Development and validation of machine-learning diagnostic models for identifying frailty in older adults with tuberculosis: a multicentre observational study protocol.

backgroundThe Frailty in Tuberculosis (FIT) study aims to assess frailty in older adults with tuberculosis (TB) using machine learning (ML) to develop diagnostic models to identify prevalent frailty. Despite its relevance, limited research exists on frailty among this population, and the potential of ML in identifying frailty remains underexplored.

methodsThe study will enroll older adults with TB hospitalized at multiple centers in Guizhou Province, China from June 2026 to May 2027. Frailty, which increases vulnerability to poor outcomes, is understudied despite being particularly relevant in the older TB population. The collected data will include clinical, laboratory, and demographic variables, which will be analyzed using ML algorithms such as logistic regression, random forest, and XGBoost. Rigorous validation, including repeated stratified five-fold cross-validation and external validation, will ensure model reliability and generalizability. The primary outcome is the baseline frailty status assessed within 1 week after admission, whereas secondary outcomes include hospital stay, healthcare-seeking delay, and 30-day mortality. DISCUSSION: By integrating ML methods, the FIT study aims to support early frailty identification, promote targeted interventions, and enhance clinical decision-making. Ultimately, the findings may support clinical decision-making for older adults with TB by providing an accessible and efficient approach to frailty assessments that guide personalized care.