Simply Adjusting for Schedulers’ Bias in Estimated Case Durations Can Accomplish the Same Objectives of Improving Predictions as Use of Machine Learning

To the Editor Strömblad et al prospectively evaluated a machine learning method to predict operating room times of 863 cases using a hospital’s current process; 21% of the performed cases were excluded.¹ Among the 327 cases in the machine learning arm (intervention), the mean absolute error was 50 minutes and the mean error was −4 minutes.¹ Among the 356 cases using the current process (control) arm, the mean absolute error was 59 minutes and the mean error was −34 minutes, the latter being a remarkably large amount of bias in case duration prediction.¹ The authors describe the control method as “…the electronic health record [EHR] standard method for deriving case duration estimates supplemented with the surgeon or scheduler’s estimate.”¹