Differentiation between septic arthritis and transient synovitis of the hip in children with clinical prediction algorithms.

BACKGROUND Differentiation between septic arthritis and transient synovitis of the hip in children can be difficult. Kocher et al. recently developed a clinical prediction algorithm for septic arthritis based on four clinical variables: history of fever, non-weight-bearing, an erythrocyte sedimentation rate of >or=40 mm/hr, and a serum white blood-cell count of >12000/mm(3) (>12.0 x 10(9)/L). The purpose of this study was to apply this clinical algorithm retrospectively to determine its predictive value in our patient population. METHODS A retrospective review was performed to identify all children who had undergone a hip arthrocentesis for the evaluation of an irritable hip at our institution between 1992 and 2000. One hundred and sixty-three patients with 165 involved hips satisfied the criteria for inclusion in the study and were classified as having true septic arthritis (twenty hips), presumed septic arthritis (twenty-seven hips), or transient synovitis (118 hips). RESULTS Patients with septic arthritis (true and presumed; forty-seven hips) differed significantly (p < 0.05) from patients with transient synovitis (118 hips) with regard to the erythrocyte sedimentation rate, differential of serum white blood-cell count, total white blood-cell count and differential in the synovial fluid, gender, previous health-care visits, and history of fever. If the four independent multivariate predictors of septic arthritis proposed by Kocher et al. were present, the predicted probability of the patient having septic arthritis was 59% in our study, in contrast to the 99.6% predicted probability in the patient population described by Kocher et al. Statistical analyses demonstrated that the best model to describe our patient population was based on three variables: a history of fever, a serum total white blood-cell count of >12000/mm(3) (>12.0 x 10(9)/L), and a previous health-care visit. When all three variables were present, the predicted probability of the patient having septic arthritis was 71%. CONCLUSIONS Although the use of a clinical prediction algorithm to differentiate between septic arthritis and transient synovitis may have improved the utility of existing technology and medical care to facilitate the diagnosis at the institution at which the algorithm originated, application of the algorithm proposed by Kocher et al. or of our three-variable model does not appear to be valid at other institutions.