Lavelle S M, Kanagaratnam B
Department of Experimental Medicine, University College, Galway, Republic of Ireland.
Int J Biomed Comput. 1990 Sep;26(3):203-9. doi: 10.1016/0020-7101(90)90043-t.
Decisional aids are dependent on the accuracy, comprehensiveness and decisional value of the observations in the database and the test subject. Diagnostic data stem from the observations of the patient, the doctor and third party tests. An information-utility index was used to compare the diagnostic contribution of these sources. In a diagnostic database of 314 cases of jaundice, twenty-two findings of high diagnostic usefulness were discovered among 6 diseases. Sixteen of these (73%) were patient's observations, 4 were doctor's (18%), one the patient's age and one a ward test. Five laboratory tests had lesser average scores. In three databases--on jaundice, abdominal pain and low back pain the 7 commonest diseases were examined, totalling 1018 patients. In a simple Bayesian procedure, the patient's observations alone correctly classified 50%, the doctor's observations a further 16%, the patient's age and sex another 4% and the prior probability of the diseases, an additional 4%. These figures varied considerably between the diseases. When all 21 diseases with more than 10 cases in the database were considered, the respective figures were 53, 15, 3 and 3%. A European Community medical research initiative is collecting data on 10,000 cases of jaundice and of acute abdominal pain to establish an adequate diagnostic database for the development of decision support systems.
决策辅助工具依赖于数据库和测试对象中观察结果的准确性、全面性和决策价值。诊断数据源于对患者、医生的观察以及第三方检测。使用信息效用指数来比较这些来源的诊断贡献。在一个包含314例黄疸病例的诊断数据库中，在6种疾病中发现了22项具有高诊断价值的发现。其中16项（73%）是患者的观察结果，4项是医生的（18%），1项是患者年龄，1项是病房检测结果。5项实验室检测的平均得分较低。在三个数据库——关于黄疸、腹痛和腰痛的数据库中，对7种最常见的疾病进行了检查，共有1018名患者。在一个简单的贝叶斯程序中，仅患者的观察结果就能正确分类50%，医生的观察结果再增加16%，患者的年龄和性别又增加4%，疾病的先验概率再增加4%。这些数字在不同疾病之间差异很大。当考虑数据库中病例数超过10例的所有21种疾病时，相应的数字分别为53%、15%、3%和3%。一项欧洲共同体医学研究计划正在收集10000例黄疸和急性腹痛病例的数据，以建立一个足够的诊断数据库，用于开发决策支持系统。
