Machine learning based on routine laboratory indicators promoting the discrimination between active tuberculosis and...

Area of Science:

• Medical Diagnostics
• Machine Learning in Healthcare
• Infectious Disease Research

Background:

• Distinguishing active tuberculosis (ATB) from latent tuberculosis infection (LTBI) presents a significant clinical challenge.
• Routine laboratory indicators are often insufficient for accurate differentiation.
• Novel diagnostic approaches are needed to improve ATB/LTBI discrimination.

Purpose of the Study:

• To evaluate machine learning models for differentiating ATB from LTBI.
• To identify routine laboratory indicators that best predict active tuberculosis.
• To develop and validate a predictive model for ATB diagnosis.

Main Methods:

• Machine learning models were developed using routine laboratory data from two independent cohorts.
• Gradient Boosting Machine (GBM) was identified as the optimal model.
• Model performance was assessed using sensitivity and specificity in discovery and validation sets.

Main Results:

• The GBM model achieved high accuracy in differentiating ATB from LTBI.
• In the validation cohort, the GBM model demonstrated a sensitivity of 87.63% and specificity of 91.34%.
• Key differentiating laboratory indicators included tuberculosis-specific antigen/phytohemagglutinin ratio, red blood cell volume distribution width, albumin, and lymphocyte count.

Conclusions:

• Machine learning, specifically the GBM model, offers a promising tool for accurate ATB diagnosis.
• The developed model effectively utilizes routine laboratory indicators for ATB/LTBI differentiation.
• This approach has the potential to significantly benefit clinical practice in tuberculosis management.