We tape-recorded and transcribed the interviews and focus groups and analysed them using thematic analysis.
Three main themes emerged: limitations of risk scores to generate concern about CHD risk and motivation to reduce it; the advantages of the ‘heart age’ format of risk score presentation in communicating a message of non-optimal lifestyle; and intentions and attempts to make moderate lifestyle changes which were prompted by the lifestyle advice regardless of concerns about the risk score. These findings suggest that there are a number of limitations to the use of risk scores to communicate a message about the need for a lifestyle change. Of the formats used, the ‘heart age’, if noticed, appears to convey a powerful message about how far from optimal risk an individual person is. An interactive, user friendly, goal setting based lifestyle website can act as a trigger to initiate moderate lifestyle changes, regardless of concerns about risk scores. We expect the findings from the qualitative studies within INFORM to be published in early 2016 and disseminated via conferences and to participants via Newsletters.
WP15 – Evaluation of screening strategies and cost-effectiveness
The main objectives of this Work Package were to estimate the costs and benefits of different risk assessment strategies focusing on: 1) when is stepwise screening preferable to mass screening? 2) what is the optimum frequency of screening? 3) how should age best be incorporated into CVD risk assessment? 4) what are the most cost-effective approaches across countries and subpopulations?
1) Development and validation of a multi-state model
The purpose of the decision model developed is to quantify the long-term natural history of CVD, and to simulate the impact of different screening and treatment strategies on fatal and non-fatal outcomes, life-expectancy, quality-adjusted life-years (QALYs) and costs. We developed a Markov multistate model in Excel and validated it using data from other epidemiological cohorts while the EPIC-CVD data was being prepared.
2) Development of methods to extract transition rates from the epidemiological data and use these rates to derive transition probabilities to populate the decision model
We used Cox proportional hazards regression, with age-at-risk as the timescale, to estimate the age-specific transition rates as a function of covariates, allowing for left truncation or delayed entry into age-at-risk groups depending on when individuals enter the study or move to a different state (i.e. changing risk-sets). This involved choosing appropriate smoothing methods for the estimated annual hazards. The initial work was based on two European epidemiological cohort studies.
As an example for men, based on a population cohort aged 50 at baseline, we estimated that the average life-years gained if all men take a statin from age 50 onwards is 1.11 years, if from age 60 onwards it is 0.82 years, and if from age 70 onwards it is 0.41 years. The corresponding figures for expected life-years gained free of CVD are 1.82 1.34 and 0.65 years. If only the men aged 50 in the top 20% of predicted risk (according to the Framingham risk score) take a statin from age 50 onwards, their estimated average life-years gained is 1.56 years, compared to the 1.11 years estimated for all men.
In a second example, using another population-based cohort of men with a wide age range at baseline, we estimated that the average life-years gained from taking statins for those in the top 20% of the Framingham risk score is 0.71 years, while that for those in the top 20% of predicted risk given their age it is 1.75 years. In other words, in terms of life-expectancy gained when 20% of the population is given statins, it is better to choose those who are in the top fifth based on high predicted risk given their age. These methods and results are being prepared for publication.
We have also developed methods to estimate transition rates in case-cohort data with multiple centres, and applied this to EPIC-CVD with the CHD outcomes.
3) Elaboration of systematic reviews and meta-analyses to derive key inputs for cost-effectiveness evaluation, and development of modelling methods.
A structured literature review was carried out in order to identify long term cost-effectiveness models of strategies for the prevention of CVD. The results of studies that evaluated screening policies were quite varied. For example, Wald NJ et al 2011 found that screening based on age is more cost-effective than Framingham score based screening. Rapsomaniki et al 2012 found that adding risk variables such as region, age and year of birth, systolic blood pressure, total cholesterol and smoking status, was cost-effective with a cost effectiveness ratio of less than $20,000 per year free of CVD. Cobiac et al 2012 showed that CVD prevention based on absolute risk is more cost-effective than single risk factor thresholds, and recommended statin drugs to everyone with at least 10% absolute risk. Ruitjer et al 2013 showed that performing carotid intima-media thickness (CIMT) measurements induced 1% lower absolute risk of myocardial infarction, and adding this variables to the Framingham equation would be cost-effective.
The studies that evaluated statin treatment (on its own or in combination with other treatments) in identified high risk individuals also showed quite varied results. Results varied depending on: age, gender, the presence or absence of other risk factors and the type of statin used. In general, the incremental cost-effectiveness ratio was lower in younger patients (<50 years) and higher in older patients (>70 years) for a given level of baseline risk (that is, statins are more cost-effective in younger than older patients, with the same baseline risk). Likewise the incremental cost-effectiveness ratio was lower for patients with higher baseline CVD risk.
This review has helped to clarify methodological questions (for example, the structure of the model, the health states that are important) and to identify parameters. 57 papers were retrieved by the searches and 29 included for review following application of exclusion criteria.
An academic paper has been published that critically appraises the published models and examines their suitability for evaluating primary prevention of CVD in European populations in European Journal of Health Economics 2015, Epstein et al.
Additional publications prepared as part of this Work Package include:
Martin-Ruiz E, Olry de Labry Lima A, Garcia Mochon L, Espin J, Ocaña-Riola R, Epstein D.
The benefits and risks of statins for primary prevention of mortality and cardiovascular events: umbrella review. European Journal of Clinical Pharmacy (accepted for publication)
Jones E, Epstein D, Garcia Mochon L. A procedure for deriving formulas to convert transition
rates to probabilities for multi-state Markov models. (Health Economics, under review)
Martin-Ruiz E, Olry de Labry Lima A, Ocaña-Riola R, Epstein D. Systematic review of the effect of adherence to statin treatment on cardiovascular events and mortality in primary prevention. (Manuscript to be submitted to a pharmacy journal)
Martin-Ruiz E, Olry de Labry Lima A, Ocaña-Riola R, Epstein D. The effectiveness of non- pharmacological interventions for primary prevention of cardiovascular disease: umbrella review of systematic reviews (Manuscript to be submitted to a pharmacy journal)
WP16 - Implementation of new cardiovascular risk assessment tools across Europe
The main objective was to identify and analyse the societal issues relating to European CVD risk screening.
As part of EPIC-CVD, we undertook work to explore current tools, policies, perceived barriers and facilitators, and regulatory issues pertaining to the assessment of cardiovascular risk in Europe. We also developed an online web-based tool as a means of implementing a risk score based on novel biomarkers. The pre-specified work package objectives were addressed by three discrete pieces of work conducted by our three partner organisations: RAND Europe, PHG Foundation, and the University of Cambridge. This report summarises the three separate pieces of work undertaken, including key findings, and outlines plans for dissemination and on-going research.
a) Current and future cardiovascular disease risk assessment in the European Union: an international comparative study
The objectives were 1), to identify tools currently being used for calculating cardiovascular risk scores in a range of European countries; 2) to identify policies that are most likely to be acceptable in different European countries and approaches that are most likely to maximise adoption; and 3) to identify the barriers and enablers to screening for a wider range of cardiovascular risk factors in a range of European countries. For the third objective, a decision was taken to focus on the barriers and facilitators to implementation of novel cardiovascular risk factors in risk tools more generally, which was felt to be more relevant to current clinical practice and policy than examining the issue specifically in relation to screening.
RAND Europe conducted an online pan-European survey of senior academic, clinical and policy stakeholders from all EU Member States (November 2014 to March 2015). Survey invitations were sent to 560 experts from all member states; the response rate was 24%. In addition, eight in-depth country case-studies were developed, based on telephone interviews with clinicians, academics and policy makers (May to October 2015). Interviews were supported by targeted literature review both to guide the interview and help interpret findings. Eight countries were selected to reflect variations in health system, macro-economic factors, and geographic location across the EU (Bulgaria, Finland, Germany, Greece, Latvia, Spain, Sweden, UK). There were between three and five interviewees for each country.
This study confirms that the European Society of Cardiology (ESC) is the main body providing and disseminating guidelines for CVD risk assessment across Europe, and the SCORE risk assessment tool, which is endorsed in these guidelines, is the tool most widely reported in use. Implementation of ESC guidelines and the use of cardiovascular risk assessment tools in clinical practice are variable and often low. Risk assessment was found to be predominantly, but not exclusively, conducted by primary care doctors, and health system factors, particularly lack of funding and lack of time, were identified as important barriers to risk assessment use. A clear theme emerging from the work was that any novel risk assessment biomarkers identified by current research will need to be incorporated in ESC guidelines as the key initial step to translation into clinical practice. Evidence of clinical utility and cost-effectiveness will be required for implementation of novel biomarkers, but ethical, legal or regulatory issues appeared of less concern to interviewees. Findings for all the key themes were consistent across Europe, despite differences in context.
In conclusion, the ESC has a central role in informing practice across Europe, and efforts to improve practice are likely to benefit from being channelled through this organisation. Although factors such as integration of risk tools in clinical computer systems may improve practice, there remain significant barriers to the successful implementation of cardiovascular risk assessment, including lack of funding. Despite considerable interest in developing novel biomarkers to optimise cardiovascular risk assessment, it is likely that focusing efforts on increasing the extent of implementation of current approaches will offer the greatest gains in terms of improving primary prevention of cardiovascular disease.
b) Genomics and risk stratification for cardiovascular disease: regulatory implications
PHG Foundation were tasked with the objective of considering the broader societal issues (e.g. ethical, legal, social) of including more detailed information in cardiovascular risk scores, such as hereditary information. It was decided to focus on investigating the legal and regulatory implications of incorporating genomic information into risk stratification for cardiovascular disease across four EU Member States. This focus was chosen as current cardiovascular risk score systems do not take into account genomic information, and the topic has hitherto been under-examined in the literature.
It was decided to include countries that have sophisticated health systems familiar with the complexities of genomic science and medicine as well as a range of cultural and legal approaches to risk stratification, the use of genomic information, and the provision of healthcare more generally. These different perspectives provide variation in terms of healthcare delivery frameworks, implementation of EU law, and the collection, storage and use of genomic information. Four countries were selected for analysis: France, Germany, the Netherlands, and the UK.
The investigation was based on expert opinion and desk-based analysis and literature searches. Expert opinion was provided by a two-day international workshop convened by the PHG Foundation in May 2015. Experts were invited from the four countries, comprising academic lawyers, practising lawyers, and cardiologists with an interest in risk stratification. The workshop was designed to act as a foundation for wider study, with an iterative approach taken: initial work helped focus the workshop, and then expert opinion from the workshop was used to guide further investigation.
This investigation found no absolute prohibition on the incorporation of genomic information into risk stratification for cardiovascular disease at EU or Member State level. However, risk stratification tools that include genomic information are subject to more rules than those that do not. The context strongly influences the application of various rules, with the most important contexts being the purpose for which the tools are used and their mode of delivery. Data protection law strongly influences the use of risk stratification tools, especially in jurisdictions making special legal provision for the use of genomic information. It is unclear whether risk stratification tools constitute in vitro diagnostic devices under current and future EU legislation.
A key recommendation arising from this work is that those providing risk stratification tools should be aware of the need to adhere to a significant number of applicable regulatory controls. Given the complexity, formal legal advice should be sought where necessary.
c) Development of a working model of a risk assessment tool incorporating novel risk factors
A web-based tool was developed which used existing algorithms, with two key features: firstly, incorporating genetic risk, and secondly, allowing presentation of a country specific risk. Such a tool could act as a proof of concept, which could be readily adapted once a final risk algorithm was made available.
A co-design process was used to inform the functionality and appearance of the web tool. The presentation of risk to individuals was provisionally based on an approach used by the intervention website used by the INFORM clinical trial. The entering of phenotype data into the system followed a standard approach as used by other systems such as the Edinburgh cardiovascular risk calculator (cvrisk.mvm.ed.ac.uk). The site was also designed to allow manual uploading of genetic data, or through an interface with an external data source (the commercial genetic data provider 23andMe was used as a test case). The underlying risk algorithm was developed using a limited dataset available from the EPIC Norfolk patient cohort.
Interviews were held separately with clinicians and patients, whilst interacting individually with the web tool. An additional group meeting was held with the patients to share ideas. Key themes were identified from interview transcripts, and changes were made to the tool accordingly.
The web tool was designed for use in a clinical environment, to facilitate discussions of risk between clinicians and patients. Risk was displayed through three “infographics”, including a graduated “thermometer” and ten-by-ten “smiley face” chart to convey absolute risk of cardiovascular disease, and a graphic depicting heart age. Comparison with an individual with optimal risk factors was also presented. A number of changes were implemented as part of patient and clinician feedback, including a focus on a combined (rather than separate) model of genetic and phenotypic risk, provision of a means of visualising the effect of risk-reducing interventions (e.g. statins, smoking cessation), improving various aspects of the graphical design, and rewording of various phrases and explanations to increase clarity. The site has not being made publically available, as the underlying algorithm is a temporary one which has been used for development purposes only. Once the EPIC-CVD programme delivers further detailed risk score algorithms, our intention is to integrate them into the website and make the tool publically available. We will also explore the potential of interfacing with alternative genetic data repositories beyond 23andMe, and making it available in different languages. However, implementation in practice will need to take into account some of the issues raised as part of the work undertaken by PHG Foundation, highlighting the need to be aware of, and adhere to, relevant legislation.
We intend to carry out a future study using the website to examine patients’ and clinicians’ reactions to using the tool, and in particular how incorporating genetic information into such a tool influences perceptions of risk and decisions about cardiovascular risk management. This will form the basis of a mixed-methods study, which will be submitted for publication in a scientific journal.
