Health Economic Model

Health Economic Models and Cost-Effectiveness Estimates

A health economic model, often called a decision model or decision-analytic model, brings together evidence from trials, observational studies, valuation surveys and cost data to estimate the expected costs and health effects of each relevant option open to a decision maker. It is the usual vehicle for cost-effectiveness analysis in health technology assessment, because a single study rarely compares all the options, runs long enough or measures the outcomes the decision needs. For technology appraisal, the NICE manual (PMG36) sets out what it expects of a model, including how it is validated and documented.

Why a single study is rarely enough

Petrou and Gray, writing in the BMJ, note that a single trial might not compare all the available options, provide evidence on all relevant inputs, or run long enough to capture differences in economic outcomes, and that relying on one trial ignores evidence from other trials, meta-analyses and observational studies. The ISPOR-SMDM task force overview by Caro and colleagues describes policy decision models as needing an explicit budget constraint and, with it, both resource consequences and health outcomes.

A model does four jobs that a single study cannot.

1. Synthesis. The NICE manual (PMG36, section 4.6.2) calls modelling an important framework for synthesising available evidence and states that models are needed for most evaluations.
2. Extrapolation. A lifetime time horizon often means projecting beyond the duration of the clinical trials and other evidence (PMG36, section 4.2.24); the extrapolation page covers the methods.
3. Linking intermediate to final outcomes. When final end points cannot be observed, PMG36 (section 4.6.5) asks for evidence supporting the link from a surrogate endpoint to mortality and quality of life, and an explanation of how that link is quantified in the model.
4. Comparing options never tested head to head. Because one trial may omit some options, the Caro overview (VI-4) recommends formal evidence synthesis, including meta-analysis and network meta-analysis as appropriate, which lets each relevant comparator be estimated.

The elements every model must specify

Every model starts from a written decision problem. Roberts and colleagues, in the task force report on conceptualising a model, recommend a written statement covering the disease spectrum, analytic perspective, target population, alternative interventions, outcomes and time horizon.

Parameters reach the model through parameter inputs, and the simplifications made along the way form its assumption structure. Costs and QALYs for each option are then compared through an ICER or net monetary benefit.

Choosing a model type for the decision problem

The problem features that drive the choice, as listed by Roberts and colleagues, are whether individuals or groups are represented, whether individuals interact, the time horizon, whether time is continuous or discrete, whether events recur, and whether resources are constrained. They warn that modelling individuals does not automatically imply greater accuracy.

Siebert and colleagues call state-transition modelling one of the most widespread techniques in health economic evaluation; the state-transition model and Markov model pages cover it in full. Individual-level versions track each person's history, which removes the Markov assumption. Pitman and colleagues explain that static models hold infection risk constant and cannot capture herd immunity, so a dynamic model is needed when an intervention such as vaccination changes transmission; an agent-based model is another option when individuals interact. Specialised structures include the state space model and the treatment sequence model, while a budget impact model answers affordability, not value for money.

Worked example combining two sources in a one-year decision tree

The example and all numbers are illustrative. A preventive drug is compared with no treatment over one year. A registry gives a baseline stroke risk of 0.10 without treatment, and a meta-analysis gives a relative risk of 0.70, so the risk with treatment is 0.10 × 0.70 = 0.07. PMG36 (section 4.6.16) describes this step of applying relative treatment effects from trials to baseline risk from other data. The drug costs GBP 500, a stroke costs GBP 12,000, and a stroke reduces QALYs in the year from 0.80 to 0.50, a loss of 0.30.

Expected costs are 0.10 × 12,000 = GBP 1,200 without treatment and 500 + 0.07 × 12,000 = GBP 1,340 with it. Expected QALYs are 0.80 − 0.10 × 0.30 = 0.770 and 0.80 − 0.07 × 0.30 = 0.779.

$$\text{ICER} = \frac{C_T - C_N}{E_T - E_N} = \frac{1340 - 1200}{0.779 - 0.770} = \frac{140}{0.009} = 15556$$

where $C_T$ and $C_N$ are expected costs (GBP) with and without treatment and $E_T$ and $E_N$ are expected QALYs. The drug costs about GBP 15,556 per QALY gained over one year. The one-year tree omits the lifelong costs and quality-of-life loss after a stroke, so a longer horizon, usually a state-transition model, would be needed for a real appraisal.

Validating and documenting a model

Eddy and colleagues, in the task force report on transparency and validation, define transparency as the extent to which interested parties can review a model's structure, equations, parameter values and assumptions. They recommend a nontechnical description available to anyone, and technical documentation detailed enough for an expert to evaluate and potentially reproduce the model, released openly or under agreements that protect intellectual property.

The same report names five types of validation, namely face validity, verification (internal validity), cross validity, external validity and predictive validity, and regards the last two as the strongest. The model validation and calibration pages cover the methods. AdViSHE, a validation-assessment tool by Vemer and colleagues, has 13 items covering the conceptual model, input data, implemented software and model outcomes. In AdViSHE, developers report what validation was done and what it found, so users can judge whether more is needed.

