More than 11 million people may need to reconsider taking medications to avoid heart attack and stroke, according to new research that says current guidelines overestimate risk for some people, but underestimate risk for others, especially African-Americans.
Right now, doctors can consult a calculator found online or in electronic health records to decide whether patients might benefit from aspirin, statins, or blood pressure medications. Those estimates of 10-year risk for cardiovascular disease were derived in 2013 and endorsed by the American College of Cardiology and the American Heart Association. They were based on statistical analyses that combined data from large studies such as the original Framingham Heart Study, whose participants were 30 to 62 years old in 1948.
A team from Stanford has created a new calculator by updating data sources — adding the more recent Jackson Heart Study and the Multi-Ethnic Study of Atherosclerosis, among others — and applying newer statistical methods. They say their findings, published Monday in Annals of Internal Medicine, improve the accuracy of risk estimates among multiple populations. They also predict that their calculator, and others like them, will also need to be updated with changing times.
“I think this model is a move in the right direction” compared to the 2013 calculator, said Nancy Cook, a Harvard Medical School professor of medicine at Brigham and Women’s Hospital in Boston who studies and develops risk models, including the Reynolds Risk Score. “I think it’s a more accurate model, it calibrated better, and the estimated risk fit the observed risk better.”
Dr. Sanjay Basu, assistant professor of primary care outcomes research at Stanford and senior author of the study, said he was motivated to investigate the current guidelines because they didn’t fit the patients in front of him, particularly if they were African-American. One of his African-American patients, for instance, a man who had high cholesterol and smoked tobacco, was classified as low risk.
That seemed “bizarre” to him. The explanation? Because there were few African-Americans included in the studies chosen for the 2013 analysis, statistical methods generalized conclusions for that small group of people across the whole data set. That approach can lead to random errors called “overfitting,” which overgeneralizes from small populations — leading to unlikely results like the one he saw with his patient, Basu said.
For their analysis, Basu and his Stanford colleagues applied newer statistical tools derived from machine learning and not often used in epidemiology: They usually predict how consumers respond to advertising or classify YouTube videos of dogs versus wolves, for example. Basu said the goal is the same: to distinguish subtle characteristics and relationships, in this case, clinical measures that lead to a higher risk for heart attack and stroke.
To test their methods, the researchers created two models. In the first one, they updated the cohorts they used. The populations were more diverse racially and ethnically, and they also reflected factors that can’t necessarily be captured by blood tests. Nutrition, physical activity, or even secondhand smoke — which wasn’t recognized in the 1940s — can affect a person’s baseline risk for disease. That’s why risk calculators don’t stay accurate forever.
Updating the cohorts alone didn’t make much difference in the risk estimates. But in their second model, they used updated cohorts along with the newer statistical methods. That showed substantial improvement.
How did they know their risk estimates were better? They compared their second model of who was at risk for a heart attack or stroke to data from 10 years later, showing whether those people actually did suffer a heart attack or stroke.
The second model correctly reclassified 13 people as low risk for every one person it incorrectly reclassified. They defined “high risk” as a predicted 7.5 percent chance of heart attack or stroke over the next 10 years.
Among African-Americans, the researchers found, their new calculator removed the “bizarre” estimates Basu had noticed in his primary care practice. The 2013 calculator made mistakes in both directions for African-Americans with the same blood pressure, cholesterol, age, gender, and smoking and diabetes status as whites. Sometimes it predicted they were at 70 percent less risk of heart attack or stroke than whites; other times it found a 250 percent increased risk — both of which are biologically unlikely, Basu said.
In their new model, fewer than 1 percent of African-Americans had such implausibly different risk estimates compared to white people with the same characteristics.
While calling the new work statistically sound, Dr. Andrew DeFilippis and Patrick Trainor of the University of Louisville also asked how race might change the way risk factors contribute to heart attack or stroke. In other words, there might be genuine differences in risk according to race, but the new calculator doesn’t account for those, they said.
“This question is timely because a growing Hispanic population and 40 percent increase in Asian-Americans account for more than half of the U. S. population growth between 2000 and 2019,” they wrote in an editorial published with the Stanford study. “We clearly need an accurate risk assessment tool for these growing American populations.”
So what should patients and their doctors do now?
First, the study needs to be repeated and evaluated by independent researchers, and then tested in clinical settings, Basu said. Even so, there won’t be a one-size-fits-all decision on treatment.
“We found that a person’s risk might be higher than previously believed and in other cases might be much lower,” Basu said. “While the choice of therapy is a personal decision between the patient and a doctor based on a number of factors, that decision and that discussion could be informed by much more accurate risk calculations.”
Dr. Donald Lloyd-Jones, chairman of the department of preventive medicine at Northwestern University and co-chair of the task force that created the 2013 risk guidelines, said the Stanford researchers had “applied some very interesting and new methods about how we can improve risk equations. And this is what should happen.”
In his view, any risk equation is just a starting point if a patient’s score is ambiguous. That’s when another test is called for, he said, such as a CT scan to reveal coronary calcium levels before prescribing a statin
“The risk equation is there to start a conversation, not to make a definitive decision,” Lloyd-Jones said.
The study was funded by the National Institutes of Health, which has posted the statistical code, the data sets, and the calculator itself.
This story has been updated to add comments from Dr. Donald Lloyd-Jones.
Leave a Reply