Predictive Model for Bacteremia with Emphasis on Procalcitonin in Patients with Physician-Based Blood Cultures at Emergency Department
Date Issued
2008
Date
2008
Author(s)
Su, Chan-Ping
Abstract
Background: Bloodstream infection or bacteremia is one of the most serious infectious diseases in emergency department (ED). Inappropriate or lacking of empirical antimicrobial therapy may be associated with a poorer outcome in bacteremic patients. How to identify patients with bacteremia timely becomes a great challenge for an emergency physician. A reliable predictive tool for bacteremia is needed to help emergency physicians in reducing the amount of unnecessary blood cultures, and in detecting high risk patients to avoid the sequale as a result of bacteremia. Despite numerous studies on predictive models for bacteremia, there was short of Procalcitonin-incorporated predictive model. bjectives: The objectives of this study are therefore to build up a predictive model for the risk of bacteremia to aid emergency physicians in identifying the high-risk patients earlier in order to reduce the chance of delaying appropriate antimicrobial therapy, and in reducing the unnecessary blood cultures collected at ED.aterial and Methods: We conducted a prospective cohort study at the ED of National Taiwan University Hospital (NTUH) from October 1, 2004 to November 30, 2004. All adult patients aged 15 years or older who had at least two sets of blood cultures collected during the study period were recruited. Factors affecting the risk for bacteremia included five categories: demographic characteristics; predisposing conditions such as underlying diseases, invasive procedures, immunosuppressive therapies; clinical presentations; laboratory tests; and presumptive diagnosis by emergency physicians. The primary outcome was true bacteremia adapted from definitions of the Centers for Disease Control and Prevention (CDC) and MacGregor and Beaty guidelines. To minimize all the possible negative confounding factors that are insignificant in the presence of other significant factors based on model selection criteria, we adopted an iterative procedure to build up a predictive model not to miss the possible negative confounding factors, and then simplified the clinical prediction rule into a coefficient-based scoring system. esults: We enrolled 558 patients with 84 episodes of true bacteremia. Predictors identified for bacteremia and their assigned scores were: (1) liver cirrhosis (adjusted odds ratio [aOR] 0.255; 95% confidence interval [CI] 0.076 to 0.851), -2 point; (2)fever>38.3℃ (aOR 2.94; 95% CI, 1.537 to 5.625), 2 point ; (3) tachycardia (aOR 3.113; 95% CI, 1.618 to 5.990), 2 point; (4) lymphocytopenia (aOR 4.241; 95% CI, 2.144 to 8.391), 2 points; (5) AST>40 IU/L (aOR 3.216; 95% CI, 1.695 to 6.100), 2 point; (6) C-reactive protein (CRP)>10 mg/dL (aOR 1.722; 95% CI, 0.849 to 3.492), 1 point; (7) procalcitonin (PCT)>0.5 ng/mL (aOR 3.837; 95% CI, 1.951to 7.549), 2 points; and (8) presumptive diagnosis of respiratory tract infections (aOR 0.205; 95% CI, 0.077 to 0.543), -3 points. The Hosmer-Lemeshow test revealed a goodness-of-fit of 8.5813 (P=0.3788). The areas under receiver operating characteristic curves (AUC) of original logistic model and the simplified scoring model were 0.861 (95% CI, 0.825 to 0.892) and 0.859 (95% CI, 0.823 to 0.890), respectively. Cross validation with 1,000 bootstraps of half cases for model training and another half for validation revealed a reduction of AUC to 0.664 (95% CI, 0.593 to 0.734). onclusion: We developed a predictive model with scoring system for bacteremia at ED by application of the risk factors associated with bacteremia. However, its generalizabilty needs further corroboration.
Subjects
emergency medicine
emergency departemnt
bacteremia
predictive model
SDGs
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-97-R93846016-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):7e33678d0dc32c6d3aae76e4e99c7842