31. Identifying Intraoperative Event Patterns in Advanced Minimally Invasive Surgery using Bioinformatics Techniques

150 150 TECHNA Symposium

Lauren Gordon, MD, MSc 1, Frank Rudzicz, PhD 2, Teodor Grantcharov, MD, PhD 3

1Division of Vascular Surgery, University of Toronto, Canada
2Department of Computer Science, University of Toronto, Canada
3Department of Surgery, University of Toronto, Canada

Introduction
By identifying patterns of events during a surgical procedure, we can better target educational interventions. Machine learning algorithms have been developed to identify patterns in and similarities between protein sequences. This proof of concept study uses one of these algorithms, Clustal Omega, to identify similar patterns of events across multiple surgical procedures.

Methods
Videos were manually reviewed for deviations in protocol resulting in patient injury or potential harm. These events were coded as free text. Keyword analysis identified the nature of events.

Analysis was performed using MATLAB R2017a and the Clustal Omega algorithm, available from the European Molecular Biology Laboratory. We aligned sequences of events across procedures, and clustered these procedures to find similar groups.

Results
From 2015 to 2016, 77 laparoscopic Roux-en-Y gastric bypass procedures were recorded, with a mean of 14.25 events and 5.7 rectification attempts per procedure. Events fell into five categories: bleeding (66.3%), thermal injury (3.1%), mechanical injury (14.9%), anastomosis issues (1.6%), and other (14.1%).

Sequence alignment identified groups of procedures with similar event patterns, representing initial bleeding and tissue injury with delayed rectification, bleeding and tissue injury with immediate and interspersed rectification, and late injury without rectification.

Hierarchical clustering identified seven groups of procedures with similar patterns. 30-day complication rate was not significantly different between groups on chi-square analysis (p=0.96).

Conclusions
The Clustal Omega algorithm can identify patterns of events. Further work is needed to conclusively identify specific event patterns and their causal errors. Identifying these patterns could lead to interventions which improve surgical education and quality of care.