PURPOSE: Misdiagnosis or delayed diagnosis of paravertebral and/or iliopsoas abscess (PVIPA) has been frequently reported to be associated with unfavorable prognosis. We aimed to develop a scoring algorithm that can easily and accurately identify patients at greater risk for PVIPA among individuals with community-onset bloodstream infections. METHODS: In a multicenter, retrospective cohort study, the score was developed with the first four study years and validated with the remaining two years. Applying logistic regression, the score values of prediction determinants were derived from the adjusted odds ratios (AOR). The performance of the scoring algorithm was assessed with the receiver operating characteristic (ROC) curve. RESULTS: In the derivation (3869 patients) and validation (1608) cohorts, patients with PVIPA accounted for 1.7% and 1.4%, respectively. In the derivation cohort, five independent predictors of PVIPA were recognized using multivariable analyses: time-to-defervescence >
5 days (AOR, 7.00
2 points), Panton-Valentine Leukocidin (PVL)-producing Staphylococcus aureus (AOR, 5.98
2 points), intravenous drug users (AOR, 2.60
1 points), and comorbid hemato-oncology (AOR, 0.41
-1 point) or liver cirrhosis (AOR, 2.56
1 points). In the derivation and validation cohorts, areas under ROC curves (95% confidence intervals) of the prediction algorithm are 0.83 (0.77-0.88) and 0.85 (0.80-0.90), and a cutoff score of + 2 represents sensitivity of 83.3% and 95.7%, specificity of 68.6% and 67.7%, positive predictive values of 4.4% and 4.1%, and negative predictive values of 99.6% and 99.9%, respectively. CONCLUSIONS: Of a scoring algorithm with substantial sensitivity and specificity in predicting PVIPA, PVL-producing S. aureus and Time-to-defervescence >
5 days were crucial determinants.