Package {gcomputation}

Type:	Package
Title:	Causal Inference by using G-Computation
Version:	0.34
Depends:	R (≥ 4.0.0), survival, hdnom, glmnet, MASS, mice
Imports:	graphics, utils, methods, grDevices, stats
Description:	Several functions and S3 methods for G-computation and emulation of clinical trials. It allows for flexible estimation of the outcome model, especially penalized regressions (Lasso, Ridge, or Elasticnet) for binary, continuous, counting, or right-censored time-to-event outcomes. Average treatment effect among the entire population (ATE) or among the treated population (ATT) can be estimated. The method for time-to-events is described by Chatton et al. (2020) <doi:10.1038/s41598-020-65917-x>. For a binary outcome, details are available in the paper proposed by Chatton et al. (2022) <doi:10.1177/09622802211047345>.
License:	GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]
Encoding:	UTF-8
LazyLoad:	yes
NeedsCompilation:	no
Maintainer:	Yohann Foucher <yohann.foucher@univ-poitiers.fr>
BugReports:	https://github.com/chupverse/gcomputation/issues
RoxygenNote:	7.3.3
Packaged:	2026-05-06 13:05:33 UTC; foucher-y
Author:	Yohann Foucher [aut, cre], Joe De Keizer [aut]
Repository:	CRAN
Date/Publication:	2026-05-11 19:20:02 UTC

Simulated Real Word Data Similar to the PROPHYVAP Study

Description

This dataset dataCOHORT is a simulated observational cohort designed to reflect a real-world data similar to the PROPHYVAP clinical trial (see dataPROPHYVAP).

Usage

data(dataCOHORT)

Format

A data frame with 2000 observations for the following variables:

GROUP

This character vector represents the treatment group (1 for Ceftriaxone and 0 otherwise)

AGE

This numeric vector represents the patient age in years

SEX

This character vector represents the gender (F=female/M=male)

BMI

This numeric vector represents the body mass index in kg/m2

DIABETES

This character vector represents the diabetes status (Yes/No)

ALCOHOL

This character vector represents the alcohol consumption (Yes/No)

SMOKING

This character vector represents the tabaco status (Yes/No)

INJURY

This character vector represents the cause of the injury in 4 classes

GLASGOW

This character vector represents the Glasgow scale in 3 classes

PAO2FIO2

This character vector represents the PAO2-FIO2 ratio in 2 classes

LEUKO

This character vector represents leukocytosis at admission per mm3 in 2 classes

TIME_INTUBATION

This numeric vector represents the time to intubation in hours

VAP

This character vector represents ventilatory associated pneumonia (1 for event and 0 otherwise)

TIME_DEATH

This numeric vector represents the time to death in days (follow-up of 60 days)

DEATH

This character vector represents status at follow-up end (1 for event and 0 otherwise)

References

Dahyot-Fizelier et al. Ceftriaxone to prevent early ventilator-associated pneumonia in patients with acute brain injury: a multicentre, randomised, double-blind, placebo-controlled, assessor-masked superiority trial. Lancet Respir Med, 12:375-385, 2024. <doi:10.1016/S2213-2600(23)00471-X>.

Examples

data(dataCOHORT)

### Kaplan and Meier estimation of the survival at day 60
plot(survfit(Surv(TIME_DEATH, DEATH) ~ 1, data = dataCOHORT),
 xlab="Time in days", ylab="Patient survival", conf.int=TRUE,
 mark.time=FALSE)
 

A Simulated Randomized Clinical Trial from the PROPHYVAP Study

Description

Ventilator-associated pneumonia (VAP) is the first cause of healthcare-associated infections in intensive care units. The PROPHYVAP is a French multicenter, randomized, double-blind, placebo-controlled, clinical trial. The main objective of this study was to determine whether a single dose of Ceftriaxone within the 12 hours post-intubation can decrease the risk of early-onset VAP and mortality at 60 days in patients admitted for severe brain injury. All variables in this dataset were simulated according to the original trial. The number of episodes of severe hypotension was not observed and completely generated by Poisson regression for illustrative purposes of this package.

