Article Text
PDF
Statistics from Altmetric.com
Learning objectives
To understand the rationale for using cardiovascular risk prediction tools to make effective and appropriate risk factor treatment decisions in clinical practice.
To understand which online cardiovascular risk prediction tools are available, and to be able to choose which to use for a given patient in clinical practice.
To be able to interpret and communicate different prediction outcomes of cardiovascular prediction tools.
Introduction
Cardiovascular disease (CVD) remains the most important cause of morbidity and mortality worldwide.1 For prevention of CVD, cardiovascular risk management is advocated in international guidelines.2 3 Many cohort studies and randomised controlled clinical trials (RCTs) have demonstrated the benefits of risk factor management, including smoking cessation, lipid lowering, blood pressure lowering, antithrombotic therapy, glucose lowering and more recently, anti-inflammatory therapies, on CVD risk.4–9 Besides these interventions, healthy lifestyle behaviour should always be promoted at individual and population level. With this growing plethora of choices in cardiovascular prevention, it can be difficult for both healthcare professional and patient to make the most appropriate treatment decisions for each individual person.
Identifying those patients who will benefit most from risk factor treatment is pivotal in the global CVD prevention effort. Risk stratification is a cornerstone in international CVD prevention guidelines, aiming to identify those at highest risk of future CVD in order to most effectively apply preventive strategies. Risk assessment using risk prediction tools can thus play a highly important part in global CVD prevention efforts in choosing the right treatment and the right treatment goals, for the right patient. This narrative review aims to guide clinicians in using risk stratification tools as decision support tool in CVD prevention.
Why should CVD risk prediction be used in clinical practice?
Prevention is better than cure, also in the context of CVD. Recommendations in international guidelines are rooted in this simple concept: the higher the absolute risk, the higher the absolute benefit of risk factor treatment, and the lower the number needed to treat (NNT) to prevent one CVD event during a certain time period.10 In line with this thinking, international guidelines state that the level of risk in an individual patient should guide the decision whether or not to treat risk factors, or how intensively to treat them.2 3 This train of thought is supported by findings from RCTs that the relative treatment effect of risk factor treatment is usually similar for various patient groups and not dependent on patient characteristics such as age, sex and comorbidities, while the absolute treatment effect is substantially larger in high-risk patients. Thus, stricter or more intensified risk factor management is recommended in the highest-risk patients compared with lower-risk patients. However, while this concept of ‘the higher the risk, the higher the benefit’ is useful for making treatment decisions in certain groups of patients, it has some important downsides: 10-year risk is largely age-driven, and is thus almost invariably low in young persons and high in older persons, severely limiting its use for clinical decision-making in these age groups. This will be discussed in more detail later in this review.
To select patients at high CVD risk, clinical intuition may not be sufficient. Using risk prediction tools helps to objectively identify patients at high risk, and thus presumably with a high benefit of risk factor treatment. Poor risk stratification may lead to both overtreatment with risk of unnecessary therapy-related adverse events and costs, and undertreatment which may lead to CVD that could have been prevented with adequate risk factor management.
Risk stratification can thus aid in making informed personalised treatment decisions for cardiovascular risk management, and can be used to inform patients about their prognosis. Risk prediction tools are not a replacement of clinical expertise, but should rather be seen as clinical decision support tools that aid healthcare professionals and patients to choose the most appropriate treatments or treatment goals, to achieve the best clinical outcomes for the individual patient, and to make the best use of healthcare resources. Importantly, using individualised risk prediction may improve patient engagement, involvement in shared decision-making and patient satisfaction,11 and may increase patient motivation for lifestyle changes and therapy adherence.
Risk prediction models: a short theoretical background
Risk prediction models are mathematical functions that predict the occurrence of an event of interest based on certain predictors, such as patient demographics, medical history and medication use, physical examination, disease characteristics and laboratory values. To be useful in clinical practice, risk prediction models should be easily accessible and easy to use, include clinically relevant and readily available predictors and reliably estimate risk in a way that improves treatment decision-making and patient outcomes. Knowing how risk prediction models are developed and validated can be useful for understanding the applicability and validity of their use in clinical practice. Therefore, a very brief summary of the most important principles of model development and assessment of model performance are provided below.10 12 13
Developing a risk model
Risk prediction models are developed in a study population of interest to daily practice and of sufficient size, preferably in high-quality prospective cohort studies with systematically obtained baseline and outcome data to minimise the risk of bias. Predictors known to be—or suspected of being—associated with the outcome of interest can be selected a priori based on previous literature, or selected using statistical methods. Which predictors are selected may be different across various groups of patients and may depend on the outcome of interest. The predicted outcome should be defined clearly and usually includes fatal and/or non-fatal vascular events, for example, myocardial infarction and stroke.
