AUTHOR=Yi Dang , Liu Huanqing , Lei Qian , Li Tingting 
  
TITLE=Machine learning-guided clinical pharmacist interventions improve treatment outcomes in tuberculosis patients: a precision medicine approach
  
JOURNAL=Frontiers in Artificial Intelligence
  
VOLUME=Volume 8 - 2025
  
YEAR=2025
  
URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1679837
  
DOI=10.3389/frai.2025.1679837
  
ISSN=2624-8212
  
ABSTRACT=BackgroundThe heterogeneity in tuberculosis (TB) treatment responses necessitates a precision medicine approach. This study employed machine learning techniques to identify patient subtypes and optimize clinical pharmacist interventions.MethodsWe conducted a prospective cohort study involving 467 TB patients (218 in the intervention group receiving machine learning-guided pharmacist care and 249 in the control group receiving standard care). Primary outcomes included time to sputum conversion (smear, culture, TB-RNA) and duration of hospitalization; secondary outcomes encompassed adverse event rates (hepatotoxicity, renal impairment, etc.), cost-effectiveness, and biomarker dynamics. Patient stratification was performed using unsupervised learning (k-means/PCA) on clinical and laboratory parameters. Treatment outcomes were assessed via Kaplan–Meier survival analysis and Cox proportional hazards modeling, with prespecified subgroup analyses by risk clusters. Post hoc analyses (e.g., correlation heatmaps of biomarkers) were explicitly labeled as exploratory. Cost-effectiveness was evaluated using incremental cost per quality-adjusted hospital day saved (ICER).ResultsMachine learning identified 2 distinct patient subtypes (inflammatory vs. immunologic profiles). The intervention group showed significantly shorter hospital stays (primary outcome: median 49.0 vs. 57.0 days; log-rank p = 0.040). Adverse event rates were lower in the intervention group (26.1% vs. 27.7%). Cost analysis demonstrated potential savings of 5,000 CNY per patient in the intervention group. Limitations: Single-center design and modest sample size may limit generalizability. Unmeasured confounders (e.g., socioeconomic factors) could influence outcomes. Post hoc biomarker correlations require validation in independent cohorts.ConclusionMachine learning-guided pharmacist interventions improved TB treatment outcomes and reduced costs. Future multicenter studies should validate subtype-specific benefits.Clinical trial registrationhttps://www.chictr.org.cn/ identifier ChiCTR2300074328.