Potential Impact:
4. Potential impact, main dissemination activities and exploitation of results.
4.1 Potential impact
EPIC-CVD aimed to build upon previous important efforts in CVD risk prediction by using population resources, biomarker technologies, and translational studies that have only recently become practicable for large-scale application across Europe. Our multidisciplinary research consortium, therefore, focused world-leading expertise in translational research, epidemiological and diagnostic technology and knowledge to produce innovative outputs, such as modifiable and detailed risk scores, risk calculation and communication tools, customised stepwise screening strategies and cost-effective policy recommendations.
The single biggest impact of EPIC-CVD will be to provide clinicians and policy-makers across Europe with an evidence-based menu of contemporary options for targeted and cost-effective CVD risk approaches tailored to the needs of Europe's diverse populations. Through this, EPIC-CVD will contribute to health, wealth, innovation, and research competitiveness within the EU.
4.1(a) Better health
Public health burden of CVD in Europe CVD is Europe's leading cause of death and a major source of disability, with a burden expected to increase as a result of the ageing population and increasing prevalence of obesity and diabetes. Such considerations have impelled FP7 to adopt as a thematic priority area "the development and validation of...methods for health promotion and prevention including the promotion of healthy ageing, diagnostic tools and medical technologies, and sustainable and efficient healthcare systems."
Calibration of scores to different European populations EPIC-CVD has developed and calibrated existing and new risk scores to diverse European populations. For risk scores already in clinical use, the outcome of this is greater predictive accuracy. This is because most existing risk scores have been derived in and calibrated to populations in the USA (eg, Framingham, Reynolds) or a single European country (eg, ASSIGN and QRISK2 from the UK or PROCAM from Germany), or lacked information on dietary and genetic factors that prevail in Europe (eg, SCORE). By addressing these limitations in existing risk scores, EPIC-CVD should encourage greater use of risk scores in European populations because patients and healthcare professionals will appropriately regard the scores as being of greater local relevance. Thus, EPIC-CVD is likely to enhance ability to identify people at high CVD risk and manage them appropriately, leading to improvements in identification, risk assessment, clinical decision making and clinical outcomes.
Modifiable risk scores EPIC-CVD's contribution, however, goes beyond providing useful refinements to existing risk scores by creating Europe's first modifiable CVD risk scores containing objective biomarkers of lifestyle. When accompanied by personalised feedback that translates risk estimates into easily understood advice about lifestyle factors that can be readily self-monitored, these novel scores should contribute to the prevention of CVD in several ways:
●to enhance predictive accuracy, especially because we have studied objective and reproducible nutritional biomarkers, resulting in improved identification and risk assessment
●to promote behaviour change so that patients reduce adverse lifestyles, leading to better clinical outcomes
●to promote change in the behaviour of healthcare professionals so that CVD risk is monitored more effectively, leading to improved clinical decision making
●to contribute to the prevention of other conditions of major public health importance in Europe, such as obesity, T2DM and certain cancers, again leading to better clinical outcomes
●to be genuinely complementary to public health activities that aim to reduce the overall population risk of cardiovascular disease by promoting a healthy lifestyle (diet, exercise, avoidance of smoking).
Detailed risk scores EPIC-CVD has also developed innovative detailed risk scores that, for the first time, are able to capture the interplay of nature and nurture. Again, this output is likely to result in improvements in identification, risk assessment, clinical decision making and clinical outcomes.
Content of scores: safe, practicable, and ready for translation Although EPIC-CVD has studied a large range of biomarkers (≈300 soluble biomarkers and 215,000 carefully selected genetic variants) on an unprecedentedly powerful population scale across Europe, this project was not simply an "-omics" discovery effort. Rather, EPIC-CVD's work will yield clinical usefully biomarkers because it has also assessed a carefully selected set of existing and emerging biomarkers. In particular, EPIC-CVD has studied biomarkers that already have standardised high-throughput blood tests and a reasonable likelihood of emerging as cost effective, safe, and valid (eg, biomarkers already in routine clinical use or recommended by guideline statements or regulatory agencies). Because these biomarkers have rapid translational potential, EPIC-CVD's efforts are more likely to lead to improved cardiovascular risk prediction and contribute to the development of personalised predictive medicine in a short timeframe.
4.1(b) Greater wealth
Efficient strategies EPIC-CVD has put a high priority on developing approaches that enhance the cost effectiveness and operational efficiency for targeting preventive resources. For example, it seems unlikely that many countries in the EU (particularly the resource-constrained countries of southern and eastern Europe) will be able to follow the example of the UK, which has introduced a universal national CVD screening programme for adults aged 40-74 years at an estimated cost of €2 billion between 2010 and 2015. Consequently, EPIC-CVD will provide cost-effective alternatives to such un-targeted approaches by building on our experience in developing sequential or "stepwise" screening strategies and by tailoring such approaches to the various budgets, health information systems, and cultures across Europe.
4.1(c) Evidence-based use of innovative technologies
There is a worldwide trend toward more demanding scrutiny of predictive tests by health technology assessment agencies, exemplified by the US FDA’s warnings to companies that sell genetic tests directly to consumers. In Europe, the 2002 European Parliament's report on Life Sciences and Biotechnology Strategy called on the Commission "to take the necessary steps for an EU-wide regulation on DNA-testing, choosing, if possible, a legal basis (e.g. Article 152 (health) or Article 153 (consumer protection))...". In 2008, the Council of Europe's Additional Protocol on Human Rights and Biomedicine, Concerning Genetic Testing for Health Purposes stated (Article 5): "Parties shall take the necessary measures to ensure that genetic services are of appropriate quality ... [and] genetic tests meet generally accepted criteria of scientific validity and clinical validity". However, a clear regulatory framework has not yet emerged in Europe for prognostic biomarkers as many products, such as genetic tests, are currently left unregulated because they are classified as low risk and require only self-certification (making them liable to multiple conflicting interpretations).
EPIC-CVD will make an important contribution to evidence-based standards for innovative prognostic technologies in 2 main ways. First, through the PHG Foundation, a partner in EPIC-CVD and a leading independent genetics think-tank, we have considered the broader ethical, legal, and social implications of prognostic biomarker technologies for Europe. Second, the scientific framework of EPIC-CVD has contributed to new evidence-based paradigms for prognostic biomarkers. This is because, for the first time, EPIC-CVD has provided a rigorous and integrated evaluation in CVD across Europe that combines comprehensive study of predictive performance of risk scores, detailed health economic modelling, and qualitative and quantitative translational studies of patient-centred outcomes. In aggregate, these approaches have been able to reliably test whether a new approach improves risk assessment, clinical decision making, and clinical outcomes as well as cost effectiveness, safety, validity and incremental benefit over existing prediction methods. The net effect of these efforts should be to better protect the public, reduce uncertainty in product development for industry, and accelerate clinical translation by avoiding wasteful effort in suboptimal studies. We should also be able to provide a much clearer path than now exists in Europe to encourage SME efforts towards research and innovation in the prognostic biomarkers sector.
4.1(d) Greater European research competitiveness
Strategic scientific context In the document “Life sciences and biotechnology – a strategy for Europe”, the European Commission describes its aim to restore European leadership in life sciences and biotechnology research. In the 2010 document "Europe 2020 Flagship Initiative: Innovation Union", the EC emphasises the need for researchers and innovators "to work and cooperate across the EU to get more innovation out of our research" and launch research partnerships that improve "the quality of life of an ageing population by new innovative solutions, clinical tests and diagnostics". A major growth area in biomedicine worldwide is "the development of personalised and predictive medicine”. Europe is strategically well-positioned to take a leading role in this endeavour because of the region’s: (1) natural advantages of striking lifestyle, cultural, and genetic diversity (2) longstanding scientific strengths in key disciplines such as public health, nutrition, and genetic epidemiology (3) cluster of excellent “research intensive SMEs” in the life sciences that provide "research and innovation" (4) track record of cohesive collaboration in major pan-European scientific projects.
Because EPIC completed its recruitment of 520,000 participants in the 1990s, it has about a 5 year head start over other large “biobank” initiatives that have been established, or that are currently recruiting in China, Australia, the USA, and individual European countries. Hence, EPIC-CVD has been able to take advantage of a “window of opportunity” and has been able to contribute to European and international leadership in this area.
4.2. Dissemination strategy
At the outset of the EPIC-CVD project, with representatives from the partner organisations, we developed a dissemination plan. The key to a dissemination plan in a project of this nature was to recognise that there are multiple target audiences and that each audience needs to be addressed in a different manner, using different media.
Communication plan The project deliverables included a “Communication Plan” which included the main dissemination materials and tools for EPIC-CVD. This plan aimed to:
(1) select the appropriate modes of communication and communication tools
(2) define key messages
(3) tailor information and deliver to the intended recipients
(4) maximise the exposure of messages
(5) evaluate the results of our dissemination efforts
(6) capitalise on the European Commission’s dissemination resources.
Interactive website The Project Management Office was responsible for the development and maintenance of our interactive website (www.epiccvd.eu) which will be maintained after the project ends, given the potential clinical importance of the material that will be added. This website has been integral to EPIC-CVD throughout its lifetime. Its provision of tools, strategies, information, publications, policy documents etc has intensified as the project has gone on.
Target audiences included organisations such as the EC, the European Medicines Agency, the European Society of Cardiology, the European Atherosclerosis Society, the European Association of Cardiovascular Prevention and Rehabilitation, the European Stroke Initiative, European Heart Network, companies that manufacture diagnostic and biomarker products, national governments.
Scientific publications With help from the Project Management Office, the Study Coordinator has overseen the work of an Analysis Coordination Group and the reporting of findings. The Project Management Office has recorded and coordinated the publication process with all partners by means of the EPIC-CVD intranet (with a dedicated publications section and a searchable database). Press releases for high-impact papers are prepared by the Project Management Office in conjunction with the organisations of the relevant stakeholders.
4.3. Dissemination to date
Dissemination of the findings from EPIC-CVD has to date taken 5 main forms:
- peer-reviewed scientific journal publications
- oral presentations at workshops, meetings and conferences
- policy documents
- online tools and resources
- participant newsletters
a) Journal publications
To date, there have been 15 publications enabled by the EPIC-CVD project. In addition to publications relating to the specific Work Packages of the project, the development of the EPIC-CVD dataset has enabled many further projects related to aetiological analyses and methodological work that weren’t necessarily anticipated at the start of the project, providing excellent added value to the European Commission’s investment. Evidence of this comes from the EPIC-CVD extranet, which has received 67 proposals for studies from investigators within EPIC-CVD to date. Scientific papers from each of these proposals will be written and submitted to peer-reviewed journals during the course of 2016/17.
b) Oral presentations
Findings from EPIC-CVD Work Packages have been presented at a number of national and international scientific conferences, workshops and meetings, including:
- Society for Academic Primary Care, January 2015
- FENS conference, June 2015
- European Epidemiology Congress, June 2015
- European Society of Cardiology, August 2015
- Cochrane Colloquium meeting, October 2015
Future oral presentations already planned include:
- American Heart Association Scientific Sessions, 2016
- European Society of Cardiology Congress, 2016
- Society for Academic Primary Care, 2016
Additional oral presentations of findings from EPIC-CVD will be decided as near-final versions of scientific manuscripts emerge.
c) Policy documents
Some aspects of EPIC-CVD work are amenable to publication as policy documents that can be distributed to relevant stakeholder groups (eg, patients, regulatory bodies, practitioners etc). For example, as part of WP16, the PHG Foundation investigated the regulatory implications of using genomics in cardiovascular risk stratification. This work was published as a policy report in 2015 and widely disseminated to key stakeholder groups identified as part of WP16. Further findings from EPIC-CVD, particularly those from WP16 (eg, results of the case studies comparing CVD risk screening across 8 European countries), will also be disseminated as policy reports if deemed appropriate.
d) Online tools and resources
In addition to the public website of EPIC-CVD (www.epiccvd.eu) which we plan to maintain and update with new findings for the foreseeable future, we will continue to support the EPIC-CVD ‘extranet’, which serves as a mechanism for investigators and collaborators within EPIC-CVD to propose analyses, upload statistical analysis plans, organise analytical efforts, share draft scientific manuscripts and log accepted papers. Additional online resources that have emerged as part of the EPIC-CVD project include the INFORM trial website (www.informstudy.org.uk) and our genetic CVD risk score tool.
e) Participant newsletters
To ensure that we continue to engage participants involved in research conducted as part of the EPIC-CVD project, we will provide feedback on our findings to participants via newsletters. This has already been successfully demonstrated during the INFORM trial (WP14), during which INFORM participants received regular newsletter updates by e-mail to keep them abreast of progress with the trial.
List of Websites:
http://www.epiccvd.eu/
1. Executive summary
EPIC-CVD was an ambitious project involving 28 partner organisations across 11 European countries with the overall aim to provide clinicians and policy-makers across Europe with an evidence-based menu of options for targeted and cost-effective CVD risk approaches tailored to the needs of Europe's diverse populations. To achieve this broad objective, EPIC-CVD intended to engage and cross-link 4 broad areas of scientific endeavour:
(1) population health science (epidemiology, public health, biostatistics)
(2) laboratory science (biochemistry, molecular genetics, nutrition)
(3) translational science (medical anthropology, primary care and cardiovascular medicine, health economics)
(4) implementation science (risk communication tools, comparative analysis, policy development).
