Anne van Dalen

222 I Chapter 8 Procedure Duration Prediction Due to the high density and non-singularity of information in a video stream, extracting its data for evaluation purposes is a challenging process. In comparison to the video stream, data from surgical instruments provide information that is easier to quantify. Whether or not such data provide sufficient information to make presumptive predictions on surgery duration is uncertain to date. Bodenstedt et al. 22 proposed and compared methods, based on CNNs to predict procedure duration based on data from surgical devices or video streams. The input was acquired from 80 recorded laparoscopic interventions of which the necessary data were available. Overall, the combined method (both video and surgical device data) performed best with an average error of 37% and an average halftime error of approximately 28%. This is an improvement to the baseline method with an average error and average halftime error of both 124%. 22 Zhao et al. sought to accurately predict procedure duration of robot-assisted surgery cases using multiple machine learning (ML) models, using case characteristics (scheduled duration, age, gender, and comorbidities of the patient, tumor location, month of year, time of day, day of the week etc.) as data input. They compared the ML models to the baseline model, which is the time scheduled for the procedure determined by former case duration averages and changes by the surgeon. The following ML models were used: (1) multivariable linear regression, (2) ridge regression, (3) lasso regression, (4) random forest, (5) boosted regression tree, and (6) ANNs. The average root-mean-squared error (RMSE), a measure for the imperfection of the fit of the estimator to the data, was lower for all the ML models than the baseline model. The average RMSE was lowest with the boosted regression tree (80.2 minutes, 95% confidence interval 74.0–86.4), which was significantly lower than the baseline model (100.4 minutes, 95% Confidence interval 90.5–110.3). The use of a boosted regression tree, a predictive modeling approach used in ML, increased the amount of correctly booked procedures from 148 to 219 (34.9% to 51.7%, P <.001). 28 Devi et al. researched several techniques to estimate procedure duration in an ophthalmology department by taking the surgical environment into account (experience of surgeon in years, experience of anesthetists in years, type of anesthesia, etc.).