Boyko E J
Medical Service, VA Medical Center, Seattle, WA 98108.
Med Decis Making. 1994 Apr-Jun;14(2):175-9. doi: 10.1177/0272989X9401400210.
Previous publications have advocated that clinicians choose the most sensitive diagnostic test to rule out disease and the most specific diagnostic test to rule in disease. This paper critically examines the validity of these recommendations. First, the author finds that following these recommendations does not lead to the highest disease probability for a positive test result (thereby best ruling in disease) or the lowest disease probability for a negative test result (thereby best ruling out disease). In general, the ability of a diagnostic test to lead to the highest (rule in) or lowest (rule out) disease probability should be judged based on likelihood ratios. Next, by comparing expected utilities, the author considers whether the most specific test leads to the best clinical outcome when a rule-in strategy is clinically advisable, i.e., when the costs of false-positive results are high, and whether the most sensitive test leads to the best clinical outcome when a rule-out strategy is clinically advisable, i.e., when the costs of false-negative results are high. The author again demonstrates that the greatest clinical utility is not always achieved by using the most specific test in a rule-in decision or the most sensitive test in a rule-out decision. Tradeoffs between sensitivity, specificity, disease probability, and utilities of correct and incorrect disease classifications by the diagnostic test must be simultaneously captured to determine which strategy maximizes clinical utility.
以往的文献主张临床医生选择最敏感的诊断试验来排除疾病，选择最特异的诊断试验来确诊疾病。本文批判性地审视了这些建议的有效性。首先，作者发现遵循这些建议并不能使阳性检测结果的疾病概率最高（从而最好地确诊疾病），也不能使阴性检测结果的疾病概率最低（从而最好地排除疾病）。一般来说，诊断试验导致最高（确诊）或最低（排除）疾病概率的能力应根据似然比来判断。接下来，通过比较预期效用，作者考虑当确诊策略在临床上可行时，即假阳性结果的成本很高时，最特异的试验是否能带来最佳临床结果；以及当排除策略在临床上可行时，即假阴性结果的成本很高时，最敏感的试验是否能带来最佳临床结果。作者再次证明，在确诊决策中使用最特异的试验或在排除决策中使用最敏感的试验并不总是能实现最大的临床效用。必须同时考虑诊断试验的敏感性、特异性、疾病概率以及正确和错误疾病分类的效用之间的权衡，以确定哪种策略能使临床效用最大化。
