getCalibrationSummary: Get a sparse summary of the calibration

In OHDSI/PatientLevelPrediction: Develop Clinical Prediction Models Using the Common Data Model

Get a sparse summary of the calibration

Description

Get a sparse summary of the calibration

Usage

getCalibrationSummary(
  prediction,
  predictionType,
  typeColumn = "evaluation",
  numberOfStrata = 10,
  truncateFraction = 0.05
)

Arguments

prediction	A prediction object as generated using the predict functions.
predictionType	The type of prediction (binary or survival)
typeColumn	A column that is used to stratify the results
numberOfStrata	The number of strata in the plot.
truncateFraction	This fraction of probability values will be ignored when plotting, to avoid the x-axis scale being dominated by a few outliers.

Details

Generates a sparse summary showing the predicted probabilities and the observed fractions. Predictions are stratified into equally sized bins of predicted probabilities.

Value

A dataframe with the calibration summary

Examples

# simulate data
data("simulationProfile")
plpData <- simulatePlpData(simulationProfile, n=500)
# create study population, split into train/test and preprocess with default settings
population <- createStudyPopulation(plpData, outcomeId = 3)
data <- splitData(plpData, population, createDefaultSplitSetting())
data$Train$covariateData <- preprocessData(data$Train$covariateData)
saveLoc <- file.path(tempdir(), "calibrationSummary")
# fit a lasso logistic regression model using the training data
plpModel <- fitPlp(data$Train, modelSettings=setLassoLogisticRegression(seed=42),
                   analysisId=1, analysisPath=saveLoc)
calibrationSummary <- getCalibrationSummary(plpModel$prediction, 
                                            "binary", 
                                            numberOfStrata = 10,
                                            typeColumn = "evaluationType")
calibrationSummary
# clean up
unlink(saveLoc, recursive = TRUE)

OHDSI/PatientLevelPrediction documentation built on Feb. 14, 2025, 9:44 a.m.