More accurate, unbiased predictions of operating room times increase labor productivity with the same staff scheduling provided allocated hours are increased

Abstract

BackgroundSuppose that before surgery starts predictions of cases’ operating room (OR) times are biased. Summing among cases in ORs, bias causes over-utilized or under-utilized time, reducing the efficiency of use of OR time and labor productivity. Correcting for such bias requires just arithmetic. We focus on benefits of improving the accuracy of unbiased predictors of OR times. Globally, for small or large surgical suites, can machine learning and other computational (e.g., Markov-Chain Monte-Carlo) methods for predicting OR times increase labor productivity? MethodsDiscrete-event simulations were performed using a realistic but inherently non-explanatory OR model. Then an analytical model of one OR was used with normally distributed times per case. ResultsDiscrete-event simulation results were consistent quantitatively with the several earlier empirical studies. When mean absolute predictive errors of cases’ OR times were decreased while retaining no bias, there was a negligible but significant reduction in productivity. The analytical model showed that there could equally have been negligible but significant increases in productivity. Managers and clinicians should have no expectation that implementing machine learning software to decrease mean absolute predictive errors will increase productivity if without increases to allocated times (i.e., longer durations of the workday into which cases are scheduled). Suppose that allocated times are increased while maintaining the same (longer) staff shift length (e.g., increasing the hours into which cases are scheduled from 10.5 to 11.5 h while staff continue to work 12 h shifts with <10% risk working late). Then, increasing the accuracy of estimates of cases’ OR times can achieve large (≈6%) increases in productivity, but less so (≈4%) for facilities already (appropriately) using case resequencing decisions to increase productivity. ConclusionsEarlier empirical studies of reductions in predictive error have shown small benefit to unbiased estimators because they used current cases and allocated time. Organizations cannot benefit from modern predictive methods unless allocated times are adjusted statistically, differentiated from the longer hours for staff scheduling. Surgical suites in provinces, organizations, etc., with labor contracts based on coverage of allocated hours (e.g., two 4 h sessions) should not expect benefit, from a labor costs perspective, from investing in newer methods for predicting average operating room time.