In dincerti/cea: Health Economic Simulation Modeling and Decision Analysis

Health economic simulation modeling

Overview

hesim is a modular and computationally efficient R package for health economic simulation modeling and decision analysis that provides a general framework for integrating statistical analyses with economic evaluation. The package supports cohort discrete time state transition models (DTSTMs), N-state partitioned survival models (PSMs), and individual-level continuous time state transition models (CTSTMs), encompassing both Markov (time-homogeneous and time-inhomogeneous) and semi-Markov processes. It heavily utilizes Rcpp and data.table, making individual-level simulation, probabilistic sensitivity analysis (PSA), and incorporation of patient heterogeneity fast.

Features of the current version can be summarized as follows:

Cohort DTSTMs, individual-level CTSTMs, and N-state PSMs that encompass Markov and semi-Markov processes
Options to build models directly from fitted statistical models or by defining them in terms of expressions
Parameter estimates from either an R based model or from an external source
Convenience functions for sampling model parameters from parametric distributions or via bootstrapping
Parameter uncertainty propagated with PSA
Modeling patient heterogeneity
Performing cost-effectiveness analyses and representing decision uncertainty from PSAs
Simulation code written in C++ to boost performance

Installation

You can install the current release from CRAN or the most up to date development version from GitHub.

# Install from CRAN:
install.packages("hesim")

# Install the development version from GitHub:
# install.packages("devtools")
devtools::install_github("hesim-dev/hesim")

Getting started

There are two good places to start:

The Introduction to hesim article provides a quick introduction.
Our preprint describes the package (including mathematical details) more thoroughly.

You might also want to explore our example analyses which can be found in the preprint and web articles. They are summarized in the table below, with some drawn from the Decision Modeling for Health Economic Evaluation textbook. Key areas of focus are the (i) statistical models of disease progression (in terms of the baseline risk and relative treatment effects) and (ii) the available data (either individual patient data (IPD) or aggregate-level data).

library("knitr")
library("kableExtra")
level <- c("Cohort", "Cohort", "Cohort", "Cohort", "Individual", "Individual", "Cohort")
links <- c(
  rep("https://arxiv.org/pdf/2102.09437.pdf", 3),
  "https://hesim-dev.github.io/hesim/articles/markov-cohort.html",
  "https://hesim-dev.github.io/hesim/articles/markov-inhomogeneous-cohort.html",
  "https://hesim-dev.github.io/hesim/articles/mlogit.html",
  "https://hesim-dev.github.io/hesim/articles/markov-inhomogeneous-indiv.html",
  "https://hesim-dev.github.io/hesim/articles/mstate.html",
  "https://hesim-dev.github.io/hesim/articles/psm.html"
)
name <- c(
  "Preprint 4.1", "Preprint 4.2", "Preprint 4.3",
  "Simple Markov cohort", "Time inhomogeneous Markov (cohort)", "Multinomial logit",
  "Time inhomogeneous Markov (individual)", "Semi-Markov multi-state", "4-state PSM"
)
name <- cell_spec(name, link = links)
model <- c("iCTSTM", "PSM", "cDTSTM", "cDTSTM", "cDTSTM", "cDTSTM", "iCTSTM", "iCTSTM", "PSM")
num <- 1:length(model)
application <- c(rep("Oncology", 3),"HIV", "Hip replacement", "Generic", "Hip replacement", "Generic", "Oncology")
dismod1 <- c("Multi-state model", "Survival models", "Multi-state model (panel data)", 
             "Multinomial", "Custom", "Multinomial logit", "Custom", "Multi-state model", 
             "Survival models")
dismod2 <- c("Coefficient (AFT)", "Coefficient (AFT)" , "RR",
             "RR", "Coefficient (HR)", "Coefficient (OR)", "Coefficient (HR)", 
             "Coefficient (AFT)", "Coefficient (AFT)")
data1 <- c(rep("IPD", 3), "Aggregate", "Aggregate", "IPD", "Aggregate", "IPD", "IPD")
data2 <- c("IPD", rep("Aggregate", 2), "Aggregate", "Aggregate", "IPD", "Aggregate",
           "IPD", "IPD")
tbl <- cbind(num, name, model, dismod1, data1, dismod2, data2, application)
colnames(tbl) <- c("", "Name", "Model", 
                   rep(c("Disease model", "Disease data"), 2),
                   "Application")
rownames(tbl) <- 1:nrow(tbl)
kable(tbl, row.names = FALSE, escape = FALSE)  %>% 
  kable_styling() %>%
  add_header_above(c(rep("", 3), "Baseline risk" = 2, "Treatment effect" = 2, "")) %>%
  column_spec(2, width = "15em") %>%
  footnote(general = paste0(
    "iCTSTM = Individual-level continuous time state transition model; ",
    "PSM = partitioned survival model; ", 
    "cDTSTM = Cohort discrete time state transition model. ",
    "AFT = accelerated failure time; RR = relative risk; HR = hazard ratio; OR = odds ratio. ",
    "IPD = individual patient data. "),
           footnote_as_chunk = TRUE)