Also the time for which the risk applies needs to be specified, in CVD risk prediction usually 5-year or 10-year risks. More recently, lifetime models have been developed that can estimate individual lifetime risk and remaining CVD-free life expectancy.14–18 The statistical methodology of these lifetime models have been explained in detail elsewhere.19 20
Assessing the predictive performance of a risk model
After the risk model development, the reliability of the predictions (ie, model performance) needs to be tested, both in the dataset used for model development (internal validation), and in other populations (external validation). Table 1 summarises some of the metrics used to assess model performance.
Metrics describing predictive model performance
Although the C-statistic for discrimination is most often presented, it could be argued that model calibration (‘goodness of fit’) is a more clinically relevant measure of model performance, as clinician and patient want to know if the predicted risk resembles the actual risk.21 If baseline event risk (ie, which percentage of people in a population will experience a CVD event) is very different in a population compared with the study population in which a model was derived, the predicted risks will systematically underestimate or overestimate the actual individual risks in that population. In that case, it is necessary to recalibrate the model to better reflect the baseline risk in the external population.13 This may, for example, be necessary in different countries (geographical recalibration), or when a model is older to better reflect a contemporary population (temporal recalibration). Good model calibration, that is, accurate risk prediction, is important, because the estimated risks will be used to make the eventual treatment decisions, for example, by established risk thresholds.21 International guidelines decide what the best model is for each region or situation based on (geographical) validation studies.
Considerations and recommendations for clinical practice
There is an abundance of CVD risk models available in medical literature, making it difficult for clinicians to choose which model to use. The majority of published risk models are not suitable for use in clinical practice, due to lack of external validation, the use of predictors that are not usually available in daily practice, methodological limitations and incomplete presentation (eg, when no calculator is presented to be able to apply the model in clinical practice).22 In the scope of the current review, we will only consider externally validated models that have an online tool available, as these are easily accessible for use in daily practice. These risk prediction models have been summarised in table 2. Many of these algorithms are recommended by (inter)national guidelines.
Overview of cardiovascular risk prediction models available in an online risk tool
Which outcome is predicted?
Different CVD risk models predict different outcomes: for example, the Systemic Coronary Risk Estimation (SCORE) model estimates the risk of fatal CVD; the Atherosclerotic CVD (ASCVD) Risk Estimator predicts the risk of a composite of coronary death, non-fatal myocardial infarction and fatal or non-fatal stroke and the Framingham score predicts the risk of coronary artery disease alone. Recent methodological advancements allow outcome predictions in a lifetime perspective,19 which will be discussed later. When interpreting the results of a risk model, it is important to inform oneself of the definition of the predicted outcome (table 2).
Additionally, as shown in table 2, several models allow the estimation of individual effects of preventive treatment.23 This may help in individualised clinical decision-making, and may increase patient motivation for lifestyle changes and therapy adherence. However, in order to effectively do so, clinicians should be able to correctly interpret and communicate prediction outcomes from different prediction tools. Several prediction outcomes and their interpretation are summarised in table 3.
Prediction outcomes from cardiovascular risk models and their interpretation
Which algorithm or tool to choose for each patient in clinical practice?
For each patient, the most suitable CVD risk assessment tool needs to be considered based on clinical characteristics (eg, medical history and age). Different risk models are available for various patient populations, which may include risk predictors specifically aimed at these populations. For example, although several risk models for primary prevention also include presence of diabetes as a predictor, using a model targeted at patients with diabetes specifically will likely result in more accurate estimations, as diabetes-specific risk factors (eg, HbA1c, duration of diabetes) are taken into consideration. Moreover, the chosen model should be appropriate for the baseline risk of the population the individual patient hails from, both in terms of geographical region and timeliness of the data used for development and validation or recalibration of the model. Finally, as discussed above, the predicted outcome should be relevant when making treatment decisions for the individual patient.
