Machine learning-based ASCVD risk calculator outperforms ACC/AHA standard

A machine learning (ML)-based risk calculator developed to assess an individual’s long-term risk for atherosclerotic cardiovascular disease (ASCVD) identified 13 percent more high-risk patients and recommended unnecessary statin therapy 25 percent less often than standard risk assessment tools in initial tests, researchers reported in the Journal of the American Heart Association.

First author Ioannis A. Kakadiaris, PhD, and colleagues with the Society for Heart Attack Prevention and Eradication (SHAPE) wrote in JAHA that the current gold standard for ASCVD risk assessment—the American College of Cardiology and American Heart Association’s Pooled Cohort Equations Risk Calculator—is flawed in its accuracy.

“Studies have demonstrated that the current U.S. guidelines based on the ACC/AHA risk calculator may underestimate risk of atherosclerotic CVD in certain high-risk individuals, therefore missing opportunities for intensive therapy and preventing CVD events,” Kakadiaris and coauthors said. “Similarly, the guidelines may overestimate risk in low-risk populations, resulting in unnecessary statin therapy. The existing approach to CVD risk assessment desperately needs an overhaul.”

According to a consensus report from SHAPE, comprehensive ASCVD risk assessment should include evaluation of plaque, blood and myocardial vulnerability factors if it’s going to be anywhere near accurate.

“Clearly, the available screening and diagnostic methods are insufficient to identify the victims before the event occurs; therefore, short-term risk prediction is needed,” the authors wrote. “To reach the goal, a stepwise multi-phase approach is warranted that includes maximizing the long-term predictive value of traditional risk factors using machine learning, gathering unique data on asymptomatic patients who, shortly after an exam with blood testing, experience an ASCVD event, and applying ML to all available clinical data, including genomic, proteomic and others.”

Kakadiaris’ team developed an ML risk calculator based on support vector machines using 13-year follow-up data from the Multi-Ethnic Study of Atherosclerosis (MESA) cohort. The dataset included 6,459 individuals who were ASCVD-free at the study’s baseline.

The researchers input the same patient data into the ACC/AHA risk calculator and their own ML calculator, both of which used the same risk variables. After validating the model in an external cohort—the Flemish FLEMENGHO registry—they found their tool detected more potential ASCVD sufferers and recommended unnecessary drug therapy for fewer low-risk patients.

Based on a 7.5 percent 10-year risk threshold, the ACC/AHA risk calculator recommended statin therapy to 46 percent of patients. But 23.8 percent of the 480 “hard” CVD events, which included MI, stroke, stroke death and fatal coronary heart disease, occurred in participants who weren’t recommended for statins. That translated to 0.76 sensitivity and 0.56 specificity in the ACC/AHA model.

In contrast, the ML risk calculator recommended statins to 11.4 percent of patients, and 14.4 percent of hard CVD events took place in those not recommended for statin therapy, resulting in a sensitivity of 0.86 and a specificity of 0.95 for the ML tool.

Kakadiaris et al. said that overall, machine learning-based prediction models will always be more “versatile and capable” than statistical ones.

“We attribute the superior performance of our ML model to its flexibility and nonlinear function,” they wrote. “ML maps the data into a multidimensional space where various separating planes are evaluated and ultimately a ‘hyperplane’ is found. Additionally, the ability to train the ML model with artificially created events using data augmentation techniques such as NEATER can further empower ML over the traditional statistical methods.”

Kakadiaris and colleagues said further studies are underway to validate and expand their risk calculator.

“As we introduce more data to our ML risk calculator, particularly to cases in which events occurred weeks or months following data collection instead of years or decades, the ‘holy grail’ of short-term CVD risk prediction may be within our reach,” they wrote.