AI can help cardiologists predict death after TAVR—but there is a catch
Artificial intelligence (AI) can predict when patients face an increased risk of mortality after transcatheter aortic valve replacement (TAVR), even outperforming risk models that have been in use for many years, according to a new analysis published in Frontiers in Cardiovascular Medicine.[1] However, these advanced AI models may still be too complex for many specialists to use on a regular basis.
“In-depth patient data analysis by AI algorithms has the ability to uncover pertinent trends and risk factors that can help forecast post-TAVR death,” wrote first author Faizus Sazzad, MBBS, a cardiac surgeon with the National University of Singapore, and colleagues. “These models can help identify high-risk individuals who might need more measures or more intensive postoperative surveillance. On the other hand, AI models could also identify patients who may not benefit from TAVR, given that the risk of mortality outweighs the benefits of undergoing the procedure.”
Sazzad et al. explored data from ten different studies. Algorithms used in these studies included random forest (RF), gradient boosting (GB), artificial neural network (ANN), multilayer perceptron (MLP) and logistic regression (LR). As the authors explained, RF algorithms include multiple decision trees. GB algorithms, meanwhile, use “weak learner decision trees” that progress over time and learn from one another. ANN algorithms are advanced mathematical models that mimic the human brain, meanwhile, and MLP algorithms are a type of ANN that include an input layer, at least one hidden layer and an output layer. LR algorithms, finally, examines datasets and then predict the probability of a specific variable based on dependent variables.
In total, the area under the ROC curve (AUC) for these algorithms was 0.75 for 30-day mortality and 0.79 for one-year mortality. Traditional risk scores, meanwhile, had an AUC of 0.75 for 30-day mortality and 0.68 for one-year mortality.
“This suggests that AI models have the potential to aid clinicians in predicting post-TAVR mortality in patients and act as an adjunct or alternative to traditional clinical scores,” the authors wrote. “AI models are able to process large amounts of diverse patient data and can be modelled to continuously analyze new data in order to make accurate predictions in real time.”
AI linked to strong performance—but is that enough?
AI’s potential to make a massive impact on patient care is clear, but the authors also emphasized that it may not be all that helpful in certain scenarios.
“Such models require a large number of patient variables to generate predictions, some of which may not be readily available to clinicians in the immediate clinical setting,” the group wrote. “Hence, at present, AI-based prediction models may be less user-friendly than simple traditional risk scores. In the future, further research into simpler AI-based models that are able to use easily-available clinical parameters in predictions is needed to increase the clinical utility of these models.”
Overall, though, the researchers were clear that AI will contribute significant value to clinicians in the years ahead, helping “predict and monitor patient outcomes, which will hopefully result in better decision-making and more desirable post-TAVR outcomes.”
Click here to read the full analysis.
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.