Usage

data(dataPROPHYVAP)

Format

A data frame with 319 observations for the following variables:

GROUP

This character vector represents the treatment (1 for Ceftriaxone and 0 for Placebo)

AGE

This numeric vector represents the patient age in years

SEX

This character vector represents the gender (F=female/M=male)

BMI

This numeric vector represents the body mass index in kg/m2

DIABETES

This character vector represents the diabetes status (Yes/No)

ALCOHOL

This character vector represents the alcohol consumption (Yes/No)

SMOKING

This character vector represents the tabaco status (Yes/No)

INJURY

This character vector represents the cause of the injury in 4 classes

GLASGOW

This character vector represents the Glasgow scale in 3 classes

PAO2FIO2

This character vector represents the PAO2-FIO2 ratio in 2 classes

LEUKO

This character vector represents leukocytosis at admission per mm3 in 2 classes

TIME_INTUBATION

This numeric vector represents the time to intubation in hours

TIME_TRT

This numeric vector represents the time to treatment in hours

VAP

This character vector represents ventilatory associated pneumonia (1 for event and 0 otherwise)

TIME_DEATH

This numeric vector represents the time to death in days (follow-up of 60 days)

DEATH

This character vector represents status at follow-up end (1 for event and 0 otherwise)

HYPOTENSION

This numeric vector represents the number of episodes of severe hypotension.

References

Dahyot-Fizelier et al. Ceftriaxone to prevent early ventilator-associated pneumonia in patients with acute brain injury: a multicentre, randomised, double-blind, placebo-controlled, assessor-masked superiority trial. Lancet Respir Med, 12:375-385, 2024. <doi:10.1016/S2213-2600(23)00471-X>.

Examples

data(dataPROPHYVAP)

### Kaplan and Meier estimation of the survival at day 60
plot(survfit(Surv(TIME_DEATH, DEATH) ~ GROUP, data = dataPROPHYVAP),
 xlab="Post-randomization time in days", ylab="Patient survival",
 mark.time=FALSE, col=c("red3","blue3"), lty=c(2,1))
 
legend("bottomleft", c("Placebo", "Ceftriaxone"), col=c("red3","blue3"), lty=1:2) 

G-Computation to Estimate a Marginal Effect for a Binary Outcome

Description

This function computes G-computation (GC) with different working models or algorithms (Q-models) for a binary outcome and a 2-class exposure/treatment.

Usage

gc_binary(formula, data, group, effect="ATE", model,
            param.tune=NULL, cv=30, boot.type="bcv", boot.number=500,
            boot.tune=FALSE, progress=TRUE, seed=NULL, boot.mi=FALSE, m=5, ...)

Arguments

Details

The option effect="ATE" corresponds to the Average Treatment effect on the Entire population, i.e. the marginal effect if the entire sample were treated versus untreated. The "ATT" modality allows the estimation of the Average Treatment effect on the Treated, i.e. the marginal effect if the treated subjects (group = 1) would have been untreated. The "ATU" modality allows the estimation of the Average Treatment effect on the Untreated, i.e. the marginal effect if the untreated subjects (group = 0) would have been treated.

Several modelling methods can be used for the Q-model estimation:

The param.tune argument allows users to specify tuning parameters of penalized regression. If NULL (the default), the tuning parameters of each algorithm are estimated by cv-fold cross-validation and the default grid of the glmnet package is used. Otherwise, the user can propose a specific grid. For "lasso" and "ridge" penalizations, it should be a vector representing the lambda penalization parameter. For "elasticnet", it should be a list or a vector of length 2, containing lambda (penalization parameter) and alpha (mixing parameter between L1 and L2 regularizations) values. The alpha value typically ranges from 0 to 1. The user may propose a single value of each tuning parameter if she/he aims to define her/his own penalty.

