Build a Health Economic Model

Building a health economic model requires you to move from a clearly defined decision problem and clinical evidence to an appropriate model structure, analyses, validation, and communication of results.

The following steps provide a practical framework for developing a health economic model and show how TreeAge Pro supports the modeling process from beginning to end.

Define the Decision Problem

Before building a model, clearly define what you want to learn. A model can only produce meaningful results if its objectives are well understood from the start.

Consider the following key elements:

• Population
  • Who is affected by the health issue?
  • Examples:
    • Patients diagnosed with a specific type of cancer
    • Women with high risk pregnancies
• Interventions
  • What treatment, diagnostic, or screening options are being evaluated?
  • Examples:
    • Radiation therapy vs. surgery vs. combination therapy
    • Breast cancer screening strategies
    • Different colonoscopy screening intervals
• Perspective
  • Whose perspective will guide the analysis?
  • Examples:
    • Healthcare payer
    • Societal
    • Patient
• Time horizon
  • How long should patients be followed?
  • Examples:
    • Short-term follow-up
    • Lifetime horizon
• Outcomes
  • What outcomes will be used to compare strategies?
  • Examples:
    • Cost per quality-adjusted life year (QALY)
    • Cost per death avoided
  • Are there additional outcomes we want to report?
    • Infections prevented
    • Healthy births
    • Cases detected

Understand Your Clinical Evidence

Before choosing how to represent disease progression, understand what your available evidence tells you and how risk and outcomes change over time.

Clinical evidence may include Kaplan-Meier curves or tables, patient-level data, fitted distributions, published probabilities, hazard ratios, treatment effects, and other sources.

Consider:

• What outcomes or events does the evidence represent?
• How does risk change over time?
• How far does the observed evidence extend?
• Will outcomes need to be extrapolated beyond the observed period?
• Does the evidence describe overall outcomes or specific events that should drive model transitions?

TreeAge Pro provides tools to explore and interpret clinical evidence and translate it into the modeling approach that fits your problem.

Select the Appropriate Model Type

Choose the modeling approach that best represents the decision problem, available evidence, and disease process.

• Decision Tree
  • Best suited for short-term analyses with a limited number of outcomes.
  • Examples:
    • Emergency department visits
    • Orthopedic procedures
• Markov Model
  • Designed for long-term disease progression involving recurring events and health states.
  • Examples:
    • Cancer treatment with adverse events
    • Asthma management
    • Chronic pain management
• Partitioned Survival Model
  • Used for long-term disease progression based on survival outcomes.
  • Example:
    • Oncology treatments
• Patient Simulation
  • Useful when patient characteristics and treatment history influence outcomes.
  • Examples:
    • Colorectal cancer screening based on prior screening history
    • Subgroup analyses based on patient characteristics
• Discrete Event Simulation
  • Models disease progression through the timing of individual events.
  • Examples:
    • Complex care pathways where event timing is critical

Build Patient Pathways

Define the model structure and patient pathways before focusing on the data needed to populate them.

Build pathways that represent the important health states, events, decisions, and disease progression in your model. Starting with the structure helps clarify what the model needs to represent—and what inputs will actually be required.

• Decision Nodes and Strategies
  • The model typically begins with a decision node containing one branch for each strategy under evaluation.
  • Each strategy is modeled independently to estimate average patient outcomes.
• Patient Pathways
  • Patient pathways represent all possible health states and events a patient may experience.
  • Pathways branch at chance nodes, where probabilities determine the likelihood of moving along each route.
• Markov Models
  • Markov models consist of health states, events, and cycles that represent disease progression over time.
  • Key characteristics:
    • The overall time horizon is divided into cycles.
    • Health states represent a patient’s current condition.
    • During each cycle, patients may experience events that move them to other health states.
    • The process repeats until the end of the time horizon.
• Partitioned Survival Models
  • Partitioned survival models use survival curves to determine membership in health states over time.
  •  Typically:
    • Progression-Free Survival (PFS) and Overall Survival (OS) curves are modeled.
    • Survival functions describe how state membership changes over tim
• Patient-Level Simulations
  • Patient simulations use a structure similar to a Markov model but simulate individual patients.
  • Advantages include:
    • Incorporation of patient characteristics
    • Tracking of patient history
    • Greater modeling flexibility
• Discrete Event Simulation
  • Discrete event simulations resemble Markov models but are driven by sampled event times rather than fixed cycles.

Add Model Inputs

Once the model structure is established, add the clinical, economic, and other inputs needed to drive the model.

Inputs may include probabilities, survival functions, treatment effects, costs, utilities, and other parameters. Add inputs where they apply within the model structure so their role in calculations and outcomes remains clear.

• Disease Progression
  • These inputs describe how the disease evolves within the target population.
  • Examples:
    • Markov models: transition probabilities
    • Partitioned survival models: survival or hazard functions
    • Discrete event simulations: event-time distributions
• Costs
  • Include costs associated with:
    • Treatments
    • Diagnostics
    • Hospitalizations
    • Follow-up care
• Utilities
  • Utilities measure health-related quality of life associated with different health states.
• Additional Inputs
  • Examples include:
    • Test sensitivity and specificity
    • Hazard ratios
    • Relative risks
    • Epidemiological parameters

Validate the Model

Model validation is essential before relying on results for decision-making.

• Visualize/Review Disease Progression
  • Compare modeled disease progression with observed clinical data to ensure reasonable behavior over time.
• Verify Outcome Calculations
  • Confirm that costs, utilities, and other outcomes are accumulated correctly throughout the model.
• Examine Patient Traces
  • Review patient traces to understand how patients move through the model.
  • This can be done at:
    • The cohort level
  • The individual patient level (for simulation models)
• Test Complex Calculations
  • Carefully review and test formulas, assumptions, and custom calculations to ensure accuracy.

Conduct a Cost-Effectiveness Analysis

The Incremental Cost-Effectiveness Ratio (ICER) represents the ration of increased cost (IC) vs. increased effectiveness (IE) – ICER = IC/IE. If the ICER is less that the Willingness-to-Pay (WTP), then the strategy provides better outcomes for less that we were willing to pay for them, and we would recommend the more expensive/effective strategy.

Incremental Cost-Effectiveness Ratio (ICER)

The Incremental Cost-Effectiveness Ratio (ICER) measures the additional cost required to achieve an additional unit of effectiveness.

ICER = Incremental Cost (IC) ÷ Incremental Effectiveness (IE)

A strategy is generally considered cost-effective when its ICER falls below the willingness-to-pay (WTP) threshold. In that case, the additional health benefits are considered worth the additional cost.

Assess the Impact of Uncertainty

After drawing conclusions from the base-case analysis, evaluate how uncertainty in model inputs may affect the results.

Many model parameters are estimated and therefore uncertain. Sensitivity analyses help determine whether conclusions remain consistent when assumptions change.

Deterministic Sensitivity Analysis

Deterministic sensitivity analysis evaluates the effect of changing one parameter at a time across a plausible range of values.

Results for multiple parameters are often consolidated and presented using tornado diagrams.

Probabilistic Sensitivity Analysis

Probabilistic sensitivity analysis evaluates uncertainty across multiple parameters simultaneously.

Parameter values are sampled from probability distributions to estimate the overall uncertainty surrounding model results.

Present Your Results

Once model development and analysis are complete, results should be communicated clearly through reports, presentations, and publications.

TreeAge Pro simplifies this process by allowing users to:

• Copy and export graphs for use in reports and manuscripts
• Share models through TreeAge Pro Web
  • Provide stakeholders with interactive access to explore model results
  • Access permissions remain under your control, allowing you to determine who can view and analyze your model.