Abstract:
|
New and evolving medical technologies have motivated the development of treatment decision rules (TDRs) that incorporate complex, expensive data such as imaging (e.g. EEGs). In practice, we aim for TDRs to be valuable for physicians and patients, such that we reduce unnecessary, costly testing while still assigning decisions. For a given TDR, a patient's characteristics lie some distance from the optimal decision boundary separating treatment classes. We expect less certainty or confidence in decisions for patients near the boundary, and more certainty for those farther from the boundary. We propose measuring confidence by estimating the probability of ultimately receiving a particular treatment decision given a patient’s data, as well as the probability that a patient’s final decision will agree with the current decision. As a patient's data accumulates, the decision is sequentially updated and the probabilities are reassessed until high confidence is achieved. We present results from extensive simulation studies and discuss the relationship between probability and distance to a decision boundary. Lastly, we give recommendations for practical use of the confidence measures.
|