SBIR Phase I: Machine Learning Software for Situation Awareness in the Operating Room for Improved Patient Flow

这个小型企业创新研究（SBIR）第一阶段项目的更广泛的影响/商业潜力是医院运营效率的显着提高。 该项目的重点是在规划，人员配备，协调和外科手术的执行效率低下的来源。 对医院数据的初步分析表明，在人员配置相同的情况下，围手术期单位的吞吐量可以显著增加，同时减少患者经历的延误并改善工作人员的工作条件。 这些改进有可能每年节省10亿美元以上。在获取和传达有关手术状态和工作人员应执行的行动的更新方面的延迟是效率低下的主要原因，任务持续时间的随机性也是如此。这些消息的时间取决于对系统状态的了解以及对其未来演变的预测。收集信息的延迟和对行动效果的错误预测导致效率降低。 拟议的创新改进了消息的选择和定时。拟议的项目基于机器学习方法，用于使用实际医院数据优化实时消息传递。该方法结合了新的参数模型的实时调度，随机梯度下降，和无穷小扰动分析。在这个公式中，扰动分析计算目标函数的梯度相对于消息的定时，并在一个有效的算法的结果。 该算法发现发送消息的最佳时间，以优化手术室效率和患者等待时间的组合。该方法能够评估调度决策的复杂级联影响，并确定处理突发事件时的最佳行动方案。及时了解情况将有助于改善医院的病人流量。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is a significant increase in the efficiency of hospital operations. The project focuses on the sources of inefficiencies in the planning, staffing, coordination, and execution of surgical procedures. Preliminary analysis of hospital data shows that the throughput of perioperative units can be increased significantly with the same staffing while reducing the delays that patients experience and improving the working conditions of the personnel. Such improvements have the potential of saving upwards of $1B per year. Delays in obtaining and communicating updates on the status of surgeries and on actions that personnel should perform are major causes of inefficiency, as is the randomness of the tasks' duration. The timing of those messages depends on the knowledge of the state of the system and a prediction about its future evolution. Delays in gathering information and incorrect predictions of the effect of actions result in reduced efficiency. The proposed innovation improves the selection and timing of messages.The proposed project is based on a machine learning approach for the optimization of real-time messaging using actual hospital data. The approach combines new parametric models of real-time scheduling, stochastic gradient descent, and infinitesimal perturbation analysis. In this formulation, perturbation analysis computes the gradient of the objective function with respect to the timing of messages and results in an efficient algorithm. The algorithm discovers the best time to send messages to optimize a combination of operating room efficiency and patient waiting times. The methodology is able to evaluate the complex cascading impact of scheduling decisions and to identify the best course of action when dealing with contingencies. Timely situation awareness will contribute to improved patient flow in the hospital.