,
Glen P Martin [fnd, rev] | Title: | Calibration Plots for the Transition Probabilities from Multistate Models |
|---|---|
| Description: | Assess the calibration of an existing (i.e. previously developed) multistate model through calibration plots. Calibration is assessed using one of three methods. 1) Calibration methods for binary logistic regression models applied at a fixed time point in conjunction with inverse probability of censoring weights. 2) Calibration methods for multinomial logistic regression models applied at a fixed time point in conjunction with inverse probability of censoring weights. 3) Pseudo-values estimated using the Aalen-Johansen estimator of observed risk. All methods are applied in conjunction with landmarking when required. These calibration plots evaluate the calibration (in a validation cohort of interest) of the transition probabilities estimated from an existing multistate model. While package development has focused on multistate models, calibration plots can be produced for any model which utilises information post baseline to update predictions (e.g. dynamic models); competing risks models; or standard single outcome survival models, where predictions can be made at any landmark time. Please see Pate et al. (2024) <doi:10.1002/sim.10094> and Pate et al. (2024) <https://alexpate30.github.io/calibmsm/articles/Overview.html>. |
| Authors: | Alexander Pate [aut, cre, cph]
,
Glen P Martin [fnd, rev] |
| Maintainer: | Alexander Pate <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 1.1.1 |
| Built: | 2025-02-25 16:26:44 UTC |
| Source: | https://github.com/alexpate30/calibmsm |
t.Estimates the inverse probability of censoring weights by fitting a cox-propotinal hazards model in a landmark cohort of individuals. Primarily used internally, this function has been exported to allow users to reproduce results in the vignette when estimating confidence intervals using bootstrapping manually.
calc_weights( data_ms, data_raw, covs = NULL, t, s, landmark_type = "state", j = NULL, max_weight = 10, stabilised = FALSE, max_follow = NULL )calc_weights( data_ms, data_raw, covs = NULL, t, s, landmark_type = "state", j = NULL, max_weight = 10, stabilised = FALSE, max_follow = NULL )
data_ms |
Validation data in msdata format |
data_raw |
Validation data in data.frame (one row per individual) |
covs |
Character vector of variable names to adjust for when calculating inverse probability of censoring weights |
t |
Follow up time at which to calculate weights |
s |
Landmark time at which predictions were made |
landmark_type |
Whether weights are estimated in all individuals uncensored at time s ('all') or only in individuals uncensored and in state j at time s ('state') |
j |
Landmark state at which predictions were made (only required in landmark_type = 'state') |
max_weight |
Maximum bound for weights |
stabilised |
Indicates whether weights should be stabilised or not |
max_follow |
Maximum follow up for model calculating inverse probability of censoring weights. Reducing this to |
Estimates inverse probability of censoring weights (Hernan M, Robins J, 2020).
Fits a cox proportional hazards model to individuals in a landmark cohort, predicting the probability of being censored
at time t. This landmark cohort may either be all individuals uncensored at time s, or those uncensored
and in state j at time s. All predictors in w_covs are assumed to have a linear effect on the hazard.
Weights are estimated for all individuals in data_raw, even if they will not be used in the analysis as they do not meet the landmarking
requirements. If an individual enters an absorbing state prior to t, we estimate the probability of being censored
before the time of entry into the absorbing state, rather than at t. Details on all the above this are provided in
vignette overview.
A data frame with two columns. id corresponds to the patient ids from data_raw. ipcw contains the inverse probability
of censoring weights (specifically the inverse of the probability of being uncesored). If stabilised = TRUE was specified,
a third variable ipcw_stab will be returned, which is the stabilised inverse probability of censoring weights.
Hernan M, Robins J (2020). “12.2 Estimating IP weights via modeling.” In Causal Inference: What If, chapter 12.2. Chapman Hall/CRC, Boca Raton.
# Estimate inverse probability of censoring weights for individual in cohort ebmtcal. # Specifically the probability of being uncensored at t = 1826 days. # Weights are estimated using a model fitted in all individuals uncensored at time s = 0. weights_manual <- calc_weights(data_ms = msebmtcal, data_raw = ebmtcal, covs = c("year", "agecl", "proph", "match"), t = 1826, s = 0, landmark_type = "state", j = 1) str(weights_manual)# Estimate inverse probability of censoring weights for individual in cohort ebmtcal. # Specifically the probability of being uncensored at t = 1826 days. # Weights are estimated using a model fitted in all individuals uncensored at time s = 0. weights_manual <- calc_weights(data_ms = msebmtcal, data_raw = ebmtcal, covs = c("year", "agecl", "proph", "match"), t = 1826, s = 0, landmark_type = "state", j = 1) str(weights_manual)
Calculates the underlying data for calibration plots of the predicted transition probabilities from a multistate model using three methods.
BLR-IPCW: Binary logistic regression with inverse probability of censoring weights.
