Prediction of bacteremia using TREAT, a computerized decision-support system

Area of Science:

• Clinical decision support systems
• Infectious disease diagnostics
• Sepsis management

Background:

• Accurate prediction of bloodstream infections at sepsis onset is crucial for effective and economical patient management.
• Timely identification of bacteremia aids in appropriate treatment strategies and resource allocation.

Purpose of the Study:

• To evaluate the performance of the TREAT (TREATment) computerized decision-support system in predicting bacteremia in patients with suspected sepsis.
• To assess the system's ability to stratify patients into different risk categories for bacteremia.

Main Methods:

• The TREAT system, utilizing a causal probabilistic network, was prospectively implemented in two independent cohorts across Israel, Italy, and Germany.
• Data were collected at the time of blood culture sampling, and system performance was evaluated using receiver-operating characteristic (ROC) curves.
• Patients were stratified into low-, intermediate-, and high-risk groups based on TREAT's bacteremia prediction values.

Main Results:

• The TREAT system demonstrated good discriminative power, with an area under the ROC curve of 0.68 in the first cohort (790 patients) and 0.70 in the second cohort (1724 patients).
• Bacteremia prevalence varied significantly across risk groups: low-risk (1.3%-2.4%), intermediate-risk (12.8%-13.2%), and high-risk (28.1%-29.9%).
• The system's performance remained stable across the three international sites, and low-risk groups constituted 15%-17% of all patients.

Conclusions:

• The TREAT system effectively stratifies patients with suspected sepsis based on bacteremia risk using readily available clinical variables.
• The system's predictive accuracy is consistent across different geographical locations.
• TREAT can identify a low-risk patient cohort, potentially obviating the need for blood cultures and optimizing clinical workflows.