External Validation of an Algorithm to Predict Adjacent Musculoskeletal Infection in Pediatric Patients With Septic Arthritis. Issue 10 (November 2020)

• Record Type: Journal Article
• Authors: Hunter, Sarah; Kennedy, Jim; Baker, Joseph F.
• Is Part Of: Journal of pediatric orthopaedics. Volume 40:Issue 10(2020)
• Publication Date: 2020-11
• Subjects: pediatric septic arthritis -- external validation -- pediatric bone and joint infection -- adjacent infection; Pediatric orthopedics -- Periodicals; 618.927
• DOI: 10.1097/BPO.0000000000001618
• Languages: English
• ISSNs: 0271-6798

Abstract

Abstract : Background: Septic arthritis (SA) remains a potentially morbid disease in the pediatric population. Magnetic resonance imaging (MRI) is the most sensitive tool for recognizing associated osteomyelitis and intramuscular abscess, but is a limited resource. The aim of this study is to externally validate a previously developed algorithm (Rosenfeld and colleagues) to predict adjacent infection in pediatric patients diagnosed with SA. Methods: We identified 120 children under 16 with presumed SA presenting to a tertiary referral center between 2008 and 2018. Patients without confirmed SA, those with insufficient data, and patients who did not receive perioperative MRI were excluded, leaving 53 patients. The previous algorithm suggests that patient age (above 4 y), C-reactive protein (>8.9 mg/L), platelet count (<310×10 3 cells/µL), duration of symptoms (>3 d), and absolute neutrophil count (>7.2×10 3 cells/µL) are risk factors for adjacent infection, with 3 or more variables signifying a "positive" result. Comparing against the gold standard of MRI, the accuracy of the algorithm was validated in terms of sensitivity, specificity, likelihood ratio (LR), and positive and negative predictive value. Discrimination and calibration of this algorithm have been assessed using receiver operating curve analysis and calibration plots. Results: The sensitivity and specificity of criteria from Rosenfeld algorithm were 73% and 44%, respectively. Receiver operating curve showed poor discrimination [area under the curve=0.54, confidence interval (CI): 0.26-0.83]. The positive predictive value was 55.9% and the negative predictive value was 63.1% with LR +1.23 (CI: 0.87-1.98) and LR −0.61 (CI 0.28-1.30). Only 53% of patients with 4 or more criteria had an adjacent infection on MRI. Examining our cohort, children with a positive MRI finding had higher mean C-reactive protein (77 vs. 122 mg/L, P =0.04) and were more likely to have waited >72 hours days between symptom onset and hospital presentation ( P =0.03). Conclusion: Although treatment algorithms are an attractive tool to guide clinicians and resource allocation, they need to take into account the local population characteristics before routine implementation. Level of Evidence: Level IV–retrospective cohort study.