MLR-IPCW: Multinomial logistic regression with inverse probability of censoring weights, based on the nominal calibration framework of van Hoorde et al. (2014, 2015)
Pseudo-values: Pseudo-values estimated using the Aalen-Johansen estimator (Aalen OO, Johansen S, 1978).
calib_msm( data_ms, data_raw, j, s, t, tp_pred, tp_pred_plot = NULL, calib_type = "blr", curve_type = "rcs", rcs_nk = 3, loess_span = 0.75, loess_degree = 2, loess_surface = c("interpolate", "direct"), loess_statistics = c("approximate", "exact", "none"), loess_trace_hat = c("exact", "approximate"), loess_cell = 0.2, loess_iterations = 4, loess_iterTrace = FALSE, mlr_smoother_type = c("sm.ps", "sm.os", "s"), mlr_ps_int = 4, mlr_degree = 3, mlr_s_df = 4, mlr_niknots = 4, weights = NULL, w_function = NULL, w_covs = NULL, w_landmark_type = "state", w_max = 10, w_stabilised = FALSE, w_max_follow = NULL, pv_group_vars = NULL, pv_n_pctls = NULL, pv_precalc = NULL, pv_ids = NULL, CI = FALSE, CI_type = "bootstrap", CI_R_boot = NULL, CI_seed = NULL, transitions_out = NULL, assess_moderate = TRUE, assess_mean = TRUE, ... )calib_msm( data_ms, data_raw, j, s, t, tp_pred, tp_pred_plot = NULL, calib_type = "blr", curve_type = "rcs", rcs_nk = 3, loess_span = 0.75, loess_degree = 2, loess_surface = c("interpolate", "direct"), loess_statistics = c("approximate", "exact", "none"), loess_trace_hat = c("exact", "approximate"), loess_cell = 0.2, loess_iterations = 4, loess_iterTrace = FALSE, mlr_smoother_type = c("sm.ps", "sm.os", "s"), mlr_ps_int = 4, mlr_degree = 3, mlr_s_df = 4, mlr_niknots = 4, weights = NULL, w_function = NULL, w_covs = NULL, w_landmark_type = "state", w_max = 10, w_stabilised = FALSE, w_max_follow = NULL, pv_group_vars = NULL, pv_n_pctls = NULL, pv_precalc = NULL, pv_ids = NULL, CI = FALSE, CI_type = "bootstrap", CI_R_boot = NULL, CI_seed = NULL, transitions_out = NULL, assess_moderate = TRUE, assess_mean = TRUE, ... )
data_ms |
Validation data in |
data_raw |
Validation data in |
j |
Landmark state at which predictions were made |
s |
Landmark time at which predictions were made |
t |
Follow up time at which calibration is to be assessed |
tp_pred |
Data frame or matrix of predicted transition probabilities at time t, if in state j at time s. There must be a separate column for the predicted transition probabilities into every state, even if these predicted transition probabilities are 0. |
tp_pred_plot |
Data frame or matrix of predicted risks for each possible transition over which to plot the calibration curves. Argument provided to enable user to apply bootstrapping manually. |
calib_type |
Whether calibration plots are estimated using BLR-IPCW ('blr'), MLR-IPCW ('mlr') or pseudo-values ('pv') |
curve_type |
Whether calibration curves are estimated using restricted cubic splines ('rcs') or loess smoothers ('loess') |
rcs_nk |
Number of knots when curves are estimated using restricted cubic splines |
loess_span |
Span when curves are estimated using loess smoothers |
loess_degree |
Degree when curves are estimated_ using loess smoothers |
loess_surface |
see |
loess_statistics |
see |
loess_trace_hat |
see |
loess_cell |
see |
loess_iterations |
see |
loess_iterTrace |
see |
mlr_smoother_type |
Type of smoothing applied. Takes values |
mlr_ps_int |
the number of equally-spaced B spline intervals in the vector spline smoother (see |
mlr_degree |
the degree of B-spline basis in the vector spline smoother (see |
mlr_s_df |
degrees of freedom of vector spline (see |
mlr_niknots |
number of interior knots (see |
weights |
Vector of inverse probability of censoring weights |
w_function |
Custom function for estimating the inverse probability of censoring weights |
w_covs |
Character vector of variable names to adjust for when calculating inverse probability of censoring weights |
w_landmark_type |
Whether weights are estimated in all individuals uncensored at time s ('all') or only in individuals uncensored and in state j at time s ('state') |
w_max |
Maximum bound for inverse probability of censoring weights |
w_stabilised |
Indicates whether inverse probability of censoring weights should be stabilised or not |
w_max_follow |
Maximum follow up for model calculating inverse probability of censoring weights. Reducing this to |
pv_group_vars |
Variables to group by before calculating pseudo-values |
pv_n_pctls |
Number of percentiles of predicted risk to group by before calculating pseudo-values |
pv_precalc |
Pre-calculated pseudo-values |
pv_ids |
Id's of individuals to calculate pseudo-values for |
CI |
Size of confidence intervals as a % |
CI_type |
Method for estimating confidence interval (currently restricted to |
CI_R_boot |
Number of bootstrap replicates when estimating the confidence interval for the calibration curve |
CI_seed |
Seed for bootstrapping procedure |
transitions_out |
Transitions for which to calculate calibration curves. Will do all possible transitions if left as NULL. |
assess_moderate |
TRUE/FALSE whether to estimate data for calibration plots |
assess_mean |
TRUE/FALSE whether to estimate mean calibration |
... |
Extra arguments to be passed to w_function (custom function for estimating weights) |
Observed event probabilities at time t are estimated for predicted
transition probabilities tp_pred out of state j at time s.
