cif_matching {adjustedCurves}R Documentation

Using Propensity-Score Matching to Calculate Adjusted CIFs

Description

This page explains the details of estimating adjusted cumulative incidence functions using propensity-score matching in a competing risks setting (method="matching" in the adjustedcif function). All regular arguments of the adjustedcif function can be used. Additionally, the treatment_model argument has to be specified in the adjustedcif call. Further arguments specific to this method are listed below.

Arguments

treatment_model

[required] Must be either a model object with variable as response variable or a vector of previously estimated propensity scores.

gtol

Tolerance at which estimated treatment assignment probabilities are truncated. Every propensity score bigger than 1 - gtol is set to 1 - gtol and every propensity score smaller than gtol is set to gtol. Useful when there are extreme propensity scores close to 0 or 1. Defaults to 0.001,

...

Further arguments passed to the Match function of the Matching Package.

Details

Using the estimated propensity score, the individual observations in the dataset are matched to each other creating a new dataset in which the covariate distributions are balanced in respect to the two groups defined by variable. A simple Aalen-Johansen estimator is then used to calculate the confounder-adjusted CIFs. This corresponds to the method described in Austin & Fine (2019). Details on the algorithm used for matching can be found in the documentation of the Matching package.

Simulation results showed that this specific implementation of this method is the least efficient method contained in this R-Package. While it does produce unbiased estimates, the variation in these estimates is very high. We strongly suggest using one of the other methods implemented here.

Value

Adds the following additional objects to the output of the adjustedcif function:

Author(s)

Robin Denz

References

Peter C. Austin and Jason P. Fine (2019). "Propensity-Score Matching with Competing Risks in Survival Analysis". In: Statistics in Medicine 38, pp. 751-777

See Also

Match, cuminc

Examples

library(adjustedCurves)
library(survival)

if (requireNamespace("Matching")) {

library(Matching)

set.seed(42)

# simulate some data as example
sim_dat <- sim_confounded_crisk(n=50, max_t=5)
sim_dat$group <- as.factor(sim_dat$group)

# estimate treatment assignment model
glm_mod <- glm(group ~ x1 + x2 + x4 + x6, data=sim_dat, family="binomial")

# calculate adjusted CIFs
adjcif <- adjustedcif(data=sim_dat,
                      variable="group",
                      ev_time="time",
                      event="event",
                      cause=1,
                      method="matching",
                      treatment_model=glm_mod)
plot(adjcif)

# Alternatively, supply the propensity score directly
# Here we use the logistic regression to calculate it, so we get
# exactly the same result. The propensity score can be calculated in
# any other way in practice, allowing flexibility
ps_score <- glm_mod$fitted.values

adjcif <- adjustedcif(data=sim_dat,
                      variable="group",
                      ev_time="time",
                      event="event",
                      cause=1,
                      method="matching",
                      treatment_model=ps_score)

# plot the curves
plot(adjcif)
}

[Package adjustedCurves version 0.11.2 Index]