Skip to contents

Introduction to testhta

Source: vignettes/introduction.Rmd
introduction.Rmd

testhta provides a software-engineering-inspired framework for building and validating Health Technology Assessment (HTA) cost-effectiveness models in R. By combining cohort Markov models with structured unit tests, it ensures model transparency, correctness, and reproducibility.

This vignette covers: 1. Setting up a Markov model. 2. Running the model under baseline parameters. 3. Accessing model outputs using getters.

1. The HTA Markov Model Setup

The core function in testhta is ce_markov(), which runs a cohort Markov model. The model expects: - start_pop: Initial counts of the cohort across health states. - p_matrix: A 3-dimensional array of transition probabilities (states ×\times states ×\times treatments). - state_c_matrix: A matrix of costs for state occupancy. - trans_c_matrix: A matrix of costs incurred during state transitions. - state_q_matrix: A matrix of quality-of-life weights (utilities) for each state.

The package includes a built-in baseline dataset called test_data. Let’s inspect its structure:

# Load baseline dataset
data(test_data)
names(test_data)
#>  [1] "start_pop"      "p_matrix"       "state_c_matrix" "trans_c_matrix"
#>  [5] "state_q_matrix" "n_cycles"       "init_age"       "s_names"       
#>  [9] "t_names"        "discount_rate"

It contains: - start_pop: Vector of length 3 (Asymptomatic disease, Progressive disease, Dead) - p_matrix: 3×3×23 \times 3 \times 2 array for without_drug and with_drug - state_c_matrix: State cost matrix for each treatment - state_q_matrix: Utility matrix for each treatment - discount_rate: Standard HTA discount rate (typically 3.5% or 0.035) - n_cycles: Time horizon of the model

2. Running the Model

We can run the model by passing test_data parameters directly to ce_markov(), or using the wrapper helper run_model().

Here is how you execute the Markov simulation:

# Run with the helper wrapper
results <- run_model(test_data)

The output of run_model() is a list containing the population distributions, cycle-specific costs and QALYs, and summary metrics.

3. Extracting Results using Getters

testhta provides clean getter functions to extract the key outcomes from your model results:

  • get_qalys(): Extracts total Quality-Adjusted Life Years (QALYs).
  • get_costs(): Extracts total costs.
  • get_le(): Extracts total Life Expectancy (undiscounted life-years).
  • get_icer(): Calculates the Incremental Cost-Effectiveness Ratio (ICER) between the two arms.

Let’s retrieve these values from our results:

# Get total QALYs
get_qalys(results)
#> without_drug    with_drug 
#>     9.048517     9.589129

# Get total Costs
get_costs(results)
#> without_drug    with_drug 
#>     9205.744    16977.576

# Get life expectancy
get_le(results)
#> without_drug    with_drug 
#>     12.06416     12.68332

# Calculate the ICER
get_icer(results)
#> [1] 14375.98

4. Parameter Modification with Setters

The package provides setter functions that follow a tidy workflow using the base R pipe (|>). These allow you to easily modify input parameters for scenario or sensitivity analyses before running the model:

  • set_discount_rate(input, discount_rate)
  • set_time_horizon(input, n_cycles)
  • set_costs(input, scenario, values)
  • set_utilities(input, scenario, values)

For example, to calculate the ICER with no discounting and a shorter time horizon (20 cycles):

scenario_results <- test_data |>
  set_discount_rate(0) |>
  set_time_horizon(20) |>
  run_model()

get_icer(scenario_results)
#> [1] 7392.202

This piping framework makes scenario analysis clean, readable, and less prone to copy-paste errors.