A simple prediction algorithm for bacteraemia in patients with acute febrile illness

Area of Science:

• Clinical Medicine
• Infectious Diseases
• Epidemiology

Background:

• Existing bacteraemia prediction models are often complex and impractical for clinical use.
• Physician adoption of prediction models depends on simplicity and high sensitivity.

Purpose of the Study:

• To develop a straightforward classification algorithm for predicting bacteraemia risk.
• To create a tool that is both simple and sensitive for clinical application.

Main Methods:

• A hospital-based study involving 526 adult patients with acute febrile illness.
• Recursive partitioning analysis and V-fold cross-validation were used to build prediction algorithms.
• Two scenarios were evaluated: one without laboratory tests and one with laboratory tests.

Main Results:

• Two algorithms identified three risk groups for bacteraemia.
• Scenario 1 (no labs): Chills, pulse, and low-risk site diagnosis predicted low risk (87.5% sensitivity).
• Scenario 2 (with labs): Chills, C-reactive protein, and low-risk site diagnosis predicted low risk (92.5% sensitivity).
• Both algorithms demonstrated high negative predictive values (98.6% and 99.1%).

Conclusions:

• A simple, sensitive prediction algorithm can effectively identify patients at low risk of bacteraemia.
• The developed algorithm shows potential utility in clinical settings for managing febrile illness.
• Further prospective validation in diverse settings is recommended.