Main Menu Browse by Day or Program Author Keywords

306 Prediction Model for Central Line Associated Bloodstream Infections (CLABSI) Across 14 Acute Care Facilities: Moving Towards Automated CLABSI Surveillance

Saturday, April 2, 2011: 3:00 PM
Cortez Ballroom (Hilton Anatole)
Bala Hota, MD, MPH , Stroger Hospital of Cook County/Rush University Medical Center, Chicago, IL
Jonathan Edwards, MStat , Centers for Disease Control and Prevention, Atlanta, GA
Scott Fridkin, MD , Centers for Disease Control and Prevention, Atlanta, GA
Paul Malpiedi, MPH , Centers for Disease Control and Prevention, Atlanta, GA
Rosie Lyles, MD, MS , Stroger Hospital of Cook County, Chicago, IL
John Martin, MPH , Premier, Inc., Philadelphia, PA
Chris Craver, MA , Premier, Inc., Charlotte, NC
Robert A. Weinstein, MD , Stroger Hospital of Cook County/Rush University Medical Center, Chicago, IL
William Trick, MD , Stroger Hospital of Cook County/Rush University Medical Center, Chicago, IL

Background:

Surveillance for CLABSIs is time-consuming and labor intensive. To inform automated surveillance approaches, we used data from the electronic record to predict the likelihood that a positive blood culture represented a CLABSI.

Objective:

To develop regression models that could predict the presence of a CLABSI.

Methods:

NHSN hospitals using Premier Inc.'s SafetySurveillorTM software system were recruited to participate.  All microbiology culture results from 1/2008–6/2009 were extracted from the SafetySurveillorTM data warehouse.  A positive blood culture episode eligible for expert review was defined as a patient's first positive blood culture in a 30-day period obtained > 2 days after admission, excluding single positive cultures for common skin contaminants.  Two experienced IPs trained in use of NHSN definitions independently reviewed a random sample of eligible episodes for the presence of CLABSI.  Episodes in which the given reviewer confirmed a central line present were used to conduct regression analyses.  Generalized estimating equations with sampling weights were used for analysis.  Variables with p<0.15 were retained in the model. 

Results:

From the 14 hospitals, 5,929 blood culture episodes were eligible for sampling, 1626 reviews were performed (n=771, reviewer 1; n=855, reviewer 2).  The unit types were: Hematology Specialty Care Area n=371(23%); Medical ICU n=347(21%), Medical-surgical ICU n=224 (14%); Surgical ICU n=382(23%); and General ward n=302(19%).  Organism categories were: gram-positive bacteria, n=828(51%); gram-negative bacteria, n=590(36%), and yeasts, n=196(12%).  Of the 1626 reviews, 181 (11%) were categorized as CLABSI. In a final multivariable model predicting CLABSIs, independent predictors included location outside the ICU (OR 1.8 (95%CI 1.2-2.6)), Staphylococcus aureus (OR 2.4 (1.2-4.8)), Escherichia coli (OR 0.5 (0.2-1.4)), Acinetobacter spp.  (OR 0.2 (0.0-0.8)), Streptococcus spp.  (OR 0.0 (0.0-0.3)), and having a matched culture from another anatomic site (0.3 (0.1-0.6)).  The model performed well (c statistic=0.70).  The predicted probabilities for any single episode ranged from 0.3% - 34%; summing the probabilities overall resulted in a CLABSI prevalence in the sampled population comparable to that derived by expert review (figure).

Conclusions:

Using a regression model predicting the presence of CLABSI, we demonstrated that estimated prevalence of CLABSI approximated manually observed rates.  These automated rates were obtained using easily accessible variables found in the electronic record.  Simplified, automated surveillance of CLABSI may be achievable through innovative uses of electronic health record data.

Figure:  Measured and Predicted Prevalence of CLABSI by Unit Location Type