Our research was divided into 4 main Research Lines of work, namely:
RL1 - Creation of a framework to develop new risk assessment approaches
RL2 - Analysis of factors related to nutrition in a large prospective study
RL3 - Analysis of intermediate biomarkers and genetic make-up in incident CVD
RL4 - Testing of clinical and public health utility
Major successes of the project to date include:
- creation of the world’s largest harmonised case-cohort study of incident coronary heart disease and stroke with >25,000 incident outcomes;
- detailed phenotyping of participants including >1million genetic variants, 300 circulating biomarkers (including metabolomics, nutritional biomarkers and clinical biomarkers), extensive dietary and lifestyle information, permitting aetiological and methodological analyses in addition to studies of risk prediction;
- development of novel risk scores that replace invasive measures with non-invasive modifiable lifestyle factors without greatly sacrificing predictive accuracy;
- development of novel risk scores that increase predictive accuracy through inclusion of non-traditional biomarkers/genetic variants;
- conducting the largest trial of behaviour change in relation to provision of CVD risk estimates (phenotypic and genotypic) with objective outcome measures (eg, physical activity using accelerometry);
- comparison of the regulatory and societal issues associated with CVD risk screening across European countries, particularly with respect to inclusion of biomarkers and genetic information;
- establishment of a framework for capturing information on the costs and benefits of CVD interventions (eg, statins, lifestyle modifications) and utilising this information to assess cost-effectiveness of novel risk scores;
- creation of novel methods to analyse multi-centre case-cohort studies, particularly with respect to risk prediction.
By March 2016, work in EPIC-CVD has led to 15 peer-reviewed scientific publications, with a further 6 already submitted, and another ~40 to be submitted in 2016/2017.
Project Context and Objectives:
2. Summary of project context and main objectives
Coronary heart disease (CHD) and stroke are Europe's first and second leading causes of death, respectively, as well as major causes of disability. Cardiovascular disease (CVD) is currently responsible for more than €170 billion annually in economic costs in the increasingly resource-constrained countries of the EU. Despite reductions in age-specific CVD death rates in many European countries in recent decades, the overall future CVD burden in the EU is predicted to increase as a result of the ageing population and increasing prevalence of obesity and diabetes. It has been estimated that tens of thousands of additional CVD outcomes could be prevented in the EU per year by better screening and management of risk. These potential gains are more likely to be realised if more compelling evidence and policies emerge to support future approaches to primary prevention of CVD. Furthermore, we believe that it is unlikely that a single approach will be optimum or acceptable across the diverse healthcare systems, budgets, cultures, and lifestyles of Europe.
The overall aim of the EPIC-CVD project was therefore to provide clinicians and policy-makers across Europe with an evidence-based menu of options for targeted and cost-effective CVD risk approaches tailored to the needs of Europe's diverse populations.
To achieve this overall aim we targeted 5 key areas:
1) Creating synergy between population-wide and personalised prevention
• EPIC-CVD aimed to create Europe’s first risk assessment approaches focusing on “modifiable” risk scores with behavioural risk factors and objective lifestyle biomarkers
Our hypothesis was that approaches based on modifiable scores should cost-effectively:
(1) enhance accuracy of risk prediction, resulting in improved identification and risk assessment
(2) maximise potential for behaviour change in patients and physicians through personalised lifestyle feedback, leading to better clinical decision making and clinical outcomes
(3) contribute to the prevention of other conditions influenced by lifestyle choices and of major public health importance in Europe, such as obesity, type 2 diabetes and certain cancers, again leading to better clinical outcomes
(4) achieve synergy with population-wide approaches.
Modifiable risk scores should genuinely be complementary to public health activities that aim to reduce the overall population risk of cardiovascular disease by promoting a healthy lifestyle (diet, exercise, avoidance of smoking).
2) Making existing CVD risk tools more relevant to European populations
• EPIC-CVD aimed to provide the first systematic comparison of existing major risk scores in a common set of participants across European populations
By exploiting data and samples in the prospective 10-country EPIC study of 520,000 people in whom extensive information on lifestyle, socioeconomic factors, blood samples, and other characteristics had already been collected in a standardised manner, we aimed to directly address the original EC call’s mandate: “The impact of biomarkers on cardiovascular disease risk prediction will need to be assessed across different European populations as they have different lifestyles (eg, dietary patterns) and varying biomarker levels”.
By achieving this aim, we hoped to:
(1) improve the performance of existing risk scores in Europe
(2) lead to stratified risk scores for subpopulations that “contribute to the development of personalised predictive medicine”
(3) enhance the adoption of these methods by patients and healthcare professionals across Europe because of their greater actual (and perceived) relevance to local circumstances.
Thus, EPIC-CVD aimed to enhance our ability to identify people at high CVD risk and manage them appropriately, leading to improvements in identification, risk assessment, clinical decision making and clinical outcomes.
3) Developing clinically useful new CVD risk scores and tools
• EPIC-CVD aimed to create the first risk score that considered biomarkers of 'nature' (eg, genetic variation), 'nurture' (ie, lifestyle biomarkers), and intermediate biological pathways (eg, lipids, inflammation etc), and to be the first study capable of incorporating the interplay of nature and nurture for CVD risk prediction across Europe.
We hypothesised that detailed risk scores based on parsimonious sets of these markers would be likely to confer a much higher degree of predictive accuracy than existing scores and be especially amenable to tailoring to subpopulations, such as younger people who may be contemplating lifelong treatment, resulting in improvements in identification, risk assessment, clinical decision making and clinical outcomes. In partnership with research intensive SMEs (eg, VITAS in Oslo, an international leader in biomarker assays), we aimed to develop innovative methods for high-throughput blood sample collection and biomarker assays, contributing to research and innovation and yielding approaches and potential products of interest and potential benefit to SME(s).
4) Designing acceptable and cost-effective approaches for a diverse Europe
• EPIC-CVD aimed to provide the first systematic evaluation of clinical and public health utility of modifiable and detailed risk scores across Europe
We aimed to exploit our expertise in devising guidelines for the evaluation of risk scores to ensure that our development and validation of new risk scores would be robust. By capitalising on our existing multidisciplinary framework, we aimed to conduct qualitative fieldwork and a randomised trial of the impact of different risk scores on "patient-centred" outcomes in 1000 people across Europe. We will build on our previous experience with sequential or "stepwise" screening strategies in which periodic invitations are made to individuals for a preventive visit using already recorded determinants (“pre-stratification”). These approaches should enhance cost effectiveness and organisational efficiency of targeting of preventive resources and reduce anxiety in the population by focusing efforts on subsets of the population most likely to benefit from detailed screening. We intended to tailor these approaches to the diverse parts of Europe. We aimed to capitalise on health economics expertise we previously used to shape national screening policies for vascular conditions in the UK and to conduct detailed health economic modelling that considered alternative approaches taking into account long-term or even lifetime risk to determine the overall costs and benefits of using various different risk markers, risk scores, and screening strategies for CVD, leading to the emergence of operational approaches that show cost effectiveness, safety, validity and incremental benefit over existing prediction methods. EPIC-CVD hoped to benefit from its partnership with RAND-Europe, an independent not-for-profit policy research organisation with specific expertise in comparisons of European health systems, to develop state-of-the-art policy for CVD risk assessment tailored to Europe’s diverse needs.
5) Addressing broader societal concerns
• EPIC-CVD will provide consideration of the broader societal implications of modifiable and detailed CVD risk scores across Europe
In partnership with the PHG Foundation, a genetics policy think-tank, EPIC-CVD aimed to explore potential ethical, legal, and social issues related to targeted CVD prevention testing methods. This work was building on a major work-package of research the PHG Foundation previously carried out as part of the FP7-funded COGS (Collaborative Oncological Gene-environment Study) project. In EPIC-CVD, we intended to apply similar considerations and to extend research in new directions to address considerations distinctive to CVD screening.
Across these areas, EPIC-CVD listed the following objectives:
● To develop and evaluate modifiable risk scores that promote behaviour change and personalised lifestyle feedback
● To compare and calibrate major existing risk scores across diverse European populations
● To develop and evaluate detailed risk scores that maximise predictive accuracy and target subpopulations
● To develop and evaluate user-friendly clinical tools to support risk assessment
● To develop and evaluate cost-effective strategies for the organisation of screening
● To compare the costs and benefits of different risk assessment and screening approaches
● To develop evidence-based policies tailored to diverse European settings
● To disseminate these tools, strategies, and policies widely to relevant audiences across Europe to maximise clinical and public health impact.
To achieve these objectives, EPIC-CVD intended to engage and cross-link 4 broad areas of scientific endeavour:
(1) population health science (epidemiology, public health, biostatistics)
(2) laboratory science (biochemistry, molecular genetics, nutrition)
(3) translational science (medical anthropology, primary care and cardiovascular medicine, health economics)
(4) implementation science (risk communication tools, comparative analysis, policy development).
The EPIC-CVD consortium consists of 28 partner organisations from 11 European countries, including 2 SMEs, a genetics policy think-tank, and an independent not-for-profit policy research organisation.
Project Results:
3. Main findings
The EPIC-CVD project is divided into 4 main interlinking Research Lines:
RL1 - Creation of a framework to develop new risk assessment approaches
RL2 - Analysis of factors related to nutrition in a large prospective study
RL3 - Analysis of intermediate biomarkers and genetic make-up in incident CVD
RL4 - Testing of clinical and public health utility
The 4 Research Lines were further divided into 16 scientific Work Packages.
a) Results from RL1 - Creation of a framework to develop new risk assessment approaches
Research Line 1 had the following objectives:
• To identify and verify diagnoses of incident stroke within the EPIC cohort
• To retrieve and prepare biological samples from stroke cases for assays
• To assess the impact of 22 routinely measured clinical biomarkers on the risk of incident CVD
• To coordinate the activities for designing the analysis of data collected in the project and to supervise the preparation of all analytical plans.
The EPIC-CVD project is underpinned by the European Prospective Investigation into Cancer and Nutrition (EPIC) study. EPIC involves 366,521 women and 153,457 men, mostly aged 35–70 years, recruited by 23 centres in 10 European countries (Denmark, France, Germany, Greece, Italy, the Netherlands, Norway, Spain, Sweden, and the United Kingdom) between 1991 and 1999. Within this large pan-European prospective cohort, a nested case-cohort study of type 2 diabetes (EPIC-InterAct) was created by identification of >12,000 incident type 2 diabetes cases and selection of a random subcohort of >15,000 referents. EPIC-InterAct was funded by a EC Framework Progamme 6 award. Taking advantage of the use of the case-cohort design and the selection of the random subcohort for comparison with disease cases, more than 15,000 incident coronary heart disease cases were subsequently identified through funding from the British Heart Foundation and the UK Medical Research Council (“EPIC-CVD”). We aimed to extend EPIC-CVD to include incident stroke cases and then to conduct assays in the CHD cases and stroke cases, creating a framework to allow study of cardiovascular disease.
WP1 - Stroke case verification
The overall objective of this Work Package was to identify and verify diagnoses of incident stroke within the EPIC cohort.
A group at UCAM, including stroke physicians and epidemiologists, devised a protocol for the standardised verification of stroke cases. The protocol outlined the methods of stroke case verification from medical record review and was circulated to all EPIC centres. All 23 EPIC administrative centres, which were all partner organisations in EPIC-CVD, were then asked to identify potential stroke cases that had occurred after study baseline, using whichever methods they had available. Stroke cases were identified through a range of means, including morbidity and mortality registers, participant surveys, and active follow-up using medical records. Identification numbers for all >10,000 suspected incident stroke cases were then provided to the EPIC-CVD Coordination Centre, along with additional information including family history of cardiovascular disease and medication usage. Other details of baseline measurements (eg, anthropometry and blood pressure), lifestyle information (eg, smoking behaviours and alcohol consumption), and dietary information were then retrieved for these participants from the central EPIC database at IARC (Lyon, France).
EPIC centres then retrieved relevant information to verify the stroke diagnoses, typically by retrieving further information on the suspected stroke cases or by conducting substudies to verify the validity of the stroke case ascertainment approach. Information was then curated, collated and harmonised across the 23 centres at the EPIC-CVD Coordinating Centre to develop a common framework for definition of stroke cases. After exclusion of prevalent cases (since our study of risk prediction was designed to assess first-ever events rather than recurrent disease), case events that fell outside of our stroke definition (eg, transient ischaemic attack), and events that turned out not to be valid or only based on self-report, we successfully identified and validated more than 8000 incident cerebrovascular events, creating one of the largest studies of incident stroke cases to date. Stroke cases were subclassified, according to ICD codes into ischaemic stroke events, haemorrhagic stroke events, subarachnoid haemorrhage events, other stroke events or unclassified stroke events. We also described the certainty of the validated stroke case events according to the sources of information used and depth of information provided.
WP2 - Retrieval and preparation of samples
The overall objective of this Work Pacakge was to retrieve and prepare biological samples from stroke cases for assays.
Once lists of suspected stroke cases had been identified by EPIC centres (see WP1), a complete list of stroke cases with stored blood samples was sent to IARC and local biorepositories in Scandinavia by the EPIC-CVD Coordinating Centre for sample retrieval. The retrieval process at IARC consisted of retrieving 0.5ml straws of red blood cells, plasma (x2), serum, and buffy coat (x2), before extracting DNA from buffy coat samples using a high-throughput automated DNA extraction procedure. The processes used were the same as those used previously to retrieve samples for incident type 2 diabetes, incident CHD cases and subcohort members. Once samples from all participants had been successfully retrieved and had DNA extracted for them, sample straws and DNA aliquots were delivered to the MRC Epidemiology Unit in Cambridge. The overall yields of DNA were high and >90% of samples had DNA above the concentration required for downstream assays (50ng/ul).
Similar procedures were followed in Denmark, Malmo and Umea, where samples from Scandinavian participants are stored. From these centres, aliquots in tubes/plates were provided rather than straws, and were shipped to Cambridge or directly to the assay laboratory.
