Main Menu Browse by Day or Program Author Keywords

420 Use of an Electronic Data Warehouse for Cardiac Surgical Site Infection Surveillance

Sunday, April 3, 2011
Trinity Ballroom (Hilton Anatole)
Gregory Walter Rose, MD, FRCPC , The Ottawa Hospital/The University of Ottawa, Ottawa, ON, Canada
Virginia R. Roth , The Ottawa Hospital/The University of Ottawa, Ottawa, ON, Canada
Kathryn N. Suh, MD, FRCPC , The Ottawa Hospital/The University of Ottawa, Ottawa, ON, Canada
Monica Taljaard, PhD , The University of Ottawa/ Ottawa Hospital Research Institute, Ottawa, ON, Canada
Carl Van Walraven, MD, FRCPC , The Ottawa Hospital/The University of Ottawa, Ottawa, ON, Canada
Alan J. Forster, MD, FRCPC , The Ottawa Hospital/The University of Ottawa, Ottawa, ON, Canada
Background: “Trigger mechanisms” concentrate labour-intensive surgical site infection (SSI) surveillance on patients with high probability of SSI. Most trigger mechanisms prioritize true positive rate (TPR = sensitivity) at the cost of higher false positive rates (FPR = 1 – specificity). More complex trigger mechanisms may allow for lower FPR without sacrificing TPR. We performed a nested case-control study of cardiac surgery patients to construct complex electronic trigger mechanisms for SSI surveillance, using a sophisticated data structure called a data warehouse.

Objective: To derive electronic trigger mechanisms for cardiac SSI that offer non-inferior TPR and superior FPR, compared to our current manual surveillance methodology.

Methods: Retrospective study of cases of cardiac SSI, with uninfected controls, nested in a cohort of all adult patients undergoing specific cardiac surgery procedures at our centre from July 1 2004 to June 30 2007.  We confirmed case and control status by manual chart review of all subjects, using National Healthcare Surveillance Network SSI definition criteria. We defined potential “trigger factors” for inclusion in trigger mechanisms, using the results of an earlier comprehensive literature review. We harvested trigger factors from our data warehouse, and used semi-Bayesian logistic regression modelling to derive two electronic trigger mechanisms – one including all data sources within the data warehouse, and one limited to time-sensitive data sources.

Results: In a cohort of 3744 cardiac procedures in 3 years, we identified 202 cases of SSI, and selected 513 uninfected controls. 33.9% of patients were female, mean age was 65.4 years, and 80.1% of patients had coronary artery bypass grafting. Incidence of all-depth SSI in the cohort was 9.4% (95% Confidence Limits [CL] 7.8, 10.9%). Incidence of deep incisional or organ space SSI was 3.2% (95% CL 2.4, 4.0%).

 We created 159 individually defined trigger factors and eliminated 125 prior to model building. From the remaining 34 trigger factors we derived two trigger mechanism models – one with 11 trigger factors drawn from all data sources in our data warehouse, one with 9 trigger factors drawn only from time-sensitive data sources.  Areas under the receiver operator curve were 0.972 and 0.96 respectively. When logit probability was dichotomized, these models provided TPR of 0.9356 and 0.9307 (at least equal to our current mechanism) and FPR of 0.0633 and 0.1033 (superior to current.)  Estimated positive predictive values are 0.60 and 0.48.

Conclusions: We have derived two electronic mechanisms to trigger cardiac SSI surveillance, with non-inferior TPR (sensitivity) and superior FPR (specificity) in comparison to our current manual method. We have demonstrated the potential value of an electronic data warehouse in infection control activities.