calib_type = 'blr' estimates calibration curves using techniques for assessing
the calibration of a binary logistic regression model (Van Calster et al., 2016).
A choice between restricted cubic splines and loess smoothers for estimating the
calibration curve can be made using curve_type. Landmarking (van Houwelingen HC, 2007)
is applied to only assess calibration in individuals who are uncensored and in state j at time s.
Calibration can only be assessed in individuals who are also uncensored at time t,
which is accounted for using inverse probability of censoring weights (Hernan M, Robins J, 2020).
See method BLR-IPCW from Pate et al., (2024) for a full explanation of the approach.
calib_type = 'mlr' estimates calibration scatter plots using a technique for assessing
the calibration of multinomial logistic regression models, namely the nominal
calibration framework of van Hoorde et al. (2014, 2015). Landmarking (van Houwelingen HC, 2007)
is applied to only assess calibration in individuals who are uncensored
and in state j at time s. Calibration can only be assessed in individuals
who are also uncensored at time t, which is accounted for using inverse probability
of censoring weights (Hernan M, Robins J, 2020). See method BLR-IPCW from Pate et al., (2024)
for a full explanation of the approach.
calib_type = 'pv' estimates calibration curves using using pseudo-values (Andersen PK, Pohar Perme M, 2010)
calculated using the Aalen-Johansen estimator (Aalen OO, Johansen S, 1978).
Calibration curves are generated by regressing the pseudo-values on the predicted transition probabilities.
A choice between restricted cubic splines and loess smoothers for estimating the
calibration curve can be made using curve_type. Landmarking (van Houwelingen HC, 2007)
is applied to only assess calibration in individuals who are uncensored and in state j at time s.
The nature of pseudo-values means calibration can be assessed in all landmarked individuals,
regardless of their censoring time. See method Pseudo-value approach from Pate et al., (2024)
for a full explanation of the approach.
Two datasets for the same cohort of inidividuals must be provided. Firstly, data_raw
must be a data.frame with one row per individual containing the variables for
the time until censoring (dtcens), and an indicator for censoring dtcens_s,
where (dtcens_s = 1) if an individual is censored at time dtcens, and dtcens_s = 0
otherwise. When an individual enters an absorbing state, this prevents censoring
from happening (i.e. dtcens_s = 0). data_raw must also contain the desired variables
for estimating the weights. Secondly, data_ms must be a dataset of class msdata,
generated using the [mstate] package. This dataset is used to apply the landmarking
and identify which state individuals are in at time t. While data_ms can be
derived from data_raw, it would be inefficient to do this within calibmsm::calib_msm
due to the bootstrapping procedure, and therefore they must be inputted seperately.
Unless the user specifies the weights using weights, the weights are
estimated using a cox-proportional hazard model, assuming a linear
functional form of the variables defined in w_covs. We urge users to
specify their own model for estimating the weights. The weights argument
must be a vector with length equal to the number of rows of data_raw.
Confidence intervals cannot be produced for the calibration scatter plots (calib_type = 'mlr').
For calibration curves estimated using calib_type = 'blr', confidence intervals
can only be estimated using bootstrapping (CI_type = 'bootstrap). This procedure uses the internal method for
estimating weights, we therefore encourage users to specify their own bootstrapping
procedure, which incorporates their own model for estimating the weights. Details
on how to do this are provided in the vignette BLR-IPCW-manual-bootstrap.
For calibration curves estimated using calib_type = 'pv', confidence intervals
can be estimated using bootstrapping (CI_type = 'bootstrap) or parametric formulae (CI_type = 'parametric).
For computational reasons we recommend using the parametric approach.
The calibration plots can be plotted using plot.calib_msm and plot.calib_mlr.
calib_msm returns a list containing two elements:
plotdata and metadata. The plotdata element contains the
data for the calibration plots. This will itself be a list with each element
containing calibration plot data for the transition probabilities into each of the possible
states. Each list element contains patient ids (id) from data_raw, the predicted
transition probabilities (pred) and the estimated observed event
probabilities (obs). If a confidence interval is requested, upper (obs_upper)
and lower (obs_lower) bounds for the observed event probabilities are also returned.
If tp_pred_plot is specified, column (id) is not returned.
The metadata element contains metadata including: a vector of the possible transitions,
a vector of which transitions calibration curves have been estimated for, the
size of the confidence interval, the method for estimating the calibration curve
and other user specified information.