What the NICE manual (PMG36) expects of a model

The NICE technology appraisal manual, PMG36 (published January 2022, last updated 13 October 2026), sets out its modelling expectations in section 4.6. Models should follow accepted guidelines, with structural assumptions and data inputs fully documented and justified, and sensitivity analyses where alternatives are plausible (4.6.1).

• Model type and structure should be justified for each new decision problem; prior use or acceptance by NICE is not enough, and the conceptual model development, including expert involvement, should be transparent (4.6.3).
• Quality assurance and validation methods and results should be reported, with disaggregated results and, for cost-utility analyses, life years gained and mortality rates (4.6.4).
• Models should be informed by the natural history of the disease and checked for clinical plausibility (4.6.18).
• On uncertainty, the model should quantify decision uncertainty (4.7.2), the preferred estimate should come from a probabilistic analysis when possible unless the model is linear (4.7.12), and structural uncertainty is explored through plausible scenarios (4.7.4). Probabilistic sensitivity analysis, structural uncertainty, scenario analysis and value of information have their own pages.

Model, model-based evaluation and other neighbours

The model is the tool; a model-based economic evaluation is the analysis that uses it, as distinct from an evaluation run alongside a single trial. Briggs, Claxton and Sculpher apply the same decision modelling techniques to medical devices, surgical procedures, diagnostic technologies and pharmaceuticals, so a pharmacoeconomic model is a health economic model for a medicine. Survival regressions and meta-analyses are inputs, not decision models, because they do not compare the full set of options on costs and outcomes. A cohort model and an individual patient simulation can run the same state structure, for the whole cohort at once or one patient at a time.

Where health economic models mislead

A model is a simplification, and its results hold only for the comparators, structure and evidence it contains. Caro and colleagues report strong concerns about the credibility of models used in policy, and several errors recur.

1. Choosing comparators by data. Roberts and colleagues state that comparators should be set by the problem, not by data availability or quality.
2. Reusing a structure because it was accepted before, which PMG36 (4.6.3) rules out as a justification.
3. Equating complexity with accuracy. More states or individual simulation add detail but not, by themselves, validity.
4. Treating extrapolation as fact. PMG36 (4.2.24) asks for scenarios that include no further benefit after treatment stops.
5. Using a static model for an infectious disease when the intervention changes transmission.

Sources

• Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford: Oxford University Press; 2006.
• Caro JJ, Briggs AH, Siebert U, Kuntz KM. Modeling good research practices, overview: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force-1. Medical Decision Making. 2012;32(5):667-677. https://doi.org/10.1177/0272989X12454577
• Eddy DM, Hollingworth W, Caro JJ, Tsevat J, McDonald KM, Wong JB. Model transparency and validation: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force-7. Value in Health. 2012;15(6):843-850. https://doi.org/10.1016/j.jval.2012.04.012
• Karnon J, Stahl J, Brennan A, Caro JJ, Mar J, Möller J. Modeling using discrete event simulation: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force-4. Value in Health. 2012;15(6):821-827. https://doi.org/10.1016/j.jval.2012.04.013
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). London: NICE; 2022, last updated 13 October 2026. Sections 4.2.22 to 4.2.24, 4.6 and 4.7. https://www.nice.org.uk/process/pmg36
• Petrou S, Gray A. Economic evaluation using decision analytical modelling: design, conduct, analysis, and reporting. BMJ. 2011;342:d1766. https://doi.org/10.1136/bmj.d1766
• Pitman R, Fisman D, Zaric GS, Postma M, Kretzschmar M, Edmunds J, Brisson M. Dynamic transmission modeling: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force-5. Value in Health. 2012;15(6):828-834. https://doi.org/10.1016/j.jval.2012.06.011
• Roberts M, Russell LB, Paltiel AD, Chambers M, McEwan P, Krahn M. Conceptualizing a model: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force-2. Value in Health. 2012;15(6):804-811. https://doi.org/10.1016/j.jval.2012.06.016
• Siebert U, Alagoz O, Bayoumi AM, Jahn B, Owens DK, Cohen DJ, Kuntz KM. State-transition modeling: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force-3. Value in Health. 2012;15(6):812-820. https://doi.org/10.1016/j.jval.2012.06.014
• Vemer P, Corro Ramos I, van Voorn GAK, Al MJ, Feenstra TL. AdViSHE: a validation-assessment tool of health-economic models for decision makers and model users. PharmacoEconomics. 2016;34(4):349-361. https://doi.org/10.1007/s40273-015-0327-2
• Woods B, Sideris E, Palmer S, Latimer N, Soares M. NICE DSU Technical Support Document 19: partitioned survival analysis for decision modelling in health care: a critical review. Sheffield: Decision Support Unit, ScHARR, University of Sheffield; 2017. https://doi.org/10.15131/shef.data.33848653