Once samples had been received in Cambridge, the laboratory team at the MRC-Epidemiology Unit then sub-aliquoted and arrayed the non-DNA samples for the assays of clinical biomarkers, fatty acids, and vitamins/carotenoids. For the vitamin C assays, this involved addition of a preservative to the sample once it had thawed (see WP5). Regular shipments of samples for biomarker assays were made from the MRC-Epidemiology Unit to the MRC Human Nutrition Research Unit (MRC-HNR) for fatty acids assays, to VITAS for vitamins/carotenoids assays and to Stiching Ingehousz (SHL, Netherlands) for clinical biomarker assays.
For DNA samples received at the MRC Epidemiology Unit in Cambridge, the concentration was initially tested using a Picogreen method. Once quantified, samples were then normalised to a standard concentration of 50ng/ul for assay, and were then stored in a robotic freezer. Any samples received at a concentration below 50ng/ul were dried down and resuspended to the appropriate concentration. Samples were further tested using a Taqman assay of variants on the sex chromosomes to check that they were genetically the same sex as expected according to the EPIC database. In total, >97% of samples passed DNA QC.
DNA samples were shipped to the labs at Cambridge Genomics Services for variant microarray assays and were subaliquoted for the in-house telomere length assay at the MRC Epidemiology Unit.
WP3 - 22 clinical chemistry biomarkers
The overall objective of this Work Package was to assess the impact of 22 routinely measured clinical biomarkers on the risk of incident CVD.
The 22 clinical chemistry biomarkers investigated included lipids (total cholesterol, HDL-cholesterol, triglycerides), lipoproteins (apolipoprotein AI, apolipoprotein B, lipoprotein[a]), glycaemic markers (HbA1c, glucose), inflammatory markers (C-reactive protein, albumin), iron markers (serum iron, ferritin, transferrin), electrolytes (calcium, magnesium), liver enzymes (ALT, AST, ALP, GGT, bilirubin), and renal markers (creatinine, uric acid). Following an options appraisal to assess the optimal laboratory to conduct these assays, it was decided to use SHL, the laboratory in the Netherlands that has previously conducted the same assays on the 43,000 serum samples comprising the CHD case samples, the type 2 diabetes samples and the subcohort samples.
In total, more than 8000 validated incident stroke case samples were measured and passed quality control steps. Results appeared to be consistent over time with no strong evidence of drift. Distributions of biomarkers were consistent with expectations based on previous studies. For example, triglycerides, C-reactive protein and lipoprotein(a) were all heavily skewed, whilst total cholesterol was normally distributed. Correlations between biomarkers were also of the expected magnitudes and directions. For example, HDL-cholesterol and apolipoprotein AI were strongly correlated (r=0.83) ALT and AST were strongly correlated (r=0.71) and glucose and HbA1c were strongly correlated (r=0.70). There were also inverse correlations, such as between HDL-cholesterol and triglyceride levels (r=-0.50).
An analysis plan was developed for the analysis of the data from the assays of 22 clinical biomarkers with incident CVD. This plan covered methods to assess cross-sectional correlates, assess the shape of dose-response relationships between the 22 clinical biomarkers and CVD outcomes, consistently adjust for potential confounding factors, assess the relevance of pre-specified subgroups, and finally to assess the incremental predictive value of the 22 clinical biomarkers beyond conventional CVD risk factors.
As expected from previous literature and large-scale meta-analyses, the strongest positive associations with disease risk were for pro-atherogenic lipids and lipoproteins, with glycaemic markers also showing positive associations. The strongest inverse associations were with HDL-cholesterol and apolipoprotein AI. C-reactive protein was also positively associated, although its ability to add to existing CVD risk scores has already been well documented. Other notable associations were inverse associations with bilirubin, iron and albumin, and positive associations with liver enzymes (GGT, ALT, ALP), ferritin and uric acid. These associations mostly remained significant after adjustment for traditional cardiovascular risk factors.
Although total cholesterol and HDL-cholesterol are already routinely included in many existing CVD risk scores, there is potential that other emerging biomarkers could usefully add to risk prediction. Whilst some, such as C-reactive protein and glycaemic markers, have been well studied in this context, other biomarkers have not. On the basis of the associations seen with cardiovascular events, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that it would be beneficial to investigate the predictive ability of other associated biomarkers (i.e. not those already in risk scores or previously well covered) given that their associations with disease risk appear to be largely independent of conventional risk factors.
WP4 - Analysis management
The overall objective of this Work Package was to coordinate the activities for designing the analysis of data collected in the project and to supervise the preparation of all analytical plans.
Given the number and range of different analyses to be performed as part of the EPIC-CVD project, and the challenges of analysing a multi-centre case-cohort study, we established a Central Analytical Team, led by Professor Simon Thompson, Dr Angela Wood and Dr Michael Sweeting at the University of Cambridge. This team worked closely with epidemiologists and data managers at the EPIC-CVD Coordinating Centre to produce documents describing recommended approaches to understanding and interpreting the EPIC-CVD data, conducting aetiological analyses in the EPIC-CVD dataset, and conducting risk prediction analyses in the EPIC-CVD dataset. Since not all methods were already clearly established, a number of pieces of methodological work were conducted, overseen by Dr Angela Wood. For example, methods to estimate risk prediction metrics in multi-centre case-cohort studies were developed and published (Sanderson et al., BMC Med Res Methods, 2013). Other publications already produced cover methods for stratified case-cohort studies (Jones et al., J Clin Epi, 2015), recommendations for reporting results from case-cohort studies (Sharp et al., PLoS One, 2014) and methods to assess risk prediction models using individual participant data from multiple studies (Pennells et al., Am J Epid, 2014). A number of other methodological papers will emerge in 2016 and 2017, such as a recently submitted paper looking at analysing genetic data in case-cohort studies (Staley et al., submitted).
The Central Analyatical Team was also responsible for ensuring that appropriate statistical analysis plans were produced prior to analysis for all risk prediction analyses, as well as the aetiological and methodological analyses enabled by the creation of the EPIC-CVD dataset. Hence, analytical plans were produced for the analysis of data from the assays of the 22 clinical biomarkers, the fatty acids panel, the vitamins and carotenoids, the customised SNP array and the telomere length.
b)_Results from RL2 - Analysis of factors related to nutrition in a large prospective study
Research Line 2 had the following objectives:
• to further optimise existing high-throughput assays for nutritional biomarkers
• to assess the potential for inclusion of vitamin C and 6 carotenoids in CVD risk scores.
• to assess the potential for inclusion of vitamin D and its epimers in CVD risk scores
• to assess the potential for inclusion of of 37 fatty acids in CVD risk scores.
WP5 - Optimising high-throughput assays
The overall objective of this Work Package was to further optimise existing high-throughput assays for nutritional biomarkers.
Dr Thomas Gundersen, a world authority on nutritional biomarkers, at VITAS conducted several optimisation procedures for assay of vitamin C, vitamin D (including its epimer) and a panel of carotenoids.
a) Carotenoids
The gold standard method for assay of carotenoids is a chromatographic approach known as HPLC-UV. This approach has the ability to separate out the six carotenoids (alpha-carotene, beta-carotene, beta-cryptoxanthin, lutein, lycopene and zeaxanthin). However, doing so required a lengthy run-time (28 minutes at the start of the project) which would not allow the assay of the requisite number of samples in the project timescale. VITAS therefore tested a number of column chemistries and particle technologies to
speed up the cycle time without sacrificing the resolution of the carotenoid isomers. On each column type numerous variations in conditions were investigated. These included mobile phase composition, temperature, flow, column length, particle size, pH and various modifiers. The C30 stationary phase was then looked at again to attempt to optimize condition on this chemistry. By using an YMC C30 column, optimizing the above mentioned parameters VITAS were able to reduce the cycle time of the method from 28 minutes to 12 minutes, a 64% reduction, still keeping the required resolution between the critical pairs of carotenoids. As plasma sample was also limited in the EPIC-CVD study, VITAS also optimized several parameters that might influence the signal strength and the signal to noise ratio. Altogether VITAS achieved a 6.5 time increase in sensitivity allowing analysis from as little as 30µL of sample.
b) Vitamin C
Vitamin C, also known as ascorbic acid, is very unstable in plasma and it is expected that significant degradation would have occurred in the EPIC-CVD samples, which are typically more than 10 years old. However, the EPIC samples have been stored in liquid nitrogen and the stability of vitamin C in plasma stored for this length of time under these conditions had not been previously examined. To investigate if detectable levels of vitamin C remain in the sample and to what degree these levels correlate with levels measured fresh at baseline, a subset of 96 samples from EPIC-Norfolk were analysed. For these samples baseline data for vitamin C were available. A protocol for this pilot investigation was designed and written. An SOP for the transfer of serum/plasma from the EPIC plasma straws to vials and subsequently to microtiter plates, addition of stabilizers and shipment on dry ice from Cambridge to Oslo was also written. The results from the pilot showed a surprisingly strong correlation of r=0.94. This indicates that vitamin C has been very well preserved when stored in liquid nitrogen as both labs providing data for this comparison obtained very similar results for all of these 96 samples.
c) Vitamin D and its epimer
It has previously been shown that 25-OH-Vitamin D3 also exists in its epimerised form in human plasma and serum. Separating isoforms of molecules is a challenging task and often requires prolonged analysis time. The presence of this isoform has only been studied in smaller sample cohorts. To include this isoform in the large EPIC-CVD project required development of a method with the isoforms resolved, but still retaining the high throughput properties of the method due to the large numbers of samples to be assayed. By applying modern HPLC columns with sun 3µm core shell particles and pentafluorophenylpropyl chemistry, VITAS were able to separate these and keep the cycle time of the method within five minutes.
WP6 - Vitamin C and carotenoids
The overall objective was to assess the potential for inclusion of vitamin C and 6 carotenoids in CVD risk scores. To do this required assay of this panel of 7 nutritional biomarkers in the EPIC-CVD dataset, including the incident CHD cases, incident stroke cases and subcohort members.
Regular shipments of samples were made to Dr Gundersen’s laboratory at VITAS for vitamin C assays to be performed. Dr Gundersen’s team regularly returned data, including quality control (QC) data, for the assays as they passed through the pipeline. In total, ~45,000 samples were assayed on the vitamin C and carotenoids platform, including ~12,000 incident CHD cases, ~8,000 incident stroke cases, ~15,000 subcohort members (as well as ~10,000 incident type 2 diabetes cases not included in EPIC-CVD). We additionally completed the measurement of around 2,100 samples from the EPIC-Norfolk study for a repeat measurements validation study. This included samples from ~700 participants with measurements available at each of three time points of the three health-check visits in EPIC-Norfolk study (http://www.srl.cam.ac.uk/epic/). Final analyses of these data are in progress, but initial impressions suggest that the measurements are highly repeatable, but have a reasonable degree of within-person change over time.
The consistency of results over the long time period of the assay was assessed to ensure that results were not susceptible to drift or batch effects. QC standard samples were plotted and continuously monitored throughout the project, with samples and plates showing strong deviations excluded. Overall consistency of measurements over time was high. Vitamin C and carotenoid levels were typically higher in women compared to men and varied by country. For example, lutein levels were considerably higher in Greece, Italy and France, than in central or northern European countries.
Plasma vitamin C levels showed a linear inverse relationship with risk of CHD, which attenuated somewhat after adjustment for conventional cardiovascular risk factors (age, energy intake, smoking, education, physical activity, alcohol, history of diabetes, hypertension, BMI) but remained statistically significant. Results varied for carotenoids but typically showed inverse associations, albeit generally weaker than for vitamin C. As these associations appear to be largely independent of risk factors in existing CVD risk scores, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that there is potential for vitamin C/carotenoid levels to improve risk prediction if added to risk scores.
WP7 - Vitamin D and its epimer
The overall objective was to assess the potential for inclusion of 25(OH)D3 and its epimer, and 25(OH)D2 in CVD risk scores. To do this required assay of this panel of these 3 biomarkers in the EPIC-CVD dataset, including the incident CHD cases, incident stroke cases and subcohort members.
Regular shipments of samples were made to Dr Gundersen’s laboratory at VITAS for assays of vitamin D and its epimers to be performed. Dr Gundersen’s team regularly returned data, including QC data, for the assays they completed as they passed through the pipeline.
In total, ~45,000 samples were assayed on the vitamin D platform, including ~12,000 incident CHD cases, ~8,000 incident stroke cases, ~15,000 subcohort members (as well as ~10,000 incident type 2 diabetes cases not included in EPIC-CVD). We additionally completed the measurement of around 2,100 samples from the EPIC-Norfolk study for a repeat measurements validation study. This included samples from ~700 participants with measurements available at each of three time points of the three health-check visits in EPIC-Norfolk study (http://www.srl.cam.ac.uk/epic/). Final analyses of these data are in progress, but initial impressions suggest that the measurements are highly repeatable, but have a reasonable degree of within-person change over time.
The consistency of results over the long time period of the assay was assessed to ensure that results were not susceptible to drift or batch effects. QC standard samples were plotted and continuously monitored throughout the project, with samples and plates showing strong deviations excluded. Overall consistency of measurements over time was high. Vitamin D levels in subcohort members varied by country and showed a strong expected correlation with the time of year in which the blood sample was taken, reflecting recent sunlight exposure at baseline in participants.
Similarly to previous smaller studies, a non-linear inverse association with vitamin D levels was seen with cardiovascular events, suggesting that only participants with vitamin D insufficiency are at increased cardiovascular risk. The strength of the association attenuated only slightly upon adjustment for conventional cardiovascular risk factors (i.e. age, sex, smoking status, systolic blood pressure, history of diabetes, BMI, total cholesterol, HDL cholesterol). As the association appeared to be largely independent of risk factors in existing CVD risk scores, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that there is potential for vitamin D levels to improve risk prediction if added to risk scores.