Aalen OO, Johansen S. An Empirical Transition Matrix for Non-Homogeneous Markov Chains Based on Censored Observations. Scand J Stat. 1978;5(3):141-150.
Andersen PK, Pohar Perme M. Pseudo-observations in survival analysis. Stat Methods Med Res. 2010;19(1):71-99. doi:10.1177/0962280209105020
Hernan M, Robins J (2020). “12.2 Estimating IP weights via modeling.” In Causal Inference: What If, chapter 12.2. Chapman Hall/CRC, Boca Raton.
Pate, A., Sperrin, M., Riley, R. D., Peek, N., Van Staa, T., Sergeant, J. C., Mamas, M. A., Lip, G. Y. H., Flaherty, M. O., Barrowman, M., Buchan, I., & Martin, G. P. Calibration plots for multistate risk predictions models. Statistics in Medicine. 2024;April:1–23. doi: 10.1002/sim.10094.
Van Calster B, Nieboer D, Vergouwe Y, De Cock B, Pencina MJ, Steyerberg EW (2016). “A calibration hierarchy for risk models was defined: From utopia to empirical data.” Journal of Clinical Epidemiology, 74, 167–176. ISSN 18785921. doi:10.1016/j.jclinepi.2015. 12.005. URL http://dx.doi.org/10.1016/j.jclinepi.2015.12.005
Van Hoorde K, Vergouwe Y, Timmerman D, Van Huffel S, Steyerberg W, Van Calster B (2014). “Assessing calibration of multinomial risk prediction models.” Statistics in Medicine, 33(15), 2585–2596. doi:10.1002/sim.6114.
Van Hoorde K, Van Huffel S, Timmerman D, Bourne T, Van Calster B (2015). “A spline-based tool to assess and visualize the calibration of multiclass risk predictions.” Journal of Biomedical Informatics, 54, 283–293. ISSN 15320464. doi:10.1016/j.jbi.2014.12.016. URL http://dx.doi.org/10.1016/j.jbi.2014.12.016.
van Houwelingen HC (2007). “Dynamic Prediction by Landmarking in Event History Analysis.” Scandinavian Journal of Statistics, 34(1), 70–85.
Yee TW (2015). Vector Generalized Linear and Additive Models. 1 edition. Springer New, NY. ISBN 978-1-4939-4198-8. doi:10.1007/978-1-4939-2818-7. URL https://link.springer.com/book/10.1007/978-1-4939-2818-7.
# Estimate BLR-IPCW calibration curves for the predicted transition # probabilities at time t = 1826, when predictions were made at time # s = 0 in state j = 1. These predicted transition probabilities are stored in tps0. # Extract the predicted transition probabilities out of state j = 1 tp_pred <- dplyr::select(dplyr::filter(tps0, j == 1), any_of(paste("pstate", 1:6, sep = ""))) # Now estimate the observed event probabilities for each possible transition. dat_calib <- calib_msm(data_ms = msebmtcal, data_raw = ebmtcal, j=1, s=0, t = 1826, tp_pred = tp_pred, w_covs = c("year", "agecl", "proph", "match")) # Summarise the output summary(dat_calib)# Estimate BLR-IPCW calibration curves for the predicted transition # probabilities at time t = 1826, when predictions were made at time # s = 0 in state j = 1. These predicted transition probabilities are stored in tps0. # Extract the predicted transition probabilities out of state j = 1 tp_pred <- dplyr::select(dplyr::filter(tps0, j == 1), any_of(paste("pstate", 1:6, sep = ""))) # Now estimate the observed event probabilities for each possible transition. dat_calib <- calib_msm(data_ms = msebmtcal, data_raw = ebmtcal, j=1, s=0, t = 1826, tp_pred = tp_pred, w_covs = c("year", "agecl", "proph", "match")) # Summarise the output summary(dat_calib)
A data frame of 2,279 individuals with blood cancer who have undergone a transplant.
This data is identical to the ebmt4 data, except two extra variables have
been derived, time until censoring and a censoring indicator, which are required
to assess calibration using some of the methods in calibmsm. Code for the derivation
of this dataset is provided in the source code for the package.
ebmtcalebmtcal
A data frame with 2,279 rows and 17 columns:
Patient indentifier
Time until and event indicator for recovery variable
Time until and event indicator for adverse event variable
Time until and event indicator for recovery + adverse event variable
Time until and event indicator for relapse variable
Time until and event indicator for death variable
Year of transplant
Age at transplant
Prophylaxis
Donor-recipient match
Time of censoring
Event indicator, 1:censoring occured, 0: absorbing state entered before censoring occured
This dataset was derived from data made available within the mstate package, see ebmt4.
The data was originally provided by the European Group for Blood and Marrow Transplantation (https://www.ebmt.org/).
