Wayne L G, Krichevsky M I, Portyrata D, Jackson C K
J Clin Microbiol. 1984 Oct;20(4):722-9. doi: 10.1128/jcm.20.4.722-729.1984.
A probability matrix is presented for identification of slowly growing mycobacteria that are likely to be encountered in clinical laboratories. The matrix includes 23 features that are useful for identifying members of 14 species or species complexes. The computer program identifies strains as a function of the ID (identification) score, which measures the discrimination among possible alternative identifications, and the R (ratio) score, which measures the degree of fit to the most likely taxa. It is not necessary to employ all 23 tests when initiating an identification; the program will suggest additional tests to perform when a partial data set fails to yield a definitive identification. Two independent sets of cultures comprising a total of 1,212 strains were used to test the matrix. Correct diagnoses were based on clustering behavior in numerical taxonomic analysis with larger numbers of features. The probable efficiencies with the two sets were 94.2 and 83.4%, respectively, and the accuracy of the definitive identifications for both sets exceeded 95%. A discussion is presented of situations when it may be appropriate to override an R score that has caused the rejection of an identification and to thereby enhance the efficiency.
本文给出了一个概率矩阵，用于识别临床实验室中可能遇到的生长缓慢的分枝杆菌。该矩阵包含23个特征，这些特征有助于识别14个物种或物种复合体的成员。计算机程序根据ID（鉴定）分数识别菌株，ID分数衡量可能的替代鉴定之间的区分度，以及R（比率）分数，R分数衡量与最可能的分类单元的拟合程度。开始鉴定时不必进行全部23项检测；当部分数据集未能得出明确鉴定结果时，程序会建议进行额外的检测。使用两组共1212株独立培养物来测试该矩阵。正确诊断基于具有更多特征的数值分类分析中的聚类行为。两组的可能效率分别为94.2%和83.4%，两组明确鉴定的准确率均超过95%。本文讨论了在哪些情况下可能适合忽略导致鉴定被拒绝的R分数，从而提高效率。
