Anesthesia Practice Shift Scheduling With a Generative Deep Learning Model.

Anesthesiology scheduling techniques are inadequate to appropriately deal with modern anesthesia practice demands. Anesthesiologists are increasingly dissatisfied with their jobs in the face of inflexible schedules, increasing workload, and the complexity of practice. Decreasing autonomy and inherent responsibility lead to burnout and decreased job satisfaction. Unfortunately, equitable and timely shift scheduling that meets individual provider expectations remains a distant mirage. Technology has been promised as a means to decrease workload and improve productivity. But technology has not met these expectations. In real-world anesthesia practice, scheduling remains contentious and time-consuming. These failures are somewhat attributable to current scheduling systems and software. In this paper, we present an alternative method of anesthesiology shift scheduling using advances in machine learning (ML). The development of this deep learning (DL) model for shift scheduling drastically reduces the effort required to create shift schedules that comply with the rules and regulations observed by anesthesia practices. A DL model architecture is developed, trained with shift schedule data from the Reno-Tahoe Anesthesia (RTA) group, and evaluated against the practice requirements. The DL model trained and evaluated demonstrates a Matthews Correlation Coefficient (MCC) of 0.9776 and balanced accuracy of 0.9531. The trained model reliably learns practice scheduling rules sufficient to generate new shift schedules in compliance with the rules. Furthermore, the trained model learns practice rules solely from past examples without requiring a human expert to codify the rules.