Asch D A, Patton J P, Hershey J C
Robert Wood Johnson Foundation Clinical Scholars Program, University of Pennsylvania School of Medicine, Philadelphia 19104-4283.
Med Decis Making. 1990 Jan-Mar;10(1):47-57. doi: 10.1177/0272989X9001000108.
In evaluating diagnostic tests, traditional methods in decision analysis often emphasize how the results of the test will or will not affect patient management. Clinicians are advised to avoid testing if the results will not alter treatment strategy or other management plans. But patients may be interested in the prognostic information that testing provides even if it is not used to guide treatment. The authors present a model that defines this prognostic information as the expected deviation from the prior probability of disease. The model generates utility functions that are curvilinear over prior probabilities. Whereas the traditional threshold approach to medical decision making produces at most three zones of management strategy (withhold, test, and treat), the incorporation of prognostic information into threshold analysis produces two additional zones (test but withhold anyway, and test but treat anyway). Conditions under which one or both of these additional zones will appear are described. The model justifies the practice of performing tests that cannot alter management plans; it explains the unwillingness of some patients to undergo diagnostic testing when they fear unwanted results; and it provides a method for quantifying the sensitive nature of confidential tests. The model is illustrated using the antibody test for the Smith antigen. This test has a high specificity but a low sensitivity for lupus erythematosus. Clinicians may use the test because a positive result will support their prior suspicion of disease even though they may not change their management strategy if the test result is negative. The advantage of testing in this setting lies in the test's potential for establishing with virtual certainty that the disease is present. Thus, the test is valued for the prognostic information it provides apart from its effect on patient management.
在评估诊断测试时，决策分析中的传统方法通常强调测试结果将如何或不会影响患者管理。建议临床医生如果测试结果不会改变治疗策略或其他管理计划，就避免进行测试。但是，即使测试结果不用于指导治疗，患者可能也会对测试提供的预后信息感兴趣。作者提出了一个模型，将这种预后信息定义为与疾病先验概率的预期偏差。该模型生成的效用函数在先前概率上是曲线的。传统的医学决策阈值方法最多产生三个管理策略区域（不进行测试、进行测试和进行治疗），而将预后信息纳入阈值分析会产生另外两个区域（进行测试但无论如何都不进行治疗，以及进行测试但无论如何都进行治疗）。描述了出现这两个附加区域中的一个或两个的条件。该模型证明了进行无法改变管理计划的测试的合理性；它解释了一些患者在担心得到不理想结果时不愿接受诊断测试的原因；并且它提供了一种量化保密测试敏感性的方法。使用针对史密斯抗原的抗体测试来说明该模型。该测试对红斑狼疮具有高特异性但低敏感性。临床医生可能会使用该测试，因为阳性结果将支持他们先前对疾病的怀疑，即使测试结果为阴性他们可能也不会改变管理策略。在这种情况下进行测试的优势在于该测试有可能几乎确定地确定疾病的存在。因此，该测试因其提供的预后信息而受到重视，而不仅仅是其对患者管理的影响。
