Economic Model

An economic model in health care is a simplified mathematical representation of a complex disease process, treatment pathway, or health care system used to estimate the costs and health outcomes of alternative interventions over time. Economic models synthesize evidence from multiple sources, including clinical trials, observational studies, cost databases, and epidemiological data, to project outcomes beyond the time frame of any single study and to compare strategies that have not been directly compared in head to head trials. They are the main analytical tool for the economic evaluation part of health technology assessment, and they are widely used in health economic research.

Clinical trials, while essential for establishing treatment efficacy, have limitations that economic models address. Trials rarely follow patients long enough to capture lifetime costs and outcomes. They typically compare only two or three interventions, leaving other relevant comparators unevaluated. And they are conducted in controlled settings that may not reflect real world clinical practice. Economic models extrapolate trial results over longer time horizons, integrate evidence on multiple comparators, and adjust for differences between trial populations and real world patients. Health technology assessment agencies rely on economic models to inform coverage, reimbursement, and formulary decisions that affect millions of patients and billions of dollars in health care spending.

Model development begins with defining the decision problem: the disease, the target population, the interventions being compared, the relevant outcomes, the time horizon, and the analytical perspective (payer, health system, or societal). The modeler then selects the appropriate model structure, which determines how the disease process and treatment effects are represented mathematically.

Common model structures include decision trees (for short term decisions with few health states), Markov models (for chronic conditions with distinct health states and recurring transitions), microsimulation models (which simulate individual patients, allowing for heterogeneity and history dependent transitions), discrete event simulation (which models events occurring at variable intervals), and dynamic transmission models (for infectious diseases where the probability of infection depends on the prevalence of disease in the population).

Each model is populated with input parameters estimated from the best available evidence. Clinical parameters (transition probabilities, treatment effects, event rates) come from systematic reviews, meta analyses, or individual studies. Cost parameters come from published cost studies, claims databases, or micro costing exercises. Health state utilities usually come from preference based instruments such as the EQ 5D, which patients complete and which are scored with preference weights (value sets) usually drawn from the general population, or from direct valuation studies. When direct evidence is unavailable, network meta analysis can provide indirect comparisons of treatments, and expert opinion can fill remaining gaps.

Model validation is essential for establishing credibility. The ISPOR-SMDM Modeling Good Research Practices Task Force distinguishes five kinds. Face validity is the judgment of experts that the model's structure, data sources, assumptions and results are reasonable. Verification, also called internal validity, checks that the model's calculations and coding are correct, for example by confirming that it produces expected results under known conditions. Cross validation compares the model's results with those of other models of the same problem. External validation compares model results with observed real world data, ideally data that were not used to build the model, and predictive validation compares them with events observed prospectively; the task force regards these last two as the strongest forms.

Economic models are appropriate whenever the decision problem requires projecting costs and outcomes beyond available trial data, comparing interventions not directly compared in trials, synthesizing evidence from multiple sources, or evaluating cost effectiveness under conditions of uncertainty. They are standard in health technology assessment submissions, health policy analysis, and clinical guideline development that incorporates economic considerations. They are not appropriate when a single trial provides all the necessary information about costs and outcomes over a sufficient time horizon, or when the complexity of the model would exceed the available evidence and produce results driven more by assumptions than by data.

A common error is building a model that is more complex than the available data can support, creating an illusion of sophistication that masks uncertainty. Another mistake is using an inappropriate model structure, such as a Markov model for an acute condition that resolves within weeks (better suited to a decision tree) or a decision tree for a chronic condition with recurrent events over decades (better suited to a Markov model or microsimulation). Failing to conduct adequate sensitivity analysis, or conducting it but not reporting the results transparently, prevents decision makers from understanding the uncertainty surrounding the model's conclusions. Using input parameters from non representative populations (such as applying costs from one country to another without adjustment) undermines the model's relevance. Finally, some modelers fail to report the model's structure and assumptions with sufficient detail for independent evaluation, falling short of the transparency standards set out in CHEERS 2022 and the ISPOR-SMDM good practice reports.

In a hypothetical example, a pharmaceutical company develops a microsimulation model to estimate the lifetime cost effectiveness of a new oral anticoagulant compared to warfarin for stroke prevention in atrial fibrillation. The model simulates 100,000 individual patients through monthly cycles, tracking events including stroke, intracranial hemorrhage, gastrointestinal bleeding, myocardial infarction, and death. Input parameters are drawn from three Phase III trials, a network meta analysis, a national claims database for costs, and EQ 5D data from a patient registry. The model projects that the new anticoagulant produces 0.32 additional QALYs at an additional lifetime cost of $8,400, yielding an ICER of $26,250 per QALY gained. Probabilistic sensitivity analysis across 5,000 iterations shows a 73% probability of cost effectiveness at a $50,000 per QALY threshold.