Applying decision analysis in therapeutic drug monitoring: using receiver-operating characteristic curves in...

In the first of a three-part series, decision-analysis techniques were applied to the evaluation, comparison, and selection of analytic systems for the purpose of improving the information content of serum drug concentration determinations. Receiver-operating characteristic (ROC) curves that graphically compare true-positive and false-positive rates at selected drug concentrations were developed and used to evaluate the clinical performance of analytic systems. Decision matrices were constructed using simulated toxicity and serum-concentration data from 150 subjects. These decision matrices were used to construct ROC curves and the curves were compared with prototype curves representing ideal, useful, useless, and hazardous test systems. The ROC curves were used to determine the analytic method (e.g., enzyme-multiplied immunoassay technique versus nephelometric assay) and analyte (e.g., free versus total drug concentration) that conveyed the most information and yielded the fewest errors in classifying patients as toxic or nontoxic. The use of ROC curves to compare new or modified analytic test systems, to monitor use of a given analytic method within an institution over time, and to compare the clinical performance of a given assay method in the therapeutic drug monitoring practices of various institutions is discussed. Decision-analysis techniques can assist practitioners in selecting the optimal analytic test system for reducing analytic variation as well as improving the quality and predictive ability of the serum drug concentration.