The boot.tune argument is logical value which determines whether the tuning parameter should be estimated inside of each bootstrap iteration. If FALSE (the default), the tuning parameter will be estimated once on the complete dataset.

The boot.type argument controls how bootstrap estimates are computed. With boot.type = "boot", each iteration fits the Q-model on a sampled dataset and predicts on the same patients. With boot.type = "bcv", the Q-model is fitted on the sampled patients, and predictions are made on patients not included in that sample.

If boot.mi = TRUE, multiple imputation is considered by using the MI-BOOT approach proposed by Schomaker & Heumann (2018). The dataset is first imputed using the mice function to create m complete datasets. The bootstrap samples are then generated from each imputed dataset. The Q-model and the G-computation is performed for each m x boot.number samples, and all runs are finally concatenated.

Value

References

Joe de Keizer et al. G-computation for increasing performances of clinical trials with individual randomization and binary response. ArXiv 2024-11. <doi:10.48550/arXiv.2411.10089>.

Schomaker M, Heumann C. Bootstrap inference when using multiple imputation. Statistics in medicine. 2018;37(14):2252-2266. <doi:10.1002/sim.7654>.

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * AGE + SEX + ALCOHOL + BMI + DIABETES + GLASGOW + INJURY)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_binary(formula=.f, model="lasso", data=dataPROPHYVAP,
              group="GROUP", param.tune=NULL, cv=10, boot.type="bcv",
              boot.number=10, effect="ATE", boot.tune=TRUE)

summary(gc1)

G-Computation to Estimate a Marginal Effect with a Continuous Outcome

Description

This function computes G-computation (GC) with different working models or algorithms (Q-models) for a continuous outcome and a 2-class exposure/treatment.

Usage

gc_continuous(formula, data, group, effect="ATE", model,
            param.tune=NULL, cv=30, boot.type="bcv", boot.number=500,
            boot.tune=FALSE, progress=TRUE, seed=NULL, boot.mi=FALSE, m=5, ...)

Arguments

Details

The option methods="ATE" corresponds to the Average Treatment effect on the Entire population, i.e. the marginal effect if the entire sample were treated versus untreated. The "ATT" modality allows the estimation of the Average Treatment effect on the Treated, i.e. the marginal effect if the treated subjects (group = 1) would have been untreated. The "ATU" modality allows the estimation of the Average Treatment effect on the Untreated, i.e. the marginal effect if the untreated subjects (group = 0) would have been treated.

Several modelling methods can be used for the Q-model estimation:

The param.tune argument allows users to specify tuning parameters of penalized regression. If NULL (the default), the tuning parameters of each algorithm are estimated by cv-fold cross-validation and the default grid of the glmnet package is used. Otherwise, the user can propose a specific grid. For "lasso" and "ridge" penalizations, it should be a vector representing the lambda penalization parameter. For "elasticnet", it should be a list or a vector of length 2, containing lambda (penalization parameter) and alpha (mixing parameter between L1 and L2 regularizations) values. The alpha value typically ranges from 0 to 1. The user may propose a single value of each tuning parameter if she/he aims to define her/his own penalty.

The boot.tune argument is logical value which determines whether the tuning parameter should be estimated inside of each bootstrap iteration. If FALSE (the default), the tuning parameter will be estimated once on the complete dataset.

The boot.type argument controls how bootstrap estimates are computed. With boot.type = "boot", each iteration fits the Q-model on a sampled dataset and predicts on the same patients. With boot.type = "bcv", the Q-model is fitted on the sampled patients, and predictions are made on patients not included in that sample.

If boot.mi = TRUE, multiple imputation is considered by using the MI-BOOT approach proposed by Schomaker & Heumann (2018). The dataset is first imputed using the mice function to create m complete datasets. The bootstrap samples are then generated from each imputed dataset. The Q-model and the G-computation is performed for each m x boot.number samples, and all runs are finally concatenated.

Value

References

Joe de Keizer et al. G-computation for increasing performances of clinical trials with individual randomization and binary response. ArXiv 2024-11. <doi:10.48550/arXiv.2411.10089>.

