Date of Award

Fall 9-22-2022

Embargo Period

10-19-2032

Document Type

Thesis

Degree Name

Doctor of Philosophy (PhD)

Department

Public Health Sciences

College

College of Graduate Studies

First Advisor

Alexander Alekseyenko

Second Advisor

Kenneth Catchpole

Third Advisor

Jihad Obeid

Fourth Advisor

Peter Kiessler

Fifth Advisor

Kai Liu

Abstract

Surgical technology continues to improve patient outcomes, but also introduces novel challenges and risks. The increasing complexity of surgical technology leads to workflow disruptions (FD), which are deviations from the expected operation progression. The following adverse outcomes result from high FD rates: increased patient mortality, procedure duration, OR member stress, perceived workload, and surgical errors.

Through collaboration with Catchpole lab (MUSC) and Cohen lab (Cedars-Sinai), we have access to a dataset of several hundred surgical cases with flow disruptions that were transcribed by trained observers in the OR. As part of our first aim, we present software, READ-TV (Research & Exploratory Analysis Driven Time-data Visualization), to visualize flow disruption data. Our second aim proves the utility of AI, specifically transformers, to detect causally related flow disruption cascades from disruption narratives and timings. Our third aim diverges from flow disruption data to present Red Flag/Blue Flag, X-AI (eXplainable-AI) software. Red Flag/Blue Flag enables researchers to analyze how a pervasive neural network model used throughout clinical NLP (natural language processing) determines binary classifications. We present the X-AI capability of Red Flag/Blue Flag on a neural network that was trained from 11,000 physician-authored operative notes to detect surgical device & instrument misadventures.

Rights

Copyright is held by the author. All rights reserved.

Download

Available for download on Tuesday, October 19, 2032

Share

COinS