Sachs Michael C
Michael C. Sachs, Unit of Biostatistics, Nobels väg 13, Karolinska Institutet, 17177 Stockholm, Sweden
J Stat Softw. 2017 Aug;79. doi: 10.18637/jss.v079.c02. Epub 2017 Aug 9.
Plots of the receiver operating characteristic (ROC) curve are ubiquitous in medical research. Designed to simultaneously display the operating characteristics at every possible value of a continuous diagnostic test, ROC curves are used in oncology to evaluate screening, diagnostic, prognostic and predictive biomarkers. I reviewed a sample of ROC curve plots from the major oncology journals in order to assess current trends in usage and design elements. My review suggests that ROC curve plots are often ineffective as statistical charts and that poor design obscures the relevant information the chart is intended to display. I describe my new R package that was created to address the shortcomings of existing tools. The package has functions to create informative ROC curve plots, with sensible defaults and a simple interface, for use in print or as an interactive web-based plot. A web application was developed to reach a broader audience of scientists who do not use R.
在医学研究中,受试者工作特征(ROC)曲线的图表随处可见。ROC曲线旨在同时展示连续诊断测试在每个可能值下的操作特征,在肿瘤学中用于评估筛查、诊断、预后和预测生物标志物。我查看了主要肿瘤学期刊中的一组ROC曲线图,以评估当前的使用趋势和设计元素。我的审查表明,ROC曲线图作为统计图往往效果不佳,糟糕的设计掩盖了图表旨在展示的相关信息。我介绍了为弥补现有工具的不足而创建的新R包。该包具有创建信息丰富的ROC曲线图的功能,具有合理的默认设置和简单的界面,可用于打印或作为基于网络的交互式图表。还开发了一个网络应用程序,以惠及更广泛的不使用R的科学家群体。