Related Experiment Videos
A risk prediction model for screening bacteremic patients: a cross sectional study
Franz Ratzinger1, Michel Dedeyan2, Matthias Rammerstorfer2
1Department of Laboratory Medicine, Division of Medical and Chemical Laboratory Diagnostics, Medical University of Vienna, Vienna, Austria.
Plos One
|September 4, 2014
Summary
This study developed a risk prediction model to identify patients with a very low risk of bacteraemia (bloodstream infection). The model can help improve blood culture cost-effectiveness by reducing unnecessary tests.
Area of Science:
- Clinical microbiology
- Infectious diseases
- Medical informatics
Background:
- Bacteraemia is a severe condition with high mortality.
- Current blood culture methods have low pathogen detection rates, increasing costs.
- Need for improved cost-effectiveness in diagnosing bacteraemia.
Purpose of the Study:
- Develop a risk prediction model for bacteraemia.
- Identify patients at very low risk for bacteraemia using an automated decision support tool.
- Enhance the cost-effectiveness of blood culture sampling.
Main Methods:
- Retrospective hospital-wide cohort study of 15,985 patients.
- Assessed 51 variables for diagnostic potency.
- Utilized A2DE classifier (supervised Bayesian classifier) for model development and validation.
Main Results:
- Neutrophile leukocyte proportion was the best individual predictor (ROC-AUC: 0.694).
- Two models (20 and 10 variables) achieved ROC-AUCs of 0.767 and 0.759 in derivation, and 0.800 and 0.786 in validation.
- Over 10% of patients identified as low risk (NPV > 98.8%).
Conclusions:
- The developed models can identify over 10% of patients with minimal risk for bacteraemia.
- Potential for increased cost-effectiveness of blood culture sampling via an automated decision support tool.
- External prospective validation is required to confirm generalizability and usefulness.
