Getting Started with EPIC

Introduction

This guide will help you get started with the Evaluation Platform in COPD (EPIC) using the epicR package. By the end of this vignette, you will be able to:

Run a basic COPD simulation
Understand and interpret the model outputs
Customize inputs and settings for your analysis
Choose between Canadian and US populations

What is EPIC?

EPIC is a discrete-event simulation (DES) model for Chronic Obstructive Pulmonary Disease (COPD). It models the entire pathway of care including:

Disease incidence and progression
Diagnosis and case detection
Exacerbations (mild, moderate, severe)
Treatment and medication
Healthcare resource utilization
Mortality

The model simulates individual patients (called “agents”) with demographic and clinical characteristics. Each agent progresses through time, experiencing events like birthdays, disease onset, exacerbations, and treatment changes.

Installation

If you haven’t already installed epicR:

# Install from GitHub
pak::pkg_install("resplab/epicR")

Quick Start: Your First Simulation

Simple Approach (Recommended)

The easiest way to run a simulation is using the simulate() function, which handles all session management automatically and provides progress information (configuration summary, real-time progress bar, elapsed time, and status messages):

library(epicR)

# Run with defaults - that's it!
results <- simulate()

# Access basic results
print(results$basic)

# Custom parameters
results <- simulate(
  jurisdiction = "us",
  time_horizon = 10,
  n_agents = 100000
)

# Quick test with fewer agents (faster for testing)
results <- simulate(n_agents = 10000)

# By default, you get both basic and extended results
results <- simulate()
print(results$basic)
print(results$extended)  # Included by default

# Get basic output only (faster, less memory)
results <- simulate(extended_results = FALSE)

# Get event history (automatically sets record_mode)
results <- simulate(return_events = TRUE)
head(results$events)

The basic output includes:

Output	Description
`n_agents`	Total number of agents simulated
`cumul_time`	Total person-years of follow-up
`n_deaths`	Number of deaths
`n_COPD`	Number of agents who developed COPD
`n_exac_*`	Exacerbation counts by severity
`total_cost`	Total healthcare costs
`total_qaly`	Total quality-adjusted life years

Choosing a Jurisdiction: Canada vs US

EPIC supports both Canadian and US populations with jurisdiction-specific parameters:

# For Canadian population (default)
results_canada <- simulate(jurisdiction = "canada")

# For US population
results_us <- simulate(jurisdiction = "us")

The jurisdictions differ in: - Population demographics - Smoking prevalence and trends - Healthcare costs - Disease incidence rates

Customizing Your Simulation

Understanding Inputs

Inputs control the model’s parameters. Explore them:

inputs <- get_input()

# Top-level structure
names(inputs)
# [1] "values" "help" "references"

# Value categories
names(inputs$values)

# Example: global parameters
names(inputs$values$global_parameters)
inputs$values$global_parameters$time_horizon  # Simulation duration in years

Common Input Modifications

The simulate() function provides convenient parameters for common customizations:

# Change time horizon
results <- simulate(time_horizon = 20)

# Change jurisdiction and time horizon
results <- simulate(jurisdiction = "us", time_horizon = 15)

# For quick testing
results <- simulate(n_agents = 10000, time_horizon = 5)

For more advanced input modifications, you can use get_input() to explore and modify parameters:

# Explore available inputs
input <- get_input()
names(input$values)  # See categories

# View specific parameters
input$values$cost$exac_dcost  # Exacerbation costs by severity
input$values$global_parameters$time_horizon

Understanding Settings

Settings control how the model runs (not what it simulates):

settings <- get_default_settings()
names(settings)

Key settings:

Setting	Description	Default
`n_base_agents`	Number of agents to simulate	60,000
`record_mode`	Level of output detail (0=aggregate, 2=individual)	0

Choosing the Number of Agents

More agents = more precision but longer runtime and more memory:

# Quick test run (10,000 agents)
results <- simulate(n_agents = 1e4)

# Standard run (60,000 agents - default)
results <- simulate()

# Production run (1,000,000 agents)
results <- simulate(n_agents = 1e6)

# Check memory requirements before running large simulations
estimate_memory_required(n_agents = 1e6, record_mode = 0, time_horizon = 20)

