Differentiating viral and bacterial infections: A machine learning model based on routine blood test values

Area of Science:

• Computational biology and infectious disease diagnostics.
• Application of machine learning in clinical decision support.

Background:

• Antibiotic resistance is a significant global health threat.
• Accurate differentiation between bacterial and viral infections is crucial for appropriate antibiotic use.
• Current diagnostic methods, like C-reactive protein (CRP) levels, have limitations in distinguishing infection types.

Purpose of the Study:

• To develop and evaluate a machine learning model for differentiating bacterial from viral infections.
• To assess the model's performance using routine blood test results, CRP, sex, and age.

Main Methods:

• Development of a 'Virus vs. Bacteria' machine learning model.
• Utilized a dataset of 44,120 cases from a single medical center.
• Input features included 16 routine blood test parameters, CRP concentration, biological sex, and age.

Main Results:

• The machine learning model achieved 82.2% accuracy, 79.7% sensitivity, and 84.5% specificity.
• Demonstrated a Brier score of 0.129 and an Area Under the ROC Curve (AUC) of 0.905.
• Outperformed a CRP-based decision rule, particularly in the 10-40 mg/L CRP range.

Conclusions:

• Integrating multiple blood parameters via machine learning offers superior diagnostic performance compared to single markers like CRP.
• The 'Virus vs. Bacteria' model shows promise for optimizing antibiotic administration and managing infections.
• This approach represents a step towards advanced, data-driven diagnostic tools in clinical practice.