Taking these considerations into mind, the following paragraphs discuss the available risk prediction tools for various patient populations. Figure 1 shows a decision aid for choosing the most suitable calculation tools currently available online. Important to note is that CVD risk models are not available for all patients. In case of severe comorbidity potentially limiting life expectancy, or very extreme risk factor levels such as seen in familial hypercholesterolaemia, existing CVD risk models are not accurate and should not be used.
Decision aid for deciding which calculator is most suitable for an individual patient. CVD, cardiovascular disease.
Apparently healthy persons
Various CVD risk prediction tools are available for ‘apparently healthy persons’ (ie, patients without established CVD) (table 2).3 16 18 24–30 The European Society of Cardiology (ESC)/European Atherosclerosis Society guidelines recommends using SCORE2; the American College of Cardiology/American Heart Association (ACC/AHA) guidelines recommend ASCVD,3 31 while the National Cholesterol Education Programme and Canadian Cardiovascular Society recommend Framingham32 33 and in the UK the Joint British Society (JBS) guidelines recommends the JBS3 calculator, based on the QRISK lifetime model,18 while the National Institute for Health and Care Excellence guidelines recommend the use of the QRISK2 model.34 The reasons for these many different recommendations often have to do with geographical and historical considerations.
Treatment decisions are usually based on 10-year risk thresholds, are dependent on the event predicted (eg, CVD mortality vs all fatal and non-fatal CVD events) and differ between guidelines. The 2016 ESC guidelines uses the following risk categories based on SCORE: low-risk (<1% 10-year CVD mortality risk), moderate-risk (1%–5%), high-risk (5%–10%) and very high-risk (≥10%)2; the 2019 ACC/AHA guidelines uses the following risk categories based on ASCVD: low-risk (>5% 10-year risk of ASCVD), borderline risk (5%–7.5%), intermediate risk (7.5%–20%) and high risk (≥20%).31
Patients with established CVD
Patients with clinically established CVD are in accordance with current guidelines, all considered at (very) high risk of recurrent CVD, with the associated recommendations for intensive (pharmacological) interventions and treatment goals of risk factors. The remaining individual risk for recurrent CVD after initiation of the generally recommended risk factor treatment (ie, residual risk) can be estimated in individual patients with established CVD.35 Patients with established CVD may have a relatively low residual 10-year risk for recurrent CVD, even below the ‘low risk’ cut-off for 10-year risk for patients without CVD, while others still have a very high residual 10-year risk for recurrent CVD even after risk factor management. Routine individual residual risk prediction in secondary prevention could guide intensification of risk factor treatment.
As the relationship between risk factors and outcomes may be very different in patients with established CVD compared with patients without CVD, specific risk models are necessary, also including additional disease-specific variables.36 Online risk stratification tools for secondary prevention include the SMART risk score, estimating 10-year residual risk in patients with stable vascular disease,35 37 and the TRS 2°P risk score, estimating 3-year risk in patients with previous myocardial infarction.38 There are currently no formal residual CVD risk thresholds above which intensification of risk factor treatment (eg, lower low-density lipoprotein cholesterol (LDL-c) goal, lower systolic blood pressure (SBP) goal or dual antiplatelet therapy or dual pathway inhibition) should be considered.
Patients with type 2 diabetes
In current international guidelines, patients with type 2 diabetes are all considered to be at (very) high risk for future CVD. In reality, there is a wide range of individual risks for future CVD events, especially after initial risk factor treatment.39 Diabetes-specific risk models including diabetes-specific predictors are the ADVANCE risk score, and the UKPDS risk engine which is available for use in the UK.40 41 There are no formal CVD risk thresholds yet established for treatment decisions in patients with diabetes.
Apparently healthy younger persons
Age is the most important driver of 10-year CVD risk. Younger persons (~<50 years) are therefore almost invariably at low 10-year risk even in the presence of high-risk factor levels. Based on risk thresholds for initiation of risk factor treatment, younger patients are usually not eligible for preventive pharmacotherapy. However, younger patients with a high burden of risk factors may have very high lifetime risks of developing CVD, and intuitively it seems that these patients may benefit hugely from early intervention. In the context of atherosclerosis, which is a gradual and progressive process, it makes sense to intervene at a young age to stop or slow down the atherosclerotic process early.
