Thesis: S. M. in Computation for Design and Optimization, Massachusetts Institute of Technology, Computation for Design and Optimization Program, 2017. / Cataloged from PDF version of thesis. / Includes bibliographical references (pages 101-104). / Health care quality and patient safety has gained an increasing amount of attention for the past two decades. The quality of care nowadays does not only refer to successful cure of diseases for patients, but a much broader concept involving health care community, inter-relationships among care providers, patients and family, efficiency, humanity and satisfaction. The intensive care units (ICU) typically admit and care for the most clinically complex patients. While much effort has been put into patient safety improvement, the critical care system still continuous to see many human errors occur each day, despite the fact that people who work in such environment have received exceptional training. Traditional interventions to mitigate patient harm events in ICU generally focus on individual patient harms, and highly underestimate the overall risk patient face during their stay. This thesis aims to establish a new framework that more accurately account for patient risk and is capable of providing recommendations for operational decision making in launching intervention strategies that improve care quality and patient safety. Our approach is based on theories regarding the underlying causes of human errors and a system engineering as well as analytics perspectives. We use various statistical methodologies to output rigorous but clinically intuitive insights. The core concept is to study and utilize how system-level conditions, including both human and environmental factors, can affect the likelihood of harm events in ICU. These can hopefully be used to reduce patient harms and promote patient safety by eliminating unfavorable conditions that are in higher correlation with these events, or promoting safe conditions. We first create a quantitative metric to assess the total burden of harm that patients face, including both high frequency harms, which are typically measured in ICUs today, as well as harms that can bring highly negative outcomes to the patient but ignored due to low frequency. It is an aggregated measure that aims to reflect the true risk level in the ICUs. Then, unlike the traditional approach that motivates intervention strategy to specific harms, we depend on the concept of risk drivers, which describe relevant ICU system conditions, and investigate what drivers affect the probability for harm events in the ICU. These conditions are defined as Risky States, and suggested by the model for elimination to avoid a variety of consequent risk and improve patient safety. The underlying assumption is that the same risk drivers (risky state) may affect many harms. Finally, we propose a new ensemble statistical learning algorithm based on regression trees that is not only powerful in examine the relationship between drivers and outcomes, but also being descriptive defining the risky states. The framework was applied to the retrospective data of 2012 and 2013 from 9 ICUs at the Beth Israel Deaconess Medical Center (BIDMC), with both clinical and administrative records of more than ten thousand patients. Based on our analysis, we see a strong evidence that system conditions are associated with harm events, which include, for example, ICU patient flow (e.g., how many patients are admitted to and discharged from unit), patient acuity level, nurse workload, and unit service type, etc. The model is capable of providing insights such as "when a medical unit has more than 3 newly admitted patients during a day shift, its risk level is approximately 35% higher than the average day shift risk levels in medical units", which can motivate decisions such as assigning a new patient to some other medical unit when the current one has already admitted 3 patients during the shift, in order to avoid the above risky state from occurring. The model output is further presented to BIDMC experts for validation through a clinical perspective. It is also being implemented and integrated to BIDMC ICU tablet application to provide guidance to ICU staff as an alerting system. The Risky State framework is unique in its innovative approach to assess patient risk and capability to offer leverage for overall patient safety improvement, and at same time designed to be compatible and spreadable with different hospital settings. / by Yiqun Hu. / S. M. in Computation for Design and Optimization
| Identifer | oai:union.ndltd.org:MIT/oai:dspace.mit.edu:1721.1/113968 |
| Date | January 2017 |
| Creators | Hu, Yiqun (Computational design expert) Massachusetts Institute of Technology |
| Contributors | Retsef Levi., Massachusetts Institute of Technology. Computation for Design and Optimization Program., Massachusetts Institute of Technology. Computation for Design and Optimization Program |
| Publisher | Massachusetts Institute of Technology |
| Source Sets | M.I.T. Theses and Dissertation |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | 104 pages, application/pdf |
| Rights | MIT theses are protected by copyright. They may be viewed, downloaded, or printed from this source but further reproduction or distribution in any format is prohibited without written permission., http://dspace.mit.edu/handle/1721.1/7582 |
Page generated in 0.0094 seconds