WP8 –Fatty acids
The overall objective was to assess the potential for inclusion of a panel of 37 plasma phospholipid fatty acids in CVD risk scores. To do this required assay of this panel of the 37 fatty acids in the incident stroke cases from EPIC-CVD, since assays in the incident CHD cases had been previously funded (by a European Research Council award), and assays in the subcohort members had been previously funded by the EPIC-InterAct Framework Programme 6 award.
An options appraisal was undertaken to select a laboratory to conduct the fatty acids assays and MRC Human Nutrition Research (MRC HNR) was selected as they have world leading expertise in the field with Dr Jules Griffin and Dr Albert Koulman. MRC HNR had already performed assays for T2D, CHD cases and common controls, so there was a strong scientific argument for undertaking the assays at the same laboratory, to reduce batch effects.
The measurement of plasma phospholipid fatty acids was successfully conducted by the MRC-HNR lab, using gas chromatography. Specifically, the method employed robotics to maintain a high throughput and generated results in relative concentrations of fatty acids in mol% units. Individual fatty acids within the major classes of fatty acids (saturated fatty acids, omega-3 and omega-6 polyunsaturated fatty acids, monounsaturated fatty acids and trans- fatty acids) were measured.
Quality control (QC) samples were included into every batch of samples analysed across the projects [InterAct (subcohort participants) and EPIC-CVD cases (CHD, stroke)]. This QC material was used to monitor assay precision and to assess if any drift occurred across the study or any significant differences were observed between the measurements of the study samples and the QC samples with known concentrations of fatty acids. We observed high QC stability of the fatty acids across the duration of the fatty acid measurements. In addition, no QC differences between measurement of InterAct (subcohort) and EPIC-CVD cases (CHD, Stroke) were observed across all the fatty acids. This is an essential prerequisite to be able to use the data on both cases and subcohort with confidence.
In addition to the samples measured in EPIC-CVD, we additionally completed the measurement of around 2,100 samples from the EPIC-Norfolk study for a repeat measurements validation study. This included samples from ~700 participants with measurements available at each of three time points of the three health-check visits in EPIC-Norfolk study (http://www.srl.cam.ac.uk/epic/). The analyses of repeat measures data are in progress and will help with understanding the determinants of change in fatty acid concentrations over time, and also with repeatability of measurements.
Overall, 92.5% of the EPIC-CVD participants (i.e. subcohort members, incident CHD cases and incident stroke cases) had measurements for all 37 fatty acids (FAs). FAs with a relative mean<0.05% (based on the subcohort) were excluded (C8:0, C10:0, C11:0, C12:0, C13:0, C21:0, C22:2, C14:1, C15:1 and C22:1-n9) due to their low levels leaving 27 FAs for analysis. 94 participants were excluded due to having outlying measurements. Levels of FAs varied across European countries – for example, trans fatty acid levels, although low overall, were substantially higher in the UK and the Netherlands compared to Mediterranean countries.
As expected, fatty acid levels were correlated within subgroups and clustered closely in hierarchical clustering models. For example, n-3 PUFAs showed strong positive correlations. Fatty acids were only weakly correlated with other cardiovascular risk factors, suggesting that any association with cardiovascular events is likely to be largely independent.
Associations of each FA with incident cardiovascular events were assessed using Prentice-weighted Cox proportional hazards models with robust standard errors to account for the case-cohort design. Shapes of association were assessed by classifying participants into quintiles of FA levels. Analyses were run within country and then combined using random-effects meta-analysis. Results were largely consistent with previous literature. For example, even chain saturated fatty acids were positively associated with CHD risk, whilst odd chain saturated fatty acids were inversely associated. These associations persisted after adjustment for conventional cardiovascular risk factors (i.e. age, sex, smoking, systolic blood pressure, BMI, history of diabetes, total cholesterol and HDL-cholesterol) suggesting that the associations with risk are independent. In conclusion, specific plasma phospholipid fatty acids are associated with cardiovascular events, independently from traditional CVD risk factors, hence the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that the inclusion of some plasma phospholipid fatty acids in CVD risk scores may provide predictive benefit.
c)_Results from RL3 - Analysis of intermediate biomarkers and genetic make-up in incident CVD
Research Line 3 had the following objectives:
• to assess the potential for inclusion of 225 metabolites in CVD risk scores
• to assess the potential for inclusion of common SNPs in CVD risk scores
• to assess the potential for inclusion of telomere length in CVD risk scores
WP9 –225 serum metabolites
The overall objective was to assess the potential for inclusion of a panel of 225 serum metabolites in CVD risk scores. To achieve this, an NMR metabolomics pipeline providing molar serum concentrations of 225 analytes (including lipids, lipoproteins, and low molecular weight metabolites) was implemented at an SME in Finland called “Brainshake”, a world-leader in developing and applying this technology. A pilot study of 930 samples showed reliable measurements could be obtained and that levels of analytes were mostly very similar to other population studies.
Samples were prepared and regular shipments of samples are being sent to Professor Ala-Korpela’s laboratory at Brainshake for metabolomics assays to be performed. Professor Ala-Korpela’s team regularly returned data, including QC data, for the assays they completed as they passed through the pipeline. The assay was successfully completed on ~46,000 samples.
Distributions of analytes looked broadly as expected. For example, total cholesterol was approximately normally distributed, triglycerides were positively skewed, whilst levels of XXL-VLDL-c were positively skewed with some samples having zero levels. Time-ordered measurements were plotted to assess drift over the timecourse of the project and coefficients of variation (CVs) were used to assess reliability of the assays. There was no evidence of drift in any of the analytes and CVs were below 10% for almost all analytes. For example, the CV for remant cholesterol was 5.9%, the CV for LDL-cholesterol was 3.8%, and the CV for alanine was 2.8%. Expected relationships were seen between lipoprotein fractions and lipid species, such as the strong relationship between triglyceride particles and lower density lipoproteins.
To assess the potential for inclusion in CVD risk scores, we investigated the correlation between NMR metabolites and cardiovascular risk, as predicted by a standard Framingham risk score (i.e. a model including age, sex, smoking status, history of diabetes, total cholesterol and HDL-cholesterol). Many metabolites were positively correlated with CVD risk, albeit weakly. For example, a range of types of cholesterol were all similarly positively correlated, suggesting that that, since they are themselves strongly positively correlated, they may not individually add to existing CVD risk scores. Other notable findings were: differing directions of association between individual metabolites and their ratios, and weak or null associations with low molecular weight metabolites.
In conclusion, we reliably measured 225 metabolite species in >45,000 serum samples. There were strong correlations between metabolites. Many metabolites, particularly those related to lipids and lipoproteins, were correlated with cardiovascular risk, usually in a positive direction. Hence the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that metabolites could usefully add to existing CVD risk scores, although only selected individual metabolites (or metabolite ratios) will be needed to capture the relevant predictive information.
WP10 – Common SNPs
The main objective was to assess the potential for inclusion of common SNPs in CVD risk scores. Cambridge Genomic Services (CGS, part of UCAM) was selected as the laboratory to perform the genetic assays using the 215,000 SNP CardioMetabochip+ array in the coronary disease cases and subcohort members due to their proven track record in providing high quality data from high-throughput projects using Illumina platforms, and their low costs. CGS was also used to perform the genetic assays using the 550,000 SNP HumanCoreExome array in the coronary and stroke cases, and subcohort members.
Genotyping of the 215,000 SNPs on the CardioMetabochip+ was been completed and the data delivered. The 550,000 SNP HumanCoreExome array was also assayed in all coronary disease cases, stroke cases and subcohort samples. Stringent quality control of both arrays was performed to ensure that only high-quality variants were included in analyses.
An initial report describing the possible genetic content for CVD risk prediction scores was produced and provided recommendations to leaders of other WPs in RL4 tasked with risk prediction analyses within EPIC-CVD. In summary, four methods of selecting common genetic variants for inclusion in risk prediction models were assessed. The method using variants known to be associated with CHD risk has the most immediate application and can be tested most simply. The unbiased selection of variants from across the genome shows great methodological promise but potentially at the cost of clinical translation. Using data from the CardioMetabochip, these methods have since been tested in relation to improvements in CVD risk prediction algorithms.
The ~50 known common variants associated with risk of coronary heart disease nearly all had associations in the same direction as the previous literature. Participants who had incident CHD events had slightly higher genetic risk scores than those who didn’t, regardless of the method used to weight the variants in the score. Participants with higher genetic risk scores had a higher risk of incident CHD. For example, there was a hazard ratio of 1.8 comparing participants in the top quintile of genetic risk score with those in the bottom quintile. Including the genetic risk score in a score based on conventional risk factors (age, sex, smoking, diabetes, blood pressure, total cholesterol and HDL-cholesterol) increased the C-index slightly, but did not statistically significantly improve reclassification. The change in C-index was smaller than that seen for other clinical biomarkers e.g. C-reactive protein. Further analyses will investigate inclusion of a broader panel of genetic variants.
WP11 – Relative telomere length
The main objective of this Work Package was to assess the potential for inclusion of telomere length in CVD risk scores. To do this, we aimed to measure leukocyte relative telomere length in the incident stroke cases, incident CHD cases and subcohort members in EPIC-CVD (additional funding was identified to allow concomitant assay of the incident type 2 diabetes cases also).
An options appraisal to identify suitable laboratories capable of measuring leukocyte telomere length (LTL) was carried out. The MRC Epidemiology Unit (part of the University of Cambridge) in Cambridge was identified as the best laboratory to carry out the work and this was based on technical capability, experience of high-throughput DNA applications and capacity to analyse >40,000 samples in the necessary timeframe. Following a thorough appraisal of the methods currently available for telomere length assessment, it was decided to use a quantitative PCR method (qPCR) since this is the only assay that allowed the high throughput necessary to conduct assays on >40,000 samples in the required timeframe. Relative telomere lengths were measured using the ratio between the relative quantities of the telomeric TTAGGGn repeat (T) and a single copy region of the Albumin gene (S), using a previously described monochrome multiplex qPCR method.
The initial optimisation and validation work for the assay was carried at the MRC Epidemiology Unit (EPID) in Cambridge (UCAM). Matullo’s laboratory at HUGEF provided 30 DNA samples, extracted from peripheral blood mononuclear cells, where telomere length had been previously measured by Southern Blot (considered the gold standard method for telomere length measurement). A blind validation assessment was carried out by EPID and a very good agreement between the two methods was observed (correlation coefficient r2=0.69). Excellent inter-plate agreement was obtained (correlation coefficient r2=0.99) and the intra-sample coefficient of variation was also very low (4.5% on average). Finally, an inverse and significant correlation between telomere length and age was found, as expected (r2=-0.61 p<0.0001) suggesting that the qPCR method was providing valid estimates of RTL.
Each sample was further diluted from 50ng/µl to 2ng/µl and a total of 10ng of sample was pipetted into each reaction well. The qPCR reagents used were SYBR® GreenER™ (ThermoFisher Scientific) and primers (IDT, Leuven, Belgium) in a total reaction volume of 15 µl. A ViiA™ 7 Real-Time PCR System (ThermoFisher Scientific) was used to run and measure all of the qPCR reactions. The main experiment was run in four separate batches that had either used different reagent lots or machine calibration settings. Each 384 well plate contained 358 unique samples, four non-template control wells, five intra-plate duplicates, five inter-plate duplicates, two global control duplicate samples and two identical 5 point standard curves. An average standard curve for T and S was calculated using the standard parameters within the ViiA™ 7 software.
Each sample was measured independently three times within the same batch or six times for duplicate samples. For a sample to pass the Coefficient of Variation (CV) needed to be <10%. Individual results were excluded if the T or S value was similar to the non-template control, if any result was measured in a well that had a systematic problem or if there were <3 T/S results. Full plates were excluded and (where possible) repeated if there were <4 passed intra-plate duplicates or they were identified as an outlier by the T/S ratio of the global duplicate. A duplicate was failed if there were less than 5 out of 6 measurements available or if the CV between them was <10%.
To adjust for any variation in assay batch we carried out a bridging experiment repeating the measurement of samples in batch 4 from samples that were run in batch 1, 2 and 3. Only samples that passed the QC criteria above were used for bridging comparisons. The mean value of the RTL for each batch was calculated and compared to the mean value of the same samples in batch 4 to calculate an adjustment factor of RTL for batch 1, 2 and 3. The correlation between T/S ratios in each bridging comparison was statistically significant (p<0.00005). The calculated adjustment factor was applied to every result for batches 1-3.
After exclusion of outlying plates and the bridging normalisation step, 37,281 samples passed quality control and were considered for statistical analysis. Cross-sectional associations for males and females of RTL with baseline variables were tested: an inverse and significant correlation between RTL and age was found, (r=-0.16) while females showed longer RTL than males (0.009; 95% CI,0.005-0.013 p<0.001) consistent with previous literature and therefore validating the reliability of the measurements in this study.
No other cross-sectional associations were observed in males or females with other baseline variables tested (Total cholesterol, HDL-C, ln(triglycerides), apolipoprotein A1, apolipoprotein B, ln(Lipoprotein[a]), ln(C-reactive protein), albumin, creatinine, uric acid, glucose, HbA1c, ferritin, total energy intake, smoking status, history of diabetes, physical activity index).
Associations between genetic variants in several genes and RTL have been previously discovered (Pooley et al., 2013). In addition to investigating the interindividual variability in RTL measurements due to different genetic background, we can use RTL-associated variants as instrumental variables in order to verify if our RTL measurements go in the expected direction. Within this very large study, our results confirmed the association between RTL measured and 5 variants previously reported in the literature to be associated with RTL, further supporting the reliability of the measurements.