Schomaker M, Heumann C. Bootstrap inference when using multiple imputation. Statistics in medicine. 2018;37(14):2252-2266. <doi:10.1002/sim.7654>.

Examples

data("dataPROPHYVAP")

.f <- formula(HYPOTENSION ~ GROUP * AGE + SEX + ALCOHOL + BMI + DIABETES + GLASGOW + INJURY)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_continuous(formula=.f, model="lasso", data=dataPROPHYVAP,
              group="GROUP", param.tune=NULL, cv=10, boot.type="bcv",
              boot.number=10, effect="ATE", boot.tune=TRUE)

summary(gc1)

G-Computation to Estimate a Marginal Effect with a Counting Outcome

Description

This function computes G-computation (GC) with different working models or algorithms (Q-models) for a counting outcome and a 2-class exposure/treatment.

Usage

gc_count(formula, data, group, effect="ATE", model,
            param.tune=NULL, cv=30, boot.type="bcv", boot.number=500,
            boot.tune=FALSE, progress=TRUE, seed=NULL, boot.mi=FALSE, m=5, ...)

Arguments

Details

The option methods="ATE" corresponds to the Average Treatment effect on the Entire population, i.e. the marginal effect if the entire sample were treated versus untreated. The "ATT" modality allows the estimation of the Average Treatment effect on the Treated, i.e. the marginal effect if the treated subjects (group = 1) would have been untreated. The "ATU" modality allows the estimation of the Average Treatment effect on the Untreated, i.e. the marginal effect if the untreated subjects (group = 0) would have been treated.

Several modelling methods can be used for the Q-model estimation:

The param.tune argument allows users to specify tuning parameters of penalized regression. If NULL (the default), the tuning parameters of each algorithm are estimated by cv-fold cross-validation and the default grid of the glmnet package is used. Otherwise, the user can propose a specific grid. For "lasso" and "ridge" penalizations, it should be a vector representing the lambda penalization parameter. For "elasticnet", it should be a list or a vector of length 2, containing lambda (penalization parameter) and alpha (mixing parameter between L1 and L2 regularizations) values. The alpha value typically ranges from 0 to 1. The user may propose a single value of each tuning parameter if she/he aims to define her/his own penalty.

The boot.tune argument is logical value which determines whether the tuning parameter should be estimated inside of each bootstrap iteration. If FALSE (the default), the tuning parameter will be estimated once on the complete dataset.

The boot.type argument controls how bootstrap estimates are computed. With boot.type = "boot", each iteration fits the Q-model on a sampled dataset and predicts on the same patients. With boot.type = "bcv", the Q-model is fitted on the sampled patients, and predictions are made on patients not included in that sample.

If boot.mi = TRUE, multiple imputation is considered by using the MI-BOOT approach proposed by Schomaker & Heumann (2018). The dataset is first imputed using the mice function to create m complete datasets. The bootstrap samples are then generated from each imputed dataset. The Q-model and the G-computation is performed for each m x boot.number samples, and all runs are finally concatenated.

Value

References

Joe de Keizer et al. G-computation for increasing performances of clinical trials with individual randomization and binary response. ArXiv 2024-11. <doi:10.48550/arXiv.2411.10089>.

Schomaker M, Heumann C. Bootstrap inference when using multiple imputation. Statistics in medicine. 2018;37(14):2252-2266. <doi:10.1002/sim.7654>.

Examples

data("dataPROPHYVAP")

.f <- formula(HYPOTENSION ~ GROUP * AGE + SEX + ALCOHOL + BMI + DIABETES + GLASGOW + INJURY)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_count(formula=.f, model="lasso", data=dataPROPHYVAP,
              group="GROUP", param.tune=NULL, cv=10, boot.type="bcv",
              boot.number=10, effect="ATE", boot.tune=TRUE)

summary(gc1)

G-Computation to Estimate a Marginal Effect for Time-to-Event Outcome

Description

This function computes G-computation (GC) with different working models or algorithms (Q-models) for a time-to-event outcome and a 2-class exposure/treatment.

