Abstract / Summary
This study aimed to identify factors associated with Staphylococcus aureus bloodstream infection ( S aureus BSI), compare 28-day survival after propensity score matching (PSM), and develop a prognostic model for 28-day mortality. Adult intensive care unit patients with culture-confirmed BSI were identified in Medical Information Mart for Intensive Care IV. Factors associated with S aureus BSI were assessed in the full cohort by univariable logistic regression. Patients with S aureus and non- S aureus BSI were matched 1:1 using nearest-neighbor PSM without a caliper. Survival was examined using Kaplan–Meier analysis and Cox models with matched-pair cluster-robust standard errors. Among patients with S aureus BSI, prognostic variables were screened by clinical domain and best-subset selection using minimum Bayesian Information Criterion; 10-fold cross-validated least absolute shrinkage and selection operator regression served as a sensitivity analysis. Model performance was assessed by bootstrap internal validation and refitting in an external cohort. The cohort included 2221 patients, including 297 with S aureus BSI. Shorter time from admission to confirmed BSI, absence of tumor, shorter invasive ventilation and vascular catheterization, and fewer antibiotic types before BSI were associated with S aureus BSI. After 1:1 PSM, 28-day mortality was 26.9% versus 24.2%, with no significant survival difference (hazard ratio 1.11, 95% confidence interval 0.80–1.53; P = .539). The prognostic model included age, maximum red cell distribution width, minimum white blood cell count, minimum activated partial thromboplastin time, and maximum albumin. The apparent C-statistic was 0.816 (95% confidence interval 0.757–0.870), and the optimism-corrected C-statistic was 0.804 (95% confidence interval 0.752–0.861). After refitting in 32 external-cohort patients, the same 5 predictors yielded an apparent C-statistic of 0.778 (95% confidence interval 0.591–0.929). S aureus BSI was associated with a distinct prediagnosis profile, but pathogen group was not associated with 28-day survival after matching. The 5-variable model showed good optimism-corrected discrimination. The original prediction equation requires validation in larger external cohorts.