It is indisputable that clinical medicine has entered the age of big data. Newer, better prediction methods, such as neural networks, random forests, and other algorithms, often categorized as machine learning (ML), can probe large-scale clinical data sets to discover predictive features unavailable to their more traditional counterparts. However, when Khera et al1 pit 3 of these algorithms against the most standard generalized linear model, logistic regression, to predict death after acute myocardial infarction, none of the ML algorithms emerge as a clear winner. Two of the 3 ML algorithms improved discrimination by a slim margin and yielded “more precise calibration across the risk spectrum.”1 However, these improvements are unlikely to be clinically meaningful, and it’s unclear whether they would be sufficient to justify the corresponding loss of interpretability. Furthermore, 1 ML approach (a neural network) performed worse than logistic regression. The data set in question is undeniably big, at least in sample size. Khera et al1 draw on an American College of Cardiology registry that contains more than 750 000 records; therefore, at first glance, it appears that the promise of using ML to harness big data is not being realized. What can explain this disconnect, and does it suggest that ML is more hype than substance?
Engelhard MM, Navar AM, Pencina MJ. Incremental Benefits of Machine Learning—When Do We Need a Better Mousetrap? JAMA Cardiol. 2021;6(6):621–623. doi:10.1001/jamacardio.2021.0139
Artificial Intelligence Resource Center
Customize your JAMA Network experience by selecting one or more topics from the list below.