iqLearn: Interactive Q-Learning

In DynTxRegime: Methods for Estimating Optimal Dynamic Treatment Regimes

Interactive Q-Learning

Description

The complete interactive Q-Learning algorithm.

Usage

## Second-Stage Analysis
iqLearnSS(..., moMain, moCont, data, response, txName, iter = 0L, 
          verbose = TRUE)

## First-Stage Analysis for Fitted Main Effects
iqLearnFSM(..., moMain, moCont, data, response, txName, iter = 0L, 
           verbose = TRUE)

## First-Stage Analysis for Fitted Contrasts
iqLearnFSC(..., moMain, moCont, data, response, txName, iter = 0L, 
           verbose = TRUE)

## First-Stage Analysis of Contrast Variance Log-Linear Model
iqLearnFSV(..., object, moMain, moCont, data, iter = 0L, verbose = TRUE)

Arguments

...	ignored. Provided to require named inputs.
moMain	An object of class modelObj or a list of objects of class modelObjSubset, which define the models and R methods to be used to obtain parameter estimates and predictions for the main effects component of the outcome regression. See ?modelObj and/or ?modelObjSubset for details. NULL is an acceptable value if moCont is defined.
moCont	An object of class modelObj or a list of objects of class modelObjSubset, which define the models and R methods to be used to obtain parameter estimates and predictions for the contrasts component of the outcome regression. See ?modelObj and/or ?modelObjSubset for details. NULL is an acceptable value if moMain is defined.
data	A data frame of covariates and treatment history.
response	For the second stage analysis, the response vector. For first stage analyses, the value object returned by iqLearnSS().
object	The value object returned by iqLearFSC()
txName	A character string giving column header of treatment variable in data
iter	An integer. See ?iter for details
verbose	A logical. If TRUE, screen prints are generated.

References

Laber, EB, Linn, KA, and Stefanski, LA (2014). Interactive model building for Q-Learning. Biometrika, 101, 831–847. PMCID: PMC4274394.

See Also

Other statistical methods: bowl(), earl(), optimalClass(), optimalSeq(), owl(), qLearn(), rwl()

Other multiple decision point methods: bowl(), optimalClass(), optimalSeq(), qLearn()

Examples

# Load and process data set
data(bmiData)

# define the negative 12 month change in BMI from baseline
y12 <- -100*(bmiData[,6L] - bmiData[,4L])/bmiData[,4L]

#### Full Interactive Q-Learning Algorithm

### Second-Stage Analysis

# outcome model
moMain <- buildModelObj(model = ~parentBMI+month4BMI,
                        solver.method = 'lm')

moCont <- buildModelObj(model = ~race + parentBMI+month4BMI,
                        solver.method = 'lm')

fitSS <- iqLearnSS(moMain = moMain, moCont = moCont,
                   data = bmiData, response = y12,  txName = 'A2')

### First-Stage Analysis Main Effects Term

# main effects model
moMain <- buildModelObj(model = ~parentBMI+baselineBMI,
                        solver.method = 'lm')

moCont <- buildModelObj(model = ~race + parentBMI+baselineBMI,
                        solver.method = 'lm')

fitFSM <- iqLearnFSM(moMain = moMain, moCont = moCont,
                     data = bmiData, response = fitSS,  txName = 'A1')

### First-Stage Analysis Contrasts Term

# contrasts model
moMain <- buildModelObj(model = ~parentBMI+baselineBMI,
                        solver.method = 'lm')

moCont <- buildModelObj(model = ~race + parentBMI+baselineBMI,
                        solver.method = 'lm')

fitFSC <- iqLearnFSC(moMain = moMain, moCont = moCont,
                     data = bmiData, response = fitSS,  txName = 'A1')

### First-Stage Analysis Contrasts Variance - Log-linear

# contrasts variance model
moMain <- buildModelObj(model = ~baselineBMI,
                        solver.method = 'lm')

moCont <- buildModelObj(model = ~baselineBMI,
                        solver.method = 'lm')

fitFSV <- iqLearnFSV(object = fitFSC, moMain = moMain, moCont = moCont,
                     data = bmiData)

####Available methods

  ### Estimated value
  estimator(x = fitFSC, y = fitFSM, z = fitFSV, w = fitSS, dens = 'nonpar')

  ## Estimated optimal treatment and decision functions for training data
  ## Second stage optimal treatments
  optTx(x = fitSS)

  ## First stage optimal treatments when contrast variance is modeled.
  optTx(x = fitFSM, y = fitFSC, z = fitFSV, dens = 'nonpar')

  ## First stage optimal treatments when contrast variance is constant.
  optTx(x = fitFSM, y = fitFSC, dens = 'nonpar')

  ## Estimated optimal treatment and decision functions for new data
  ## Second stage optimal treatments
  optTx(x = fitSS, bmiData)

  ## First stage optimal treatments when contrast variance is modeled.
  optTx(x = fitFSM, y = fitFSC, z = fitFSV, dens = 'nonpar', bmiData)

  ## First stage optimal treatments when contrast variance is constant.
  optTx(x = fitFSM, y = fitFSC, dens = 'nonpar', bmiData)

### The following methods are available for all objects: fitSS, fitFSM,
### fitFSC and fitFSV. We include only one here for illustration.

  # Coefficients of the outcome regression objects
  coef(object = fitSS)

  # Description of method used to obtain object
  DTRstep(object = fitFSM)

  # Value object returned by outcome regression method
  fitObject(object = fitFSC)

  # Value object returned by outcome regression method
  outcome(object = fitFSV)

  # Plots if defined by outcome regression method
  dev.new()
  par(mfrow = c(2,4))

  plot(x = fitSS)
  plot(x = fitSS, suppress = TRUE)

  # Show main results of method
  show(object = fitFSM)

  # Show summary results of method
  summary(object = fitFSV)

DynTxRegime documentation built on Nov. 25, 2023, 1:09 a.m.