Getting Detailed Results

Extended Output

For more detailed results by year and demographics:

# By default, you get both basic and extended results
results <- simulate()

# Access basic results
print(results$basic)

# Access detailed output tables
names(results$extended)

Individual Patient Data

To collect event-level data for each agent, use return_events = TRUE:

# Get event history (automatically sets record_mode = 2)
# Keep n_agents small due to memory requirements
results <- simulate(
  n_agents = 1e4,
  time_horizon = 5,
  return_events = TRUE
)

# Access events data frame
head(results$events)

# Get everything including events
results <- simulate(
  n_agents = 1e4,
  extended_results = TRUE,  # TRUE by default
  return_events = TRUE
)
# Returns: results$basic, results$extended, results$events

Warning: Recording individual events requires substantial memory. Start with a small number of agents.

Event Types

Each event in the individual data has a numeric code:

Event	Code
Start	0
Annual	1
Birthday	2
Smoking change	3
COPD incidence	4
Exacerbation	5
Exacerbation end	6
Death by exacerbation	7
Doctor visit	8
Medication change	9
Background death	13
End	14

Open vs Closed Cohort

By default, EPIC is an open population model that captures population dynamics (births, deaths, immigration, emigration). For a closed cohort analysis (fixed initial population, no new entries):

# Run closed cohort analysis
results <- simulate(closed_cohort = TRUE)

# Combine with other parameters
results <- simulate(
  closed_cohort = TRUE,
  jurisdiction = "us",
  time_horizon = 10,
  n_agents = 50000
)

Comparing Scenarios

A common use case is comparing interventions. The simulate() function makes this straightforward:

# Baseline scenario
results_baseline <- simulate(
  jurisdiction = "canada",
  time_horizon = 20,
  n_agents = 100000
)

# Intervention scenario (e.g., different time horizon or jurisdiction)
results_intervention <- simulate(
  jurisdiction = "us",
  time_horizon = 20,
  n_agents = 100000
)

# Compare outcomes
cost_diff <- results_intervention$basic$total_cost -
  results_baseline$basic$total_cost
qaly_diff <- results_intervention$basic$total_qaly -
  results_baseline$basic$total_qaly
icer <- cost_diff / qaly_diff

For more complex scenarios requiring custom input modifications, see the “Advanced Usage” section below.

Advanced Usage

For advanced users who need fine-grained control over input parameters:

Custom Input Modifications

For complex input modifications not covered by simulate() parameters:

# Get and modify inputs
input <- get_input()

# Modify specific parameters
input$values$global_parameters$time_horizon <- 5
input$values$agent$p_female <- 0.55

# Run with custom inputs
results <- simulate(input = input$values)

Running Multiple Simulations

To run multiple simulations with different parameters:

# Simulation 1
results1 <- simulate(n_agents = 50000, time_horizon = 10, seed = 123)

# Simulation 2 with different parameters
results2 <- simulate(n_agents = 50000, time_horizon = 20, seed = 123)

# Compare results
comparison <- data.frame(
  time_horizon = c(10, 20),
  n_deaths = c(results1$basic$n_deaths, results2$basic$n_deaths)
)

Error Handling

The simulate() function includes automatic error handling and cleanup. For additional safety:

results <- tryCatch({
  simulate(n_agents = 50000)
}, error = function(e) {
  message("Simulation failed: ", e$message)
  NULL
})

Troubleshooting

Memory Errors

If you get memory allocation errors: 1. Reduce the number of agents: simulate(n_agents = 10000) 2. Don’t request event history unless needed (omit return_events) 3. Check available memory with estimate_memory_required()

Session Issues

The simulate() function handles session management automatically and includes error handling.

Next Steps

Model Background: See vignette("BackgroundEPIC") for model structure details
Calibration: See vignette("Calibrate_COPD_Prevalence") for calibration methods
Validation: Explore the validate_*() functions for model validation
US Model: See vignette("Calibrate_Smoking") for US-specific calibration

References

Sadatsafavi M, et al. (2019). Development and Validation of the Evaluation Platform in COPD (EPIC): A Population-Based Outcomes Model of COPD for Canada. Medical Decision Making. doi:10.1177/0272989X18824098