PredictionTools

This package is under development and for personal use only. It includes various metrics that can be used to validate predictions. Please contact me at carolienm@live.nl if you encounter any issues.

Install

library(remotes)

remotes::install_github("CHMMaas/PredictionTools")

Note, if you want to update the package you can also use the above call.

Updates

Release: 2023

Update 25 September 2023: added features

1. indicate which metrics to show in calibration plot with show.metrics (TRUE/FALSE vector of length 8);
2. pre-specify axis limits.

Update 4 December 2023: added features

1. indicate whether to show confidence interval in calibration plot;
2. add optimism corrected C-index in calibration plot.

Update 5 December 2023: added features

1. Uno's C-index for val.surv.mi()

Update 8 December 2023: added features

1. fun.event() function added that translates linear predictor and baseline hazard for a specific time point into predicted probabilities
2. Updated the example of val.surv.mi()

Update 6 March 2024: improved speed

1. Use concordance() instead of rcorr.cens() to improve speed

Update 27 March 2024: n.iter argument removed

1. Standard error of Uno's C-index is not calculated by bootstrapping anymore, so the argument n.iter is removed

Update 26 June 2024: smoothed calibration curve added

1. When the option smoothed.curve is set to TRUE, a smoothed calibration curve is plotted

Update 3 June 2024: n.sim argument added

1. Added argument to change the number of simulations used to compute the confidence interval of the time-dependent AUC. Note, computation of 95% CI for the time-dependent AUC is very slow if n.sim is large (2000).

Update 7 October 2024: calculate.ARD.dRMST added

1. Added function to calculate absolute risk difference (ARD) and difference in restricted mean survival time (dRMST) for time-to-event data
2. Area under thetime-dependent ROC functionality is now made available even if the Surv-object is restricted to a certain time horizon

Update 10 October 2024: rename calculate.ARD.dRMST into calculate.RD.dRMST

Update 9 December 2024:

1. Smoothed curve updated
2. Added Integrated Calibration Index (https://onlinelibrary.wiley.com/doi/10.1002/sim.8570)

Load

library(PredictionTools)

val.surv.mi

This function outputs the calibration and discrimination of predictions of survival of multiple imputed data sets.

val.prob.mi

This function outputs the calibration and discrimination of predictions of outcome probabilities of multiple imputed data sets.

Rubin.combine

This function combines multiple estimates with multiples standard errors using the Rubin's rule.

mb.c

This function calculates model-based concordance.

calculate.RD.dRMST

This function calculates the risk difference (RD) and the difference in restricted mean survival time (dRMST).