Objectives: Any diagnostic test has an indication area of prior probabilities wherein the gain in diagnostic certainty outweighs its loss. Here, we investigate whether indication area and the maximum diagnostic gain are robust measures if we assume test dependence, alternative physician's heuristics, and varying patient's utilities. Study Design and Setting: Three mathematical functions for the dependence of test sensitivity (Se) and specificity (Sp) on the prior disease probability were studied. In addition, three different decision heuristics for further management were explored for the case that "no test" would be done. Finally, the valuation of test outcomes was varied. A sensitivity analysis was performed to determine the impact of the alternative assumptions on the indication area and maximum diagnostic gain. Results: By assuming test dependence, the indication area shifts to higher priors and increases the maximum diagnostic gain. Decision strategies assuming a "threshold before treat" can inadvertently widen the indication area and increase the maximum diagnostic gain. Varying patient utilities will usually reduce the net diagnostic gain. A sensitivity analysis revealed the robustness of the model. Conclusion: The indication area and the maximum diagnostic gain are robust measures of test performance and are easier to interpret than the classical performance measures. Published by Elsevier Inc.
|Journal||Journal of Clinical Epidemiology|
|Publication status||Published - 1 Jan 2015|