Development of a Machine Learning-Based Prediction Model to Differentiate Infectious and Non-Infectious Diseases in...

Area of Science:

• Internal Medicine
• Clinical Diagnostics
• Medical Informatics

Background:

• Fever has diverse causes, including infections and noninfectious inflammatory diseases (NIID).
• Serum ferritin (SF) aids differentiation but lacks sufficient specificity.
• Accurate early diagnosis of fever etiology is clinically important.

Purpose of the Study:

• To develop a diagnostic prediction model for differentiating infectious from non-infectious febrile illnesses.
• To utilize common blood tests and SF levels for early fever cause identification.
• To improve diagnostic accuracy in patients with undiagnosed fever.

Main Methods:

• Retrospective observational study of 143 adult patients with fever of unidentified origin.
• Machine learning and multivariable logistic regression for model development.
• Evaluation of model performance using AUC, shrinkage coefficient, and likelihood ratio.

Main Results:

• A five-factor model including white blood cell count, neutrophil percentage, platelet count, lactate dehydrogenase, and log-transformed SF was developed.
• The model achieved an AUC of 0.794 (95% CI: 0.721-0.867).
• Sensitivity was 77.1% and specificity was 68.5%.

Conclusions:

• A practical prediction model using readily available blood tests was successfully developed.
• This model can assist clinicians in differentiating infectious diseases from other fever causes early in the diagnostic process.
• The findings support the early application of this model for undiagnosed fever patients.