We reiterate the source statement given by the developers of mstate:
"We acknowledge the European Society for Blood and Marrow Transplantation (EBMT)
for making available these data. Disclaimer: these data were simplified for the
purpose of illustration of the analysis of competing risks and multi-state models
and do not reflect any real life situation. No clinical conclusions should be
drawn from these data."
EBMT (2023). “Data from the European Society for Blood and Marrow Transplantation.” URL https://search.r-project.org/CRAN/refmans/mstate/html/EBMT-data.html.
de Wreede LC, Fiocco M, Putter H (2011). “mstate: An R Package for the Analysis of Competing Risks and Multi-State Models.” Journal of Statistical Software, 38(7).
Used in vignette/article: Comparison-with-graphical-calibration-curves-in-competing-risks-setting.
ebmtcal_cmprskebmtcal_cmprsk
A data frame with 2,279 rows and 17 columns:
Patient indentifier
Time until and event indicator for recovery variable
Time until and event indicator for adverse event variable
Time until and event indicator for recovery + adverse event variable
Time until and event indicator for relapse variable
Time until and event indicator for death variable
Year of transplant
Age at transplant
Prophylaxis
Donor-recipient match
Time of censoring
Event indicator, 1:censoring occured, 0: absorbing state entered before censoring occured
A data frame of 2,279 individuals with blood cancer who have undergone a transplant.
This data is identical to the ebmt4 data, except two extra variables have
been derived, time until censoring and a censoring indicator, which are required
to assess calibration using some of the methods in calibmsm. Specifically, the
time until censoring ar calculated in the setting of a competing risks model out of the
first state, where no further transitions can be made. This means entry into any
state (as they are all absorbing states) will have the effect of preventing censoring
from occurring, and dtcens and dtcens_s will be different than the values found
in ebmtcal. This dataset has been designed to be used alongside
dataset msebmtcal_cmprsk, when assessing the calibration of a competing
risks model. Code for the derivation of this dataset is provided in the source code for the package.
This dataset was derived from data made available within the mstate package, see ebmt4.
The data was originally provided by the European Group for Blood and Marrow Transplantation (https://www.ebmt.org/).
We reiterate the source statement given by the developers of mstate:
"We acknowledge the European Society for Blood and Marrow Transplantation (EBMT)
for making available these data. Disclaimer: these data were simplified for the
purpose of illustration of the analysis of competing risks and multi-state models
and do not reflect any real life situation. No clinical conclusions should be
drawn from these data."
EBMT (2023). “Data from the European Society for Blood and Marrow Transplantation.” URL https://search.r-project.org/CRAN/refmans/mstate/html/EBMT-data.html.
de Wreede LC, Fiocco M, Putter H (2011). “mstate: An R Package for the Analysis of Competing Risks and Multi-State Models.” Journal of Statistical Software, 38(7).
Create S3 generic for printing metadata
metadata(x, ...)metadata(x, ...)
x |
Object generated from |
... |
Extra arguments |
msdata format.The ebmt4 data converted into msdata format (see msprep),
using the processes implemented in the mstate package. Code for the derivation
of this dataset is provided in the source code for the package.
msebmtcalmsebmtcal
A data frame in msdata format (see msprep) with 15,512 rows and 8 columns:
Patient indentifier
transition from state
transition to state
transition number
time entered state 'from'
time leaving state 'from'
time in state 'from'
event indicator, 1:transitioned to state 'to'
This dataset was derived from data made available within the mstate package, see ebmt4.
The data was originally provided by the European Group for Blood and Marrow Transplantation (https://www.ebmt.org/).
We reiterate the source statement given by the developers of mstate:
"We acknowledge the European Society for Blood and Marrow Transplantation (EBMT)
for making available these data. Disclaimer: these data were simplified for the
purpose of illustration of the analysis of competing risks and multi-state models
and do not reflect any real life situation. No clinical conclusions should be
drawn from these data."
EBMT (2023). “Data from the European Society for Blood and Marrow Transplantation.” URL https://search.r-project.org/CRAN/refmans/mstate/html/EBMT-data.html.
de Wreede LC, Fiocco M, Putter H (2011). “mstate: An R Package for the Analysis of Competing Risks and Multi-State Models.” Journal of Statistical Software, 38(7).
Used in vignette/article: Comparison-with-graphical-calibration-curves-in-competing-risks-setting.
msebmtcal_cmprskmsebmtcal_cmprsk
A data frame with 9,116 rows and 8 columns:
Patient indentifier
transition from state
transition to state
transition number
time entered state 'from'
time leaving state 'from'
time in state 'from'
event indicator, 1:transitioned to state 'to'
The ebmt4 data converted into msdata format (see msprep),
where all subsequent states are considered absorbing states. i.e. only transitions out of the initial state are considered,
meaning this data constitutes a competing risks model out of the initial state. Code for the derivation
of this dataset is provided in the source code for the package.
This dataset was derived from data made available within the mstate package, see ebmt4.
The data was originally provided by the European Group for Blood and Marrow Transplantation (https://www.ebmt.org/).
