AI predicts LBBB risk in TAVR patients prior to treatment
Artificial intelligence (AI) can help cardiologists anticipate when transcatheter aortic valve replacement (TAVR) may be associated with an elevated risk of left bundle branch block (LBBB), according to new findings published in Future Cardiology.[1] All the AI requires is data that can be gathered before the procedure.
“Despite its benefits, TAVR is frequently linked to conduction disturbances, with new-onset LBBB being the most common complication,” Vasileios Cheilas, MD, MSc, an electrophysiologist with Onassis Cardiac Surgery Center in Athens, Greece, and colleagues. “These disturbances are primarily caused by mechanical trauma to the conduction system, which is situated in close proximity to the aortic valve.”
Only data gathered prior to TAVR were used for this analysis, meaning care teams can understand the risks before the patient is even treated. Cheilas et al. explored data from 469 TAVR patients treated at a single facility. All patients had no prior history of LBBB.
Data from 328 patients were used to build a training set for the study’s various advanced AI models. Data from the remaining 141 patients were used to then test the AI models once they had been properly trained.
The authors tested out a variety of different machine learning techniques, including large language models (LLMs) and more traditional AI algorithms. The accuracy, precision and F1 score for each technique was then calculated.
Overall, XGBoost, a gradient-boosted decision tree machine learning library, was the most accurate approach for predicting a patient’s LBBB risk after TAVR. GPT-4, the popular LLM developed by OpenAI, was also quite accurate when researchers utilized chain-of-thought prompting, which is designed to more closely mimic the way an LLM mimics the reasoning of a person.
“While recognizing the limitations inherent in relying solely on pre-implantation data, this model lays the foundation for potential problem-solving methodologies in medicine through the integration of AI,” the authors wrote. “The findings underscore the significance of incorporating advanced AI techniques, such as LLMs, in predicting medicinal outcomes. Future research in AI and medicine may explore ways to enhance this predictive model, considering additional parameters and refining the model’s adaptability to further improve accuracy and clinical utility.”
Read the full analysis here.
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.