Surgical operations can put patients at high risk of infections and other complications. However, studies have shown that half or more of surgical infections are discovered after hospital discharge. The American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP) began in 2005 with the goal of assisting hospitals with identifying and preventing surgical complications. Each participating ACS NSQIP hospital assigns a surgical clinical nurse reviewer to collect preoperative through 30-day postoperative data on a sample of surgical patients in order to risk adjust postoperative complications so that they can be compared across participating hospitals. Although at large volume hospitals these samples might represent only 10-15% of all surgical cases, these data are considered to be the current gold standard for accurate identification and comparison of postoperative complications. Unfortunately, chart review is time-consuming and costly and, as a result, cannot be scaled up easily to cover all surgical patients. The goal of this project is to learn from the ACS NSQIP data in order to develop electronic algorithms for identifying postoperative infections that can be scaled up easily and inexpensively. Development of such algorithms will also permit evaluation of interventions that intend to impact large populations of patients at risk of postoperative infections. Considering the large number of available binary classification algorithms for data mining that are easy to implement, it is paramount to consider new methods for identifying surgical infections electronically. Furthermore, postoperative infections are rare, and occur in about 7% of operated patients; therefore, it is difficult to identify models that classify infections well. Sampling techniques are commonly used in conjunction with classification models in order to improve sensitivity and positive predictive value. We believe that modern statistical techniques for classification combined with strategic sampling, and the use of ICD9 and ICD10 codes and pharmacy data, will improve upon existing methods for electronically identifying postoperative infections.
The aims of this proposal are (1a) to develop algorithms for identifying surgical infections using machine learning techniques, (1b) to develop models for specific types of postoperative infections collected in the ACS NSQIP data separately, which include SSI, urinary tract infection, pneumonia and sepsis and (2) to validate these models in prospective ACS NSQIP data.

Public Health Relevance

In this application, we propose to use outcomes data from the University of Colorado American College of Surgeons National Surgical Quality Improvement Program combined with electronic health record and administrative claims data in order to develop models for identifying postoperative infections electronically. These infections are currently identified through manual chart review, and it is our hope that successful development of these models could replace costly, time-consuming manual chart review. We will use machine learning and sampling techniques to achieve better performance than models previously reported in the literature.

Agency
National Institute of Health (NIH)
Institute
Agency for Healthcare Research and Quality (AHRQ)
Type
Small Research Grants (R03)
Project #
5R03HS026019-02
Application #
9681394
Study Section
Healthcare Information Technology Research (HITR)
Program Officer
Gray, Darryl T
Project Start
2018-04-06
Project End
2020-03-31
Budget Start
2019-04-01
Budget End
2020-03-31
Support Year
2
Fiscal Year
2019
Total Cost
Indirect Cost
Name
University of Colorado Denver
Department
Internal Medicine/Medicine
Type
Schools of Medicine
DUNS #
041096314
City
Aurora
State
CO
Country
United States
Zip Code
80045
Colborn, Kathryn L; Bronsert, Michael; Amioka, Elise et al. (2018) Identification of surgical site infections using electronic health record data. Am J Infect Control 46:1230-1235
Colborn, Kathryn L; Bronsert, Michael; Hammermeister, Karl et al. (2018) Identification of urinary tract infections using electronic health record data. Am J Infect Control :