We reiterate the source statement given by the developers of mstate:
"We acknowledge the European Society for Blood and Marrow Transplantation (EBMT)
for making available these data. Disclaimer: these data were simplified for the
purpose of illustration of the analysis of competing risks and multi-state models
and do not reflect any real life situation. No clinical conclusions should be
drawn from these data."
EBMT (2023). “Data from the European Society for Blood and Marrow Transplantation.” URL https://search.r-project.org/CRAN/refmans/mstate/html/EBMT-data.html.
de Wreede LC, Fiocco M, Putter H (2011). “mstate: An R Package for the Analysis of Competing Risks and Multi-State Models.” Journal of Statistical Software, 38(7).
calib_mlr estimated using
using calib_msm.Plots calibration scatter plots for the transition probabilities of a multistate model estimated using the MLR-IPCW approach.
## S3 method for class 'calib_mlr' plot( x, ..., combine = TRUE, ncol = NULL, nrow = NULL, size_point = 0.5, size_text = 12, transparency_plot = 0.25, marg_density = FALSE, marg_density_size = 5, marg_density_type = "density", marg_rug = FALSE, marg_rug_transparency = 0.1, titles_include = TRUE, titles = NULL, axis_titles_x = NULL, axis_titles_text_x = "Predicted risk", axis_titles_y = NULL, axis_titles_text_y = "Predicted-observed risk" )## S3 method for class 'calib_mlr' plot( x, ..., combine = TRUE, ncol = NULL, nrow = NULL, size_point = 0.5, size_text = 12, transparency_plot = 0.25, marg_density = FALSE, marg_density_size = 5, marg_density_type = "density", marg_rug = FALSE, marg_rug_transparency = 0.1, titles_include = TRUE, titles = NULL, axis_titles_x = NULL, axis_titles_text_x = "Predicted risk", axis_titles_y = NULL, axis_titles_text_y = "Predicted-observed risk" )
x |
Object of class |
... |
Other |
combine |
Whether to combine into one plot using ggarrange, or return as a list of individual plots |
ncol |
Number of columns for combined calibration plot |
nrow |
Number of rows for combined calibration plot |
size_point |
Size of points in scatter plot |
size_text |
Size of text in plot |
transparency_plot |
Degree of transparency for points in the calibration scatter plot |
marg_density |
Whether to produce marginal density plots TRUE/FALSE |
marg_density_size |
Size of the main plot relative to the density plots (see |
marg_density_type |
What type of marginal plot to show (see |
marg_rug |
Whether to produce marginal rug plots TRUE/FALSE |
marg_rug_transparency |
Degree of transparency for the density rug plot along each axis |
titles_include |
Whether to include titles for each individual calibration plots |
titles |
Vector of titles for the calibration plots_ Defaults to "State k" for each plot_ |
axis_titles_x |
Position of plots for which to include title on x-axis |
axis_titles_text_x |
x-axis title |
axis_titles_y |
Position of plots for which to include title on y-axis |
axis_titles_text_y |
y-axis title |
If combine = TRUE, returns an object of classes gg, ggplot, and ggarrange,
as all ggplots have been combined into one object. If combine = FALSE, returns an object of
class list, each element containing an object of class gg and ggplot.
# Using competing risks data out of initial state (see vignette: ... -in-competing-risk-setting). # Estimate and plot MLR-IPCW calibration scatter plots for the predicted transition # probabilities at time t = 1826, when predictions were made at time # s = 0 in state j = 1. These predicted transition probabilities are stored in tp_cmprsk_j0. # To minimise example time we reduce the datasets to 150 individuals. # Extract the predicted transition probabilities out of state j = 1 for first 150 individuals tp_pred <- tp_cmprsk_j0 |> dplyr::filter(id %in% 1:150) |> dplyr::select(any_of(paste("pstate", 1:6, sep = ""))) # Reduce ebmtcal to first 150 individuals ebmtcal <- ebmtcal |> dplyr::filter(id %in% 1:150) # Reduce msebmtcal_cmprsk to first 150 individuals msebmtcal_cmprsk <- msebmtcal_cmprsk |> dplyr::filter(id %in% 1:150) # Now estimate the observed event probabilities for each possible transition. dat_calib <- calib_msm(data_ms = msebmtcal_cmprsk, data_raw = ebmtcal, j=1, s=0, t = 1826, tp_pred = tp_pred, calib_type = "mlr", w_covs = c("year", "agecl", "proph", "match"), mlr_ps_int = 2, mlr_degree = 2) # These are then plotted plot(dat_calib, combine = TRUE, nrow = 2, ncol = 3)# Using competing risks data out of initial state (see vignette: ... -in-competing-risk-setting). # Estimate and plot MLR-IPCW calibration scatter plots for the predicted transition # probabilities at time t = 1826, when predictions were made at time # s = 0 in state j = 1. These predicted transition probabilities are stored in tp_cmprsk_j0. # To minimise example time we reduce the datasets to 150 individuals. # Extract the predicted transition probabilities out of state j = 1 for first 150 individuals tp_pred <- tp_cmprsk_j0 |> dplyr::filter(id %in% 1:150) |> dplyr::select(any_of(paste("pstate", 1:6, sep = ""))) # Reduce ebmtcal to first 150 individuals ebmtcal <- ebmtcal |> dplyr::filter(id %in% 1:150) # Reduce msebmtcal_cmprsk to first 150 individuals msebmtcal_cmprsk <- msebmtcal_cmprsk |> dplyr::filter(id %in% 1:150) # Now estimate the observed event probabilities for each possible transition. dat_calib <- calib_msm(data_ms = msebmtcal_cmprsk, data_raw = ebmtcal, j=1, s=0, t = 1826, tp_pred = tp_pred, calib_type = "mlr", w_covs = c("year", "agecl", "proph", "match"), mlr_ps_int = 2, mlr_degree = 2) # These are then plotted plot(dat_calib, combine = TRUE, nrow = 2, ncol = 3)
calib_msm.Plots calibration curves for the transition probabilities of a multistate model estimated using BLR-IPCW and pseudo-value approaches.
