Determine intercept, treatment/exposure and covariate coefficients that can be used for binary data generation with a logit link and a set of covariates

Source: R/utility.R

This is an implementation of an iterative bisection procedure that can be used to determine coefficient values for a target population prevalence as well as a target risk ratio, risk difference, or AUC. These coefficients can be used in a subsequent data generation process to simulate data with these desire characteristics.

logisticCoefs(
  defCovar,
  coefs,
  popPrev,
  rr = NULL,
  rd = NULL,
  auc = NULL,
  tolerance = 0.001,
  sampleSize = 1e+05,
  trtName = "A"
)

Arguments

defCovar

A definition table for the covariates in the underlying population. This tables specifies the distribution of the covariates.

coefs

A vector of coefficients that reflect the relationship between each of the covariates and the log-odds of the outcome.

popPrev

The target population prevalence of the outcome. A value between 0 and 1.

rr

The target risk ratio, which must be a value between 0 and 1/popPrev. Defaults to NULL.

rd

The target risk difference, which must be between -(popPrev) and (1 - popPrev). Defaults to NULL

auc

The target AUC, which must be a value between 0.5 and 1.0 . Defaults to NULL.

tolerance

The minimum stopping distance between the adjusted low and high endpoints. Defaults to 0.001.

sampleSize

The number of units to generate for the bisection algorithm. The default is 1e+05. To get a reliable estimate, the value should be no smaller than the default, though larger values can be used, though computing time will increase.

trtName

If either a risk ratio or risk difference is the target statistic, a treatment/exposure variable name can be provided. Defaults to "A".

Value

A vector of parameters including the intercept and covariate coefficients for the logistic model data generating process.

Details

If no specific target statistic is specified, then only the intercept is returned along with the original coefficients. Only one target statistic (risk ratio, risk difference or AUC) can be specified with a single function call; in all three cases, a target prevalence is still required.

References

Austin, Peter C. "The iterative bisection procedure: a useful tool for determining parameter values in data-generating processes in Monte Carlo simulations." BMC Medical Research Methodology 23, no. 1 (2023): 1-10.

Examples

if (FALSE) { # \dontrun{
d1 <- defData(varname = "x1", formula = 0, variance = 1)
d1 <- defData(d1, varname = "b1", formula = 0.5, dist = "binary")

coefs <- log(c(1.2, 0.8))

logisticCoefs(d1, coefs, popPrev = 0.20) 
logisticCoefs(d1, coefs, popPrev = 0.20, rr = 1.50, trtName = "rx") 
logisticCoefs(d1, coefs, popPrev = 0.20, rd = 0.30, trtName = "rx")
logisticCoefs(d1, coefs, popPrev = 0.20, auc = 0.80)
} # }