Somoza E, Mossman D
Psychiatry Service, Veterans Administration Medical Center, Cincinnati, Ohio 45220.
Biol Psychiatry. 1990 May 1;27(9):990-1006. doi: 10.1016/0006-3223(90)90036-2.
New diagnostic techniques must be evaluated for their intrinsic accuracy and for their applicability to particular patient groups in specific clinical settings. Using receiver operating characteristic (ROC) analysis and concepts from information theory, we have developed a new mathematical and graphical method that can evaluate, compare, and optimize the performance of diagnostic tests for any value of disorder prevalence. Our analytic method is appropriate to any test that sorts disordered from nondisordered subjects using a continuous or nonbinary diagnostic variable; its characterization of the fundamental properties of such tests thus has important implications for the evaluation and optimization of diagnostic modalities used by clinicians in all medical specialties. We demonstrate our method using published data from five studies that used sleep architecture as a "biological marker" for depression. Our analysis confirms that REM latency is comparable to the dexamethasone suppression test in its ability to discriminate depressed from control subjects. For each of the five studies, we show how optimal REM latency cut-off times may be selected so that diagnostic information yield is maximized, and we compare the ability of each study to detect depressed subjects in populations where the prevalence of affective disorder can be specified.
新的诊断技术必须评估其内在准确性以及在特定临床环境中对特定患者群体的适用性。利用接受者操作特征（ROC）分析和信息论中的概念，我们开发了一种新的数学和图形方法，该方法可以针对疾病患病率的任何值评估、比较和优化诊断测试的性能。我们的分析方法适用于任何使用连续或非二元诊断变量将患病与未患病个体区分开来的测试；因此，其对这类测试基本特性的描述对于所有医学专业临床医生使用的诊断方式的评估和优化具有重要意义。我们使用五项研究的已发表数据来演示我们的方法，这些研究将睡眠结构用作抑郁症的“生物学标志物”。我们的分析证实，快速眼动睡眠潜伏期在区分抑郁症患者与对照受试者方面的能力与地塞米松抑制试验相当。对于这五项研究中的每一项，我们展示了如何选择最佳的快速眼动睡眠潜伏期截止时间，以使诊断信息产出最大化，并且我们比较了每项研究在能够明确情感障碍患病率的人群中检测抑郁症患者的能力。