## S3 method for class 'calib_msm' plot( x, ..., combine = TRUE, ncol = NULL, nrow = NULL, size_line = 0.5, size_text = 12, marg_density = TRUE, marg_density_size = 5, marg_density_type = "density", marg_rug = FALSE, marg_rug_transparency = 0.1, titles_include = TRUE, titles = NULL, axis_titles_x = NULL, axis_titles_text_x = "Predicted risk", axis_titles_y = NULL, axis_titles_text_y = "Predicted-observed risk", legend_include = TRUE, legend_seperate = FALSE, legend_title = NULL, legend_position = "bottom" )## S3 method for class 'calib_msm' plot( x, ..., combine = TRUE, ncol = NULL, nrow = NULL, size_line = 0.5, size_text = 12, marg_density = TRUE, marg_density_size = 5, marg_density_type = "density", marg_rug = FALSE, marg_rug_transparency = 0.1, titles_include = TRUE, titles = NULL, axis_titles_x = NULL, axis_titles_text_x = "Predicted risk", axis_titles_y = NULL, axis_titles_text_y = "Predicted-observed risk", legend_include = TRUE, legend_seperate = FALSE, legend_title = NULL, legend_position = "bottom" )
x |
Object of class 'calib_msm' generated from |
... |
Other |
combine |
Whether to combine into one plot using ggarrange, or return as a list of individual plots |
ncol |
Number of columns for combined calibration plot |
nrow |
Number of rows for combined calibration plot |
size_line |
Size of line plots |
size_text |
Size of text in plot |
marg_density |
Whether to produce marginal density plots TRUE/FALSE |
marg_density_size |
Size of the main plot relative to the density plots (see |
marg_density_type |
What type of marginal plot to show (see |
marg_rug |
Whether to produce marginal rug plots TRUE/FALSE |
marg_rug_transparency |
Degree of transparency for the density rug plot along each axis |
titles_include |
Whether to include titles for each individual calibration plots |
titles |
Vector of titles for the calibration plots. Defaults to "State k" for each plot. |
axis_titles_x |
Position of plots for which to include title on x-axis |
axis_titles_text_x |
x-axis title |
axis_titles_y |
Position of plots for which to include title on y-axis |
axis_titles_text_y |
y-axis title |
legend_include |
Whether to produce a legend |
legend_seperate |
= Whether to include legend in plot (FALSE) or as a seperate object (TRUE) |
legend_title |
Title of legend |
legend_position |
Position of legend |
If combine = TRUE, returns an object of classes gg, ggplot, and ggarrange,
as all ggplots have been combined into one object. If combine = FALSE, returns an object of
class list, each element containing an object of class gg and ggplot.
# Estimate and plot BLR-IPCW calibration curves for the predicted transition # probabilities at time t = 1826, when predictions were made at time # s = 0 in state j = 1. These predicted transition probabilities are stored in tps0. # Extract the predicted transition probabilities out of state j = 1 tp_pred <- dplyr::select(dplyr::filter(tps0, j == 1), any_of(paste("pstate", 1:6, sep = ""))) # Now estimate the observed event probabilities for each possible transition. dat_calib <- calib_msm(data_ms = msebmtcal, data_raw = ebmtcal, j=1, s=0, t = 1826, tp_pred = tp_pred, w_covs = c("year", "agecl", "proph", "match")) # These are then plotted plot(dat_calib, combine = TRUE, nrow = 2, ncol = 3)# Estimate and plot BLR-IPCW calibration curves for the predicted transition # probabilities at time t = 1826, when predictions were made at time # s = 0 in state j = 1. These predicted transition probabilities are stored in tps0. # Extract the predicted transition probabilities out of state j = 1 tp_pred <- dplyr::select(dplyr::filter(tps0, j == 1), any_of(paste("pstate", 1:6, sep = ""))) # Now estimate the observed event probabilities for each possible transition. dat_calib <- calib_msm(data_ms = msebmtcal, data_raw = ebmtcal, j=1, s=0, t = 1826, tp_pred = tp_pred, w_covs = c("year", "agecl", "proph", "match")) # These are then plotted plot(dat_calib, combine = TRUE, nrow = 2, ncol = 3)
Used in vignette/article: Comparison-with-graphical-calibration-curves-in-competing-risks-setting.