In conclusion, based on the independence of the telomere length measurement from other cardiovascular risk factors, the final report from this WP recommended to the Work Packages in RL4 tasked with assessing risk prediction that there is the potential for RTL to usefully add to a CVD risk score.
d)_Results from RL4 - Testing of clinical and public health utility
Research Line 4 had the following objectives:
• to identify lifestyle factors independently associated with CVD risk for use in modifiable risk scores
• to compare and calibrate existing risk scores to European populations, and recalibrate, develop and validate new risk scores
• to assess the impact on patients and physicians of new “modifiable” versus “detailed” risk scores
• to estimate the costs and benefits of different risk assessment strategies
• to identify and analyse the societal issues relating to European CVD risk screening.
WP12 - Development of modifiable risk scores
The main objective of this Work Package was to identify lifestyle factors independently associated with CVD risk for use in modifiable risk scores. In EPIC-CVD we aimed to optimize risk prediction for cardiovascular disease for the general population, by incorporating biomarkers and genetic markers on top of known risk factors for cardiovascular disease. Before developing new models, we first identified existing risk models through a systematic review of published scientific literature and validated their performance. The aim of the review was to provide an overview of all prognostic models that predict future risk of developing cardiovascular disease in the general population, assessing both modifiable and detailed risk scores.
Our systematic review of modifiable and detailed risk scores, conducted in conjunction with investigators from WP13, was conducted with 314 articles with risk scores focused on CVD outcomes being included in the literature review. Within these articles, 373 different prognostic models were developed, 519 external validations were performed and 278 incremental values were assessed. Models that are analysed in development, validation, and incremental value studies, and the populations used in these studies, show heterogeneity on many aspects, which need to be taken into account in the comparison of model performance. In the studies on validation of existing models on the EPIC-CVD dataset, and in assessing incremental value of biomarkers, genetic markers, and modifiable risk factors, this heterogeneity was taken into account. The systematic review has been completed and disseminated at conferences, with a scientific manuscript currently under revision at a major scientific journal. The results of the review have served as an input for decision regarding the design of the external validation study and the development of new detailed risk scores and risk scores with modifiable risk factors.
Having identified previous modifiable risk scores through the systematic review, and put in place a statistical analysis plan for validating existing risk scores and developing new modifiable risk scores (including a set of modifiable risk factors for evaluation), data generated within EPIC-CVD were analysed, initially using the CHD cases. The two main aims were:
1) To develop a model to predict cardiovascular events within 10 years that included only non-invasively measured lifestyle factors, including diet and physical activity
2) To assess the added predictive value of lifestyle factors when added to a set of well established cardiovascular risk factors, such as age, sex and smoking.
We performed backwards step-wise model selection based on a Wald t-test (p-value<0.05) to build a reduced prediction model including easy-to-obtain, non-invasive predictors. Criteria for consideration of lifestyle factors and non-invasively measured predictors (e.g. anthropometry, physical activity, diet) included strength and consistency of association with cardiovascular risk in the literature and availability of data within the EPIC-CVD dataset. To assess the added value of dietary information and physical activity over and above conventional predictors, we estimated the change in C-index, Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI).
Consistent with previous studies, known cardiovascular risk factors (e.g. age, smoking, blood pressure, pro-atherogenic lipids) were all higher in cases than non-cases. Consumption of fruit, vegetables and nuts was higher in non-cases, whilst red meat consumption was higher in cases. Our model including only non-invasive predictors, stratified by sex and country, included age, history of hypertension and diabetes, body mass index, waist-to-hip ratio, smoking, physical activity, alcohol consumption and three food groups: fruit, nuts (inverse association) and red meat (positive). Apparent discrimination of the combined model at 10 years was good with a C-index of 0.714 (95% confidence interval: 0.707-0.722). Heterogeneity across countries was large and statistically significant (I2=95%, p<0.0001) with country-specific C-index estimates ranging from 0.602 to 0.800.
Compared to a model including age, blood pressure, history of hypertension, total and HDL cholesterol, diabetes, and smoking, diet (alcohol, fruit, red meat) provided minimal additional predictive value: change in C-index of 0.002 and NRI of 0.72%. Further addition of physical activity did not improve discrimination or reclassification further (NRI=0.75%; change in C-index 0.002). Qualitatively similar results were found with multiple imputation.
In conclusion, there was little evidence of additional predictive value of non-invasive modifiable lifestyle factors, such as dietary factors or physical activity, when added to traditional cardiovascular risk scores in terms of discrimination, classification and calibration. However, the lifestyle-based risk score that we developed, including only non-invasively measured predictors, showed acceptable discrimination and therefore may have potential to be used for self-assessment of risk or inexpensive population screening. Use of such a score may have a range of benefits including greater acceptability to patients, cost and time saving for healthcare services (particularly those with limited resources), and greater potential to motivate behaviour change.
WP13 – Existing and detailed risk scores
This Work Package had two main objectives, namely (i) to compare and calibrate existing risk scores to European populations, and (ii) to recalibrate, develop and validate new risk scores.
Initially, we compared three varieties of the Framingham risk score (the most-often-validated risk score for CVD prediction): the original risk score, the original score but adapted to the incidence of the European population (recalibrated) and an adapted score, in which also coefficients were fit according to the European population (refit). By doing so, we assessed whether using an existing score would be sufficient, or whether adaptation would be necessary for risk prediction in the European population when assessing added value of modifiable risk factors and biomarkers).
We compared predicted probabilities by the original FRS (original) with a risk score in which the baseline hazard was recalibrated to the baseline hazard in our population (recalibrated) and an entirely new risk score using the same predictors (refitted) with the observed outcome, focusing on predictive performance. This was assessed using Harrell’s C-index, similar to the area under the receiver operator characteristic curve (AUC) adapted to survival analysis. This indicator ranges from 0.5 (no discrimination) to 1 (perfect discrimination). Validation was done per country and sex group and results were subsequently pooled by random-effects meta-analysis.
C indices with 95% confidence interval for the original, recalibrated and refit model were 0.698 (0.673-0.724) 0.693 (0.667-0.720) and 0.703 (0.676-0.731) respectively. These differences may seem small on this absolute scale, but as potential added value of biomarkers or modifiable factors may also be small, we consider the refitted model to be superior to the others. Using the existing Framingham risk score shows reasonable discrimination. Recalibration of the intercept did not improve the discriminative performance, in fact it even slightly decreased, probably due to heterogeneity among country and sex groups. The best discriminative performance was obtained by the model in which the predictors were actually refitted to the data. Therefore, in the analyses on modifiable risk scores and detailed risk scores, we used predictors from the Framingham model as a basis, but refit the coefficients of these variables.
For investigation of detailed risk scores, we developed four different models in a stepwise fashion. The first model used the easily obtainable predictors: age, smoking, history of diabetes, history of hypertension, and body mass index. Next we added systolic blood pressure. In model 3 total cholesterol and HDL were added to the model (as these factors are more often used in risk scores). Finally all other biomarkers were added to the model to investigate the maximum predictive performance that could be obtained by these biomarkers. Similarly, sex-specific models were developed for males and females separately. Performance of the four models in predicting 10-year risk was evaluated using discrimination (i.e. whether the model can distinguish between individuals who did and did not have the event) and calibration (i.e. to what extent the predicted probabilities agree with the observed risk) measures.
Consistent with previous studies, known cardiovascular risk factors (e.g. age, smoking, blood pressure, pro-atherogenic lipids) were all higher in cases than non-cases. Glycaemic markers, liver markers, CRP and lipoprotein(a) also had notably higher levels in cases. The model containing non-invasive easily measured variables only (model 1) achieved an apparent C statistic of 0.71 (95% confidence interval 0.70 - 0.72) which was similar after correction for optimism: 0.71 (0.68-0.73). However, heterogeneity across countries was very large (I2 = 93%, p < 0.0001) in particular among females. Adding systolic blood pressure (model 2) and cholesterol levels (model 3), variables frequently included in currently used risk scores, improved the optimism-corrected C index to 0.71 (0.69-0.74) and 0.73 (0.71-0.76) respectively. Additionally adding the other biomarkers (model4) further increased the optimism-corrected C statistic to 0.75 (0.72-0.77). The most important predictors in the full model which remained statistically significant were age, smoking, history of hypertension, systolic blood pressure, HDL, apolipoprotein B and lipoprotein(a), creatinine and uric acid, glucose and HbA1c, and alkaline phosphatase.
In conclusion, adding a set of routinely measured clinical biomarkers into a cardiovascular risk score including standard risk factors improved the potential of the risk score to identify persons at increased risk of future events. However, the major improvement above the non-invasive well-established risk factors was achieved by adding systolic blood pressure, total cholesterol and HDL-cholesterol to the model. Adding biomarkers related to different areas of biology (lipids and glucose metabolism, inflammation, renal function) improved discrimination further. However, the considerable heterogeneity between countries and between males and females should be taken into account and warrants further investigation. In resource-rich settings, inclusion of additional biomarkers beyond those routinely used (total cholesterol and HDL-cholesterol) may improve our ability to identify individuals at high risk of cardiovascular disease for targeting interventions.
WP14 - Assessment of impact of risk scores with differing content on behaviour
The main objective of this Work Package was to assess the impact of ‘new modifiable’ versus ‘detailed’ risk scores on patients and physicians. As a form of triangulation, and in order to collect data that is rich as possible, we conducted three inter-related studies in this working package:
a) Randomised controlled trial - Information and Risk Modification Trial (INFORM)
b) Systematic review - Impact of provision of cardiovascular disease risk estimates to healthcare professionals and patients
c) Qualitative study - The response to receiving phenotyping and genetic coronary heart disease risk scores and lifestyle advice
a) Randomised controlled trial - Information and Risk Modification Trial (INFORM)
The INFORM randomised trial was designed to contribute toward the understanding of the impact of provision of phenotypic and genetic coronary heart disease (CHD) risk information on health-related behaviour change and other important clinical outcomes. Comparison between randomised trial arms enabled estimation of: 1) the effect of providing web-based lifestyle advice; 2) the effect of providing a genetic risk score in addition to lifestyle advice and a phenotypic risk score; 3) the effect of providing risk score information and lifestyle advice and 4) the effect of providing risk score information in addition to lifestyle advice. The INFORM study will also provide information on potential moderators and mediators of associations between the intervention and health-related behaviour change and other clinical outcomes.
In a parallel-group, open randomised trial, we allocated 932 male and female blood donors with no previous history of CVD aged 40¬84 years in England to either no intervention (control group) or to one of three active intervention groups: i) lifestyle advice only; ii) lifestyle advice plus information on estimated 10-year CHD risk based on phenotypic characteristics; and iii) lifestyle advice plus information on estimated 10-year CHD risk based on phenotypic and genetic characteristics. The primary outcome was change in objectively measured physical activity. Secondary outcomes included: objectively measured dietary behaviours; cardiovascular risk factors; current medication and healthcare usage; perceived risk; cognitive evaluation of provision of CHD risk scores; and psychological outcomes. The follow-up assessment took place 12 weeks after randomisation. The protocol paper for INFORM study was published in September 2015 by BMC Public Health.
Between March 2015 and June 2015, electronic invitations approved by the local ethics committee were sent to participants who met the invitation criteria. Recruitment was completed earlier than planned on 25/06/2015 as we reached our target of 932 participants. We completed follow up of participants on 31/12/2015. 5124 participants of the INTERVAL study (study within which INFORM was nested) were available for invitation, we sent 2908 invitations (57%), 1152 participants responded (39.6%), 166 participants were not eligible/did not consent (14.4%), 986 participants consented (85.6%), 956 participants were randomised (97%).
Initial analyses have been completed and an abstract describing the preliminary results has been submitted to the SAPC (Society for Academic Primary Care) conference 2016. These results suggest that the provision of risk information (either based on phenotype or genotype) is unlikely to enhance preventive strategies but neither does it cause psychological harm in the short-term. Provision of lifestyle advice alone or alongside CHD risk information might improve fruit and vegetable intake, which needs to be confirmed once analyses of carotenoid levels are completed.
b) Systematic review - Impact of provision of cardiovascular disease risk estimates to healthcare professionals and patients
The purpose of this review was to assess whether provision of a CVD risk model estimate to either patients or practitioners, as opposed to other simultaneous or subsequent interventions, such as lifestyle advice or exercise programmes, impacts patient or practitioner behaviour or health outcomes.
An electronic literature search of Medline and PubMed from 01/01/2004 to 01/06/2013 was conducted with no language restriction and manual screening of reference lists of systematic reviews on similar topics and all included papers. Eligibility criteria for selecting studies were: (1) primary research published in a peer reviewed journal; (2) inclusion of participants with no history of CVD; (3) intervention strategy consisted of provision of a CVD risk model estimate to either professionals or patients; and (4) the only difference between the intervention group and control group (or the only intervention in the case of before-after studies) was the provision of a CVD risk model estimate.
After duplicates were removed, the initial electronic search identified 9671 papers. We screened 196 papers at title and abstract level and included 17 studies. The heterogeneity of the studies limited the analysis but together they showed that provision of risk information to patients improved the accuracy of risk perception without decreasing quality of life or increasing anxiety, but had little effect on lifestyle. Providing risk information to physicians increased prescribing of lipid lowering and blood pressure medication, with greatest effects in those with CVD risk >20% (RR for change in prescribing 2.13 (1.02 to 4.63) and 2.38 (1.11 to 5.10) respectively). Overall there was a trend towards reductions in cholesterol and blood pressure and a statistically significant reduction in modelled CVD risk (-0.39% (-0.71 to -0.07) after, on average, 12 months.
