Generate a synthetic cohort resembling a urodynamics study — cp_synthetic

A cohort with known truth, for fixtures, examples and reports. The public repository must never carry patient data, so every demonstration in this package runs on this.

Usage

cp_synthetic_cohort(
  n = 500,
  prevalence = 0.11,
  years = 2019:2024,
  repeat_frac = 0.08,
  seed = 1978L
)

Arguments

n: Rows to generate.
prevalence: Target outcome prevalence. The default, with the default n, gives roughly the 57 events the specification describes for the reference study, which is the regime where degrees of freedom have to be spent carefully.
years: Calendar years to spread the cohort over, for temporal validation.
repeat_frac: Fraction of rows that are a second visit by a patient already present, so clustered resampling has something to cluster.
seed: Single whole number.

Value

A data frame of n rows with attribute truth, a list holding the true coefficients and the generating formula.

Details

The generating model is deliberately not a plain additive one. It carries a nonlinear age effect and one interaction, because a pipeline that is only ever exercised on additive data will not reveal whether its spline and hierarchy handling are right.

The true coefficients are attached as the truth attribute, and the intercept is solved numerically so the realised prevalence lands near prevalence rather than wherever the other terms happen to put it.

Examples

cohort <- cp_synthetic_cohort(n = 200, seed = 1)
table(cohort$event)
#> 
#>   0   1 
#> 181  19 
attr(cohort, "truth")$coefficients
#>     intercept        age_sq           bmi      sex_male prior_uti_yes 
#>     -2.867895      0.001100      0.045000      0.350000      0.800000 
#>   uti_by_male 
#>      0.600000