January 11-14, 2023
CA-1 Resident Physician, Danny Kim, MD, MS, PhD candidate, and Chair, Maxime Cannesson, MD, PhD, presented two oral and poster presentations at the Society for Technology in Anesthesia (STA) 2023 Annual Meeting in Las Vegas, NV. Anesthesia researchers shared their findings and developments on topics ranging from advances in machine learning to supply chain impacts on healthcare in an effort to improve technological practices in anesthesia.
Prediction of intraoperative hypotension is critical for anesthesiologists. In this research, Dr. Kim and his research group applied a deep neural network (DNN) using continuous physiologic waveforms to predict upcoming hypotension in advance. In the prediction of 10 minutes in advance of intraoperative hypotension, their DNN model achieved 26% higher performance in the hypotension, compared to their previous works applying a logistic regression-based machine learning approach (Area Under Curve increased from 0.76 to 0.86).
Possible racial bias in the accuracy of oxygen saturation measured by pulse oximetry (SpO2) has obtained public attention recently, with previous research showing less accurate measurements for African American patients than for Caucasian patients. In Dr. Kim's research, the transfer learning-based DNN is trained by the Caucasian group first, then a fine-tuning process adjusts the network's accuracy for the African American group. Results showed that this algorithm improved 23% of the accuracy of SpO2 reading in the African American group.