Haoping Dai, Hongtao Li, Changming Xiao
Multi-method, indication-contextualized FAERS analysis distinguishes BJI-supported from general reporting signals and provides a transparent framework for prioritizing monitoring during prolonged antibiotic therapy. Findings are hypothesis-generating and require validation in controlled observational studies. They prioritize safety signals rather than providing estimates of incidence, absolute risk, causality, dose-response, or renal-function effects.
BACKGROUND: Bone and joint infections (BJI) require prolonged antibiotic therapy that may amplify cumulative toxicity risk, yet indication-contextualized safety data remain sparse.
METHODS: Using the FDA Adverse Event Reporting System (FAERS, 2004Q1-2025Q4; 20,006,981 deduplicated cases), we identified 9,072 cases involving 17 BJI-target antibiotics. Four disproportionality measures, namely reporting odds ratio (ROR), proportional reporting ratio (PRR), information component (IC), and empirical Bayes geometric mean (EBGM), were applied against the full FAERS database. We then performed BJI within-cohort contextualization, primary-suspect-only (PS-only) sensitivity analysis, exploratory XGBoost machine learning, and Weibull time-to-onset (TTO) modeling, in accordance with the READUS-PV reporting framework.
RESULTS: Of 153 drug-adverse drug event (ADE) pairs, 103 met the prespecified four-method robust full-FAERS signal criterion; 28 were retained after BJI within-cohort contextualization. Indication-supported signals included vancomycin-nephrotoxicity (within-cohort ROR 3.30, 95% CI 2.95-3.70), linezolid-hematologic toxicity (3.65, 3.21-4.16), ceftaroline-hematologic toxicity (4.81, 3.42-6.77), and ertapenem-neurotoxicity (4.76, 3.50-6.47). Several strong full-FAERS signals (e.g., fluoroquinolone-tendon and broad-spectrum-C. difficile infection) were attenuated within the BJI cohort or under PS-only restriction, highlighting the role of prescribing pattern and reporting-role confounding. Weibull modeling indicated predominantly early-onset reporting patterns.
CONCLUSION: Multi-method, indication-contextualized FAERS analysis distinguishes BJI-supported from general reporting signals and provides a transparent framework for prioritizing monitoring during prolonged antibiotic therapy. Findings are hypothesis-generating and require validation in controlled observational studies. They prioritize safety signals rather than providing estimates of incidence, absolute risk, causality, dose-response, or renal-function effects.