estimate_ess: Estimate effective sample size

In Roche/MAIC: Package to Perform Matched-adjusted Indirect Comparisons

Estimate effective sample size

Description

Estimate the effective sample size (ESS).

Usage

estimate_ess(data, wt_col = "wt")

Arguments

data	A data frame containing individual patient data from the intervention study, including a column containing the weights (derived using estimate_weights).
wt_col	The name of the weights column in the data frame containing the intervention individual patient data and the MAIC propensity weights. The default is wt.

Details

For a weighted estimate, the effective sample size (ESS) is the number of independent non-weighted individuals that would be required to give an estimate with the same precision as the weighted sample estimate. A small ESS, relative to the original sample size, is an indication that the weights are highly variable and that the estimate may be unstable. This often occurs if there is very limited overlap in the distribution of the matching variables between the populations being compared. If there is insufficient overlap between populations it may not be possible to obtain reliable estimates of the weights

Value

The effective sample size (ESS) as a numeric value.

References

NICE DSU TECHNICAL SUPPORT DOCUMENT 18: METHODS FOR POPULATION-ADJUSTED INDIRECT COMPARISONS IN SUBMSISSIONS TO NICE, REPORT BY THE DECISION SUPPORT UNIT, December 2016

See Also

estimate_weights

Examples

# This example code uses the weighted individual patient data, outputted from
# the estimate_weights function to perform weight diagnostics. The weighted data
# is saved within est_weights. To check the weighted aggregate baseline
# characteristics for 'intervention' match those in the comparator data,
# standardized data "target_pop_standard" is used. Please see the package
# vignette for more information on how to use the estimate_weights function and
# derive the "target_pop_standard" data.

library(dplyr)
library(MAIC)

# load est_weights
data(est_weights, package = "MAIC")

# load target_pop_standard
data(target_pop_standard, package = "MAIC")

# List out the uncentered variables used in the matching
match_cov <- c("AGE",
               "SEX",
               "SMOKE",
               "ECOG0")

# Are the weights sensible? ----------------------------------------------------

# The wt_diagnostics function requires the output from the estimate_weights
# function and will output:
# - the effective sample size (ESS)
# - a summary of the weights and rescaled weights (mean, standard deviation,
#   median, minimum and maximum)
# - a unique set of weights with the corresponding patient profile based on the
#   matching variables

diagnostics <- wt_diagnostics(est_weights$analysis_data,
                              vars = match_cov)

diagnostics$ESS
diagnostics$Summary_of_weights
diagnostics$Weight_profiles

# Each of the wt_diagnostics outputs can also be estimated individually
ESS <- estimate_ess(est_weights$analysis_data)
weight_summ <- summarize_wts(est_weights$analysis_data)
wts_profile <- profile_wts(est_weights$analysis_data, vars = match_cov)

# Plot histograms of unscaled and rescaled weights
# bin_width needs to be adapted depending on the sample size in the data set
histogram <- hist_wts(est_weights$analysis_data, bin = 50)
histogram


# Has the optimization worked? -------------------------------------------------

# The following code produces a summary table of the intervention baseline
# characteristics before and after matching compared with the comparator
# baseline characteristics:

# Create an object to hold the output
baseline_summary <- list('Intervention' = NA,
                         'Intervention_weighted' = NA,
                         'Comparator' = NA)

# Summarise matching variables for weighted intervention data
baseline_summary$Intervention_weighted <- est_weights$analysis_data %>%
  transmute(AGE, SEX, SMOKE, ECOG0, wt) %>%
  summarise_at(match_cov, list(~ weighted.mean(., wt)))

# Summarise matching variables for unweighted intervention data
baseline_summary$Intervention <- est_weights$analysis_data %>%
  transmute(AGE, SEX, SMOKE, ECOG0, wt) %>%
  summarise_at(match_cov, list(~ mean(.)))

# baseline data for the comparator study
baseline_summary$Comparator <- transmute(target_pop_standard,
                                         AGE,
                                         SEX,
                                         SMOKE,
                                         ECOG0)

# Combine the three summaries
# Takes a list of data frames and binds these together
trt <- names(baseline_summary)
baseline_summary <-  bind_rows(baseline_summary) %>%
  transmute_all(sprintf, fmt = "%.2f") %>% #apply rounding for presentation
  transmute(ARM = as.character(trt), AGE, SEX, SMOKE, ECOG0)

# Insert N of intervention  as number of patients
baseline_summary$`N/ESS`[baseline_summary$ARM == "Intervention"] <- nrow(est_weights$analysis_data)

# Insert N for comparator from target_pop_standard
baseline_summary$`N/ESS`[baseline_summary$ARM == "Comparator"] <- target_pop_standard$N

# Insert the ESS as the sample size for the weighted data
# This is calculated above but can also be obtained using the estimate_ess function as shown below
baseline_summary$`N/ESS`[baseline_summary$ARM == "Intervention_weighted"] <- est_weights$analysis_data %>%
  estimate_ess(wt_col = 'wt')

baseline_summary <- baseline_summary %>%
  transmute(ARM, `N/ESS`=round(`N/ESS`,1), AGE, SEX, SMOKE, ECOG0)

Roche/MAIC documentation built on Nov. 9, 2024, 5:21 p.m.