Health Economics

Health Economics Evaluation in R

Overview

Health economic evaluation methods covering cost-effectiveness analysis (CEA), quality-adjusted life years (QALYs), incremental cost-effectiveness ratios (ICERs), budget impact analysis, Markov cohort models, partitioned survival analysis, probabilistic sensitivity analysis, and value of information analysis.

Cost-Effectiveness Fundamentals

Basic Calculations

# Treatment comparison data
# Intervention vs Comparator
costs_int 
  mutate(
    delta_cost = cost_trt - cost_base,
    delta_qaly = qaly_trt - qaly_base,
    icer = delta_cost / delta_qaly
  )

# CEAC calculation
wtp_range  0)
})

# Plot CEAC
plot(wtp_range, ceac, type = "l",
     xlab = "Willingness-to-Pay ($/QALY)",
     ylab = "Probability Cost-Effective",
     main = "Cost-Effectiveness Acceptability Curve")

Value of Information Analysis

Expected Value of Perfect Information (EVPI)

library(BCEA)

# EVPI from BCEA object
evpi_result 
  mutate(
    Total_Cost = Cost_New + Cost_Current,
    Budget_Impact = Total_Cost - Cost_Reference
  )

# Summary
print(budget_impact)

# Total 5-year budget impact
total_impact 
  select(Year, Cost_New, Cost_Current, Cost_Reference) |>
  pivot_longer(
    cols = -Year,
    names_to = "Category",
    values_to = "Cost"
  )

# Stacked bar chart
ggplot(bi_long |> filter(Category != "Cost_Reference"),
       aes(x = factor(Year), y = Cost / 1e6, fill = Category)) +
  geom_bar(stat = "identity") +
  geom_line(data = bi_long |> filter(Category == "Cost_Reference"),
            aes(y = Cost / 1e6, group = 1), linetype = "dashed", size = 1) +
  labs(x = "Year", y = "Cost ($ millions)",
       title = "Budget Impact Analysis",
       fill = "Treatment") +
  scale_fill_brewer(palette = "Set2") +
  theme_bw()

Decision Trees

Using dampack Package

library(dampack)

# Define decision tree parameters
params <- list(
  p_disease = 0.10,          # Probability of disease
  p_cure_trt = 0.80,         # Cure probability with treatment
  p_cure_notrt = 0.50,       # Cure probability without treatment
  c_test = 100,              # Cost of diagnostic test
  c_treatment = 5000,        # Cost of treatment
  c_disease = 20000,         # Cost of disease (if not cured)
  u_healthy = 1.0,           # Utility healthy
  u_disease = 0.60           # Utility with disease
)

# Strategy 1: Treat all
cost_treat_all <- params$c_treatment +
  params$p_disease * (1 - params$p_cure_trt) * params$c_disease
qaly_treat_all <- params$p_disease * (
  params$p_cure_trt * params$u_healthy +
  (1 - params$p_cure_trt) * params$u_disease
) + (1 - params$p_disease) * params$u_healthy

# Strategy 2: Test then treat
cost_test_treat <- params$c_test +
  params$p_disease * (params$c_treatment +
  (1 - params$p_cure_trt) * params$c_disease)

# Calculate all strategies and compare

Discounting

# Discount costs and effects
discount <- function(values, rate, time_points) {
  values / (1 + rate)^time_points
}

# Example: 30-year costs
years <- 0:29
annual_costs <- rep(5000, 30)
discount_rate <- 0.03

# Present value
pv_costs <- sum(discount(annual_costs, discount_rate, years))
cat("Undiscounted total:", sum(annual_costs), "\n")
cat("Present value (3% discount):", round(pv_costs, 0), "\n")

# Differential discounting (costs vs QALYs)
# Some guidelines recommend different rates
discount_costs <- 0.03
discount_qalys <- 0.015

pv_costs <- sum(discount(annual_costs, discount_costs, years))
pv_qalys <- sum(discount(annual_qalys, discount_qalys, years))

Key Packages Summary

| Package | Purpose | |---------|---------| | BCEA | Bayesian cost-effectiveness analysis | | heemod | Markov cohort models for HE | | hesim | Health economic simulation modeling | | dampack | Decision-analytic modeling tools | | survHE | Survival analysis for HE | | flexsurv | Parametric survival for extrapolation | | CEAutil | CEA utility functions | | valueEQ5D | EQ-5D utility mapping |

Best Practices

1. Model structure: Match model type to disease natural history
2. Time horizon: Sufficient to capture all relevant costs and effects
3. Discounting: Apply recommended rates (often 3% for both costs and effects)
4. Uncertainty: Always conduct PSA and report CIs/CrIs
5. Transparency: Document all assumptions and data sources
6. Validation: Internal consistency checks and external validation
7. Reporting: Follow CHEERS checklist for publications
8. Perspective: Clearly state healthcare payer vs societal perspective

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: choxos
• Source: choxos/BiostatAgent
• License: MIT

Install and usage instructions live in the source repository linked above.