Algorithms used to estimate risk of heart problems are performing poorly when applied to new populations, resulting in substantial risk of harm, according to a study published in Circulation April 5.
The study, led by researchers at Medford, Mass.-based Tufts University, looked at how well clinical prediction models, used to guide patient care, perform on new populations, as well as how likely these models are to improve clinical decision making.
Out of 158 external validations of 104 clinical prediction models across three domains of cardiovascular diseases, the researchers found that more than 100 clinical prediction models used in cardiovascular care had the potential to motivate harmful clinical decisions.
The researchers suggested that model updating can mitigate the risks associated with the algorithms, but usage of clinical prediction models still needs better evaluation before deployment and monitoring afterward.