Diagnostic and Therapeutic Decision Making

Predictions from a statistical model may guide the decision to perform a diagnostic test. For instance, if the probability of an outcome, e.g. the presence of a disease, is lower than a certain threshold, no further tests have to be performed to further decrease the probability (Pauker and Kassirer, 1980). Similarly, if the probability is higher than a certain threshold, no further tests have to be performed. In contrast, for intermediate probabilities, a test may help to indicate the diagnosis.

Note that the concept of a test threshold implies that the diagnosis does not have to be 100% certain. It is sufficient to have a likelihood of the underlying disease that exceeds the threshold to initiate treatment. So, the combination of a set of symptoms may make an underlying disease or syndrome sufficiently likely to start treatment.

Risk groups may be constructed based on cut-off values of the probability of the outcome. For example, a high and low risk group may be considered, where different treatment strategies are followed. The construction of such risk groups should ideally agree with test thresholds or treatment thresholds that are justified on decision-analytic grounds.