Hartz A, McKinney W P, Centor R, Krieg A, Simms G, Henck S
Med Decis Making. 1986 Jul-Sep;6(3):145-8. doi: 10.1177/0272989X8600600303.
Thresholds have traditionally been represented by a single number; the optimal management of the patient depends on whether his probability of disease is above or below this number. The concept of a threshold as a single number, however, inadequately represents the treatment approach of a group of physicians who do not have all the same threshold or a single physician who is uncertain about the exact value of the threshold. An alternative to a single valued threshold is to consider the threshold as having a probability distribution: for every probability that the patient has the disease there is a probability that the threshold is exceeded. This "stochastic" threshold model contains information about the uncertainty of the threshold estimation. Stochastic thresholds can be useful for testing the sensitivity of a management decision to the patient's probability of disease. They can also be used for comparing the standards of practice of individual physicians or comparing the practice of an individual physician with that of a group.
传统上，阈值由单个数字表示；患者的最佳治疗方案取决于其患病概率是高于还是低于这个数字。然而，将阈值视为单个数字的概念，不足以体现一群并非都有相同阈值的医生或一个对阈值的确切值不确定的医生的治疗方法。单值阈值的一种替代方法是将阈值视为具有概率分布：对于患者患该病的每一个概率，都存在超过阈值的概率。这种“随机”阈值模型包含有关阈值估计不确定性的信息。随机阈值可用于测试管理决策对患者患病概率的敏感性。它们还可用于比较个体医生的执业标准，或将个体医生的执业情况与一组医生的执业情况进行比较。
