The Open Epidemiology Journal

2010, 3 : 24-28
Published online 2010 April 28. DOI: 10.2174/1874297101003010024
Publisher ID: TOEPIJ-3-24

A Clinical Predictive Model for Catheter Related Bloodstream Infections from the Electronic Medical Record

Gonzalo M.L. Bearman , Michael I. Oppenheim , Eneida A. Mendonca , Nathaniel Hupert , Maryam Behta , Paul J. Christos and Lewis M. Drusin
VCU Medical Center, Richmond, Box 980019, VA 23298-0019, USA

ABSTRACT

Background and Methods: Previous studies identifying risk factors for catheter related blood stream infections (BSI) have relied on labor-intensive paper chart review. The electronic medical record (EMR) can provide real-time data for profiling patients at risk for BSI. Our study's aim was to use our hospital EMR to identify risk factors and develop a clinical predictive model for catheter related BSI from June 2001-June 2002. Using a modified CDC definition for BSI, we retrospectively extracted a number of variables from the EMR data repository including demographics, medications, nutritional status, presence of diabetes (DM), skin/wound status, use of a mechanical ventilator, and presence of a urinary catheter or central/arterial/PICC catheter. We developed a Cox Proportional Hazards model using half the cohort (14,977 controls and 123 BSI cases), and validated the model using the other half. Final variables in the multivariate model were chosen by clinical relevance, significance on univariate analysis and by step-wise automated selection. Results: The final model contained: use of antibiotics (HR 4.6), central venous catheter for 0-6 days (HR 1.8), central venous catheter for greater or equal to 7 days (HR 1.3), tunneled central venous port (HR 9.3) and DM (HR 2.1). Variables from multivariate analysis were used for the predictive index; points were derived from the respective HR of the Cox model. A cutoff value that maximized sensitivity (75%) and specificity (89%) had a positive predictive value (PPV) of .05. This model was validated on the second half of the cohort (122 BSI, 14,977 ‘non-cases’) with resultant sensitivity of 69%, specificity of 88% and PPV of 0.05. Comparison of the two models revealed no significant differences in the sensitivity, specificity and PPV. Conclusion: We successfully used a hospital-based EMR in developing a Cox Proportional Hazards model for catheter related BSI. This model was used to derive a predictive index. At a cutoff point providing a sensitivity of 75% and specificity of 89% for finding patients who develop catheter related BSI, the index had a PPV 0.05. When applied to a predictive index, real time data gathered from the EMR may assist in identifying patients at risk for a catheter related BSI and may be further utilized as part of a protocol driven, preventive strategy for catheter related BSI risk reduction.