In conclusion, our review suggested that providing CVD risk model estimates to professionals and patients improves perceived CVD risk and medical prescribing, with little evidence of harmful effects on psychological well-being. The systematic review was published in October 2015 by BMJ Open.
c) Qualitative study - The response to receiving phenotypic and genetic coronary heart disease risk scores and lifestyle advice
The aim of this study was to use qualitative methods to explore the short term response to receiving different forms of coronary heart disease (CHD) risk information and lifestyle advice for risk reduction on health-related behaviours. Forty two face to face interviews and two focus groups were conducted across England with participants from the INFORM trial who received a combination of individualised phenotypic and genotypic CHD risk score and web-based lifestyle advice. The risk scores were presented in different formats, including absolute 10 year risk as a percentage, ‘heart age’ and colour visualisation of absolute and comparative risk as a thermometer. Interviews and focus groups explored participants’ understanding and reaction to the risk scores, the usefulness of the lifestyle advice and attempts to change lifestyle during the intervention. We tape-recorded and transcribed the interviews and focus groups and analysed them using thematic analysis.
Three main themes emerged: limitations of risk scores to generate concern about CHD risk and motivation to reduce it; the advantages of the ‘heart age’ format of risk score presentation in communicating a message of non-optimal lifestyle; and intentions and attempts to make moderate lifestyle changes which were prompted by the lifestyle advice regardless of concerns about the risk score. These findings suggest that there are a number of limitations to the use of risk scores to communicate a message about the need for a lifestyle change. Of the formats used, the ‘heart age’, if noticed, appears to convey a powerful message about how far from optimal risk an individual person is. An interactive, user friendly, goal setting based lifestyle website can act as a trigger to initiate moderate lifestyle changes, regardless of concerns about risk scores. We expect the findings from the qualitative studies within INFORM to be published in early 2016 and disseminated via conferences and to participants via Newsletters.
WP15 – Evaluation of screening strategies and cost-effectiveness
The main objectives of this Work Package were to estimate the costs and benefits of different risk assessment strategies focusing on: 1) when is stepwise screening preferable to mass screening? 2) what is the optimum frequency of screening? 3) how should age best be incorporated into CVD risk assessment? 4) what are the most cost-effective approaches across countries and subpopulations?
1) Development and validation of a multi-state model
The purpose of the decision model developed is to quantify the long-term natural history of CVD, and to simulate the impact of different screening and treatment strategies on fatal and non-fatal outcomes, life-expectancy, quality-adjusted life-years (QALYs) and costs. We developed a Markov multistate model in Excel and validated it using data from other epidemiological cohorts while the EPIC-CVD data was being prepared.
2) Development of methods to extract transition rates from the epidemiological data and use these rates to derive transition probabilities to populate the decision model
We used Cox proportional hazards regression, with age-at-risk as the timescale, to estimate the age-specific transition rates as a function of covariates, allowing for left truncation or delayed entry into age-at-risk groups depending on when individuals enter the study or move to a different state (i.e. changing risk-sets). This involved choosing appropriate smoothing methods for the estimated annual hazards. The initial work was based on two European epidemiological cohort studies.
As an example for men, based on a population cohort aged 50 at baseline, we estimated that the average life-years gained if all men take a statin from age 50 onwards is 1.11 years, if from age 60 onwards it is 0.82 years, and if from age 70 onwards it is 0.41 years. The corresponding figures for expected life-years gained free of CVD are 1.82 1.34 and 0.65 years. If only the men aged 50 in the top 20% of predicted risk (according to the Framingham risk score) take a statin from age 50 onwards, their estimated average life-years gained is 1.56 years, compared to the 1.11 years estimated for all men.
In a second example, using another population-based cohort of men with a wide age range at baseline, we estimated that the average life-years gained from taking statins for those in the top 20% of the Framingham risk score is 0.71 years, while that for those in the top 20% of predicted risk given their age it is 1.75 years. In other words, in terms of life-expectancy gained when 20% of the population is given statins, it is better to choose those who are in the top fifth based on high predicted risk given their age. These methods and results are being prepared for publication.
We have also developed methods to estimate transition rates in case-cohort data with multiple centres, and applied this to EPIC-CVD with the CHD outcomes.
3) Elaboration of systematic reviews and meta-analyses to derive key inputs for cost-effectiveness evaluation, and development of modelling methods.
A structured literature review was carried out in order to identify long term cost-effectiveness models of strategies for the prevention of CVD. The results of studies that evaluated screening policies were quite varied. For example, Wald NJ et al 2011 found that screening based on age is more cost-effective than Framingham score based screening. Rapsomaniki et al 2012 found that adding risk variables such as region, age and year of birth, systolic blood pressure, total cholesterol and smoking status, was cost-effective with a cost effectiveness ratio of less than $20,000 per year free of CVD. Cobiac et al 2012 showed that CVD prevention based on absolute risk is more cost-effective than single risk factor thresholds, and recommended statin drugs to everyone with at least 10% absolute risk. Ruitjer et al 2013 showed that performing carotid intima-media thickness (CIMT) measurements induced 1% lower absolute risk of myocardial infarction, and adding this variables to the Framingham equation would be cost-effective.
The studies that evaluated statin treatment (on its own or in combination with other treatments) in identified high risk individuals also showed quite varied results. Results varied depending on: age, gender, the presence or absence of other risk factors and the type of statin used. In general, the incremental cost-effectiveness ratio was lower in younger patients (<50 years) and higher in older patients (>70 years) for a given level of baseline risk (that is, statins are more cost-effective in younger than older patients, with the same baseline risk). Likewise the incremental cost-effectiveness ratio was lower for patients with higher baseline CVD risk.
This review has helped to clarify methodological questions (for example, the structure of the model, the health states that are important) and to identify parameters. 57 papers were retrieved by the searches and 29 included for review following application of exclusion criteria.
An academic paper has been published that critically appraises the published models and examines their suitability for evaluating primary prevention of CVD in European populations in European Journal of Health Economics 2015, Epstein et al.
Additional publications prepared as part of this Work Package include:
Martin-Ruiz E, Olry de Labry Lima A, Garcia Mochon L, Espin J, Ocaña-Riola R, Epstein D.
The benefits and risks of statins for primary prevention of mortality and cardiovascular events: umbrella review. European Journal of Clinical Pharmacy (accepted for publication)
Jones E, Epstein D, Garcia Mochon L. A procedure for deriving formulas to convert transition
rates to probabilities for multi-state Markov models. (Health Economics, under review)
Martin-Ruiz E, Olry de Labry Lima A, Ocaña-Riola R, Epstein D. Systematic review of the effect of adherence to statin treatment on cardiovascular events and mortality in primary prevention. (Manuscript to be submitted to a pharmacy journal)
Martin-Ruiz E, Olry de Labry Lima A, Ocaña-Riola R, Epstein D. The effectiveness of non- pharmacological interventions for primary prevention of cardiovascular disease: umbrella review of systematic reviews (Manuscript to be submitted to a pharmacy journal)
WP16 - Implementation of new cardiovascular risk assessment tools across Europe
The main objective was to identify and analyse the societal issues relating to European CVD risk screening.
As part of EPIC-CVD, we undertook work to explore current tools, policies, perceived barriers and facilitators, and regulatory issues pertaining to the assessment of cardiovascular risk in Europe. We also developed an online web-based tool as a means of implementing a risk score based on novel biomarkers. The pre-specified work package objectives were addressed by three discrete pieces of work conducted by our three partner organisations: RAND Europe, PHG Foundation, and the University of Cambridge. This report summarises the three separate pieces of work undertaken, including key findings, and outlines plans for dissemination and on-going research.
a) Current and future cardiovascular disease risk assessment in the European Union: an international comparative study
The objectives were 1), to identify tools currently being used for calculating cardiovascular risk scores in a range of European countries; 2) to identify policies that are most likely to be acceptable in different European countries and approaches that are most likely to maximise adoption; and 3) to identify the barriers and enablers to screening for a wider range of cardiovascular risk factors in a range of European countries. For the third objective, a decision was taken to focus on the barriers and facilitators to implementation of novel cardiovascular risk factors in risk tools more generally, which was felt to be more relevant to current clinical practice and policy than examining the issue specifically in relation to screening.
RAND Europe conducted an online pan-European survey of senior academic, clinical and policy stakeholders from all EU Member States (November 2014 to March 2015). Survey invitations were sent to 560 experts from all member states; the response rate was 24%. In addition, eight in-depth country case-studies were developed, based on telephone interviews with clinicians, academics and policy makers (May to October 2015). Interviews were supported by targeted literature review both to guide the interview and help interpret findings. Eight countries were selected to reflect variations in health system, macro-economic factors, and geographic location across the EU (Bulgaria, Finland, Germany, Greece, Latvia, Spain, Sweden, UK). There were between three and five interviewees for each country.
This study confirms that the European Society of Cardiology (ESC) is the main body providing and disseminating guidelines for CVD risk assessment across Europe, and the SCORE risk assessment tool, which is endorsed in these guidelines, is the tool most widely reported in use. Implementation of ESC guidelines and the use of cardiovascular risk assessment tools in clinical practice are variable and often low. Risk assessment was found to be predominantly, but not exclusively, conducted by primary care doctors, and health system factors, particularly lack of funding and lack of time, were identified as important barriers to risk assessment use. A clear theme emerging from the work was that any novel risk assessment biomarkers identified by current research will need to be incorporated in ESC guidelines as the key initial step to translation into clinical practice. Evidence of clinical utility and cost-effectiveness will be required for implementation of novel biomarkers, but ethical, legal or regulatory issues appeared of less concern to interviewees. Findings for all the key themes were consistent across Europe, despite differences in context.
In conclusion, the ESC has a central role in informing practice across Europe, and efforts to improve practice are likely to benefit from being channelled through this organisation. Although factors such as integration of risk tools in clinical computer systems may improve practice, there remain significant barriers to the successful implementation of cardiovascular risk assessment, including lack of funding. Despite considerable interest in developing novel biomarkers to optimise cardiovascular risk assessment, it is likely that focusing efforts on increasing the extent of implementation of current approaches will offer the greatest gains in terms of improving primary prevention of cardiovascular disease.
b) Genomics and risk stratification for cardiovascular disease: regulatory implications
PHG Foundation were tasked with the objective of considering the broader societal issues (e.g. ethical, legal, social) of including more detailed information in cardiovascular risk scores, such as hereditary information. It was decided to focus on investigating the legal and regulatory implications of incorporating genomic information into risk stratification for cardiovascular disease across four EU Member States. This focus was chosen as current cardiovascular risk score systems do not take into account genomic information, and the topic has hitherto been under-examined in the literature.
It was decided to include countries that have sophisticated health systems familiar with the complexities of genomic science and medicine as well as a range of cultural and legal approaches to risk stratification, the use of genomic information, and the provision of healthcare more generally. These different perspectives provide variation in terms of healthcare delivery frameworks, implementation of EU law, and the collection, storage and use of genomic information. Four countries were selected for analysis: France, Germany, the Netherlands, and the UK.
The investigation was based on expert opinion and desk-based analysis and literature searches. Expert opinion was provided by a two-day international workshop convened by the PHG Foundation in May 2015. Experts were invited from the four countries, comprising academic lawyers, practising lawyers, and cardiologists with an interest in risk stratification. The workshop was designed to act as a foundation for wider study, with an iterative approach taken: initial work helped focus the workshop, and then expert opinion from the workshop was used to guide further investigation.
This investigation found no absolute prohibition on the incorporation of genomic information into risk stratification for cardiovascular disease at EU or Member State level. However, risk stratification tools that include genomic information are subject to more rules than those that do not. The context strongly influences the application of various rules, with the most important contexts being the purpose for which the tools are used and their mode of delivery. Data protection law strongly influences the use of risk stratification tools, especially in jurisdictions making special legal provision for the use of genomic information. It is unclear whether risk stratification tools constitute in vitro diagnostic devices under current and future EU legislation.
A key recommendation arising from this work is that those providing risk stratification tools should be aware of the need to adhere to a significant number of applicable regulatory controls. Given the complexity, formal legal advice should be sought where necessary.
c) Development of a working model of a risk assessment tool incorporating novel risk factors
A web-based tool was developed which used existing algorithms, with two key features: firstly, incorporating genetic risk, and secondly, allowing presentation of a country specific risk. Such a tool could act as a proof of concept, which could be readily adapted once a final risk algorithm was made available.
A co-design process was used to inform the functionality and appearance of the web tool. The presentation of risk to individuals was provisionally based on an approach used by the intervention website used by the INFORM clinical trial. The entering of phenotype data into the system followed a standard approach as used by other systems such as the Edinburgh cardiovascular risk calculator (cvrisk.mvm.ed.ac.uk). The site was also designed to allow manual uploading of genetic data, or through an interface with an external data source (the commercial genetic data provider 23andMe was used as a test case). The underlying risk algorithm was developed using a limited dataset available from the EPIC Norfolk patient cohort.
Interviews were held separately with clinicians and patients, whilst interacting individually with the web tool. An additional group meeting was held with the patients to share ideas. Key themes were identified from interview transcripts, and changes were made to the tool accordingly.