Usage

gc_times(formula, data, group, pro.time=NULL, effect="ATE", model,
            param.tune=NULL, cv=30, boot.type="bcv", boot.number=500,
            boot.tune=FALSE, progress=TRUE, seed=NULL, boot.mi=FALSE, m=5, ...)

Arguments

Details

The option effect="ATE" corresponds to the Average Treatment effect on the Entire population, i.e. the marginal effect if the entire sample were treated versus untreated. The "ATT" modality allows the estimation of the Average Treatment effect on the Treated, i.e. the marginal effect if the treated subjects (group = 1) would have been untreated. The "ATU" modality allows the estimation of the Average Treatment effect on the Untreated, i.e. the marginal effect if the untreated subjects (group = 0) would have been treated.

Several modelling methods can be used for the Q-model estimation:

The param.tune argument allows users to specify tuning parameters of penalized regression. If NULL (the default), the tuning parameters of each algorithm are estimated by cv-fold cross-validation and the default grid of the glmnet package is used. Otherwise, the user can propose a specific grid. For "lasso" and "ridge" penalizations, it should be a vector representing the lambda penalization parameter. For "elasticnet", it should be a list or a vector of length 2, containing lambda (penalization parameter) and alpha (mixing parameter between L1 and L2 regularizations) values. The alpha value typically ranges from 0 to 1. The user may propose a single value of each tuning parameter if she/he aims to define her/his own penalty.

The boot.tune argument is logical value which determines whether the tuning parameter should be estimated inside of each bootstrap iteration. If FALSE (the default), the tuning parameter will be estimated once on the complete dataset.

The boot.type argument controls how bootstrap estimates are computed. With boot.type = "boot", each iteration fits the Q-model on a sampled dataset and predicts on the same patients. With boot.type = "bcv", the Q-model is fitted on the sampled patients, and predictions are made on patients not included in that sample.

If boot.mi = TRUE, multiple imputation is considered by using the MI-BOOT approach proposed by Schomaker & Heumann (2018). The dataset is first imputed using the mice function to create m complete datasets. The bootstrap samples are then generated from each imputed dataset. The Q-model and the G-computation is performed for each m x boot.number samples, and all runs are finally concatenated.

Value

References

Chatton et al. G-computation and doubly robust standardisation for continuous-time data: A comparison with inverse probability weighting. Stat Methods Med Res. 31(4):706-718. 2022. <doi:10.1177/09622802211047345>.

Examples

data(dataPROPHYVAP)

.f <- formula(Surv(TIME_DEATH, DEATH) ~ GROUP + AGE +
              SEX + BMI + DIABETES)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_times(formula=.f, model="lasso", data=dataPROPHYVAP,
              group="GROUP", cv=10, boot.type="bcv",
              boot.number=10, effect="ATE", progress=TRUE , pro.time=20,
              boot.tune=FALSE)

summary(gc1)

S3 Method for Plotting a 'gcbinary' Object

Description

Provides a calibration plot for an object returned by the function gc_binary. This method assesses how well the Q-model's predicted values align with the observed binary outcomes.

Usage

## S3 method for class 'gcbinary'
plot(x, n.groups=5, smooth=FALSE, ...)

Arguments

Details

The function visualizes the calibration of the Q-model by dividing predicted probabilities into n.groups quantiles. For each group, the average predicted probability is plotted against the observed proportion of events, including 95% confidence intervals for the observed values. An identity line is provided for reference and perfect calibration is indicated when the observed points are directly along this line.

Value

No return value for this S3 method.

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * AGE + SEX + BMI + DIABETES)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_binary(formula=.f, model="ridge", data=dataPROPHYVAP, group="GROUP", 
                 param.tune=NULL, boot.type="bcv", cv=10, boot.number=10, 
                 effect="ATE", progress=TRUE, boot.tune=TRUE)
                     
### Plot the calibration curve
plot(gc1, n.groups=3, col="red3")

S3 Method for Plotting a 'gccontinuous' Object

Description

Provides a calibration plot for an object returned by the function gc_continuous. This method assesses how well the Q-model's predicted values align with the observed continuous outcomes.

