Globally, cardiovascular disease (CVD) is a leading cause of disability and premature mortality, with the number of deaths attributable to CVD rising from 12.3 to 17.3 million between 1990 and 2013 [1]. Furthermore, projections indicate that CVD will remain the leading cause of mortality worldwide for decades [2]. Notwithstanding the public health importance of CVD and its risk factors, the use of proven preventative strategies to mitigate risk is far from satisfactory due to a multitude of factors including complex intra- and interregional distribution of the determinants of CVD [1,3]. For instance, deaths attributable to different types of CVD vary between China and the United Kingdom. This variation may, in part, reflect the prevalence of different underlying risk factors such as hypertension and diabetes [4]. Thus, with regions around the world undergoing major epidemiological transitions, a population strategy aimed at reducing the burden of classical risk factors can have a major impact on CVD risk reduction [5].

In England, the National Health Service (NHS) established a population-based Health Check program in 2009 to provide a systematic assessment and management of CVD risk for adults. Individuals aged 40-74 years without pre-existing CVD, diabetes, or hypertension are invited for a five-yearly check-up to identify and to receive advice and treatment for their risk of having a CVD event [6]. Commissioned by local authorities and delivered primarily through primary care practices, henceforth referred to as general practices (GP), this program aims to prevent around 2000 deaths and 9500 nonfatal myocardial infarctions (MI) and strokes annually [7,8]. As a free service for all eligible individuals, this program has the potential to reduce health inequalities associated with CVD [7,8], assuming equitable uptake and universal coverage. Available data suggest that between 2011-12, NHS Health Check coverage in east London was quite high at 73% (compared to national coverage of 30%). Of those that attended the screening, over a third belonged to ethnic minority groups such as south Asians and Black African/Caribbeans [9-13]. While modification of individual risk factors (eg, blood pressure) has generally been shown to be associated with improved health outcomes (eg, lower incidence of ischaemic stroke), favorable changes in risk factor distribution at a societal level have also contributed to overall improvements in community health [14]. For example, in 1989, in Poland, the increased accessibility and consumption of fruits and vegetables through the opening of markets resulted in a reduction in CVD mortality [15]. By contrast, Japanese migrants, who adopted the diet and lifestyle of the United States (US) population, relative to those who retained their original dietary patterns, had a higher prevalence of CVD [16]. Given the variation in ethnic-specific diet and lifestyle habits, and given genetic differences across population groups around the world, there is currently an unmet need to understand the complex relationship between ethnicity and factors as diverse as heredity, environment, diet, lifestyle, and socioeconomic factors (and their mutual interaction) in determining the burden, distribution, and temporal trends in both CVD risk factors and clinical CVD outcomes and cause-specific death.

Dietary exposures determine nutritional status, which may, in turn, impact CVD outcomes. As nutritional status is a key intermediary in the development of CVD, health services need to measure nutritional status as well as associated dietary exposures in order to guide appropriate preventative measures, particularly in multiethnic populations in primary care. The National Diet and Nutrition Survey (NDNS) and components of the National Survey of Health and Development (NSHD) have been developed to assess dietary exposure patterns, nutritional status, and associated social as well as biological factors within a sample of UK populations [17-19]. However, to date, no systematic attempts have been made to harness these data to boost or complement information gleaned from population-based studies investigating ethnic-specific CVD risk factor burden or risk [20].

The risk of CVD in an asymptomatic individual is based on risk factor data collected by their primary care physician to produce a numerical estimate. Traditional risk scores, such as those derived from the Framingham risk equation, have potential limitations, primarily because they were derived from a cohort of predominantly White individuals in the United States [21,22]. Moreover, they do not make allowances for factors such as socioeconomic status, resulting in systematic over- or underestimation of CVD risk based on the population studied [21,22]. With the emergence of obesity and diabetes as key global cardiovascular risk factors [22], the QRISK2 CVD risk calculator was developed and calibrated, enabling adjustments for more contemporary risk factors in the United Kingdom [22,23]. Despite these considerations, the representation of ethnic minorities in the cohort used to derive QRISK2 was approximately 1% for South Asians and 0.5% for Afro-Caribbean individuals, with the likelihood of underestimation of 10-year CVD event risk in these population groups [21].

