Evaluation of three theophylline dosing methods in pediatric patients

Abstract

Three methods used for individualizing theophylline dosing were prospectively evaluated in 34 pediatric patients to compare the methods' ability to accurately predict steady-state serum theophylline concentrations (STCs) from non-steady-state data. Methods evaluated included a Bayesian weighted sum of squares regression program, an algebraic method developed by Chiou, and a commonly used population-based pediatric dosing algorithm. An expanded retrospective evaluation was also done to further compare the Bayesian and Chiou methods. Patients (aged 1-11 y) admitted to the hospital for acute bronchospasm refractory to inhaled beta agonists and with physician's orders to receive intravenous aminophylline were eligible for participation. Study patients received a 6-mg/kg loading dose aminophylline, followed by a 0.8-1.0 mg/kg/h constant aminophylline infusion. STCs were obtained 0.5 and 5.5 hours postinfusion. Subjects were randomized to one of three dosing method groups (Bayesian, Chiou, or pediatric algorithm). Doses were calculated using the assigned method to attain a target steady-state STC. Predictions from each dosing method were compared with actual serum concentrations for bias and precision. Additionally, analysis of fit-to-the-line-of-identity for predicted versus observed STCs was evaluated for each method. Precision of methods was also compared with regard to their ability to predict within 20 percent of their observed steady-state STC. Results from the prospective evaluation showed no significant difference between the three methods tested. Predicted STCs fell within 20 percent of their observed steady-state concentrations for 18 percent (2/11) of patients in the algorithm group, and 45 percent (5/11) of patients in the Chiou group and 17 percent (2/12) of the patients in the Bayesian group met this criterion. Retrospective analysis of all 34 patients demonstrated that the Bayesian and Chiou methods had similar bias and precision and no statistical difference was found between them. The results from this evaluation suggest that the pediatric dosing algorithm is equivalent in predictive bias and precision to the Bayesian and Chiou methods as well as in its ability to identify doses that result in steady-state STCs within 20 percent of their target values. Given the relative inaccuracy of these methods, cautious use of these techniques is recommended when evaluating non-steady-state STCs obtained from children during the acute stages of reactive airway disease.