• Benjamin D Welling, Lee S Haruno, and Scott B Rosenfeld.
• B. D. WellingBaylor College of Medicine, Department of Orthopedic Surgery, Houston, TX, USA L. S. Haruno, S. B. RosenfeldTexas Children's Hospital, Division of Orthopedic Surgery, Houston, TX, USA.
• Clin. Orthop. Relat. Res. 2018 Jan 1; 476 (1): 153-159.

BackgroundSeptic arthritis is frequently associated with adjacent infections including osteomyelitis and subperiosteal and intramuscular abscesses. While often clinically indiscernible from isolated septic arthritis, the diagnosis of adjacent infections is important in determining the need for additional surgical intervention. MRI has been used as the diagnostic gold standard for assessing adjacent infection. Routine MRI, however, can be resource-intensive and delay surgical treatment. In this context, there is need for additional diagnostic tools to assist clinicians in determining when to obtain preoperative MRI in children with septic arthritis. In a previous investigation by Rosenfeld et al., an algorithm, based on presenting laboratory values and symptoms, was derived to predict adjacent infections in septic arthritis. The clinical applicability of the algorithm was limited, however, in that it was built from and applied on the same population. The current study was done to address this criticism by evaluating the predictive power of the algorithm on a new patient population.Questions/Purposes(1) Can a previously created algorithm used for predicting adjacent infection in septic arthritis among pediatric patients be validated in a separate population?MethodsRecords for all pediatric patients (1-18 years old) surgically treated for suspected septic arthritis during a 3-year period were retrospectively reviewed (109 patients). Of these patients, only those with a diagnosis of septic arthritis confirmed by synovial fluid analysis were included in the study population. Patients without confirmation of septic arthritis via synovial fluid analysis, Gram stain, or culture were excluded (34 patients). Patients with absence of MRI, younger than 1 year, insufficient laboratory tests, or confounding concurrent illnesses also were excluded (18 patients), resulting in a total of 57 patients in the study population. Five variables which previously were shown to be associated with risk of adjacent infection were collected: patient age (older than 4 years), duration of symptoms (> 3 days), C-reactive protein (> 8.9 mg/L), platelet count (< 310 x 10 cells/µL), and absolute neutrophil count (> 7.2 x 10 cells/µL). Adjacent infections were determined exclusively by preoperative MRI, with all patients in this study undergoing preoperative MRI. MR images were read by pediatric musculoskeletal radiologists and reviewed by the senior author. According to the algorithm we considered the presence of three or more threshold-level variables as a "positive" result, meaning the patient was predicted to have an adjacent infection. Comparing against the gold standard of MRI, the algorithm's accuracy was evaluated in terms of sensitivity, specificity, positive predictive value, and negative predictive value.ResultsIn the new population, the sensitivity and specificity of the algorithm were 86% (95% CI, 0.70-0.95) and 85% (95% CI, 0.64-0.97), respectively. The positive predictive value was determined to be 91% (95% CI, 0.78-0.97), with a negative predictive value of 77% (95% CI, 0.61-0.89). All patients meeting four or more algorithm criteria were found to have septic arthritis with adjacent infection on MRI.ConclusionsCritical to the clinical applicability of the above-mentioned algorithm was its validation on a separate population different from the one from which it was built. In this study, the algorithm showed reproducible predictive power when tested on a new population. This model potentially can serve as a useful tool to guide patient risk stratification when determining the likelihood of adjacent infection and need of MRI. This better-informed clinical judgement regarding the need for MRI may yield improvements in patient outcomes, resource allocation, and cost.Level Of EvidenceLevel II, diagnostic study.

38 keywords...

hide…