Furthermore, QRISK prediction models contain imputations due to missing data. While established statistical methods are often used, considerations such as imputing cholesterol values to individuals without vascular disease could lead to weakened associations between cholesterol and CVD [24], as was observed in the first iteration of QRISK where about 80% of cholesterol measures were imputed. Hence, additional validation of QRISK2 in a contemporaneous multiethnic population with more systematic approaches to the collation of exposures than previously possible would be beneficial [21].

These systems need to be interoperable to enable a systematic analysis of routine healthcare data collected in electronic health records (EHRs). However, over 100 commercial EHR vendors are operating within the NHS across primary, secondary, and tertiary sectors [25]. While organizations are beginning to link data from different EHR systems, there is still fragmentation and absence of vital data as, for example, GPs serving multiethnic populations are not providing their data. Moreover, given that establishing bespoke prospective studies on CVD prevention can be prohibitively expensive, using real-life routinely collected data from EHRs and subsequent linkage of these datasets across different sectors can provide invaluable systematic approaches to conducting research.

We developed the following study specific objectives:

We propose to conduct a large-scale, cross-sectional observational study of NHS Health Check attendees in London with additional virtual longitudinal follow-up established via electronic health record linkage of their primary (NHS Health Check) and secondary care (hospital) data over 10 years starting in 2009. One of the major providers of EHR software to GP practices is Egton Medical Information Systems (EMIS). GPs utilizing this EHR will be eligible to participate. Figure 1 illustrates the number of individuals belonging to ethnic minority groups within London. Such a high prevalence of diverse ethnic groups in London, relative to other areas in England, coupled with high utilization of the EMIS system (Figure 2), lends itself naturally to London as the ideal backdrop for our proposed investigation.

GP practices using EMIS Web clinical systems across London will be invited to provide practice-wide consent to participating in the study by way of sharing pseudonymized data on individuals undergoing NHS Health Checks within their practices between 2009 and 2018. GP practices across London are being informed of the study and recruited through liaising with clinical research networks through information letters and Clinical Commissioning Group meetings. Data will be remotely extracted centrally by EMIS using Read codes.

Our study, designed as a population-based registry at the outset, will collect, alongside NHS Health Check data, additional demographic variables and prescription information to help build a large and robust repository of primary care data on residents of London undergoing NHS Health Check from 2009 onwards in GP practices.

All individuals seeking healthcare in the United Kingdom are assigned a unique identifier (ie, their NHS number). This number allows the allocation of any subsequent data collected to the individual’s record and enables linkage with diverse data sources. For instance, the Office for National Statistics and Hospital Episode Statistics databases maintain records on mortality and clinical outcomes of individuals utilizing secondary healthcare services, respectively. Data from these sources can be linked through NHS numbers to the NHS Health Check data to provide a comprehensive and contemporary database on CVD risk factors (predictor variables) and outcomes across different ethnic groups. Other data sources recording clinical CVD outcomes such as the Myocardial Infarction National Audit Project and the National Institute of Cardiovascular Outcomes Research will also be harnessed to provide additional details relevant to this study.

This process of data sharing and linking, while protecting an individual’s confidentiality, is possible due to pseudonymization and will be conducted by EMIS and a third party organization familiar with bespoke linkage. Using Open Pseudonymiser Software, EMIS will create a pseudonymized version of NHS numbers for individuals who meet the inclusion criteria. The data will be encrypted with a project-specific key. By using this encrypted key, an external organization will be able to link the NHS Health Check data with outcomes data through a similar pseudonymization process and a Global Unique Identifier (GUID). The research team will only have access to completely linked anonymized data as the linkage of NHS Health Check data with outcomes data will not contain the GUID.

Although data are not available for 2009-2011, of the invited population for NHS Health Check in 2016-17, 51.6% and 49.9% were residing in London and the rest of England, respectively [27]. In 2015, EMIS utilization in GP clinics was approximately 55%. Therefore, if 50% of all NHS Health Check attendees visit GP clinics utilizing the EMIS system, then it can be assumed that 587,273 individuals in London are eligible for consideration in this study between 2011 and 2017 [28,29]. Due to a lack of visibility of the true number of GP practices using a particular system, Figure 2 highlights that the majority of the areas in London do utilize the EMIS system, and, therefore, we anticipate that the intended sample size will be met.

