Abstract
Operating room (OR) scheduling relies on estimating surgical durations, while intraoperative care increasingly depends on AI-driven risk prediction systems, both of which are shaped by uncertainty and variability that can significantly impact outcomes. My technical project focuses on using machine learning to improve surgical duration predictions at University of Virginia Health System (UVA Health), with the goal of reducing inefficiencies such as delays, staff overtime, and resource misallocation. Alongside this, my STS research investigates how intraoperative AI-driven risk prediction systems create a misalignment between decision making authority and accountability in surgical environments. While these projects focus on different stages of surgical care, they are connected by a shared theme of uncertainty within the OR, where every move carries significant weight. This connection highlights the importance of STS perspectives in engineering. As AI becomes more embedded in surgical practice, it shapes decision making without changing how accountability is assigned.
In the technical portion of my thesis, my capstone team developed machine learning models that identify the most influential factors affecting surgery duration and improve time estimates at UVA Health. Existing scheduling approaches rely heavily on surgeon estimates and historical averages, often failing to capture variability across procedures. To address these limitations, my team and I trained XGBoost and Random Forest models on surgical data spanning July 2022 to June 2025. A key distinguishing feature of this work is the use of clustering to group similar surgical procedures based on both their descriptions and typical duration patterns, allowing the models to better account for variation across cases. By incorporating features such as scheduled start time, day of the week, and procedure groupings, our models produce more reliable duration estimates than the current approach at UVA Health, achieving over 90% accuracy within a 30 minute tolerance compared to approximately 75% for current estimates. These results show that machine learning can significantly improve surgical duration prediction accuracy, producing more reliable estimates that may support more effective OR scheduling.
In my STS research, I examine how intraoperative AI risk prediction systems redistribute decision making authority while accountability remains concentrated on surgeons, creating a structural misalignment in surgical practice. Drawing on Actor-Network Theory (ANT), I view surgical decision making as a process shaped by interactions between human and nonhuman actors that include surgeons, AI systems, clinical data streams, and institutional protocols. Rather than treating AI systems as neutral tools, my analysis shows how they shape decision making by interpreting patient data, prioritizing risks, and influencing the timing of surgical responses. Despite this distributed influence, existing legal and professional frameworks continue to assign full responsibility to the surgeon. As a result, my research shows how a “responsibility gap” emerges, where decision making authority is shared but accountability remains concentrated. This finding highlights a structural disconnect between how surgical decisions are actually produced and how responsibility is assigned in practice.
Considering both my technical and STS projects together shows why it is important to look at engineering problems beyond just the technical solution. My capstone’s machine learning models improve surgical duration prediction accuracy, reducing average prediction error by nearly 60%. However, my STS research shows that systems do not operate in isolation, and that these technologies are part of larger networks of actors, including hospital staff, surgeons, and clinical data systems. Looking at my technical project through the lens of my STS research makes me think about how this machine learning model would be embedded within existing hospital structures. This perspective highlights the role of STS in engineering by showing that engineers need to consider not only how systems perform and what efficiencies they create, but also how they fit within and interact with the environments in which they are used.
Notes
School of Engineering and Applied Science
Bachelor of Science in Systems Engineering
Technical Advisor: Robert Riggs, Daniel Otero-Leon, Constanza Lorca
STS Advisor: William Davis
Technical Team Members: Elise Williamson, Mike Dertke, Mackenzie Craig, Jacob Singer, Nathaniel Kusic