New Machine Learning Models Help Predict Heart Disease Risk in Women

The commonly used Framingham Risk Score, which predicts the likelihood of developing cardiovascular disease within the next 10 years, is based on specific criteria including age, sex, cholesterol levels, and blood pressure. However, this does not account for anatomical differences between sexes; female hearts, for example, are typically smaller and have thinner walls. Consequently, using the same diagnostic standards for both sexes means that women's hearts need to increase disproportionally more than men’s to meet the same risk criteria. A team of researchers has now built more accurate cardiovascular risk models than the Framingham Risk Score using a large dataset and have also quantified the underdiagnosis of women compared to men.

Researchers at Stanford University (Stanford, CA, USA) quantified the underdiagnosis of women compared to men and found that the use of sex-neutral criteria results in significant underdiagnosis of female patients. To achieve more accurate predictions for both sexes, they incorporated four additional metrics absent in the Framingham Risk Score: cardiac magnetic resonance imaging, pulse wave analysis, EKGs, and carotid ultrasounds. Utilizing data from over 20,000 individuals in the UK Biobank—a comprehensive biomedical database of around half a million UK residents aged 40 and over—they applied machine learning techniques. They found that EKGs were particularly effective in enhancing cardiovascular disease detection in both sexes. Despite this, traditional risk factors remain valuable for assessing risk, according to the researchers.

This study marks the first step towards reevaluating risk factors for heart disease by incorporating advanced technologies to improve risk prediction. Nevertheless, the study faces limitations that future research should address. One such limitation is the binary treatment of sex in the UK Biobank, ignoring the complex nature of sex involving hormones, chromosomes, and physical traits that may not fit neatly into 'male' or 'female' categories. Moreover, the study's focus on middle-aged and older UK residents may limit the applicability of the findings to other demographic groups and geographical locations.

“We found that that sex-neutral criteria fail to diagnose women adequately. If sex-specific criteria were used, this underdiagnosis would be less severe,” said Skyler St. Pierre, a researcher at Stanford University's Living Matter Lab. “We also found the best exam to improve detection of cardiovascular disease in both men and women is the electrocardiogram (EKG).”

“While traditional clinical models are easy to use, we can now use machine learning to comb through thousands of other possible factors to find new, meaningful features that could significantly improve early detection of disease,” added St. Pierre.

Related Links:
Stanford University