External Validation of Prediction Algorithm Using Non-invasive Monitoring Device for Intraoperative Hypotension

The hypotension that occurs during surgery is associated with the poor prognosis of patients after surgery. Previous studies have reported that even a short period of time of hypotension increases the risk of postoperative complications such as kidney injury. If anesthesiologists can predict intraoperative hypotension in advance, they can prevent or minimize the damage.

Recently, there are many reports on medical artificial intelligence models that predict the intraoperative hypotension. Among them, the Hypotension Prediction Index (HPI) model has already been commercialized and used in clinical practice. However, HPI has limitations in that it is necessary to perform invasive techniques (arterial cannulation) or to use dedicated equipment at high cost. However, since many of the general anesthesia are performed without invasive monitoring devices, the use of HPI medical devices is subject to considerable restrictions.

The investigators have reported the prediction algorithm for intraoperative hypotension using five non-invasive monitoring devices commonly used in general anesthesia: 1) blood pressure (NBP, number), 2) electrocardiogram (ECG, waveform), 3) end-oxygen saturation waveform (PPG, waveform), 4) end-stage carbon dioxide waveform (ETCO2, waveform), and 5) an anesthesia depth (BIS, number) By conducting a retrospective external validation process using public clinical data from other institutions (tertiary hospital in Korea), the final model was able to have good predictability with an Area Under the Receiver-Operating Characteristic Curve (AUROC) value of 0.917.

However, investigators did not externally validate that algorithm through a prospective designed study. This study intends to externally validate the "hypertension prediction model during surgery using non-invasive monitoring device", which has already reported It is expected that the usefulness and limitations of the prediction model can be evaluated again, and the model can be advanced based on the results.