Algorithm enables continuous forecasting of intraoperative blood pressure

Editor's Note

Researchers at the Medical University of Vienna developed a new method using the Temporal Fusion Transformer (TFT) model to predict intraoperative hypotension in patients under general anesthesia. According to findings published August 30 in eClinical Medicine, part of The Lancet, the model utilizes routine vital sign data, including heart rate, blood pressure, and intraoperative medication information, to forecast mean arterial pressure (MAP) up to 7 minutes in advance.

Tested on over 80,000 anesthesia cases, the model achieved a low prediction error of 4 mmHg in internal tests and 7 mmHg in external validation, researchers write. The model demonstrated strong accuracy in predicting episodes of hypotension (MAP < 65 mmHg) with an AUROC score above 0.9. By forecasting entire MAP trajectories, the model provides anesthesiologists with early warnings of hypotension, allowing them to intervene preemptively to prevent outcomes such as myocardial or kidney injury.

The model requires only low-resolution data, making it accessible for use in routine surgical settings without the need for invasive monitoring such as arterial pressure measurement, researchers write. Future studies will explore how this approach can be incorporated into clinical practice to enhance patient care.