Abstract
Vancomycin-intermediate Staphylococcus aureus (VISA) is a multi-drug-resistant pathogen of significant clinical concern. Various S. aureus strains can cause infections, from skin and soft tissue infections to life-threatening conditions such as bacteremia and pneumonia. VISA infections, particularly bacteremia, are associated with high mortality rates, with 34% of patients succumbing within 30 days. This study aimed to develop predictive models for VISA (including hVISA) bacteremia outcomes using data mining techniques, potentially improving patient management and therapy selection. We focused on three endpoints in patients receiving traditional vancomycin therapy: VISA persistence in bacteremia after 7 days, after 30 days, and patient mortality. Our analysis incorporated 29 risk factors associated with VISA bacteremia. The resulting models demonstrated high predictive accuracy, with 82.0-86.6% accuracy for 7-day VISA persistence in blood cultures and 53.4-69.2% accuracy for 30-day mortality. These findings suggest that data mining techniques can effectively predict VISA bacteremia outcomes in clinical settings. The predictive models developed have the potential to be applied prospectively in hospital settings, aiding in risk stratification and informing treatment decisions. Further validation through prospective studies is warranted to confirm the clinical utility of these predictive tools in managing VISA infections.