The web tool was designed for use in a clinical environment, to facilitate discussions of risk between clinicians and patients. Risk was displayed through three “infographics”, including a graduated “thermometer” and ten-by-ten “smiley face” chart to convey absolute risk of cardiovascular disease, and a graphic depicting heart age. Comparison with an individual with optimal risk factors was also presented. A number of changes were implemented as part of patient and clinician feedback, including a focus on a combined (rather than separate) model of genetic and phenotypic risk, provision of a means of visualising the effect of risk-reducing interventions (e.g. statins, smoking cessation), improving various aspects of the graphical design, and rewording of various phrases and explanations to increase clarity. The site has not being made publically available, as the underlying algorithm is a temporary one which has been used for development purposes only. Once the EPIC-CVD programme delivers further detailed risk score algorithms, our intention is to integrate them into the website and make the tool publically available. We will also explore the potential of interfacing with alternative genetic data repositories beyond 23andMe, and making it available in different languages. However, implementation in practice will need to take into account some of the issues raised as part of the work undertaken by PHG Foundation, highlighting the need to be aware of, and adhere to, relevant legislation.
We intend to carry out a future study using the website to examine patients’ and clinicians’ reactions to using the tool, and in particular how incorporating genetic information into such a tool influences perceptions of risk and decisions about cardiovascular risk management. This will form the basis of a mixed-methods study, which will be submitted for publication in a scientific journal.
Potential Impact:
4. Potential impact, main dissemination activities and exploitation of results.
4.1 Potential impact
EPIC-CVD aimed to build upon previous important efforts in CVD risk prediction by using population resources, biomarker technologies, and translational studies that have only recently become practicable for large-scale application across Europe. Our multidisciplinary research consortium, therefore, focused world-leading expertise in translational research, epidemiological and diagnostic technology and knowledge to produce innovative outputs, such as modifiable and detailed risk scores, risk calculation and communication tools, customised stepwise screening strategies and cost-effective policy recommendations.
The single biggest impact of EPIC-CVD will be to provide clinicians and policy-makers across Europe with an evidence-based menu of contemporary options for targeted and cost-effective CVD risk approaches tailored to the needs of Europe's diverse populations. Through this, EPIC-CVD will contribute to health, wealth, innovation, and research competitiveness within the EU.
4.1(a) Better health
Public health burden of CVD in Europe CVD is Europe's leading cause of death and a major source of disability, with a burden expected to increase as a result of the ageing population and increasing prevalence of obesity and diabetes. Such considerations have impelled FP7 to adopt as a thematic priority area "the development and validation of...methods for health promotion and prevention including the promotion of healthy ageing, diagnostic tools and medical technologies, and sustainable and efficient healthcare systems."
Calibration of scores to different European populations EPIC-CVD has developed and calibrated existing and new risk scores to diverse European populations. For risk scores already in clinical use, the outcome of this is greater predictive accuracy. This is because most existing risk scores have been derived in and calibrated to populations in the USA (eg, Framingham, Reynolds) or a single European country (eg, ASSIGN and QRISK2 from the UK or PROCAM from Germany), or lacked information on dietary and genetic factors that prevail in Europe (eg, SCORE). By addressing these limitations in existing risk scores, EPIC-CVD should encourage greater use of risk scores in European populations because patients and healthcare professionals will appropriately regard the scores as being of greater local relevance. Thus, EPIC-CVD is likely to enhance ability to identify people at high CVD risk and manage them appropriately, leading to improvements in identification, risk assessment, clinical decision making and clinical outcomes.
Modifiable risk scores EPIC-CVD's contribution, however, goes beyond providing useful refinements to existing risk scores by creating Europe's first modifiable CVD risk scores containing objective biomarkers of lifestyle. When accompanied by personalised feedback that translates risk estimates into easily understood advice about lifestyle factors that can be readily self-monitored, these novel scores should contribute to the prevention of CVD in several ways:
●to enhance predictive accuracy, especially because we have studied objective and reproducible nutritional biomarkers, resulting in improved identification and risk assessment
●to promote behaviour change so that patients reduce adverse lifestyles, leading to better clinical outcomes
●to promote change in the behaviour of healthcare professionals so that CVD risk is monitored more effectively, leading to improved clinical decision making
●to contribute to the prevention of other conditions of major public health importance in Europe, such as obesity, T2DM and certain cancers, again leading to better clinical outcomes
●to be genuinely complementary to public health activities that aim to reduce the overall population risk of cardiovascular disease by promoting a healthy lifestyle (diet, exercise, avoidance of smoking).
Detailed risk scores EPIC-CVD has also developed innovative detailed risk scores that, for the first time, are able to capture the interplay of nature and nurture. Again, this output is likely to result in improvements in identification, risk assessment, clinical decision making and clinical outcomes.
Content of scores: safe, practicable, and ready for translation Although EPIC-CVD has studied a large range of biomarkers (≈300 soluble biomarkers and 215,000 carefully selected genetic variants) on an unprecedentedly powerful population scale across Europe, this project was not simply an "-omics" discovery effort. Rather, EPIC-CVD's work will yield clinical usefully biomarkers because it has also assessed a carefully selected set of existing and emerging biomarkers. In particular, EPIC-CVD has studied biomarkers that already have standardised high-throughput blood tests and a reasonable likelihood of emerging as cost effective, safe, and valid (eg, biomarkers already in routine clinical use or recommended by guideline statements or regulatory agencies). Because these biomarkers have rapid translational potential, EPIC-CVD's efforts are more likely to lead to improved cardiovascular risk prediction and contribute to the development of personalised predictive medicine in a short timeframe.
4.1(b) Greater wealth
Efficient strategies EPIC-CVD has put a high priority on developing approaches that enhance the cost effectiveness and operational efficiency for targeting preventive resources. For example, it seems unlikely that many countries in the EU (particularly the resource-constrained countries of southern and eastern Europe) will be able to follow the example of the UK, which has introduced a universal national CVD screening programme for adults aged 40-74 years at an estimated cost of €2 billion between 2010 and 2015. Consequently, EPIC-CVD will provide cost-effective alternatives to such un-targeted approaches by building on our experience in developing sequential or "stepwise" screening strategies and by tailoring such approaches to the various budgets, health information systems, and cultures across Europe.
4.1(c) Evidence-based use of innovative technologies
There is a worldwide trend toward more demanding scrutiny of predictive tests by health technology assessment agencies, exemplified by the US FDA’s warnings to companies that sell genetic tests directly to consumers. In Europe, the 2002 European Parliament's report on Life Sciences and Biotechnology Strategy called on the Commission "to take the necessary steps for an EU-wide regulation on DNA-testing, choosing, if possible, a legal basis (e.g. Article 152 (health) or Article 153 (consumer protection))...". In 2008, the Council of Europe's Additional Protocol on Human Rights and Biomedicine, Concerning Genetic Testing for Health Purposes stated (Article 5): "Parties shall take the necessary measures to ensure that genetic services are of appropriate quality ... [and] genetic tests meet generally accepted criteria of scientific validity and clinical validity". However, a clear regulatory framework has not yet emerged in Europe for prognostic biomarkers as many products, such as genetic tests, are currently left unregulated because they are classified as low risk and require only self-certification (making them liable to multiple conflicting interpretations).
EPIC-CVD will make an important contribution to evidence-based standards for innovative prognostic technologies in 2 main ways. First, through the PHG Foundation, a partner in EPIC-CVD and a leading independent genetics think-tank, we have considered the broader ethical, legal, and social implications of prognostic biomarker technologies for Europe. Second, the scientific framework of EPIC-CVD has contributed to new evidence-based paradigms for prognostic biomarkers. This is because, for the first time, EPIC-CVD has provided a rigorous and integrated evaluation in CVD across Europe that combines comprehensive study of predictive performance of risk scores, detailed health economic modelling, and qualitative and quantitative translational studies of patient-centred outcomes. In aggregate, these approaches have been able to reliably test whether a new approach improves risk assessment, clinical decision making, and clinical outcomes as well as cost effectiveness, safety, validity and incremental benefit over existing prediction methods. The net effect of these efforts should be to better protect the public, reduce uncertainty in product development for industry, and accelerate clinical translation by avoiding wasteful effort in suboptimal studies. We should also be able to provide a much clearer path than now exists in Europe to encourage SME efforts towards research and innovation in the prognostic biomarkers sector.
4.1(d) Greater European research competitiveness
Strategic scientific context In the document “Life sciences and biotechnology – a strategy for Europe”, the European Commission describes its aim to restore European leadership in life sciences and biotechnology research. In the 2010 document "Europe 2020 Flagship Initiative: Innovation Union", the EC emphasises the need for researchers and innovators "to work and cooperate across the EU to get more innovation out of our research" and launch research partnerships that improve "the quality of life of an ageing population by new innovative solutions, clinical tests and diagnostics". A major growth area in biomedicine worldwide is "the development of personalised and predictive medicine”. Europe is strategically well-positioned to take a leading role in this endeavour because of the region’s: (1) natural advantages of striking lifestyle, cultural, and genetic diversity (2) longstanding scientific strengths in key disciplines such as public health, nutrition, and genetic epidemiology (3) cluster of excellent “research intensive SMEs” in the life sciences that provide "research and innovation" (4) track record of cohesive collaboration in major pan-European scientific projects.
Because EPIC completed its recruitment of 520,000 participants in the 1990s, it has about a 5 year head start over other large “biobank” initiatives that have been established, or that are currently recruiting in China, Australia, the USA, and individual European countries. Hence, EPIC-CVD has been able to take advantage of a “window of opportunity” and has been able to contribute to European and international leadership in this area.
4.2. Dissemination strategy
At the outset of the EPIC-CVD project, with representatives from the partner organisations, we developed a dissemination plan. The key to a dissemination plan in a project of this nature was to recognise that there are multiple target audiences and that each audience needs to be addressed in a different manner, using different media.
Communication plan The project deliverables included a “Communication Plan” which included the main dissemination materials and tools for EPIC-CVD. This plan aimed to:
(1) select the appropriate modes of communication and communication tools
(2) define key messages
(3) tailor information and deliver to the intended recipients
(4) maximise the exposure of messages
(5) evaluate the results of our dissemination efforts
(6) capitalise on the European Commission’s dissemination resources.
Interactive website The Project Management Office was responsible for the development and maintenance of our interactive website (www.epiccvd.eu) which will be maintained after the project ends, given the potential clinical importance of the material that will be added. This website has been integral to EPIC-CVD throughout its lifetime. Its provision of tools, strategies, information, publications, policy documents etc has intensified as the project has gone on.
Target audiences included organisations such as the EC, the European Medicines Agency, the European Society of Cardiology, the European Atherosclerosis Society, the European Association of Cardiovascular Prevention and Rehabilitation, the European Stroke Initiative, European Heart Network, companies that manufacture diagnostic and biomarker products, national governments.
Scientific publications With help from the Project Management Office, the Study Coordinator has overseen the work of an Analysis Coordination Group and the reporting of findings. The Project Management Office has recorded and coordinated the publication process with all partners by means of the EPIC-CVD intranet (with a dedicated publications section and a searchable database). Press releases for high-impact papers are prepared by the Project Management Office in conjunction with the organisations of the relevant stakeholders.
4.3. Dissemination to date
Dissemination of the findings from EPIC-CVD has to date taken 5 main forms:
- peer-reviewed scientific journal publications
- oral presentations at workshops, meetings and conferences
- policy documents
- online tools and resources
- participant newsletters
a) Journal publications
To date, there have been 15 publications enabled by the EPIC-CVD project. In addition to publications relating to the specific Work Packages of the project, the development of the EPIC-CVD dataset has enabled many further projects related to aetiological analyses and methodological work that weren’t necessarily anticipated at the start of the project, providing excellent added value to the European Commission’s investment. Evidence of this comes from the EPIC-CVD extranet, which has received 67 proposals for studies from investigators within EPIC-CVD to date. Scientific papers from each of these proposals will be written and submitted to peer-reviewed journals during the course of 2016/17.
b) Oral presentations
Findings from EPIC-CVD Work Packages have been presented at a number of national and international scientific conferences, workshops and meetings, including:
- Society for Academic Primary Care, January 2015
- FENS conference, June 2015
- European Epidemiology Congress, June 2015
- European Society of Cardiology, August 2015
- Cochrane Colloquium meeting, October 2015
Future oral presentations already planned include:
- American Heart Association Scientific Sessions, 2016
- European Society of Cardiology Congress, 2016
- Society for Academic Primary Care, 2016
Additional oral presentations of findings from EPIC-CVD will be decided as near-final versions of scientific manuscripts emerge.
c) Policy documents
Some aspects of EPIC-CVD work are amenable to publication as policy documents that can be distributed to relevant stakeholder groups (eg, patients, regulatory bodies, practitioners etc). For example, as part of WP16, the PHG Foundation investigated the regulatory implications of using genomics in cardiovascular risk stratification. This work was published as a policy report in 2015 and widely disseminated to key stakeholder groups identified as part of WP16. Further findings from EPIC-CVD, particularly those from WP16 (eg, results of the case studies comparing CVD risk screening across 8 European countries), will also be disseminated as policy reports if deemed appropriate.
d) Online tools and resources
In addition to the public website of EPIC-CVD (www.epiccvd.eu) which we plan to maintain and update with new findings for the foreseeable future, we will continue to support the EPIC-CVD ‘extranet’, which serves as a mechanism for investigators and collaborators within EPIC-CVD to propose analyses, upload statistical analysis plans, organise analytical efforts, share draft scientific manuscripts and log accepted papers. Additional online resources that have emerged as part of the EPIC-CVD project include the INFORM trial website (www.informstudy.org.uk) and our genetic CVD risk score tool.
e) Participant newsletters
To ensure that we continue to engage participants involved in research conducted as part of the EPIC-CVD project, we will provide feedback on our findings to participants via newsletters. This has already been successfully demonstrated during the INFORM trial (WP14), during which INFORM participants received regular newsletter updates by e-mail to keep them abreast of progress with the trial.
List of Websites:
http://www.epiccvd.eu/