Concept Architecture
Concept
Theoretically, Structural Uncertainty is the uncertainty arising from assumptions about the structure, formulation and conceptual design of a health economic model rather than from uncertainty in numerical parameter values. It reflects incomplete knowledge regarding the appropriate representation of disease progression, treatment pathways, model states or analytical methods. In health economics, structural uncertainty is recognised as a major source of decision uncertainty because different plausible model structures may produce different estimates of costs, health outcomes and cost-effectiveness.
Mathematically, structural uncertainty is represented by comparing alternative mathematical formulations of the same decision problem rather than modifying parameter values within a single model. Competing model structures may differ in their state definitions, transition mechanisms, survival extrapolation methods or analytical frameworks, leading to different model outputs despite using similar input data.
In practice, structural uncertainty is evaluated through structural sensitivity analysis, scenario analysis or model averaging by comparing results from multiple plausible model specifications. It is routinely assessed in health technology assessment to determine whether reimbursement recommendations remain robust across alternative modelling assumptions and conceptual frameworks.
Purpose
Used to quantify the uncertainty arising from alternative model structures and conceptual assumptions, supporting robust health economic decision-making and transparent assessment of model validity.
Mathematical Formulae
Primary Formula
There is no universally recognised canonical mathematical formula.
Supporting Formulae
Incremental Cost-Effectiveness Ratio:
ICER = ?C / ?E
Net Monetary Benefit:
NMB = ?E ? C
Structural comparison:
?Y = Y? ? Y?
where:
Y? and Y? are outcomes generated by alternative model structures.
Related Mathematical Methods
- Structural Sensitivity Analysis
- Scenario Analysis
- Model Averaging
- Sensitivity Analysis
- Markov Model
- Partitioned Survival Model
- Decision Tree
Example
A health technology assessment evaluates an oncology intervention using both a Markov model and a partitioned survival model. Although both models use identical clinical trial data, the Markov model estimates an ICER of �25,200 per QALY whereas the partitioned survival model estimates �31,000 per QALY. The difference illustrates structural uncertainty arising from alternative representations of disease progression.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| CHOOSE | =CHOOSE(A2,MarkovResult,PSMResult,DESResult) | Select alternative structural model outputs for comparison. |
| INDEX | =INDEX(ResultRange,ModelNumber) | Retrieve results from different model structures. |
| MATCH | =MATCH("Markov",ModelList,0) | Identify the structural model used in the analysis. |
| IF | =IF(B2<=30000,"Cost-effective","Not cost-effective") | Compare reimbursement decisions across alternative model structures. |
VBA (Optional)
VBA can automate execution of alternative model structures, compare outcomes across competing modelling approaches and generate reports summarising structural uncertainty.
Sources
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press.
- Philips Z, Ginnelly L, Sculpher M, et al. Review of guidelines for good practice in decision-analytic modelling in health technology assessment. Health Technology Assessment. 2004;8(36).
- ISPOR-SMDM Modeling Good Research Practices Task Force Reports.
- NICE. NICE Health Technology Evaluations: The Manual.
Related Concepts (2)
Institutional Perspectives (2)
- PBAC
Parameterise Structural Assumptions or Use Scenario Analyses
PSA characterises parameter uncertainty but cannot address structural (or translational) uncertainty. Where multiple plausible model structures exist, the submission must assess their impact and, if substantial, characterise it formally — parameterising structural assumptions where clinical evidence or expert opinion allows, otherwise using scenario analyses.
Pharmaceutical Benefits Advisory Committee, Guidelines for Preparing a Submission to the PBAC, Section 3A.9View source → - NICE
Structural Assumptions Justified and Explored via Scenario Analysis
Structural assumptions and the choice of model structure should be justified and made transparent; their impact on results should be explored through scenario analyses (preferably probabilistic), and model structures that limit feasibility of probabilistic analysis must be specified and justified.
