Machine Learning-based Prediction of Active Tuberculosis in People With HIV Using Clinical Data

Area of Science:

• Medical Informatics
• Infectious Disease Epidemiology
• Machine Learning in Healthcare

Background:

• Coinfection with Mycobacterium tuberculosis (MTB) and human immunodeficiency virus (HIV) presents a significant global health challenge.
• Individuals with MTB infection have an increased risk of developing active tuberculosis (TB), a progression that can be prevented with timely therapy.
• Existing diagnostic methods often fail to identify individuals at high risk for subsequent active TB development.

Purpose of the Study:

• To develop and validate machine learning models for predicting incident active TB in people with HIV (PWH).
• To assess the performance of these models against standard diagnostic tests for TB risk stratification.

Main Methods:

• Random forest models were developed using routinely collected clinical data from the Swiss HIV Cohort Study (SHCS) for training.
• The training dataset included 55 PWH who developed active TB and 1432 matched PWH without TB.
• External validation was performed using data from the Austrian HIV Cohort Study (AHIVCOS), comprising 43 PWH with incident active TB and 1005 PWH without TB.

Main Results:

• The model achieved an area under the receiver operating characteristic (ROC) curve (AUC) of 0.83 in the SHCS.
• After adjustments and re-fitting, AUC values of 0.72 (SHCS) and 0.67 (AHIVCOS) were obtained.
• The machine learning model demonstrated superior performance compared to standard care (tuberculin skin test and interferon-gamma release assay), with a lower number needed to diagnose (1.96 vs. 4).

Conclusions:

• Machine learning models show significant potential for enhancing the care of PWH by improving the identification of individuals who could benefit from preventive TB treatment.
• These models require no additional data collection and incur minimal costs, offering a cost-effective solution.
• The findings support the integration of predictive modeling into clinical practice for better management of TB/HIV coinfection.