Our long term goal is to reduce postoperative infections. We will start by developing a system to accurately and completely identify their occurrence by applying machine learning algorithms to electronic health record (EHR) data. We will utilize a comprehensive audit and feedback system to create reports of risk-adjusted rates and specific details of postoperative infectious complications that are shared with surgeons and other healthcare providers to facilitate their awareness. We call this system the Automated Surveillance of Postoperative Infections (ASPIN). ASPIN will be piloted in the four major hospitals of the University of Colorado Health system (UCHealth) with a combined surgical volume of approximately 80,000 patients per year. We expect this will supersede the costly and laborious manual partial sampling of postoperative infectious complications which is current utilized by many hospitals.
Specific Aim 1. Expand and enhance models for preoperative risk prediction and postoperative identification of surgical infections using EHR and ACS NSQIP data from patients who underwent operations at four UCHealth hospitals.
Specific Aim 1 a) Enhance previously-developed models for identification of postoperative infections by controlling Type-I errors via ?knockoffs,? a recent statistical innovation for high dimensional model selection using false discovery rate correction.
Specific Aim 1 b) Deploy natural language processing methods using EHR text reports of these patients to identify additional indicators of postoperative infections and further refine the models.
Specific Aim 1 c) Create preoperative risk models for infection using EHR data - similar to the models implemented in the AHRQ-funded Surgical Risk Preoperative Assessment System - but that do not require additional data entry by the health care providers.
Specific Aim 2. From the beginning of the study, develop ASPIN with input from an Advisory Committee composed of administrators and surgeons from all four UCHealth hospitals. Additional feedback from surgeons will be obtained through focus groups and semi-structured interviews at several steps of ASPIN development and implementation planning.
Specific Aim 3. A pilot implementation of ASPIN will utilize the RE-AIM framework to guide and examine the preliminary effectiveness and feasibility of ASPIN at UCHealth. We will recruit 30 surgeon participants from all four UCHealth hospitals to use ASPIN, and we will evaluate the reach, effectiveness, adoption, and implementation of ASPIN. This research responds to AHRQ priorities by utilizing existing data to develop a learning health system with a distinct focus on improving surveillance and reporting of postoperative healthcare-associated infections.
Infectious complications of surgery (surgical site infection, pneumonia, urinary tract infection, and sepsis) are common, occurring in about 8% of major surgical procedures, and are costly. Currently used surveillance techniques require extensive human participation, resulting in delays of at least six months before the data are available to care providers, and the data represent only a small sample of all operations. We propose to develop a machine learning-based, near real-time recurring audit and feedback system for identification of postoperative infections that will inform providers about infectious outcomes of their patients.
