An Integrated Data-Driven Model for Clinical Phenotyping of Tuberculosis Disease Severity

Area of Science:

• Medical Informatics
• Infectious Disease Epidemiology
• Biostatistics

Background:

• Current binary classifications of tuberculosis (TB) disease severity may not fully represent the spectrum of clinical presentations.
• Disease progression in TB is associated with increased bacterial load and inflammation, correlating with worse outcomes.

Purpose of the Study:

• To develop and validate a novel method, tuberculosis SeveriTy Assessment Tool for Informed Stratification (TB-STATIS), for classifying TB disease severity phenotypes.
• To move beyond binary classifications and identify distinct, data-driven disease severity classes.

Main Methods:

• TB-STATIS integrates diverse data sources including smear microscopy, chest X-ray findings, and symptoms.
• The method employs event-based modeling, a data-driven disease progression modeling framework.
• Simulations were conducted to assess TB-STATIS performance across various sample sizes and data uncertainties.

Main Results:

• Simulations demonstrated TB-STATIS's ability to accurately identify true disease classes under different conditions.
• Application to South African TB cohort and a global clinical trial dataset revealed correlations between TB-STATIS classes and culture conversion.
• The generated disease strata were found to be clinically meaningful.

Conclusions:

• TB-STATIS provides a robust, data-driven approach to stratify tuberculosis disease severity.
• This tool enhances the understanding of TB disease phenotypes at presentation and their clinical relevance.
• The method shows potential for improving patient stratification in TB research and clinical practice.