PMCALPLOT: Stata module to produce calibration plot of prediction model performance
PMCALPLOT: Stata module to produce calibration plot of prediction model performance
复制标题
PMCALPLOT：Stata 模块，用于生成预测模型性能的校准图
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
E. Martin
中科院分区：
文献类型：
--
作者：
J. Ensor;K. Snell;E. Martin
pmcalplot生成观察概率与预期概率的校准图，用于评估预测模型性能。pmcalplot现在可以处理二元、生存或连续结果类型的预测模型。按照TRIPOD指南中的建议，在风险谱中按组绘制校准，并且还可以显示分组的置信区间（NB：不适用于连续结果）。此外，可以在图上显示事件和非事件分布的尖峰图，以及允许在个体患者水平评估校准的低平滑器[NB：生存结局的尖峰图和低平滑器正在进行中]。对于连续的结果，可以在相应的轴上显示观察值和预期值的直方图。此外，还可以在图上显示常见的预测模型性能统计信息，从而量化模型性能。pmcalplot主要用于在现有模型的外部验证中评估模型性能。然而，它也可以在模型开发过程中使用，以检查模型的表观性能（应显示完美的校准）。
pmcalplot produces a calibration plot of observed against expected probabilities for assessment of prediction model performance. pmcalplot can now handle prediction models with binary, survival or continuous outcome types. Calibration is plotted in groups across the risk spectrum as recommended in the TRIPOD guidelines, and confidence intervals for the groupings can also be displayed (NB: not for continuous outcomes). Further, a spike plot of the distribution of events and non-events can be displayed on the plot, as well as a lowess smoother allowing assessment of the calibration at the individual patient level [NB: Spike plot and lowess smoother for survival outcomes is work in progress]. For continuous outcomes a histogram of observed and expected values can be displayed on the corresponding axes. Additionally, common prediction model performance statistics can also be displayed on the plot, quantifying the models performance. pmcalplot is primarily useful for assessment of model performance in an external validation of an existing model. However it can also be used during model development to check the apparent performance of the model (which should show perfect calibration).