Structure follows the disease and the decision
The right model structure depends on the disease, the data and the decision. Partitioned survival models are common in oncology but rest on assumptions about the relationship between progression and survival that assessors increasingly test. State transition and individual-level models capture history-dependence and sequences more naturally. We choose and justify the structure explicitly, and test how much the choice matters.
What we deliver
- De novo cost-effectiveness and cost-utility models
- Budget impact models to HIQA and NICE expectations
- Cost-comparison and cost-minimisation analyses
- Early economic models for portfolio, pricing and evidence planning decisions
- Survival analysis and extrapolation, including flexible and cure models
- Utility analysis, mapping and severity modifier calculations
- Adaptation of global models to local reference cases
- Independent model review, validation and quality control
- Interactive payer tools and R Shiny interfaces
Validation and transparency
Every model comes with a technical report, a validation log and a clear audit trail of inputs and sources. We use recognised frameworks such as TECH-VER and AdViSHE for verification and validation reporting. Our analysis of adapting global models for NICE and NCPE and survival extrapolation sets out our approach.
Why Evidaxia
We build models for NICE and NCPE submissions and have reviewed company models from the assessment side, so we know where models are most often challenged and how to make them robust.
What NICE asks of a submission model
NICE's technology appraisal and highly specialised technologies manual (PMG36) sets out expectations that apply to almost every submission model. Section 4.6 asks for full documentation and justification of structural assumptions and data inputs, and for sensitivity analyses wherever alternative plausible assumptions exist. The conceptual modelling process that led to the structure should be transparent, including how experts were involved. The manual is clear that it is not enough to state that a structure has been used or accepted before: the type of model and its structure should be justified for each new decision problem.
The manual also asks for the methods and results of quality assurance and validation, and for results in disaggregated form with a table of key results, including life years gained for cost-utility analyses. Parameter uncertainty is captured through probabilistic sensitivity analysis, with distributions chosen to represent the evidence and justified. Structural uncertainty is explored through scenarios that are consistent with the evidence.
Where a model relies on a surrogate end point, NICE describes three levels of evidence for the surrogate relationship, from biological plausibility, through consistent observational association, to randomised evidence that an effect on the surrogate corresponds to a commensurate effect on the final outcome. Planning which level the evidence can reach is part of designing the model. SMC and HIQA set out their own reference cases; see model adaptation for how they compare.
Choosing a model structure
| Structure | Often suits | Design consideration |
|---|---|---|
| Partitioned survival | Oncology, where progression-free and overall survival data are available | State occupancy comes from separately fitted survival curves, so the plausibility of each extrapolation, and of the relationship between them, needs explicit support (NICE DSU TSD 19) |
| State transition (Markov) cohort | Chronic conditions with well-defined health states | Transition probabilities depend only on the current state (the Markov assumption), unless tunnel states or time-in-state dependence are built in |
| Decision tree | Short-term decisions, diagnostics and one-off interventions | Often combined with a longer-term model when consequences extend over time |
| Individual patient simulation (microsimulation or discrete event simulation) | Treatment sequences, history dependence and heterogeneous populations | Greater flexibility, with more data and computation needed (NICE DSU TSD 15) |
Survival modelling draws on a wider set of DSU guidance: TSD 14 on extrapolation with patient-level data, TSD 21 on flexible methods, TSD 24 on adjusting for treatment switching and TSD 26 on expert elicitation for long-term survival outcomes. Our analysis of survival extrapolation in HTA sets out how these methods combine.
How the model connects to the rest of the evidence
A model is only as coherent as the evidence that feeds it. Relative effects come from the systematic review and the indirect treatment comparisons, and the same assumptions should carry through to the budget impact model. Utility inputs follow each agency's preferred measure and value set: for NICE technology appraisals and highly specialised technologies evaluations with an invitation to participate issued after 27 August 2026, the interim methods statement PMG51 makes the EQ-5D-5L descriptive system and the new UK value set the preferred approach for adults. See our analyses of the NICE EQ-5D-5L value set and the NICE severity modifier, both of which change how QALYs are counted.