Department of Data Science, The Institute of Statistical Mathematics, Tachikawa, Tokyo, Japan.
Clin Cancer Res. 2012 Nov 1;18(21):6065-73. doi: 10.1158/1078-0432.CCR-12-1206. Epub 2012 Aug 27.
It is highly challenging to develop reliable diagnostic tests to predict patients' responsiveness to anticancer treatments on clinical endpoints before commencing the definitive phase III randomized trial. Development and validation of genomic signatures in the randomized trial can be a promising solution. Such signatures are required to predict quantitatively the underlying heterogeneity in the magnitude of treatment effects.
We propose a framework for developing and validating genomic signatures in randomized trials. Codevelopment of predictive and prognostic signatures can allow prediction of patient-level survival curves as basic diagnostic tools for treating individual patients.
We applied our framework to gene-expression microarray data from a large-scale randomized trial to determine whether the addition of thalidomide improves survival for patients with multiple myeloma. The results indicated that approximately half of the patients were responsive to thalidomide, and the average improvement in survival for the responsive patients was statistically significant. Cross-validated patient-level survival curves were developed to predict survival distributions of individual future patients as a function of whether or not they are treated with thalidomide and with regard to their baseline prognostic and predictive signature indices.
The proposed framework represents an important step toward reliable predictive medicine. It provides an internally validated mechanism for using randomized clinical trials to assess treatment efficacy for a patient population in a manner that takes into consideration the heterogeneity in patients' responsiveness to treatment. It also provides cross-validated patient-level survival curves that can be used for selecting treatments for future patients.
在开始进行确定性 III 期随机试验之前，开发能够可靠地预测患者对癌症治疗反应的临床终点诊断测试极具挑战性。在随机试验中开发和验证基因组特征是一种很有前途的解决方案。此类特征需要定量预测治疗效果的潜在异质性。
我们提出了一种在随机试验中开发和验证基因组特征的框架。预测和预后特征的共同开发可以允许预测患者水平的生存曲线，作为治疗个别患者的基本诊断工具。
我们将我们的框架应用于来自大规模随机试验的基因表达微阵列数据，以确定添加沙利度胺是否可以改善多发性骨髓瘤患者的生存。结果表明，大约一半的患者对沙利度胺有反应，而有反应的患者的平均生存改善具有统计学意义。开发了交叉验证的患者水平生存曲线，以预测单个未来患者的生存分布，作为他们是否接受沙利度胺治疗以及他们的基线预后和预测特征指数的函数。
所提出的框架代表了可靠的预测医学的重要一步。它提供了一种内部验证的机制，用于以考虑患者对治疗反应的异质性的方式，使用随机临床试验来评估患者人群的治疗效果。它还提供了交叉验证的患者水平生存曲线，可用于为未来的患者选择治疗方法。
