Compare population covariate distributions — compare

Covariate distributions for studies in the network may be compared by overlap effective sample size (via propensity scores) or by (standardised) Euclidean distance between the covariate means. The overlap effective sample size accounts for the full multivariate joint distribution of the covariates, whereas the Euclidean distance between the covariate means only accounts for the location and not the range of variation or joint structure.

Usage

compare_populations(
  network,
  covariates = NULL,
  method = c("propensity", "euclidean")
)

# S3 method for class 'pop_comp'
print(
  x,
  order = c("decreasing", "increasing"),
  simplify = TRUE,
  ...,
  digits = 2L,
  n = 10
)

# S3 method for class 'pop_comp'
plot(x, ...)

Arguments

network: An nma_data network object.
covariates: Character vector of covariate names to compare on.
method: Method to compare distributions, either "propensity" to calculate overlap effective sample size with propensity scores, or "euclidean" to calculate standardised Euclidean distance between means.
x: A pop_comp object produced by compare_populations()
order: String, the order in which to display the comparison summaries. Either "decreasing", to list most similar studies first (the default), or "increasing" to list most dissimilar studies first.
simplify: Logical, should the output be simplified to only show comparisons between subnetworks (TRUE, default), or between every study in the network (FALSE)? If a connected network was provided, then simplify is always FALSE.
...: Additional arguments (unused)
digits: Number of digits to print in the summary, default 2
n: Number of rows to show in the table of study summaries

Value

A pop_comp object, containing a summary data frame and comparison_matrix matrix of pairwise comparisons. When method = "propensity", a list propensity_scores of data frames of fitted propensity scores will also be included.

Aggregate data setup

When aggregate data are present in the network, some setup is necessary to use this function. For method = "propensity", the network must have integration points present; specify these using add_integration(). For method = "euclidean", the network must either have integration points present, or the mean and standard deviation of covariates can be provided in the original input data when setting up the network. In the latter case, the covariate means are assumed to be provided in the named <covariate> column, and the corresponding standard deviation in a column named <covariate>_sd. For example, if the covariate is age, mean age should be provided in the age column, and the standard devation of age in age_sd column.