Background Establishing the diagnosis in a child presenting with an atraumatic limp can be challenging. There is particular difficulty distinguishing septic arthritis (SA) from transient synovitis (TS) and consequently clinical prediction algorithms have been devised to differentiate the conditions using the presence of fever, raised erythrocyte sedimentation rate (ESR), raised white cell count (WCC) and inability to weight bear. Within Europe measurement of the ESR has largely been replaced with assessment of C-reactive protein (CRP) as an acute phase protein. We have evaluated the utility of including CRP in a clinical prediction algorithm to distinguish TS from SA.
Method All children with a presentation of ‘atraumatic limp’ and a proven effusion on hip ultrasound between 2004 and 2009 were included. Patient demographics, details of the clinical presentation and laboratory investigations were documented to identify a response to each of four variables (Weight bearing status, WCC >12,000 cells/m3, CRP >20mg/L and Temperature >38.5 degrees C. The definition of SA was based upon microscopy and culture of the joint fluid collected at arthrotomy.
Results 311 hips were included within the study. Of these 282 were considered to have transient synovitis. 29 patients met criteria to be classified as SA based upon laboratory assessment of the synovial fluid. The introduction of CRP eliminated the need for a four variable model as the use of two variables (CRP and weight bearing status) had similar efficacy. An algorithm that indicated a diagnosis of SA in individuals who could not weight-bear and who had a CRP >20mg/L correctly classified SA in 94.8% individuals, with a sensitivity of 75.9%, specificity of 96.8%, positive predictive value of 71.0%, and negative predictive value of 97.5%. CRP was a significant independent predictor of septic arthritis.
Conclusions CRP was a strong independent risk factor of septic arthritis, and its inclusion within a regression model simplifies the diagnostic algorithm, such that a two-variable model correctly classified 95% individuals with SA. Nevertheless, this and similar algorithms are generally more reliable in excluding SA, than confirming SA, and therefore a clinician’s acumen remains important in identifying SA in those individuals with a single abnormal variable.