文档介绍:诊断试验与ROC曲线分析
目录
一、 基本概念
诊断试验四格表基本统计基本指标
ROC曲线:
二、 实例分析
各诊断项目(变量)分别诊断效果分析:
诊断模型分析:
比较两预测模型:
时间依赖的ROC曲线(Time-depe患者和非患者都有同样的“机会”被诊断为阳性。ROC曲线越接近机会线,即曲 ,表明诊断试验区分患者和非患者的能力越弱;越接近1,表明诊断 试验的准确度越强。一般认为, -,诊断价值较小;-,诊断 价值中等;>,
诊断价值较高。最理想的诊断试验的ROC曲线是从坐标原点出发,沿 着Y轴到(0,1)点,再沿着X轴的水平线到(1,1)点。在比较ROC曲线下面积时,还 应考虑到实际临床应用情况。比如,某项诊断试验主要用于排除疾病时,则需要较高的特 异度,这是我们仅对左侧的ROC曲线(即高特异度的ROC曲线部分)下的面积感兴趣。通 。
诊断界值确定:实际工作中,人们希望找到灵敏度和特异度均接近“1”的点。横轴为
(1-特异度),所以横轴原点就是特异度为1的点,因此我们要找的点就是距ROC曲线图 中左上角最近的点,也就是(灵敏度+特异度)取最大值的点。如果认为灵敏度的重要性 是特异度的a倍,此时可选取(a*灵敏度+1 *特异度)取值最大的点。在实际应用中,可 以根据不同的研究目的确定阈值,如果诊断试验目的是筛查本病时,宜选在误诊率充许的 范围内灵敏度较高的截断点,此时保证了漏诊率低;若试验目的为确诊本病,则宜选在漏 诊率充许范围内特异度较高的截断点,此时误诊率低。
Cut point selection
One of the best-known methods is based on selecting the cutpoint that provides the same value for the sensitivity and specificity. This point is known as the equivalence or symmetry point (Greiner, 1995; Defreitas et al., 2004; Adlhoch et al., 2011). Graphically, it corresponds with the operating point on the ROC curve that intersects the perpendicular to the positive diagonal line, that is, y = 1 - x, where x is the false positive rate. The symmetry point can also be seen as the point that maximizes simultaneously both types of correct classifications (Riddle and Stratford, 1999; Gallop et al., 2003), that is, it corresponds to the probability of correctly classifying any subject, whether it is healthy or diseased (Jimenez-Valverde, 2012, 2014). Additionally, the incorporation of costs for the misclassification rates in the estimation of optimal cutpoints is crucial for evaluating not only the test accuracy but also its clinical efficacy, although this aspect is not taken into account most of the times. So, an interesting generalization of the equivalence or symmetry point, cS, that takes into account the costs associated to the false positive and false negative misclassifications, CFP and CFN, respectively, is the generalized equivalence p