resilience (version 2025.1.1)

prepost: Identifying Predictors of Resilience to Stressors in Single-Arm Studies of Pre-Post Change

Description

Studies of resilience in older adults are typically conducted with a single-arm where everyone experiences the stressor. The simplistic approach of regressing change versus baseline yields biased estimates due to mathematical coupling and regression-to-the-mean. This function provides a method to correct the bias.

Usage

prepost(
  formula,
  data,
  change = TRUE,
  k = c(1, 1.5, 2),
  m = 1,
  nboot = 1000,
  ci.level = 0.95,
  boot.method = c("perc", "norm", "basic", "bca"),
  ncores = 2
)

Value

A list with the following components:

naive.beta

Unadjusted, naive estimates of regression coefficients

corrected.beta

The corrected coefficients of the variables. A matrix with one column of parameter values for each value of sensitivity parameter k

CI

A list of length equal to the number of sensitivity values. Each component of the list is a matrix with two columns of lower and upper confidence interval for each parameter

Arguments

formula

Formula object where LHS is Post response variable and RHS is pre-response variable plus the covariates. Note: the first variable of the RHS of the formula must be the pre-response variable. For example, `y2 ~ y1 + x1 + x2`.

data

Data frame containing all the variables. Only complete cases are used in the analysis, i.e. rows of dataframe with missing values in any of the predictors are automatically deleted.

change

A logical variable. If `TRUE` (default) the dependent variable of regression is the pre-post change. If `FALSE`, the post response is used as the dependent variable.

k

A sensitivity analysis parameter. Typically, it is greater than or equal to 1.0. It is recommended that the user provide at least two values to examine how the effects vary with `k`. Default setting allows three values: k = 1.0, 1.5, and 2.0. For more details about this parameter refer to the manuscript.

m

Another sensitivity analysis parameter. It is set equal to 1.0. It is recommended that the user not change this unless there is information from an external study to justify a different value.

nboot

Number of bootstrap samples for calculating the confidence intervals of corrected regression coefficients. Default is 1000.

ci.level

Confidence coefficient for confidence interval. Default is 0.95 (95% confidence intervals).

boot.method

The bootstrap method for confidence interval. Four options are provided: "perc" (percentile), "norm" (normal approximation), "basic", and "bca" (bias-corrected accelerated bootstrap). Default is "perc".

ncores

Number of cores available for parallel computing. Default is set to 2 due to CRAN requirements. If more cores are available, the user can utilize all available cores with the command: `ncores = parallel::detectCores()`

Missing Data

This function uses complete-case analysis only. If your data contains missing values, the function will issue a warning and exclude cases with any missing values. For more robust inference with missing data, consider using multiple imputation followed by prepost_resilience_mi_list(), which pools results across multiply imputed datasets using Rubin's rules.

Parallel Processing

The function uses the parallel and foreach packages to perform parallel computations of bootstrap confidence intervals for different values of the sensitivity parameter, k.

Author

Ravi Varadhan

References

Varadhan, R., Zhu, J., and Bandeen-Roche, K (2024). Identifying Predictors of Resilience to Stressors in Single-Arm Studies of Pre-Post Change. Biostatistics. 25(4): 1094-1111.

See Also

prepost_mi for analysis with multiply imputed data.

Examples

if (FALSE) {
data(tkr)

names(tkr.dat)
dim(tkr.dat)

# pre-post change regression
ans1 <- prepost(post.Y ~ pre.Y + I(age-mean(age)) + I((age - mean(age))^2) + 
  bmi + gender + as.factor(smoker), data=tkr.dat, k=c(1.2, 1.5), nboot=200)
print(ans1)

# Post regression
ans2 <- prepost(post.Y ~ pre.Y + I(age-mean(age)) + I((age - mean(age))^2) + 
  bmi + gender + as.factor(smoker), data=tkr.dat, 
  k=c(1.2, 1.5), change=FALSE, nboot=200, boot.method="norm")
print(ans2)

# without any covariates
ans3 <- prepost(post.Y ~ pre.Y, data=tkr.dat, k=c(1.2, 2.0), nboot=200)
print(ans3)

# Bootstrapping using "bca" - relatively slow
# Not run
# ans4 <- prepost(post.Y ~ pre.Y, data=tkr.dat, k=c(1.2, 2.0), change=FALSE,
#   boot.method = "bca")
}