Editorial 

May 13, 2024

Can Cardiovascular Risk Assessment Be Improved in the 21st Century?

Thomas A. Gaziano, J. Michael Gaziano

JAMA. Published online May 13, 2024. doi:10.1001/jama.2024.7644

Cardiovascular risk assessment has become an essential part of preventive strategies designed to target risk factor interventions and has contributed to dramatic reductions in cardiovascular disease mortality during the past 60 years in the US and other countries. Beginning in the 1960s, investigators from the Framingham Heart study^1,2 identified key physiological, behavioral, and biochemical risk factors for atherosclerotic cardiovascular disease and were the first to develop multivariable risk equations using major risk factors to predict first-time fatal and nonfatal myocardial infarction and later atherosclerotic cardiovascular disease events during a period of up to 12 years.³ These risk equations are able to explain about 70% to 80% of the variability of cardiovascular risk in the populations in which they were derived.⁴ The generalizability of these early tools has been enhanced by using large, diverse, pooled cohorts in the US (American Heart Association/American College of Cardiology Pooled Cohort Risk Equation),⁵ Europe (European Society of Cardiology Systematic Coronary Risk Evaluation 2 [SCORE2]),⁶ and elsewhere. Since their development, many attempts to improve the models have been pursued, yet the basic models have stood the test of time.

In this issue of JAMA, Neumann et al⁷ evaluate the incremental utility of several clinically available biomarkers when added to a standard risk equation. The study explores whether troponins, natriuretic peptides, and high-sensitivity C-reactive protein enhance the intermediate and long-term prediction of atherosclerotic cardiovascular disease events, heart failure, and cardiovascular and all-cause mortality when added to a standard prediction model. The authors compiled data from 28 cohorts,⁷ increasing the power to provide stable estimates for the associations with many outcomes.

For each analysis, there were tens of thousands of individuals to more than 100 000 individuals included in the specific biomarker analyses. Each of the biomarkers was associated with the cardiovascular outcomes studied and, when added to the basic prediction tool, they demonstrated significant, but modest, incremental improvements in the discriminatory power with the changes generally in the second or third decimal place of the C statistic. The model improved the C statistic from 0.81 to 0.82 for 10-year incident atherosclerotic cardiovascular disease when 3 biomarkers (high-sensitivity cardiac troponin I, N-terminal pro-B-type natriuretic peptide, and high-sensitivity C-reactive protein) were simultaneously included, and there was little decline in the C statistic over time from 1 year to 5 years to 10 years.⁷ There were also improvements in reclassification of risk by adding the biomarkers. This study explored a variety of outcomes, and these models improved prediction of not only fatal and nonfatal atherosclerotic cardiovascular disease events, but also heart failure and importantly all-cause mortality.

One major advantage of this analysis⁷ is the collection of individual-level data from many cohorts from around the globe spanning 4 continents. Another advantage of this study is the excellent use of existing cohorts with long follow-up averaging about 12 years and as long as 28 years. The median risk in the population was about 5% over 10 years, making this a useful pooled cohort for modeling risk in the range where aggressive lipid and blood pressure management is warranted. For example, aggressive lipid management is generally recommended at a 10-year risk of about 7.5%. There was good representation of individuals older than 65 years; however, it would be useful to use these data to look further at age groups older than the age stratification of 65 years. More data among those older than 75 years or 80 years are needed. Even though the pooled data are impressive in scale, they are small compared with the new generation of mega-cohorts, such as the UK Biobank,⁸ the Million Veteran Program,⁹ and All of Us.¹⁰ Notably, in the current study, key ancestry groups are still underrepresented.

A major shortcoming of this and other similar studies is that risk is assessed at only a single point in time. In the study by Neumann et al,⁷ the single measure was used to predict risk over as long as 28 years of follow-up. This is in stark contrast to the reality of the clinical setting where there are opportunities to assess risk repeatedly over time. In the clinic, risk is not assessed just once and then the patient is followed up for the next 28 years. If the assessment is not quite perfect in the classification of risk at one moment, all is not lost. That patient will be back many times over the years, providing many opportunities to assess risk and determine risk factor modification strategies. Repeated measurement and following the trajectory of risk is another strategy to improve the accuracy of risk assessment. This clinical reality needs to be incorporated into research studies on risk assessment to enhance their clinical relevance.

