Methods: In this retrospective, case-control study at a single, high-volume, tertiary-care, academic medical center, we used ICD-9 diagnostic codes to identify all adult inpatients from 2005-2015 with potential SEA. Records were systematically abstracted and designated as cases or controls based on an imaging and microbiologic algorithm. Logistic regression was used to develop a predictive model.
Results: Univariate screening of 250 eligible patients identified several covariates that varied significantly between cases (n = 162) and controls (n = 88). A multivariable model identified seven as independent predictors of SEA in this population (Figure), associated with 90% sensitivity, 84% specificity, and an AUC 0.91 (95% CI 0.87, 0.95). Removing bacteremia from the model, as it may not be apparent at initial clinical presentation, had little effect on sensitivity but reduced the specificity to 76%. ESR >50 (OR 6.7, 95% CI 2.1, 20.8) was a strong predictor of SEA in the 73% of cases and 32% of controls with available data.
Conclusion: Our model reliably discriminated SEA from other spinal pathologies in this retrospective study. It requires prospective validation in a multicenter study.
A. Holers, None
D. Lewis, None
J. Fitzgerald, None
P. Visintainer, None