Prediction of intraoperative hypotension during Caesarean delivery with deep learning models from intraoperative non-invasive monitor

During Caesarean sections (CS), hypotension is a common complication that can lead to adverse fetal outcomes. Early prediction and intervention can help prevent intraoperative hypotension, but there is currently no available method to accurately predict hypotension. This study aims to predict intraoperative hypotension during CS with deep learning-based models from intraoperative non-invasive hemodynamic data which are routinely monitored. A multicentre retrospective study was conducted on pregnant women who underwent CS under spinal anesthesia: (1) Seoul National University Hospital (SNUH, n = 987) and (2) Boramae Medical Center (BMC, n = 180). Intraoperative hypotension was defined as a mean arterial pressure of less than 65mm Hg at any point during surgery. A deep learning-based prediction models (DLPM) for intraoperative hypotension was developed, using preoperative variables and non-invasive hemodynamic monitoring data. The prediction model calculated the risk of intraoperative hypotension after 1 and 5 minutes in real-time. In SNUH, 39,239 and 37,852 were extracted at 1 and 5 minutes prior to the event, respectively, and the incidence of hypotension for each segment was 8.54% and 8.77%. In BMC, the incidence of hypotension was found to be 3.58% and 3.56% from 5,643 and 5,614 segments analysed at 1 and 5 minutes prior to the event, respectively. The DLPM successfully predicted the risk of hypotension at 1 and 5 minutes prior to occurrence, with areas under the receiver operating characteristic curve of 0.80 and 0.74 in SNUH internal validation, and 0.76 and 0.72 in BMC external validation set, respectively. The deep learning-based algorithm has been successful in predicting intraoperative hypotension during CS and might be used for early intervention. However, further studies are required to explore the clinical usefulness of prediction model for prevention of adverse outcomes.