Skip to contents

One-step continuous-time Targeted Minimum Loss-Based Estimator (TMLE) for outcome-specific absolute risk estimands in right-censored survival settings with or without competing risks

concrete-package.Rd

Implements the methodology described in Rytgaard et al. (2023) <doi:10.1111/biom.13856> and Rytgaard and van der Laan (2023) <doi:10.1007/s10985-022-09576-2>. Currently can be used to estimate the effects of static or dynamic interventions on binary treatments given at baseline. Cross-validated treatment estimation is done using the 'SuperLearner' package, and conditional hazards are estimated from event-specific candidate libraries using Cox regressions from 'survival', Coxnet from 'glmnet', random survival forests from 'randomForestSRC', additive hazards from 'survival', or HAL from 'hal9001'. Hazard libraries currently use cross-validated discrete selection. The TMLE update includes configurable stopping rules and adaptive update controls for numerically difficult rare-event settings.

Details

formatArguments() many check...(), format...() functions getInitialEstimates() getPropScores() getHazEstimates() getEIC() getIC() doTMLEUpdate() getOutput()

Author

David McCoy, <david_mccoy@berkeley.edu> Maintainer: David McCoy <david_mccoy@berkeley.edu>

References

Rytgaard et al. (2023) <doi:10.1111/biom.13856> Rytgaard and van der Laan (2023) <doi:10.1007/s10985-022-09576-2>

See also

SuperLearner coxph glmnet rfsrc fit_hal

Examples

library(concrete)
library(data.table)
#> 
#> Attaching package: ‘data.table’
#> The following object is masked from ‘package:base’:
#> 
#>     %notin%
set.seed(12345)
data <- as.data.table(survival::pbc)
data <- data[!is.na(trt), ][, trt := trt - 1]
data <- data[, c("time", "status", "trt", "age", "sex", "albumin")]
# \donttest{
ConcreteArgs <- formatArguments(DataTable = data,
                                EventTime = "time",
                                EventType = "status",
                                Treatment = "trt",
                                Intervention = 0:1,
                                TargetTime = 1500,
                                TargetEvent = 1:2,
                                MaxUpdateIter = 250,
                                CVArg = list(V = 10),
                                Verbose = FALSE)
ConcreteEst <- doConcrete(ConcreteArgs)
#> 
#> Estimating Treatment Propensity:
#> Loading required package: nnls
#> Loading required namespace: glmnet
#> Loading required namespace: xgboost
#> 
#> Estimating Hazards:
#> 
#> Starting TMLE Update:
ConcreteOut <- getOutput(ConcreteEst)
# }

## Joint Intervention
data <- data[, trt2 := sample(0:1, .N, replace = TRUE, prob = c(0.3, .7))]
Intervention <- makeITT("A1" = data.frame(trt = rep_len(1, nrow(data)), 
                                          trt2 = rep_len(1, nrow(data))), 
                        "A0" = data.frame(trt = rep_len(0, nrow(data)), 
                                          trt2 = rep_len(0, nrow(data))))
# \donttest{
ConcreteArgs <- formatArguments(DataTable = data,
                                EventTime = "time",
                                EventType = "status",
                                Treatment = c("trt", "trt2"),
                                Intervention = Intervention,
                                TargetTime = 2000,
                                TargetEvent = 1:2,
                                MaxUpdateIter = 250,
                                CVArg = list(V = 10),
                                Verbose = FALSE)
ConcreteEst <- doConcrete(ConcreteArgs)
#> 
#> Estimating Treatment Propensity:
#> 
#> Estimating Hazards:
#> 
#> Starting TMLE Update:
#> Warning: TMLE has not converged by step 250 - Estimates may not have the desired asymptotic properties
ConcreteOut <- getOutput(ConcreteEst)
# }