Skip to content

dampack

Decision-analytic modeling in R

An R package from the DARTH workgroup for analyzing and visualizing the health economic outputs of mathematical models: cost-effectiveness frontiers, sensitivity analysis, and value of information.

dampack

R package · GPL-3 · CRAN + GitHub

First CRAN release · March 2021

Alarid-Escudero, F., Knowlton, G., Easterly, C. W., Garibay, D., Clements, M., Enns, E.

Johannes Vermeer — Woman Holding a Balance (c. 1664) National Gallery of Art · Washington

What it does

dampack is a suite of R functions for analyzing and visualizing the health economic outputs of mathematical models. It comes from the DARTH workgroup, and its theory is laid out in Hunink et al. (2014).

Give it the costs and effects that a decision model produces for each strategy. It finds the cost-effectiveness frontier and the incremental cost-effectiveness ratios (ICERs). It propagates parameter uncertainty through probabilistic and deterministic sensitivity analysis. It also measures what resolving that uncertainty would be worth, through value of information.

Every result type has a plot() method, so the same ggplot2-based figures come out of each analysis.

4 Kinds of analysis ICERs, probabilistic, deterministic, value of information

24 Exported functions for building, running and summarizing analyses

9 plot() methods, one per result type acceptability curves, tornado plots, EVPI and more

Which function to use

	Functions
Costs, effects and ICERs	calculate_icers(), calculate_icers_psa()
Probabilistic analysis	make_psa_obj(), run_psa(), gen_psa_samp(), ceac(), calc_exp_loss()
Deterministic analysis	run_owsa_det(), run_twsa_det(), owsa(), twsa(), owsa_tornado()
Value of information	calc_evpi(), calc_evppi(), calc_evsi(), metamodel()

A six-strategy example

Install the release from CRAN, or the development version from GitHub. The example dataset used below ships with the development version:

01-install.R

install.packages("dampack")                          # CRAN
remotes::install_github("DARTH-git/dampack",         # development version
                        ref = "dev")

df_example_psa_elc is a de-identified sample of 10,000 probabilistic analysis draws from a published decision model, with six strategies. As willingness to pay rises, its cost-effectiveness frontier changes the optimal strategy three times. Hand the cost half and the effect half straight to make_psa_obj(), then compute the acceptability curves:

02-ceac.R

library(dampack)
library(ggplot2)

psa <- make_psa_obj(
  cost          = df_example_psa_elc[, 1:6],
  effectiveness = df_example_psa_elc[, 7:12],
  strategies    = paste0("S", 1:6),
  currency      = "$"
)

ceac_obj <- ceac(wtp = seq(0, 60000, by = 500), psa = psa)
plot(ceac_obj, frontier = TRUE, points = FALSE)

Each curve is the probability that a strategy is cost-effective at a given willingness to pay. The white squares mark the strategy on the frontier, the one with the highest expected net benefit, at each value.

The frontier changes hands three times. S5 is optimal first, then S6 from $11,000, S3 from $30,000 and S1 from $41,000 per QALY. A strategy can be optimal without having the highest probability of being cost-effective, which is why the squares do not always sit on the top curve.

The sample supports cost-effectiveness analysis only. The parameter draws behind it were removed, so it cannot be used for metamodeling or EVPPI. Strategy labels are generic.

This figure is not rendered by the page. The dataset is not on CRAN, so the code runs from make-ceac-figure.R in this folder against the development version, and the PNG is committed. That script adds a transparent dark theme to the plot shown above.

Acceptability curves for six strategies, with the frontier marked.

Summarizing the switches

summary() of a ceac object returns the willingness-to-pay interval over which each strategy is optimal, with one row per interval however many times the frontier switches. For the example above:

range_min	range_max	optimal_strategy
0	11,000	S5
11,000	30,000	S6
30,000	41,000	S3
41,000	60,000	S1

The third column, optimal_strategy, names the strategy with the highest expected net benefit over that interval.

Beyond the example

The package vignettes cover the rest, in this order: basic_cea, psa_analysis, voi, psa_generation and dsa_generation. Open one with vignette("basic_cea", package = "dampack").

  Source on GitHub     CRAN page