PatientLevelPrediction (version 6.7.0)

simulatePlpBenchmarkData: Generate benchmark PLP data with known outcome risk

Description

simulatePlpBenchmarkData creates semi-synthetic PLP data by resampling rows from an existing plpData object and generating outcomes from a known logistic outcome model. The source covariate patterns are preserved, so this function is intended for benchmark simulations rather than realistic synthetic patient generation. Rows without observable time at riskWindowStart are excluded before resampling, and generated event times are sampled within each row's observed risk window. When daysToCohortEnd is present and contains positive follow-up, event times are also kept within the target cohort end.

The attached truth is defined for each simulated row's observable risk window. For benchmark evaluation against this truth, use population settings with requireTimeAtRisk = FALSE; otherwise PLP can drop short-follow-up non-outcome rows and change the estimand. For a strict fixed-horizon benchmark, first restrict the source plpData to rows with complete follow-up through riskWindowEnd.

Usage

simulatePlpBenchmarkData(
  plpData,
  outcomeModel,
  n = nrow(plpData$cohorts),
  riskWindowStart = 1,
  riskWindowEnd = 365,
  outcomeId = NULL,
  targetOutcomeRate = NULL,
  seed = NULL,
  returnTruth = TRUE
)

Value

An object of type plpData.

Arguments

plpData

An object of type plpData.

outcomeModel

A named numeric vector of logistic model coefficients. Use "(Intercept)" for the intercept and covariate IDs for covariate coefficients.

n

The number of rows to generate.

riskWindowStart

The earliest generated daysToEvent for outcome cases.

riskWindowEnd

The latest generated daysToEvent for outcome cases.

outcomeId

The outcome ID to use. If NULL, the first outcome ID in plpData$metaData$databaseDetails$outcomeIds is used.

targetOutcomeRate

Optional target mean true risk. If provided, the intercept is shifted so the generated population has this mean risk.

seed

An optional seed for the random number generator.

returnTruth

If TRUE, a truth table with source row IDs, linear predictors, true risks, and generated outcome status is attached as the simulationTruth attribute.

Examples

data("simulationProfile")
plpData <- simulatePlpData(simulationProfile, n = 100, seed = 42)
benchmarkData <- simulatePlpBenchmarkData(
  plpData = plpData,
  outcomeModel = c("(Intercept)" = -2, "1002" = 0.04),
  n = 100,
  riskWindowStart = 1,
  riskWindowEnd = 365,
  seed = 42
)
attr(benchmarkData, "simulationTruth")