Tools for checking calibration of a Cox model in external validation: Approach based on individual event probabilities
Tools for checking calibration of a Cox model in external validation: Approach based on individual event probabilities
复制标题
DOI：
10.1177/1536867x1401400403
复制
发表时间：
2014-01-01
期刊：
影响因子：
4.8
通讯作者：
Royston, Patrick
中科院分区：
文献类型：
--
作者：
Royston, Patrick
在过去的40年里，Cox比例风险模型在医学上得到了广泛的应用。一个流行的应用是开发一个多变量预测模型，通常是一个预测模型，根据在某个初始时间点测量的“基线”因素来预测特定疾病患者的临床结果。为了使这种模型在实践中有用，它必须经过“验证”；也就是说，它必须在患者的外部样本中表现得令人满意，而不是依赖于最初开发该模型的样本。性能的一个关键方面是校准，这是预测的准确性，特别是在时间起点之后的任何时间的存活(或等效地，故障或事件)概率的准确性。我们认为，文献中对COX模型校准的系统评估在很大程度上被忽视了。在本文中，我们建议了一种使用在不同时间点估计的单个事件概率来评估校准的方法。我们通过对原发性胆汁性肝硬变的两个数据集的详细分析来举例说明该方法；这些数据集包括一个派生数据集和一个验证数据集。我们描述了一个新命令stcoxcal，它执行必要的计算。Stcoxcal的结果可以图形显示，这使得用户更容易根据后续时间来描绘校准(或缺少校准)。
The Cox proportional hazards model has been used extensively in medicine over the last 40 years. A popular application is to develop a multivariable prediction model, often a prognostic model to predict the clinical outcome of patients with a particular disorder from "baseline" factors measured at some initial time point. For such a model to be useful in practice, it must be "validated"; that is, it must perform satisfactorily in an external sample of patients independent of the sample on which the model was originally developed. One key aspect of performance is calibration, which is the accuracy of prediction, particularly of survival (or equivalently, failure or event) probabilities at any time after the time origin. We believe systematic evaluation of the calibration of a Cox model has been largely ignored in the literature. In this article, we suggest an approach to assessing calibration using individual event probabilities estimated at different time points. We exemplify the method by detailed analysis of two datasets in the disease primary biliary cirrhosis; the datasets comprise a derivation and a validation dataset. We describe a new command, stcoxcal, that performs the necessary calculations. Results for stcoxcal can be displayed graphically, which makes it easier for users to picture calibration (or lack thereof) according to follow-up time.