NICE Health Technology Evaluations: The Manual (PMG36), Section 4 (Economic Evaluation)View source →
Library
Tools & Resources
1
SAVI — Sheffield Accelerated Value of Information — Mark Strong, Jeremy Oakley & Penny Breeze (University of Sheffield), Web application ed., 2024 (University of Sheffield)
A free, open-access web calculator that computes value-of-information measures (EVPI, partial EVPI/EVPPI and EVSI) directly from a model’s probabilistic sensitivity analysis output — no need to re-run the model. Also reports payer strategy-specific and uncertainty burden.
Web Tool (R Shiny)View source →
Process for Incorporating Economic Evidence into Federal Vaccine Recommendations — National Advisory Committee on Immunization, Current online guidance ed., 2021 (Public Health Agency of Canada)
Canadian federal guidance explaining when lifetime horizons are appropriate, when shorter horizons may suffice, and why extrapolation and horizon scenarios create decision uncertainty.
Web GuidanceView source →
Frequently Asked Questions (6)
What is structural uncertainty?
Uncertainty from the choice of a model's fundamental structure or method, such as which health states to include, unlike parameter uncertainty within a given structure.
Source: Bojke et al. 2009
Why is structural uncertainty easy to overlook?
Structural uncertainty concerns whether the model is built the right way, its choice of states, events, and functional forms, rather than the values put into a given structure. It is easy to overlook because a single model, once built, presents its structure as settled, and the analysis proceeds as if that shape were certain. Yet a differently but equally defensibly structured model might give another answer, and that possibility leaves no trace in the results of the one built. Making it visible requires deliberately comparing alternatives. Bojke and colleagues (2009) examine this.
Source: Bojke et al. 2009
How does structural uncertainty differ from parameter uncertainty?
Structural uncertainty concerns the model's form and structural assumptions, such as its states, transitions, or extrapolation function, while parameter uncertainty concerns imperfect knowledge of the numerical input values within a chosen structure. Parameter uncertainty is represented by distributions and sampled probabilistically, whereas structural uncertainty involves discrete choices between model forms, addressed by comparing alternative structures or averaging over them. So the two differ in kind: one is about how the model is built, the other about the values used within a fixed build.
Source: Bojke et al. 2009
How is structural uncertainty addressed?
Structural uncertainty is addressed mainly by structural sensitivity analysis, running the model under alternative plausible structures and comparing the results, and by model averaging, which combines the results of several candidate structures weighted by their support, so no single structure is relied upon. Parameterising structural choices, where possible, allows some structural uncertainty to be represented within the model. These approaches show how conclusions depend on the structural assumptions, since structural uncertainty cannot be captured simply by sampling parameter distributions within one fixed structure.
Source: Briggs, Claxton & Sculpher 2006
Why does structural uncertainty matter?
Structural uncertainty matters because the choice of model structure can affect results as much as, or more than, the parameter values, particularly in areas such as survival extrapolation, so relying on a single structure without examining alternatives can give overconfident conclusions. Recognising structural uncertainty prompts testing alternative structures and, where appropriate, averaging over them, giving decision makers a fuller account of how robust the conclusions are. Ignoring it would understate the true uncertainty, since a different reasonable structure might lead to a different decision.
Source: Bojke et al. 2009
What are examples of structural uncertainty?
Examples of structural uncertainty include the choice of survival extrapolation function beyond the observed data; the set of health states and the transitions allowed in a model; whether to model a process at the cohort or individual level; assumptions about how long a treatment effect persists; and how adverse events or comorbidities are represented. Each is a choice about the model's form rather than a parameter value. These structural choices can influence results substantially, so they are examined through structural sensitivity analysis or model averaging.
Source: Bojke et al. 2009
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 30 Oct 2025
Content version: 1.0.0
Canonical Identity
- Term code
- HE-EM-UA-073
Stable URI · Machine-readable · Resolvable · CC BY 4.0