Lifetime CVD models can be used in shared decision-making for individual younger patients. For example, lipid lowering can be considered in younger patients with elevated cholesterol and a high lifetime CVD risk, and thus a high lifetime benefit from treatment. Even when pharmacotherapy is not (yet) considered, lifetime risk estimations can facilitate in communicating the importance of healthy lifestyle and risk factor management at a young age, especially smoking cessation. An alternative method is ‘heart age’ (ie, the chronological age of someone with the same CVD risk score but with normal levels of modifiable risk factors), a method found to have impact on patient motivation for healthier lifestyle42 43; however, this method can not be used to estimate treatment effects.
Apparently healthy older persons
Traditional 10-year risk models are also of limited use in apparently healthy persons at older age (>65 years). First, the relationship between classical risk factors such as lipids and blood pressure with CVD risk attenuates with age. Second, many risk models mathematically do not take into account that older persons are at a high risk of dying of non-cardiovascular causes, after which they are no longer at risk for CVD (the so-called ‘competing risk’ of non-CVD mortality). This can lead to overestimation of the actual CVD risk in older persons.44 Indeed, traditional risk models have been shown to have limited predictive value in older persons.45–47
An older person-specific risk score, adjusted for competing risk of non-CVD mortality is currently available online for the prediction of 10-year CVD risk in patients aged 70 years or older.48 Alternatively, risk algorithms are available that can estimate lifetime CVD risk, remaining CVD-free life expectancy and lifetime treatment benefit in older persons and are also competing risk adjusted.14 15 18 49 As in older persons remaining life expectancy may be limited, the lifetime benefit of treatment in terms of additional CVD-free life expectancy may also be limited.
Moving from prediction of CVD risk to prediction of treatment benefit
Clinical practice and guidelines move towards a more individualised approach in CVD prevention. Especially the availability of novel, but often costly, preventive treatments such as PCSK9-monoclonal antibodies, dual antithrombotic pathway inhibition and anti-inflammatory therapy make identification of patients that would benefit most from (intensified) risk factor treatment and have a low risk of adverse events (such as bleeding) of utmost importance. By combining novel lifetime prediction models with relative treatment effects from RCTs or large meta-analyses, it is possible to estimate treatment benefit expressed as extra CVD-free life in years in individual patients (table 3).19
Estimation of lifetime risk and benefit may have several advantages over 10-year risk predictions. First, the individual lifetime benefit—that is, gain in CVD-free life from, for example, smoking cessation, risk factor treatment or antithrombotic therapy—is easier to interpret for patients than a percentage of absolute risk reduction. This may improve the communication of therapy benefits in a shared decision-making process. As lifetime benefit is an easy to interpret concept, it may also help increase patient motivation to adhere to preventive lifestyle changes and pharmacotherapy.
Furthermore, lifetime benefit may have advantages in younger patients, who generally have a low 10-year risk (and high 10-year NNT) even in the presence of high-risk factor levels and would not qualify for treatment in current guidelines despite having large absolute benefits from risk factor treatment during their lifetime. Older persons generally have a high absolute 10-year CVD risk (and low 10-year NNT), but limited gain from risk factor treatment in terms of extra CVD-free life, as the remaining lifespan is often <10 years, especially in the presence of high risk of competing non-CVD mortality. This shows that the statement ‘the higher the absolute CVD risk, the higher the absolute benefit’ is a simplification of reality. Figure 2 shows an example of the difference between a 10-year risk approach versus a lifetime benefit approach, illustrating how 10-year risk and associated absolute risk reduction from treatment increases with age, while the lifetime benefit decreases.
Patient example: 10-year risk and treatment effects compared with the CVD-free life expectancy and lifetime benefit for a younger versus an older individual. Ten-year predictions were estimated using the SCORE risk model30; lifetime predictions using the LIFE-CVD lifetime model.49 CVD, cardiovascular disease.
Lifetime models exist for apparently healthy persons (LIFE-CVD or JBS-3),16 18 for patients with established CVD (SMART-REACH)14 and for patients with diabetes mellitus (DIAL).15
Future perspectives in CVD risk stratification
As treatment predictions in individual patients become available, it is likely that common practice will move from a risk-based perspective to a more benefit-based approach.
