validation of the calibration of predictions
to obtain unbiased estimates of the difference between Cox predicted and Kaplan–Meier
survival estimates at a fixed time u. Here is one sequence of steps.
Obtain cutpoints (e.g., deciles) of predicted survival at time u so as to have a given number of subjects (e.g., 50) in each interval of predicted survival. These cutpoints are based on the distribution of ˆ S(u|X)inthe whole sample for the“final”model (for data-splitting, instead use the model developed in the training sample). Let k denote the number of intervals used.
对于时刻u，关于cox预测生存概率进行分层。k代表分出来的区间的个数。
Compute the average ˆ S(u|X)ineachinterval.
计算每个区间的平均的s.
Compare this with the Kaplan–Meier survival estimates at time u,stratified by intervals of ˆ S(u|X). Let the differences be denoted by d =(d1,...,dk)
与KM进行比较，时刻u处的生存估计由S的区间进行分层的，差异记为d
Use bootstrapping or cross-validation to estimate the overoptimism in d and then to correct d to get a more fair assessment of these differences.
d是什么？然后纠正d以得到一个更合理的对于差别的评估。
For each repetition, repeat any stepwise variable selection or stagewise significance testing using the same stopping rules as were used to derive the “final”model. No more than B = 200 replications are needed to obtain accurate estimates.
对于每次重复，重复stepwise variable selection 或者stagewise显著性检验，使用相同的停止法则。重复不超过200次。
If desired, the bias-corrected d can be added to the original stratified Kaplan–Meier estimates to obtain a bias-corrected calibration curve.
survival estimates at a fixed time u. Here is one sequence of steps.
Obtain cutpoints (e.g., deciles) of predicted survival at time u so as to have a given number of subjects (e.g., 50) in each interval of predicted survival. These cutpoints are based on the distribution of ˆ S(u|X)inthe whole sample for the“final”model (for data-splitting, instead use the model developed in the training sample). Let k denote the number of intervals used.
对于时刻u，关于cox预测生存概率进行分层。k代表分出来的区间的个数。
Compute the average ˆ S(u|X)ineachinterval.
计算每个区间的平均的s.
Compare this with the Kaplan–Meier survival estimates at time u,stratified by intervals of ˆ S(u|X). Let the differences be denoted by d =(d1,...,dk)
与KM进行比较，时刻u处的生存估计由S的区间进行分层的，差异记为d
Use bootstrapping or cross-validation to estimate the overoptimism in d and then to correct d to get a more fair assessment of these differences.
d是什么？然后纠正d以得到一个更合理的对于差别的评估。
For each repetition, repeat any stepwise variable selection or stagewise significance testing using the same stopping rules as were used to derive the “final”model. No more than B = 200 replications are needed to obtain accurate estimates.
对于每次重复，重复stepwise variable selection 或者stagewise显著性检验，使用相同的停止法则。重复不超过200次。
If desired, the bias-corrected d can be added to the original stratified Kaplan–Meier estimates to obtain a bias-corrected calibration curve.
还没人转发这篇日记
