Automatic Updating of Times Remaining in Surgical Cases Using Bayesian Analysis of Historical Case Duration Data

BACKGROUND: Operating room (OR) whiteboards (status displays) communicate times remaining for ongoing cases to perioperative stakeholders (e.g., postanesthesia care unit, anesthesiologists, holding area, and control desks). Usually, scheduled end times are shown for each OR. However, these displays are inaccurate for predicting the time that remains in a case. Once a case scheduled for 2 h has been on-going for 1.5 h, the median time remaining is not 0.5 h but longer, and the amount longer differs among procedures. METHODS: We derived the conditional Bayesian lower prediction bound of a case's duration, conditional on the minutes of elapsed OR time. Our derivations make use of the posterior predictive distribution of OR times following an exponential of a scaled Student t distribution that depends on the scheduled OR time and several parameters calculated from historical case duration data. The statistical method was implemented using Structured Query Language (SQL) running on the anesthesia information management system (AIMS) database server. In addition, AIMS workstations were sent instant messages displaying a pop-up dialog box asking for anesthesia providers' estimates for remaining times. The dialogs caused negotiated interruptions (i.e., the anesthesia provider could reply immediately, keep the dialog displayed, or defer response). There were no announcements, education, or efforts to promote buy-in. RESULTS: After a case had been in the OR longer than scheduled, the median remaining OR time for the case changes little over time (e.g., 35 min left at 2:30 pm and also at 3:00 pm while the case was still on-going).