Whole-Blood 3-Gene Signature as a Decision Aid for Rifapentine-based Tuberculosis Preventive Therapy

Area of Science:

• Pharmacogenomics
• Infectious Disease Epidemiology
• Biomarker Discovery

Background:

• Systemic drug reactions (SDRs) are a significant safety concern with the 3HP regimen (weekly rifapentine plus isoniazid for 12 doses) used for latent tuberculosis infection (LTBI).
• Predicting SDRs and identifying at-risk individuals before treatment initiation is crucial for enhancing the cost-effectiveness and safety of LTBI programs.

Purpose of the Study:

• To develop and validate a predictive model for SDRs associated with the 3HP regimen.
• To identify key biomarkers for predicting SDR risk in patients undergoing LTBI treatment.

Main Methods:

• Prospective recruitment of 187 participants receiving 3HP, with transcriptomic data generated from a pilot cohort.
• Selection of candidate genes using a hierarchical systems biology model and therapy-biomarker pathway approach.
• Development and validation of interpretable machine learning models (SHAP values) for SDR risk prediction using RT-qPCR data.

Main Results:

• A 3-gene predictive model for 3HP-related SDRs demonstrated high performance with a sensitivity of 0.972 and specificity of 0.947.
• The model's accuracy was validated across combined training and testing cohorts and performed consistently across subgroups.
• SHapley Additive exPlanations (SHAP) values were utilized for model interpretability and risk assessment.

Conclusions:

• The developed prediction model for 3HP-related SDRs can guide the establishment of safe and personalized treatment regimens.
• This tool supports the broader implementation of LTBI programs by mitigating drug safety concerns.
• The findings hold potential translational value for future research into drug-related hypersensitivity reactions.