Another consistent challenge with the current study and previous studies is the approach to reclassification. The categorical and continuous reclassification measures explored reclassification across the full range of risk in the population. However, reclassification is clinically relevant around a single decision boundary, usually at an intermediate risk level used, for example, to determine whether to start treatment with a statin or how aggressively to manage blood pressure. Reclassification at the higher or lower ends of the risk spectrum does not generally change clinical decision-making. Therefore, rather than assessing reclassification into 4 risk categories around 3 risk boundaries or continuously across the full risk range, assessment at a single decision boundary is needed. This approach to reclassification is more relevant to clinical decision-making and would certainly decrease the number of improvements in reclassification observed because there is more opportunity for reclassification with more risk boundaries.

The study by Neumann et al⁷ demonstrates statistically significant improvements in the C statistic and reclassification that are consistent with previous similar studies. Are these incremental improvements clinically useful? How do they compare with other ways to enhance risk prediction and classification? To answer these questions, it is necessary to consider other ways that risk prediction might be improved. Other biomarkers have demonstrated similar improvements. Measures of kidney function have been shown to improve risk prediction, and measures of kidney function are readily available for many patients.

It is surprising that Neumann et al⁷ did not add the widely available marker of estimated glomerular filtration rate to their models before adding the markers that would have to be newly measured in most patients for primary prevention. The use of estimated glomerular filtration rate in the recently published Predicting Risk of Cardiovascular Events (PREVENT) equations marginally improved the C statistic compared with the Pooled Cohort Equations.¹¹ Many other biomarkers including various lipid and inflammatory markers, gene variants, polygenic risk scores, and proteomic scores, among others, have demonstrated similar modest improvements in prediction as well. In addition to biomarkers, risk prediction can be aided by other parameters including physical measures such as ankle-brachial index, exercise testing, and imaging. Lin et al,¹² in a review for the US Preventive Services Task Force, demonstrated greater improvements in the C statistic with coronary artery calcium scores and ankle-brachial index than other biomarkers when added to traditional models.

With so many options for improvement in risk assessment, how do clinicians or guideline committees decide which ones to use or recommend? The currently available risk prediction equations are handy because the data elements are usually readily available in the clinical setting and can easily be calculated from available elements in the electronic medical record. Any additional parameters should also be available, feasible, and worth the cost for the added value. Cost-effectiveness is often lacking for the incremental improvement in risk prediction. Rather than applying any novel measure to the entire population, another way to improve risk prediction is to use additional measures as secondary screening tools only when there is a need, specifically for those who are close to a level of risk used for clinical decision-making. For example, coronary calcium, ankle-brachial index, or other biomarkers may be most useful for those at an intermediate level of risk and reclassification could be helpful in determining who should have more aggressive therapy or additional testing. This would likely greatly enhance the efficiency and cost-effectiveness of using any additional measure because it would be used for the small fraction of the population in whom the benefit would be greatest.

In the clinical setting, risk assessment is relatively straight forward. First, identify if a patient has any atherosclerotic cardiovascular disease, in which case the most aggressive risk factor management is needed. If diabetes is present, risk scores may be less important given the already high risk associated with this condition. Age older than 75 years also generally confers sufficient atherosclerotic cardiovascular disease risk to consider aggressive risk factor modification. In this case, a frailty index may be useful in making intermediate- and longer-term preventive decisions.¹³

For the rest of the general population, the tried-and-true standard risk tool may be sufficient as an initial screening to identify a primary prevention strategy, focusing aggressive risk factor management in those at highest risk and beginning with lifestyle counseling for those at lower risk. For those who are near an important clinical decision boundary, a secondary screening instrument can be used. Until more comparative cost-effectiveness data are available, a biomarker that may already be available, such as estimated glomerular filtration rate¹⁴ derived from creatinine, could be used before turning to biomarkers that need to be measured anew. One can also consider ankle-brachial index or coronary calcium in selected patients as other secondary screening tools. Risk scores can also be repeated and a patient followed up over time to improve understanding of a particular patient’s risk trajectory.

In summary, even though the study by Neumann et al⁷ demonstrates that clinically available biomarkers improved risk prediction when added to a standard atherosclerotic cardiovascular disease prediction tool, the magnitude of that improvement, like in previous biomarker studies, is modest. Further study regarding the discrimination in risk with these biomarkers is needed in other populations beyond the predominantly US and European populations used for this study. The cost and access to the biomarkers in low-resource settings will need to be overcome before wider use could be considered. The cost-effectiveness and clinical utility of this strategy are still uncertain. Future studies of risk prediction should consider how risk assessment is used in the clinical setting rather than the artificial research setting. This general approach should be taken for assessment of all novel predictive measures and biomarkers, including gene variants and proteomic scores.