tp_cmprsk_j0tp_cmprsk_j0
A data frame with 2,279 rows and 13 columns:
Patient indentifier
Predicted transition probabilities of transitions into states 1 to 6
Standard error of the predicted transition probabilities of transitions into states 1 to 6
Data frame containing the predicted transition probabilities out of state j = 1
made at time s = 0, for a competing risks model out of the initial state (see msebmtcal_cmprsk).
The predicted transition probabilities were estimated by fitting
a competing risks model to the msebmtcal_cmprsk data using a leave-one-out approach.
Code for deriving this dataset is provided in the source code for calibmsm. Code for the derivation
of this dataset is provided in the source code for the package.
This dataset was derived from data made available within the mstate package, see ebmt4.
The data was originally provided by the European Group for Blood and Marrow Transplantation (https://www.ebmt.org/).
We reiterate the source statement given by the developers of mstate:
"We acknowledge the European Society for Blood and Marrow Transplantation (EBMT)
for making available these data. Disclaimer: these data were simplified for the
purpose of illustration of the analysis of competing risks and multi-state models
and do not reflect any real life situation. No clinical conclusions should be
drawn from these data."
EBMT (2023). “Data from the European Society for Blood and Marrow Transplantation.” URL https://search.r-project.org/CRAN/refmans/mstate/html/EBMT-data.html.
de Wreede LC, Fiocco M, Putter H (2011). “mstate: An R Package for the Analysis of Competing Risks and Multi-State Models.” Journal of Statistical Software, 38(7).
Data frame containing the predicted transition probabilities out of state j = 1
made at time s = 0. The predicted transition probabilities were estimated by fitting
a multistate model to the ebmt4 data using a leave-one-out approach.
Code for deriving this dataset is provided in the source code for calibmsm. Code for the derivation
of this dataset is provided in the source code for the package.
tps0tps0
A data frame with 13,674 (CHANGE) rows and 14 columns:
Patient indentifier
Predicted transition probabilities of transitions into states 1 to 6
Standard error of the predicted transition probabilities of transitions into states 1 to 6
State from which the predicted transition probabilities are estimated from
This dataset was derived from data made available within the mstate package, see ebmt4.
The data was originally provided by the European Group for Blood and Marrow Transplantation (https://www.ebmt.org/).
We reiterate the source statement given by the developers of mstate:
"We acknowledge the European Society for Blood and Marrow Transplantation (EBMT)
for making available these data. Disclaimer: these data were simplified for the
purpose of illustration of the analysis of competing risks and multi-state models
and do not reflect any real life situation. No clinical conclusions should be
drawn from these data."
EBMT (2023). “Data from the European Society for Blood and Marrow Transplantation.” URL https://search.r-project.org/CRAN/refmans/mstate/html/EBMT-data.html.
de Wreede LC, Fiocco M, Putter H (2011). “mstate: An R Package for the Analysis of Competing Risks and Multi-State Models.” Journal of Statistical Software, 38(7).
Data frame containing the predicted transition probabilities out of states 1 (transplant),
2 (adverse event), 3 (recovery) and 4 (adverse event + recovery), made at time s = 100.
The predicted transition probabilities were estimated by fitting a multistate model
to the ebmt4 data using a leave-one-out approach. Code for deriving
this dataset is provided in the source code for calibmsm. Code for derivation
of this dataset is provided in the source code for the package.
tps100tps100
A data frame with 13,674 (CHANGE) rows and 14 columns:
Patient indentifier
Predicted transition probabilities of transitions into states 1 to 6
Standard error of the predicted transition probabilities of transitions into states 1 to 6
State from which the predicted transition probabilities are estimated from
This dataset was derived from data made available within the mstate package, see ebmt4.
The data was originally provided by the European Group for Blood and Marrow Transplantation (https://www.ebmt.org/).
We reiterate the source statement given by the developers of mstate:
"We acknowledge the European Society for Blood and Marrow Transplantation (EBMT)
for making available these data. Disclaimer: these data were simplified for the
purpose of illustration of the analysis of competing risks and multi-state models
and do not reflect any real life situation. No clinical conclusions should be
drawn from these data."
EBMT (2023). “Data from the European Society for Blood and Marrow Transplantation.” URL https://search.r-project.org/CRAN/refmans/mstate/html/EBMT-data.html.
de Wreede LC, Fiocco M, Putter H (2011). “mstate: An R Package for the Analysis of Competing Risks and Multi-State Models.” Journal of Statistical Software, 38(7).