Usage

## S3 method for class 'gccontinuous'
plot(x, n.groups=5, smooth=FALSE, ...)

Arguments

Details

The function visualizes the calibration of the Q-model by dividing predicted probabilities into n.groups quantiles. For each group, the average predicted probability is plotted against the observed proportion of events, including 95% confidence intervals for the observed values. An identity line is provided for reference and perfect calibration is indicated when the observed points are directly along this line.

Value

No return value for this S3 method.

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * (AGE + SEX + ALCOHOL + BMI + DIABETES))

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_continuous(formula=.f, model="all", data=dataPROPHYVAP, group="GROUP", 
                     cv=10, boot.type="boot", boot.number=10, boot.tune=TRUE, 
                     effect="ATE", progress=TRUE, seed=5192)
                     
### Plot the calibration curve
plot(gc1, n.groups=5, col="red3")

S3 Method for Plotting a 'gccount' Object

Description

Provides a calibration plot for an object returned by the function gc_count. This method assesses how well the Q-model's predicted values align with the observed counting outcomes.

Usage

## S3 method for class 'gccount'
plot(x, n.groups=5, smooth=FALSE, ...)

Arguments

Details

The function visualizes the calibration of the Q-model by dividing predicted probabilities into n.groups quantiles. For each group, the average predicted probability is plotted against the observed proportion of events, including 95% confidence intervals for the observed values. An identity line is provided for reference and perfect calibration is indicated when the observed points are directly along this line.

Value

No return value for this S3 method.

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * (AGE + SEX + ALCOHOL + BMI + DIABETES))

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_count(formula=.f, model="all", data=dataPROPHYVAP, group="GROUP", 
                cv=10, boot.type="boot", boot.number=10, boot.tune=TRUE, 
                effect="ATE", progress=TRUE, seed=5192)
                     
### Plot the calibration curve
plot(gc1, n.groups=5, col="red3")

S3 Method for Plotting a 'gctimes' Object

Description

Provides calibration and survival plots for an object returned by the function gc_times. This method assesses how well the Q-model's predicted survival aligns with observed time-to-event outcomes.

Usage

## S3 method for class 'gctimes'
plot(x, method="calibration", n.groups=5, pro.time=NULL, smooth=FALSE, ...)

Arguments

Details

Methods:

• "calibration": Visualizes calibration at pro.time by dividing predicted probabilities into n.groups quantiles. For each group, the average predicted probability is plotted against the observed proportion of events at pro.time, including 95% confidence intervals. An identity line is provided for reference, and perfect calibration is indicated when the observed points lie directly along this line.

• "calibration2": Plots the Kaplan-Meier estimations of the overall survival against the mean of the survival predictions derived from the Q-model. Note that this method cannot be used when boot.mi=TRUE.

• "survival": Plots the predicted survival curves derived from the Q-model for each treatment/exposure group.

Survival predictions are computed using the linear predictor from the fitted model and the baseline cumulative hazard estimated via Breslow's method. Bootstrap iterations are used in gc_times to compute adjusted and unadjusted survival curves, RMST, and differences between groups.

Value

No return value for this S3 method.

Examples

data("dataPROPHYVAP")

.f <- formula(Surv(TIME_DEATH, DEATH) ~ GROUP * (AGE + SEX + ALCOHOL + BMI + DIABETES))

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_times(formula=.f, model="lasso", data=dataPROPHYVAP, group="GROUP",
                cv=10, boot.type="boot", boot.number=10, boot.tune=TRUE,
                effect="ATE", progress=TRUE, pro.time=20, seed=5192)

### Calibration plot at 20 days
plot(gc1, method="calibration", n.groups=5, pro.time=10)

### Kaplan-Meier vs mean predicted survival
plot(gc1, method="calibration2")

### Predicted survival curves for treatment groups
plot(gc1, method="survival", col=c("red3","blue3"))
legend("bottomleft", c("Placebo", "Ceftriaxone"), col=c("red3","blue3"), lty=1)

S3 Method for Printing an 'gcbinary' Object

Description

Print a summary of an object returned from gc_binary function.

