837 The Identification of Gaze Behaviour and Physiological Markers Associated With Making An Error During Laparoscopic Cholecystectomy

Abstract

Abstract Introduction Adverse surgical events remain at an unacceptably high level despite multiple global safety initiatives being introduced. As yet, however there is no conclusive evidence to identify whether physiological markers can be used to predict whether a surgeon will make an error Method Surgeons were asked to complete a simulated laparoscopic cholecystectomy task while physiological metrics and gaze behaviour was tracked. LightGBM and CatBoost were used to predict the physiological metric most useful in predicting whether a surgeon was about to make an error. The binary task used a boolean value of “does an error occur in the next 5 seconds” as the dependent variable, while the multiclass task classified the severity of error (0, 1, 2, 3). Results Autocorrelation with lag (eventually calculated with a lag of timestep 2) measured the tendency of this timeseries to correlate with itself. The degree of correlation, or lack of correlation, and sudden changes in correlation over time were gleaned from this feature. Conclusions Skin conductance was most likely to successfully predict impending error. However when gaze features were added, overall model performance improved by 6.4%. The potential for reduction in surgical error rate and improvement in patient safety are important factors to consider.