Objective: To create a set of patient level prediction models for patients undergoing non-maternal, non-cardiac surgeries, examining a set of post-operative outcomes of interest
Rationale: Surgical procedures are frequently performed in large health care systems, with over 15 million invasive surgeries per year in the United States (1). Serious complication rates arise in this population (2). In an effort to counsel patients and reduce their cardiac and non-cardiac surgical risks, the field of perioperative medicine often looks to multivariate prediction models across outcomes of interest. Point of care deployments of these often favour parsimonious models (e.g. the 6 point Revised Cardiac Risk Index (3)). These could potentially be outperformed or complemented by machine learning approaches to prediction that utilize a comprehensive representation of the patient record as a feature source, especially as point of care application becomes automated in the era of the electronic medical record.
Project Lead(s): Evan Minty, Lichy Han, Nigam Shah
Coordinating Institution(s): Stanford University
Additional Participants: Collaborators Welcome
Full Protocol: in development
Initial Proposal Date: April 20 2018
Launch Date: TBA
Study Closure Date: TBA
Results Submission: TBA
CDM: V5, uses Feature Extraction 2.0
Table Accessed: TBA
Database Dialects: SQL Server, Postgres, Oracle
Software: R