Development of a predictive machine learning model for pathogen profiles in patients with secondary immunodeficiency

Area of Science:

• Infectious Diseases
• Machine Learning
• Immunology

Background:

• Secondary immunodeficiency (SID) increases susceptibility to infections.
• Causes of SID include HIV, chronic diseases, malignancy, and immunosuppressants.
• Pathogen profiles in non-HIV SID are largely unknown.

Purpose of the Study:

• To develop predictive models for infectious pathogens in patients with SID from various causes (excluding HIV).
• To aid clinicians in early diagnosis and treatment decisions.

Main Methods:

• Utilized machine learning techniques.
• Developed predictive models using patient medical records from the First Affiliated Hospital of Nanchang University.
• Focused on secondary immunodeficiency patients with diverse etiologies, excluding HIV.

Main Results:

• The Gradient Boosting Machine model demonstrated superior performance.
• Achieved the highest accuracy of 91.01%.
• Outperformed other models by a significant margin (13.48%, 7.14%, 4.49%).

Conclusions:

• Developed models effectively predict potential infectious pathogens in non-HIV secondary immunodeficiency.
• Facilitates prompt antibiotic administration before culture results.
• Supports timely clinical decision-making for improved patient outcomes.