New Equation May Better Estimate Cardiovascular Disease Risk

June 06, 2018

A new equation for estimating future risk for cardiovascular events appears to give more accurate results than does the existing equation currently recommended — and this has major implications for selecting patients eligible for statins, aspirin, and antihypertensive therapy.

The new equation is described in a paper published online June 5 in Annals of Internal Medicine.

The authors say that their equation could substantially improve the accuracy of cardiovascular risk estimates and reduce the overestimation of risk among modern populations and the problem of "extreme (and highly implausible) risk estimates" for black adults that occur with the current risk calculator.

They estimate that with use of this equation, approximately 11.8 million US adults previously labeled high risk by the current risk calculator would be relabeled lower risk, substantially reducing the number of people recommended for treatment with statins, aspirin, or antihypertensive agents.

"We prescribe statins, aspirin, and now also antihypertensive medication on the basis of the patient's cardiovascular risk, so it is very important to have an accurate assessment of this risk," senior author Sanjay Basu, MD, Stanford University, California, explained to Medscape Medical News.

"Many physicians are not confident in the accuracy of the risk calculator currently recommended by the ACC/AHA [American College of Cardiology/American Heart Association] and US Preventive Services Taskforce, which was released in the 2013 ACC/AHA lipid guidelines."  

Basu said he first became aware of the inadequacy of the 2013 risk equation in his own clinical practice. "I was worried about an increased risk of stroke in an older African American man who had high cholesterol and was a smoker, but when I did the risk assessment using the 2013 calculator, it gave a very low risk estimate, which seemed totally inappropriate.

"Many others have also reported that this calculator seems to give estimates that are way too high or low for some patients," he added. "It is believed that it mainly overestimates risk but in some cases it can underestimate risk."

Basu and colleagues tried to update the risk calculator using more recent cohort data and more sophisticated statistical methods.

Their new models suggested that the 2013 calculator overestimated 10-year cardiovascular risk by an average of 20% across risk groups. Misestimation of risk was particularly prominent among black adults, of whom 33% had extreme risk estimates (<70% or >250% those of white adults with otherwise-identical risk factor values).

Basu says many African American people will have an increased risk for cardiovascular disease with the new calculator and older white adults will more likely have a lower estimate of risk compared with the 2013 calculator.

A "Major Step Forward"

The new equation has been welcomed by outside observers. Steve Nissen, MD, Cleveland Clinic, Ohio, told Medscape Medical News that the 2013 risk calculator was rolled out in the guidelines without being previously published, peer reviewed, or investigated by outside groups.

"Multiple studies have been published showing it to be highly inaccurate," Nissen said. "Here we are, 5 years after the 2013 ACC/AHA lipid guidelines were released and we are still talking about the fact that they are based on a flawed risk calculator."

"This new risk calculator is an attempt to fix the situation," he said. "That is a major step forward. Dr Basu and colleagues have taken a responsible approach of publishing their new equation and asking others to validate it.  That is the process that should have been used before."

Coauthor of an accompanying editorial, Andrew Paul DeFilippis, MD, University of Louisville, Kentucky, previously published a paper highlighting the failings of the 2013 risk calculator.

"Dr Basu et al's new risk calculator found a significant improvement in risk score," DeFilippis said. "This is very hopeful, although it must be validated in independent cohorts."

To develop the new risk equation, Basu and colleagues used six different datasets that included a total of 26,689 people aged 40 to 79 years without prior cardiovascular disease who were followed for more than 10 years between 1983 and 2014.

The population was divided randomly into two cohorts: one used to derive the risk equation and the other used to validate the equation.

The researchers used two models to develop new equations. The first just used updated cohort data without new statistical techniques, but this model did not find much of an improvement over the 2013 equation. The second version included both more recent datasets and a new statistical model known as an "elastic net," which uses a machine learning method that takes samples from cohorts, learns from errors, are updates itself.

Basu compared the statistical methods used to formulate the new risk calculators to technology used in new dictation software.

"When it first came out, it was an advance, but it would get more than half the words wrong. But the newer version learns and adapts to a person's voice," he said. "In the same way, the statistical methods used to develop this new risk calculator learn from each new dataset analyzed. This sort of software is often applied to Internet problems —  we adapted it to a medical context."  

Basu stressed that the equation needs to be further validated in other populations.

"We used publicly available datasets. Before the equation gets adopted into clinical practice, I would recommend that others validate it in their own datasets — this could include private healthcare organizations using their private datasets."

In their editorial, DeFilippis and his coauthor, Patrick Trainor, MS, also from University of Louisville, note that compared to the 2013 equation the new equation correctly reclassified 13 persons who did not have a cardiovascular event as low risk (10-year risk for a cardiovascular event < 7.5%) for each 1 person who was reclassified as low risk but went on to have an event.

"So we would be saving exposure to unnecessary chronic drug therapy in 13 out of 14 patients at the expense of missing 1 out 14 who would go on to have an event," DeFilippis told Medscape Medical News. "Predicting the future is hard, but I do think that is a substantial improvement in risk prediction."  

At a more conservative definition of "low risk" (a 10-year risk < 5%), the ratio improved to 23:1, whereas at a more liberal definition (10-year risk < 10%), the ratio was 8:1, the editorial notes.

DeFilippis points out that Basu and colleagues used populations from the same datasets to derive and validate their equation.

"It would be better to have used a completely different population for validation. But this is a good first step and others can now try and confirm the results in different groups. It is very important to create a risk calculator that holds up in many different populations," he said.

"The committee working on the new ACC/AHA lipid guidelines will be looking around for a new risk score to use with the new guidelines," he added. "Several other new proposed risk equations have been developed, and we now have to wait and see which ones are validated further." 

The study was funded by the National Institutes of Health. Basu has disclosed no relevant financial relationships. Disclosures for coauthors appear in the paper. DeFilippis reports grants from the National Institute of General Medical Sciences and National Institutes of Health. 

Ann Intern Med. Published online June 5, 2018.  Abstract, Editorial

For more Medscape Neurology news, join us on Facebook and Twitter


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.