Usage

## S3 method for class 'gcbinary'
print(x, digits=4, ...)

Arguments

Value

No return value for this S3 method.

Examples

data(dataPROPHYVAP)

.f <- formula(VAP ~ GROUP * AGE + SEX + BMI + DIABETES)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
gc1 <- gc_binary(formula=.f, model="ridge", data=dataPROPHYVAP,
              group="GROUP", param.tune=NULL, boot.type="bcv", cv=10,
              boot.number=10, effect="ATE", progress=TRUE, boot.tune=TRUE)

print(gc1)

S3 Method for Printing an 'gccontinuous' Object

Description

Print a summary of and object returned from gc_continuous function.

Usage

## S3 method for class 'gccontinuous'
print(x, digits=4, ...)

Arguments

Value

No return value for this S3 method.

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * (AGE + SEX + ALCOHOL + BMI + DIABETES))

### In practice use larger values of boot.number (e.g., 500)
### We set boot.number and cv at 10 for speed in CRAN checks
gc1 <- gc_continuous(formula=.f, model="all", data=dataPROPHYVAP, group="GROUP",
                    cv=10, boot.type="boot", boot.number=10, boot.tune=TRUE,
                    effect="ATE", progress=TRUE, seed=5192)

print(gc1)

S3 Method for Printing an 'gccontinuous' Object

Description

Print a summary of and object returned from gc_count function.

Usage

## S3 method for class 'gccount'
print(x, digits=4, ...)

Arguments

Value

No return value for this S3 method.

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * (AGE + SEX + ALCOHOL + BMI + DIABETES))

### In practice use larger values of boot.number (e.g., 500)
### We set boot.number and cv at 10 for speed in CRAN checks
gc1 <- gc_count(formula=.f, model="all", data=dataPROPHYVAP, group="GROUP",
                    cv=10, boot.type="boot", boot.number=10, boot.tune=TRUE,
                    effect="ATE", progress=TRUE, seed=5192)

print(gc1)

S3 Method for Printing an 'gctimes' Object

Description

Print a summary of the gctimes object returned from gc_times function

Usage

## S3 method for class 'gctimes'
print(x,  digits=4, ...)

Arguments

Value

No return value for this S3 method.

Examples

data(dataPROPHYVAP)

.f <- formula(Surv(TIME_DEATH, DEATH) ~ GROUP * AGE + SEX +
              BMI + DIABETES)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
gc1 <- gc_times(formula=.f, model="lasso", data=dataPROPHYVAP,
group="GROUP", param.tune=NULL, boot.type="bcv", cv=10, boot.number=10,
              effect="ATE", progress=TRUE , pro.time=10, boot.tune=FALSE)

print(gc1)

S3 Method for Summarizing an 'gcbinary' Object

Description

Summarize an object returned by the function gc_binary.

Usage

## S3 method for class 'gcbinary'
summary(object, digits=4, ci.type=NULL, ci.level=0.95,
                         unadjusted=TRUE, ...)

Arguments

Value

Examples

data(dataPROPHYVAP)

.f <- formula(VAP ~ GROUP * AGE + SEX + BMI + DIABETES)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_binary(formula=.f, model="ridge", data=dataPROPHYVAP,
              group="GROUP", param.tune=NULL, boot.type="bcv", cv=10,
              boot.number=10, effect="ATE", progress=TRUE, boot.tune=TRUE)

summary(gc1, ci.type="norm", ci.level=0.95)

S3 Method for Summarizing an 'gccontinuous' Object

Description

Summarize an object returned by the function gc_continuous.