While treatment thresholds exist based on 10-year CVD risk estimations in apparently healthy persons by using SCORE or ASCVD, these still need to be established for risk prediction algorithms for patients with established CVD or diabetes, and for the lifetime risk prediction models. The obvious consequence of a lifetime approach is that patients will be eligible for treatment at a much younger age, thus increasing the average treatment duration. The benefits of this approach need to be weighed against disadvantages and costs, and this should be discussed in future guidelines.
Furthermore, potential benefits from preventive risk factor management always need to be weighed against potential downsides. In shared decision-making, this weighing can include subjective considerations (eg, the burden of having to take daily medication, past experiences, experiences from family or friends), but where available also quantifiable risks of adverse events. For example, estimation of (lifetime) bleeding risk from antithrombotic treatment allows us to estimate and weigh treatment benefit against treatment harm in individual patients.17 50 The Aspirin Guide (http://www.aspiringuide.com) is an online decision support tool that helps in making treatment decisions regarding aspirin use by balancing the ASCVD benefits of aspirin against the risk of harm due to gastrointestinal or other bleeding.51
The easy use of risk and treatment effect prediction algorithms as decision support tools could be largely enhanced in daily clinical practice by incorporating these tools into electronic health records (EHR), so that risk calculators can be automatically filled and predictions automatically presented. Furthermore, EHRs may be an important source of individual-level data, which in future research can be used for machine learning, a technique that could allow for regular updating and recalibration of risk calculators to local practices. Finally, genetic studies have suggested that have lifelong genetic exposure to even small differences in SBP and LDL-c is associated with marked differences in CVD risk, even with a magnitude larger than expected from the combined treatment effects from pharmacological risk factor reduction.52 Future studies combining information from risk models with data from these genetic studies may improve lifetime risk estimation.
Conclusions
The number of treatment options and treatment goals in CVD prevention are increasing, which is exciting but on the other hand makes it increasingly difficult to make treatment decisions for individual patients. Multivariate CVD risk prediction models quantify the CVD risk over a defined period or during a lifetime in (younger and older) apparently healthy persons, patients with established CVD or patients with type 2 diabetes, and may aid in identifying which patients will benefit most from (intensified) risk factor treatment. Predictions of CVD risk thus support informed treatment decisions for preventive treatment and aid in choosing the right treatments and right treatment goals for the right patient. Estimation of individual lifetime benefit of lifestyle changes and risk factor treatment, expressed as gain in CVD-free life, is a new development that may support the process of shared decision-making and may increase patient motivation to adhere to lifestyle changes and medication use. The most-used CVD risk algorithms are freely available as online risk tools.
Key messages
With the plethora of treatment options and treatment targets available for cardiovascular disease (CVD) prevention, it can be difficult to make treatment decisions for individual patients.
Freely online accessible risk stratification tools using predictors that are widely available in clinical practice can easily be integrated into daily clinical practice for risk stratification and can be used a decision support tools for making treatment decisions.
Risk stratification tools are available for apparently healthy persons, older persons, patients with established CVD and patients with diabetes mellitus.
Individualised risk prediction can be used in clinical care by healthcare professionals to support informed treatment decisions for cardiovascular risk factor management, to inform patients about their prognosis and to involve patients in the decision-making process.
Prediction of treatment effects from preventive risk factor management can help in individualised decision making in the prevention of (recurrent) cardiovascular disease, to select the right treatment for the right patient and to improve allocation of healthcare resources.
Estimations of lifetime CVD risk and benefit from treatment may improve patient communication, and can help in patient motivation for lifestyle changes and therapy adherence.
CME credits for Education in Heart
Education in Heart articles are accredited for CME by various providers. To answer the accompanying multiple choice questions (MCQs) and obtain your credits, click on the 'Take the Test' link on the online version of the article. The MCQs are hosted on BMJ Learning. All users must complete a one-time registration on BMJ Learning and subsequently log in on every visit using their username and password to access modules and their CME record. Accreditation is only valid for 2 years from the date of publication. Printable CME certificates are available to users that achieve the minimum pass mark.
Supplemental material
References
Footnotes
Contributors Both authors have contributed to the manuscript according to the ICMJE Recommendations on authorship.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Patient consent for publication Not required.
Provenance and peer review Commissioned; externally peer reviewed.
Data availability statement There are no data in this work
Author note References which include a * are considered to be key references.