A MACHINE learning-based model has demonstrated the ability to predict active tuberculosis (TB) in people with HIV (PWH) using clinical data, offering a promising tool for improving early detection and preventive treatment. The findings were presented at the Conference on Retroviruses and Opportunistic Infections (CROI) in San Francisco, highlighting how artificial intelligence can enhance patient care without requiring additional testing.
TB remains a major health concern for individuals with HIV, as co-infection with Mycobacterium tuberculosis (MTB) increases the risk of disease progression. While preventive therapy can help reduce this risk, existing screening methods often fail to accurately identify those who will develop active TB. To address this gap, researchers developed a machine learning model using random forest algorithms, trained on data from the Swiss HIV Cohort Study (SHCS). The study analyzed 55 PWH who developed active TB within six months of enrollment and 1,432 matched controls without TB, spanning data collected from 2000 to 2023. External validation was conducted using the Austrian HIV Cohort Study (AHIVCOS), which included 43 TB cases and 1,005 controls.
The model demonstrated strong predictive performance, achieving an area under the receiver operating characteristic (ROC) curve (AUC) of 0.83 (95% CI 0.8-0.86) in the SHCS. When adjusted for demographic variables and refined with fewer parameters, the model maintained an AUC of 0.72 in the Swiss cohort and 0.67 in the Austrian cohort. The study found that demographic factors such as region of origin and ethnicity influenced performance due to varying TB incidence rates, while socioeconomic factors, including profession, education, and mode of HIV transmission, also played a role. Additionally, key laboratory metrics—such as CD4 cell count and HIV RNA levels—proved crucial in predicting TB progression, alongside other health indicators like body mass index (BMI), creatinine, and hemoglobin levels.
Crucially, the machine learning model outperformed standard TB diagnostic tools, including the tuberculin skin test and interferon-gamma release assay, with a lower number needed to diagnose (1.96 vs. 4). By leveraging routinely collected clinical data, the model offers a cost-effective and efficient approach to identifying high-risk individuals, potentially improving TB prevention strategies for people with HIV.
These findings highlight the potential of artificial intelligence in transforming disease prediction and preventive care, paving the way for further research and implementation in clinical practice.
Reference: Bartl L et al. Machine Learning-Based Prediction of Active Tuberculosis in People With HIV Using Clinical Data. Abstract 260. The Conference on Retroviruses and Opportunistic Infections (CROI), San Francisco, CA, USA, March 9-12, 2025.
Anaya Malik | AMJ
