Department of Molecular Carcinogenesis, Medical University of Lodz, Lodz, Poland.
Department of Molecular Carcinogenesis, Medical University of Lodz, Lodz, Poland.
Comput Methods Programs Biomed. 2019 Aug;177:133-139. doi: 10.1016/j.cmpb.2019.05.023. Epub 2019 May 23.
Growing evidence of transcriptional and metabolomic differentiation induced many studies which analyze such differentiation in context of outcome of disease progression, treatment or influence of many different factors affecting cellular and tissue metabolism. Particularly, cancer researchers are looking for new biomarkers that can serve as a diagnostic/prognostic factor and its further corresponding relationship regarding clinical effects. As a result of the increasing interest in use of dichotomization of continuous variables involving clinical or epidemiological data (gene expression, biomarkers, biochemical parameters, etc.) there is a large demand for cutoff point determination tools with simultaneous lack of software offering stratification of patients based on continuous and binary variables. Therefore, we developed "Evaluate Cutpoints" application offering wide set of statistical and graphical methods for cutpoint optimization enabling stratification of population into two or three groups.
Application is based on R language including algorithms of packages such as survival, survMisc, OptimalCutpoints, maxstat, Rolr, ggplot2, GGally and plotly offering Kaplan-Meier plots and ROC curves with cutoff point determination.
All capabilities of Evaluate Cutpoints were illustrated with example analysis of estrogen, progesterone and human epidermal growth factor 2 receptors in breast cancer cohort. Through ROC curve the cutoff points were established for expression of ESR1, PGR and ERBB2 in correlation with their immunohistochemical status (cutoff: 1301.253, 243.35, 11,434.438, respectively; sensitivity: 94%, 85%, 64%, respectively; specificity: 93%, 86%, 91%, respectively). Through disease-free survival analysis we divided patients into two and three groups regarding expression of ESR1, PGR and ERBB2. Example algorithm cutp showed that lowered expression of ESR1 and ERBB2 was more favorable (HR = 2.07, p = 0.0412; HR = 2.79, p = 0.0777, respectively), whereas heightened PGR expression was correlated with better prognosis (HR = 0.192, p = 0.0115).
This work presents application Evaluate Cutpoints that is freely available to download at http://wnbikp.umed.lodz.pl/Evaluate-Cutpoints/. Currently, many softwares are used to split continuous variables such as Cutoff Finder and X-Tile, which offer distinct algorithms. Unlike them, Evaluate Cutpoints allows not only dichotomization of populations into groups according to continuous variables and binary variables, but also stratification into three groups as well as manual selection of cutoff point thus preventing potential loss of information.
越来越多的证据表明转录组学和代谢组学的分化，这促使许多研究分析这种分化与疾病进展、治疗结果或影响细胞和组织代谢的许多不同因素之间的关系。特别是，癌症研究人员正在寻找新的生物标志物，作为诊断/预后因素，并进一步研究其与临床效果的关系。由于人们越来越关注将涉及临床或流行病学数据（基因表达、生物标志物、生化参数等）的连续变量进行二分法处理，因此需要同时确定能够根据连续和二进制变量对患者进行分层的软件，而提供这种软件的工具却非常缺乏。因此，我们开发了“Evaluate Cutpoints”应用程序，它提供了广泛的统计和图形方法来优化切点，从而将人群分为两组或三组。
该应用程序基于 R 语言，包括 survival、survMisc、OptimalCutpoints、maxstat、Rolr、ggplot2、GGally 和 plotly 等包的算法，提供了用于确定切点的 Kaplan-Meier 图和 ROC 曲线。
我们通过分析乳腺癌队列中雌激素、孕激素和人表皮生长因子 2 受体的实例分析，展示了“Evaluate Cutpoints”应用程序的所有功能。通过 ROC 曲线，确定了 ESR1、PGR 和 ERBB2 表达与免疫组织化学状态的切点（切点：分别为 1301.253、243.35、11434.438；灵敏度：分别为 94%、85%、64%；特异性：分别为 93%、86%、91%）。通过无病生存分析，我们根据 ESR1、PGR 和 ERBB2 的表达将患者分为两组和三组。示例算法 cutp 表明，ESR1 和 ERBB2 的低表达更有利（HR=2.07，p=0.0412；HR=2.79，p=0.0777），而 PGR 表达升高与预后较好相关（HR=0.192，p=0.0115）。
本研究介绍了可免费下载的应用程序“Evaluate Cutpoints”，其网址为 http://wnbikp.umed.lodz.pl/Evaluate-Cutpoints/。目前，许多软件（如 Cutoff Finder 和 X-Tile）都用于将连续变量进行二分法处理，它们提供了不同的算法。与这些软件不同的是，Evaluate Cutpoints 不仅允许根据连续变量和二进制变量将人群分为两组，还允许分为三组，并手动选择切点，从而防止潜在的信息丢失。