To develop novel risk prediction tools for population groups, we anticipate that data will be available on approximately 200,000 individuals across London between 2011 and 2017. During 2009-13, of those with a recorded ethnic group, uptake of the NHS Health Check across England was greatest amongst South Asians (19.2%), Black Caribbeans (19.6%), Black Africans (15.7%), and Chinese (15.3%). Attendance was 16.9% in all other recorded ethnic groups and 17.4% in white Caucasians. Extrapolating the current rate of CVD incidence (2.4%) for people aged below 79 years in England on our estimated study sample of 200,000, we expect to encounter 15,000 CVD events giving 100% power to observe relative risks as modest as 1.2-1.5 [23]. Within population subgroups, the power to detect overall association is expected to range from 49% to 100% [38,39]. Given the large Black and minority ethnic (BME) population in London and considering the above numbers, we estimate that our study sample will be highly representative of a diverse cohort [38,39].

Cardiovascular risk factors (predictor variables) used to calculate QRISK2 include age, sex, ethnicity, index of multiple deprivations, smoking status, diabetes status, family history of CVD, chronic kidney disease (stage 4 or 5), atrial fibrillation, hypertension treatment, rheumatoid arthritis, lipids, blood pressure, and BMI. EMIS will also extract prescription data. CVD outcomes will include fatal and nonfatal coronary heart disease, fatal and nonfatal stroke, heart failure, all CVD-related hospitalizations (including those for revascularization), and all-cause mortality.

Since the data for our study will be extracted at a single source (EMIS), this is likely to obviate the need for harmonizing, although we will undertake detailed quality checks (both at baseline and on follow-up) and undertake data cleaning, where necessary.

By using appropriate statistical tests to perform univariate comparisons, the collected data will be initially analyzed to explore the cross-sectional characteristics of this population. Skewed data will be transformed to an approximately normal distribution, with the calculation of Pearson correlation coefficients and partial correlation coefficients for relevant factors of interest.

Cross-sectional associations of baseline risk factors will be studied with adjustments for appropriate confounders using logistic regression analyses. Prospective associations between individual risk factors and CVD outcomes will be evaluated using Kaplan-Meier survival curves and further quantified by fitting Cox proportional hazards regression models with multivariable adjustments. Where possible, associations will be studied after making allowances for the time-varying effects of both exposures and potential confounders. Detailed analyses of the associations will be conducted in prespecified subgroups of participants (including, but not restricted to, age, sex, conventional risk factors, and ethnicity). Established methods of calibration (Hosmer-Lemeshow), discrimination (C or D statistic), and reclassification (net reclassification index, integrated discrimination improvement) will be used to study the predictive ability of individual risk factors on CVD outcomes, with prespecified analyses involving multiple ethnic groups to enable direct, head-to-head comparisons of any differences.

Additional analyses will be conducted to quantify the impact of factors such as access to healthcare and utilization rates (of NHS Health Checks), prescription rates and medication compliance, and patterns of healthy behavior on CVD risk (both overall and, where possible, within subgroups). Standardized rates of prescription drug use will be calculated to investigate the extent of correlation between the present use of cardiovascular medication and cardiovascular risk. Variations in prescription patterns of CVD prevention and risk factor control, discontinuation of prescribed medication(s), and adverse effects to prescription medications from various therapeutic categories such as anti-hypertensive agents, lipid-lowering drugs (statins), and antidiabetic medication will be further assessed for their impact on CVD as time-dependent covariates in Cox regression models. Physical measurements within the NHS Health Check program will be obtained by the GP practices according to standard operating procedures, anticipating minimum inter-observer variability. Analyses to explore potential sources of heterogeneity concerning physical measurements (besides other characteristics) will also be conducted.

New and existing risk prediction models will be compared in terms of their ability to predict the onset of CVD across diverse population groups to develop improved risk prediction tools for primary prevention of CVD using outcomes data up to 2020. The resulting CVD risk algorithm will be calibrated by further analyses of CVD outcomes data of these patients through to 2027.

The study protocol underwent an external peer review before the ethics submission. This study has received a favorable ethics opinion by the East Scotland Research Ethics Committee (Ref: 17-ES-0104), Health Research Authority, and adopted by the clinical research network portfolio in November 2018.