Prediction Rules for Selective Radiographic Assessment of Extremity Injuries in Children and Adolescents

Abstract

To assess the potential for selective use of roentgenography in evaluating extremity injuries, prediction rules were developed based on prospective observations for 617 injured children and adolescents examined in our Emergency Department (phase 1) and tested on 601 examined 1 year later (phase 2). Logit analysis produced best-fitting statistical models for phase 1 data with significant (P less than 0.05) direct effects of gross signs, point tenderness, activity not routine, swelling moderate or severe, time from injury less than 6 hours, and pain with motion for upper extremity injuries; and, for lower extremity injuries, not knee injury, activity not routine, point tenderness, and foot injury. Prediction rules developed in phase 1 performed equally well when tested on phase 2 injuries. Data from both phases were combined, therefore, in analysis that produced risk estimates. For all injury types (ie, for injuries with all possible combinations of presence or absence of these findings), risk for fracture was derived. For upper extremity injuries, with a threshold risk for fracture of 20% used to select specific injury types for roentgenography, prediction rule outcomes were 18.1% of roentgenograms avoided and 5.3% of fractures missed. For lower extremity injuries, using a threshold risk of 10% to select injury types for roentgenography, outcomes were 25.8% of roentgenograms avoided and 5.3% of fractures missed. Alternative prediction rules allowed still greater roentgenogram avoidance, although missed fractures also increased. Risk of adverse functional outcome from missed fractures appeared small. Annual national cost savings from the elimination of 18.1% of upper and 25.8% of lower extremity roentgenographic evaluations was estimated at $103 million.