In n8thangreen/CEdecisiontree: Cost-effectiveness decision tree analysis

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Introduction

PSA is a core part of any cost-effectiveness analysis [@Briggs2012]. Here we will carry this out for a simple decision tree. This involves repeatedly sampling from a distribution for each branch probability and cost and calculating the total expected value for each set of realisations.

Example

library(CEdecisiontree, quietly = TRUE)
library(assertthat, quietly = TRUE)
library(tibble, quietly = TRUE)
library(tidyverse, quietly = TRUE)
library(purrr, quietly = TRUE)

We first define the decision tree. The difference to previous trees is that we now use the list-column feature to define distributions rather than point values.

tree_dat <-
  list(child = list("1" = c(2, 3),
                    "2" = NULL,
                    "3" = NULL),
       dat = tibble(
         node = 1:3,
         prob =
           list(
             NA_real_,
             list(distn = "unif", params = c(min = 0, max = 1)),
             list(distn = "unif", params = c(min = 0, max = 1))),
         vals =
           list(
             0L,
             list(distn = "unif", params = c(min = 0, max = 1)),
             list(distn = "unif", params = c(min = 0, max = 1)))))

tree_dat

We can now loop over this tree and generate samples of values for the given distributions. We use the sample_distributions() function to do this. This is all wrapped up in the convenience function create_psa_inputs.

library(purrr)

tree_dat_sa <- create_psa_inputs(tree_dat, n = 1000)

This results in a list of trees.

head(tree_dat_sa, 2)

Now it is straightforward to map over each of these trees to obtain the total expected values

res <- map_dbl(tree_dat_sa, dectree_expected_values)
head(res)


hist(res, breaks = 20)

References

n8thangreen/CEdecisiontree documentation built on Sept. 13, 2022, 5:25 a.m.