Diagnosing intensive care units and hyperplane cutting for design of optimal production systems

Thesis: S.M., Massachusetts Institute of Technology, Department of Mechanical Engineering, 2015. In conjunction with the Leaders for Global Operations Program at MIT.Thesis: M.B.A., Massachusetts Institute of Technology, Sloan School of Management, 2015. In conjunction with the Leaders for Global Operations Program at MIT.Cataloged from PDF version of thesis.Includes bibliographical references (pages 101-107).This thesis provides a new framework for understanding how conditions, people, and environments of the Intensive Care Unit (ICU) effect the likelihood the preventable harm will happen to a patient in the ICU. Two years of electronic medical records from seven adult ICUs totalling 77 beds at Beth Israel Deaconess Medical Center (BIDMC) were analysed. Our approach is based on several new ideas. First, instead of measuring safety through frequency measurement of a few relatively rare harms, we leverage electronic databases in the hospital to measure Total Burden of Harm, which is an aggregated measure of a broad range of harms. We believe that this measure better reflects the true level of harm occurring in Intensive Care Units and also provides hope for more statistical power to understand underlying contributors to harm. Second, instead of analysing root causes of specific harms or risk factors of individual patients, we focus on what we call Risk Drivers, which are conditions of the ICU system, people (staff, patients, families) and environments that affect the likelihood of harms to occur, and potentially their outcomes. The underlying premise is that there is a relatively small number of risk drivers which are common to many harms. Moreover, our hope is that the analysis will lead to system level interventions that are not necessarily aiming at a specific harm, but change the quality and safety of the system. Third, using two years of data that includes measurements of harms and drivers values of each shift and each of seven ICUs at BIDMC, we develop an innovative statistical approach that identifies important drivers and High and Low Risky States. Risky States are defined through specific combinations of values of Risk Drivers. They define environmental characteristics of ICUs and shifts that are correlated with higher or lower risk level of harms. To develop a measurable set of Risk Drivers, a survey of current ICU quality metrics was conducted and augmented with the clinical experience of senior critical care providers at BIDMC. A robust machine learning algorithm with a series of validation techniques was developed to determine the importance of and interactions between multiple quality metrics. We believe that the method is adaptable to different hospital environments. Sixteen statistically significant Risky States (p < .02) where identified at BIDMC. The harm rates in the Risky States range over a factor of 10, with high risk states comprising more that 13.9% of the total operational time in the ICU, and low risk states comprise 38% of total operating shifts. The new methodology and validation technique was developed with the goal of providing a basic tools which are adaptable to different hospitals. The algorithm described within serves as the foundation for software under development by Aptima Human Engineering and the VA Hospital network with the goal of validation and implementation in over 150 hospitals. In the second part of this thesis, a new heuristic is developed to facilitate the optimal design of stochastic manufacturing systems. The heuristic converges to optimal, or near optimal results in all test cases in a reasonable length of time. The heuristic allows production system designers to better understand the balance between operating costs, inventory costs, and reliability.by J Adam Traina.S.M.M.B.A