Abstract
My technical capstone project and my STS research are connected through their shared focus on technology in hospital environments and the broader systems in which that technology operates. My technical capstone focuses on the design and implementation of a machine learning model to improve surgical duration predictions, emphasizing the practical benefits of new technology in a healthcare setting. In contrast, my STS research analyzes the failure of the Therac-25 radiation therapy machine, using Actor-Network Theory to understand how breakdowns across a network of actors led to catastrophic consequences. While one project centers on creation and the other on failure, both highlight the importance of viewing technology as part of a larger sociotechnical system. Together, they reinforce the need for careful, systems oriented thinking when developing and evaluating technology in high-stakes environments like healthcare.
My technical project focuses on developing a machine learning model to better predict surgical durations at UVA Health. Using two years of historical surgical data that was screened for HIPAA compliance, we evaluated multiple models, with XGBoost and Random Forest emerging as the top performers. These models improved prediction accuracy compared to existing hospital methods, with performance varying across surgical departments. By providing more reliable time estimates, the model is intended to support more efficient operating room scheduling and resource allocation.
In my STS research on the Therac-25 accidents, I argued that the failures of the machine were not solely the result of faulty code or operator error, but rather a breakdown in the broader sociotechnical system. Through the lens of Actor-Network Theory, I showed how both human actors (engineers, operators, hospital staff) and nonhuman actors (software, hardware, interface design) contributed to a series of small but compounding failures. These misalignments ultimately led to catastrophic outcomes, including patient deaths. This analysis emphasizes that responsibility in complex systems is distributed, and that failures emerge from interactions within the network rather than from a single point of fault.
Working on both projects simultaneously deepened my understanding of the responsibilities involved in designing technology for healthcare settings. Studying the Therac-25 failures made the risks of cutting corners and overlooking system interactions much more tangible while I was actively developing my own model. It shifted my perspective from focusing solely on technical performance to considering how my work fits into a larger network of users, data, and institutional practices. Moving forward, I will carry a strong appreciation for learning from past failures and for approaching engineering problems with a systems-oriented mindset that prioritizes safety, accountability, and real-world impact.
