Objectives This study aimed to investigate pathogen-specific prevalence, 30-day mortality, and host factors associated with infection in a large tertiary-care hospital cohort, and to evaluate whether comorbidity patterns and machine-learning models can predict individual outcomes. Methods We conducted a 39-month retrospective cohort study of 16,282 hospitalized patients with positive bacterial cultures at a tertiary-care university hospital. Cox proportional hazards models identified mortality risk factors. Comorbidity-pathogen associations were examined using logistic regression. Five machine-learning classifiers were trained to predict 30-day mortality. Results Enterococcus faecium contributed deaths comparable to Staphylococcus aureus and Escherichia coli despite markedly lower prevalence, reflecting the highest case fatality rate across all infection sites (adjusted HR 1.91, 95% CI 1.65--2.22). Multidrug resistance or sex were not independently associated with mortality. Comorbidity profiles were strikingly uniform across pathogens and infection sites; only hematologic malignancies were overrepresented in E. faecium bloodstream infections. Machine-learning models achieved limited predictive performance overall, with higher sensitivity only in elderly patients and those with E. faecium infections. Conclusions Nosocomial mortality is shaped by pathogen-specific case fatality rather than prevalence alone. Uniform host vulnerability profiles constrain individual risk prediction and argue for dedicated E. faecium surveillance in tertiary-care settings.