Usage

## S3 method for class 'gccontinuous'
summary(object, digits=4, ci.type=NULL, ci.level=0.95,
                         unadjusted=TRUE, ...)

Arguments

Value

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * (AGE + SEX + ALCOHOL + BMI + DIABETES))

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_continuous(formula=.f, model="all", data=dataPROPHYVAP, group="GROUP",
                    cv=10, boot.type="boot", boot.number=10, boot.tune=TRUE,
                    effect="ATE")

summary(gc1, ci.type="norm", ci.level=0.99)

S3 Method for Summarizing an 'gccount' Object

Description

Summarize an object returned by the function gc_count.

Usage

## S3 method for class 'gccount'
summary(object, digits=4, ci.type=NULL, ci.level=0.95,
                         unadjusted=TRUE, ...)

Arguments

Value

Examples

data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * (AGE + SEX + ALCOHOL + BMI + DIABETES))

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_count(formula=.f, model="all", data=dataPROPHYVAP, group="GROUP",
                    cv=10, boot.type="boot", boot.number=10, boot.tune=TRUE,
                    effect="ATE")

summary(gc1, ci.type="norm", ci.level=0.99)

S3 Method for Summarizing an 'gctimes' Object

Description

Summarize an object returned by the function gc_times.

Usage

## S3 method for class 'gctimes'
summary(object, digits=4, ci.type=NULL, ci.level=0.95,
                         unadjusted=TRUE, ...)

Arguments

Value

Examples

data(dataPROPHYVAP)

formula <- formula(Surv(TIME_DEATH, DEATH) ~ GROUP * AGE +
              SEX + BMI + DIABETES)

### In practice use larger values for cv and boot.number
### We set boot.number and cv at 10 for speed in CRAN checks
### cv=30 and boot.number=1000 are more appropriate values
gc1 <- gc_times(formula=formula, model="lasso", data=dataPROPHYVAP,
              group="GROUP", param.tune=NULL, boot.type="bcv", cv=10,
              boot.number=10, effect="ATE", pro.time=20, boot.tune=TRUE)
              
summary(gc1, ci.type="norm", ci.level=0.99)

Transport the Marginal Effect for Another Populations

Description

Applies an already fitted Q-model to estimate marginal effects for a new population (i.e., with differentially distributed baseline characteristics used as predictors in the Q-Model).

Usage

transport(object, newdata, n.sim=500, seed=NULL)

Arguments

Details

The function employs two distinct estimation strategies:

• Monte Carlo Simulation: For unpenalized models (likelihood-based models like "all", "aic", or "bic"), the function uses the variance-covariance matrix of the Q-model regression coefficients to simulate Q-models assuming a multivariate normal distribution. This provides a distribution of adjusted results and the related confidence intervals.

• Single Point Estimate: For penalized models ("lasso", "ridge", "elasticnet") and survival models (gctimes), the function does not estimate the distribution of adjusted results and the related confidence intervals.

Value

The function returns an object of the same class with the same arguments as the input object, but with updated results for the newdata.

References

Pearl J, Bareinboim E. External validity: From do-calculus to transportability across populations. Statistical Science. 2012;29(4):579-595.

Examples

### For a binary outcome
data("dataPROPHYVAP")

.f <- formula(VAP ~ GROUP * AGE + SEX + BMI + DIABETES)

### In practice use larger values of boot.number (e.g., 500)
### We set boot.number and cv at 10 for speed in CRAN checks
gc1 <- gc_binary(formula=.f, model="all", data=dataPROPHYVAP,
              group="GROUP", param.tune=NULL, boot.type="bcv", cv=10,
              boot.number=10, effect="ATE", progress=TRUE, boot.tune=FALSE)

summary(gc1, ci.type="norm", ci.level=0.95)

### New dataset (for example a subset of younger patients)
newdata <- subset(dataPROPHYVAP, AGE <= 50)

### transport the gc_binary model to the new dataset
gc2 <- transport(object=gc1, newdata=newdata, n.sim=10)

summary(gc2, ci.type="perc", ci.level=0.95)