Background of Research
In my prior work as a data scientist working in the medical/healthcare space, I got the opportunity to work closely with Surgeons at major academic institutions across the United States. My role at the time was to partner with hospital networks or individual academic surgeons in a joint research initiative to set up a data architecture and execute on a long-term data collection effort focused only on surgical performance KPIs, namely Surgical Efficiency and Proficiency metrics1.
Most importantly, these efforts were conducted at scale and would result in the aggregation of a massive amount of longitudinal metrics describing Surgical Actions.
- Surgical Actions
-
Time-stamped records of actions performed by the surgeon over the course of a surgical procedure
- Surgical Efficiency
-
Metrics describing the operating efficiency of an OR Team and Surgeon working together to complete a surgical procedure
- Surgical Proficiency
-
Benchmarks describing the relative expertise level of a surgeon performing a procedure
About the Data Collected
The dataset ultimately collected covered thousands of surgeons across the United States over a period of 5 years. The population of Surgeons in the data included a wide range of experience levels ranging from novice, having performed fewer than 10 procedures, to expert surgeons that have completed more than 300 procedures. Additionally, the population of surgeons include a many different surgical specialties, performing a wide range of cases including gynecological, thoracic, cardiac, and more.
Producing Surgical Performance KPIs
The the raw point-in-time Surgical Actions data collected would ultimately be transformed, and the resulting data is best described as event-based and resembled the following:
The events above would then be summarized into procedure-level metrics that describe surgical proficiency, efficiency, or both
Lastly, these metrics would then be translated to KPIs, benchmarked against a target performance level2, and presented to the Executive Stakeholders of this work.
What’s Next
Once there was enough of these historical surgical actions data collected, that’s when the interesting part of my work would begin… the analysis!
Subscribe to my blog and stay tuned for upcoming posts where I dive into this work and share some fascinating insights effort to develop Surgical Learning Curve models, and further present a case study where I demonstrate how to use these models as the foundation for assessing the surgical expertise and monitoring novice surgeons as they progress along their own learning curves.
Footnotes
In this work, we were primarily interested in data about the Surgeons, and observational metrics describing their performance during the procedure, rather than healthcare outcomes. Thus, no personally identifiable information was collected. This simplified our work ensuring all HIPAA rules and regulations were satisfied.↩︎
The Target Performance for benchmarking would be derived by identifying a control group of Surgeons with high experience levels and are known to be both efficient and proficient. The metric values for these expert surgeons would guide the assessment of performance for all other Surgeons.↩︎