estimate_weights.RdEstimate propensity weights for matching-adjusted indirect comparison (MAIC).
estimate_weights(intervention_data, matching_vars, method = "BFGS", ...)A data frame containing individual patient data from the intervention study.
A character vector giving the names of the covariates to use in matching. These names must match the column names in intervention_data.
The method used for optimisation - The default is method =
"BFGS". Refer to optim for options.
Additional arguments to be passed to optimisation functions such
as the method for maximum likelihood optimisation. Refer to optim
for options.
A list containing 2 objects. First, a data frame named analysis_data containing intervention_data with additional columns named wt (weights) and wt_rs (rescaled weights). Second, a vector called matching_vars of the names of the centered matching variables used.
The premise of MAIC methods is to adjust for between-trial differences in patient demographic or disease characteristics at baseline. When a common treatment comparator or ‘linked network’ are unavailable, a MAIC assumes that differences between absolute outcomes that would be observed in each trial are entirely explained by imbalances in prognostic variables and treatment effect modifiers.
The aim of the MAIC method is to estimate a set of propensity weights based on prognostic variables and treatment effect modifiers. These weights can be used in subsequent statistical analysis to adjust for differences in patient characteristics between the population in the intervention trial and the population in a comparator study. For additional details on the statistical methods, refer to the package vignette.
The data required for an unanchored MAIC are:
Individual patient data from a single arm study of 'intervention'
Aggregate summary data for 'comparator'. This could be from a single arm study of the comparator or from one arm of a randomized controlled trial.
Psuedo patient data from the comparator study. This is not required for the matching process but is needed to derive the relative treatment effects between the intervention and comparator.
For the matching process:
All binary variables to be used in the matching should be coded 1 and 0
The variable names need to be listed in a character vector called match_cov
Aggregate baseline characteristics (number of patients, mean and SD for continuous variables and proportion for binary variables) from the comparator trial are needed as a data frame. Naming of the covariates in this data frame should be consistent with variable names in the intervention data.
Patient baseline characteristics in the intervention study are centered on the value of the aggregate data from the comparator study
The estimate_weights function can then be used to estimate propensity weights for each patient in the intervention study
For full details refer to the example below and the package vignette
NICE DSU TECHNICAL SUPPORT DOCUMENT 18: METHODS FOR POPULATION-ADJUSTED INDIRECT COMPARISONS IN SUBMSISSIONS TO NICE, REPORT BY THE DECISION SUPPORT UNIT, December 2016
# This example code estimates weights for individual patient data from a single
# arm study of 'intervention' based on aggregate baseline characteristics from
# the comparator trial
library(dplyr)
library(MAIC)
#### Prepare the data ----------------------------------------------------------
### Intervention data
# Read in relevant ADaM data and rename variables of interest
adsl <- read.csv(system.file("extdata", "adsl.csv", package = "MAIC",
mustWork = TRUE))
adrs <- read.csv(system.file("extdata", "adrs.csv", package = "MAIC",
mustWork = TRUE))
adtte <- read.csv(system.file("extdata", "adtte.csv", package = "MAIC",
mustWork = TRUE))
adsl <- adsl %>% # Data containing the matching variables
mutate(SEX=ifelse(SEX=="Male", 1, 0)) # Coded 1 for males and 0 for females
adrs <- adrs %>% # Response data
filter(PARAM=="Response") %>%
transmute(USUBJID, ARM, response=AVAL)
adtte <- adtte %>% # Time to event data (overall survival)
filter(PARAMCD=="OS") %>%
mutate(Event=1-CNSR) %>% #Set up coding as Event = 1, Censor = 0
transmute(USUBJID, ARM, Time=AVAL, Event)
# Combine all intervention data
intervention_input <- adsl %>%
full_join(adrs, by=c("USUBJID", "ARM")) %>%
full_join(adtte, by=c("USUBJID", "ARM"))
# List out the variables in the intervention data that have been identified as
# prognostic factors or treatment effect modifiers and will be used in the
# matching
match_cov <- c("AGE",
"SEX",
"SMOKE",
"ECOG0")
## Baseline data from the comparator trial
# Baseline aggregate data for the comparator population
target_pop <- read.csv(system.file("extdata", "aggregate_data.csv",
package = "MAIC", mustWork = TRUE))
# Rename target population cols to be consistent with match_cov
target_pop_standard <- target_pop %>%
rename(
N=N,
Treatment=ARM,
AGE=age.mean,
SEX=prop.male,
SMOKE=prop.smoke,
ECOG0=prop.ecog0
) %>%
transmute(N, Treatment, AGE, SEX, SMOKE, ECOG0)
#### Estimate weights ----------------------------------------------------------
### Center baseline characteristics
# (subtract the aggregate comparator data from the corresponding column of
# intervention PLD)
intervention_data <- intervention_input %>%
mutate(
Age_centered = AGE - target_pop$age.mean,
# matching on both mean and standard deviation for continuous variables (optional)
Age_squared_centered = (AGE^2) - (target_pop$age.mean^2 + target_pop$age.sd^2),
Sex_centered = SEX - target_pop$prop.male,
Smoke_centered = SMOKE - target_pop$prop.smoke,
ECOG0_centered = ECOG0 - target_pop$prop.ecog0)
## Define the matching covariates
cent_match_cov <- c("Age_centered",
"Age_squared_centered",
"Sex_centered",
"Smoke_centered",
"ECOG0_centered")
## Optimization procedure
# Following the centering of the baseline characteristics of the intervention
# study, patient weights can be estimated using estimate_weights
# The function output is a list containing (1) a data set of the individual
# patient data with the assigned weights "analysis_data" and (2) a vector
# containing the matching variables "matching_vars"
est_weights <- estimate_weights(intervention_data = intervention_data,
matching_vars = cent_match_cov)