Identifying combinations of cancer markers for further study as triggers of early intervention
Identifying combinations of cancer markers for further study as triggers of early intervention
复制标题
DOI：
10.1111/j.0006-341x.2000.01082.x
复制
发表时间：
2000-12-01
期刊：
影响因子：
1.9
通讯作者：
Baker, SG
中科院分区：
文献类型：
--
作者：
Baker, SG
在许多长期临床试验或队列研究中，研究人员反复收集和存储癌症病例的组织或血清样本以及后来的测试样本，以及癌症潜在标志物的随机对照样本。一个重要的问题是，在未来的试验中，应该研究分子标记的哪种组合，如果有的话，作为早期干预的触发因素。为了回答这个问题，我们使用接收器工作特征(ROC)曲线总结了各种组合的性能，这些曲线绘制了真阳性与假阳性的比率。为了构造ROC曲线，我们提出了一类新的非参数算法，将ROC范式扩展到多个测试。我们将各种标记组合匹配到训练样本中，并使用基于效用函数的目标区域来评估测试样本中的性能。我们将该方法应用于前列腺癌的下列标志物：总前列腺特异性抗原(PSA)的末值、总PSA与游离PSA的最后比率、总PSA的最后斜率以及比率的最后斜率。在测试样本中，最后一个总PSA的ROC曲线比四个标记组合的ROC曲线更接近靶区。在一个单独的验证样本中，最后总PSA的ROC曲线在77%的Bootstrap复制中与目标区域相交，这表明有一些希望进行进一步研究。我们还讨论了样本量的计算。
In many long-term clinical trials or cohort studies, investigators repeatedly collect and store tissue or serum specimens and later test specimens from cancer cases and a random sample of controls for potential markers for cancer. An important question is what combination, if any, of the molecular markers should be studied in a future trial as a trigger for early intervention. To answer this question, we summarized the performance of various combinations using Receiver Operating Characteristic (ROC) curves, which plot true versus false positive rates. To construct the ROC curves, we proposed a new class of nonparametric algorithms which extends the ROC paradigm to multiple tests. We fit various combinations of markers to a training sample and evaluated the performance in a test sample using a target region based on a utility function. We applied the methodology to the following markers for prostate cancer, the last value of total prostate-specific antigen (PSA), the last ratio of total to free PSA, the last slope of total PSA, and the last slope of the ratio. In the test sample, the ROC curve for last total PSA was slightly closer to the target region than the ROC curve for a combination of four markers. in a separate validation sample, the ROC curve for last total PSA intersected the target region in 77% of bootstrap replications, indicating some promise for further study. We also discussed sample size calculations.