Loren Riskin, MD, Dan E Azagury, MD, Oliver Varban, MD, Daniel J Riskin, MD MBA FACS. Stanford University, Stanford, California USA
Objectives: The last decade has seen major advances in data driven healthcare at the research, clinical, and policy levels. The HITECH Act and the Affordable Care Act drive electronic health record infrastructure while reimbursement and cost drivers promote data use. Medical societies are gearing up for data driven healthcare, assuring discrete, structured data capture appropriate to their specialties. Compared to medical fields, procedural fields can be more challenging to study, particularly because medical history, medications, and laboratory values are rarely the most critical information. Use of unstructured narrative data (e.g. H&P, operative, and discharge notes) through natural language processing has been suggested as one process for extracting high quality, granular procedural technique and outcome data. But little discussion on unstructured data use has occurred at the society level. Our goal is to understand whether procedural societies could benefit from less reliance on manually captured registries and real consideration of unstructured data solutions.
Methods: To evaluate critical processes underlying data driven healthcare in procedural fields, literature review and interviews were undertaken. Literature review focused on defined pathways to capture and use unstructured, granular data in clinical specialties. Manuscripts were separated based on reference to primary care, hospital-based care, procedural care and overall healthcare. Interviews were undertaken with experts in data mining and data-driven healthcare, as well as leaders in data use in anesthesia and surgical communities.
Results: Literature reviews and interviews revealed a strong focus on using discrete data captured through manual processes. Interviewees theorized that early surgical leadership in analytics, via the National Surgical Quality Improvement Program (NSQIP) and other programs, supported registry creation with a focus on discrete data capture through a relatively narrow set of variables. This was the only possible process at the time these systems were developed and encompasses very low levels of procedural detail. Some interviewees expressed concern that capture of limited variables represented a bias within the registry system itself, suggesting that limited data acquisition translates to limited understanding of factors influencing outcomes.
Conclusion: As healthcare advances into a more mature phase of meaningful data use, it has become clear that richer codified data are critically needed. While medical fields can effectively study clinical choices and outcomes on discrete structured data captured today by the electronic medical record, this may not be true for procedural fields that include a multitude of options in procedural technique. Narrative notes combined with natural language processing may represent a valuable option for capture of technique and post-procedural outcomes. We believe meaningful discussion within the procedural societies of effective capture and aggregation of narrative procedural and post-procedural content is warranted.
Session Number: Poster – Poster Presentations
Program Number: P616
View Poster