Creates propensity scores and inverse probability of treatment weights (IPTW) using a regularized logistic regression.
createPs( cohortMethodData, population = NULL, excludeCovariateIds = c(), includeCovariateIds = c(), maxCohortSizeForFitting = 250000, errorOnHighCorrelation = TRUE, stopOnError = TRUE, prior = createPrior("laplace", exclude = c(0), useCrossValidation = TRUE), control = createControl(noiseLevel = "silent", cvType = "auto", seed = 1, resetCoefficients = TRUE, tolerance = 2e-07, cvRepetitions = 10, startingVariance = 0.01), estimator = "att" )
A data frame describing the population. This should at least have a
rowId column corresponding to the
rowId column in the
CohortMethodData covariates object and a
If population is not specified, the full population in the
CohortMethodData will be used.
Exclude these covariates from the propensity model.
Include only these covariates in the propensity model.
If the target or comparator cohort are larger than this number, they will be downsampled before fitting the propensity model. The model will be used to compute propensity scores for all subjects. The purpose of the sampling is to gain speed. Setting this number to 0 means no downsampling will be applied.
If true, the function will test each covariate for correlation with the treatment assignment. If any covariate has an unusually high correlation (either positive or negative), this will throw and error.
If an error occur, should the function stop? Else, the two cohorts will be assumed to be perfectly separable.
The prior used to fit the model. See
Cyclops::createPrior() for details.
The control object used to control the cross-validation used to
determine the hyperparameters of the prior (if applicable). See
Cyclops::createControl() for details.
The type of estimator for the IPTW. Options are
estimator = "ate"
for the average treatment effect,
estimator = "att" for the
average treatment effect in the treated, and
estimator = "ato"
for the average treatment effect in the overlap population.
IPTW estimates either the average treatment effect (ate) or average treatment effect in the treated (att) using stabilized inverse propensity scores (Xu et al. 2010).
Xu S, Ross C, Raebel MA, Shetterly S, Blanchette C, Smith D. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health. 2010;13(2):273-277. doi:10.1111/j.1524-4733.2009.00671.x
data(cohortMethodDataSimulationProfile) cohortMethodData <- simulateCohortMethodData(cohortMethodDataSimulationProfile, n = 1000) #> Generating covariates #> Generating treatment variable #> Generating cohorts #> Generating outcomes after index date #> Generating outcomes before index date ps <- createPs(cohortMethodData) #> Removing 0 redundant covariates #> Removing 0 infrequent covariates #> Normalizing covariates #> Tidying covariates took 1.25 secs #> Warning: All coefficients (except maybe the intercept) are zero. Either the covariates are completely uninformative or completely predictive of the treatment. Did you remember to exclude the treatment variables from the covariates? #> Propensity model fitting finished with status OK #> Creating propensity scores took 4.71 secs