3A.2 Computational methods and structure of the economic evaluation
Page last updated: September 2016
- Review the literature for relevant economic references and any additional clinical or epidemiological literature relevant to the model that has not already been presented, and attach copies of studies and original sources of data used in the economic evaluation (Section 3A.2.1)
- Report and justify the model structure and its development, and justify the time horizon (Subsection 3A.2.2)
- Describe and justify the modelling technique used. If an individual-level model is used instead of a state transition model, explain why (Subsection 3A.2.3)
- Provide a fully editable electronic copy (Subsection 3A.2.4)
3A.2.1 Literature review
Present the results of a literature search for economic evaluations involving the proposed and similar medicines or alternative managements, or similar treatment algorithms, focusing on the structure of the existing models. This may include published reports and models considered by other health technology assessment agencies.
Present any additional literature (eg additional clinical trials, guidelines, natural history studies, burden of disease studies, utility studies, surveys) that informs the model structure or inputs and that has not already been presented in Subsection 1.2 or Section 2, noting what aspect of the model it informs. Provide copies of the original sources of all data not already presented in Section 2, or expert opinion used in the model, in an attachment. Cross-reference the extraction of data from each source to the page, table or figure number of the source document.
3A.2.2 Structure of the economic model
Ensure that the model structure captures all relevant health states or clinical events along the disease or condition pathway, and that it is consistent with the treatment and disease or condition algorithms presented in Subsection 1.2.
Inform the model structure using the results of the literature review of economic evaluations, and other clinical and economic literature, including clinical trials, clinical guidelines, natural history studies and burden of disease studies.
Disaggregate patient-relevant events if there are important differences in mortality, disease or condition progression, associated costs, or quality-of-life effects, and the distribution differs between the intervention and comparator.
During the model development, consider whether, for a given patient, an event experienced in the model should influence the risk of experiencing subsequent events – this may inform the choice of computational method.
Assess the model structure(s) to establish face validity. Justify the exclusion of any potentially relevant states or events identified in the literature, and reference data sources and expert input. Discuss the potential impact of any exclusions on the model outputs. Where the model structure differs from existing models, explain the basis for the selection of the submission’s approach.
If relevant, define multiple plausible model structures and test them as part of a structural sensitivity analysis. Examine and address structural uncertainty in Subsection 3A.9.
Time horizon of the evaluation
Define and justify the time horizon over which the costs and outcomes of the proposed medicine and its main comparator are estimated. Ensure that the time horizon captures all important differences in costs and outcomes between the intervention and the comparator, as a result of the choice of treatment, but does not extend unnecessarily beyond this.
Where interventions do not affect mortality, and have temporary health or quality-of-life effects, a relatively short time horizon may be appropriate.
Where there is evidence that a treatment affects mortality or long-term/ongoing quality of life, then a lifetime time horizon is appropriate. Note that a lifetime time horizon relates to the life expectancy of the relevant patient population, and reflects the time span required for nearly all of the model cohort to die. The validity of the lifetime horizon is determined by the population of the model, and the inputs; it is not an independently nominated duration. Inputs that are not realistic will result in a model predicting an implausible duration of outcomes or survival and, thus, an implausible lifetime time horizon. The assessment of plausibility should also apply to how the model extrapolates the curves to reach this time horizon (see Subsection 3A.4).
As a modelled time horizon extends – in absolute terms and relative to available data – it is associated with increasing inherent uncertainty. Therefore, economic claims based on models with very extended time horizons and predominantly extrapolated benefits will be less certain and are likely to be less convincing to the PBAC. Subsections 3A.4.3 and 3A.9 address the extrapolation of costs and outcomes for an extended time horizon and associated uncertainty.
Where possible, input data should be sourced directly from the evidence presented in Section 2.
Where relevant, applicability issues with clinical data from Section 2 are identified (see Subsection 2.7.1), these are discussed and translated to the Australian population and setting, if necessary, in Subsection 3A.3.
Describe the methods used to identify data to populate the model input parameters. For example, whether systematic or ad hoc reviews of the literature were undertaken, or how relevant primary data sources, including registries and observational studies, were identified. The method of identifying the data should be robust and transparent. Where multiple sources of data exist, the choice of the source used in the base case should be justified.
Applicability concerns (and any translation) relating to additional data should be described in the relevant subsection. For example, transition probabilities beyond the scope of the clinical trial evidence are described in Subsection 3A.4, health outcomes and utilities are described in Subsection 3A.5, and health care resource use and costs are described in Subsection 3A.6.
If adequate input data are not available to inform the model according to the initially defined structure, review the model structure in the light of the available data, and assess the face validity of alternative model structures that better conform to the available data. If a valid alternative model structure can be defined, describe the revisions to the structural model and discuss the potential effects on the model outputs.
If an alternative valid model structure cannot be defined, use expert opinion to estimate input parameters for which empirical data were not identified (see Subsection 3A.9 and Appendix 1 for more information.)
3A.2.3 Computational methods
If a trial-based economic evaluation is being undertaken using individual patient data on costs and outcomes from a clinical trial(s), describe the methods and software used to do this.
For model-based economic evaluations, identify the most appropriate modelling technique for the implementation of the final model structure(s).32 Generally, select the least complicated modelling technique for which it is feasible to implement the specified model structure, moving from decision trees to cohort-based state transition models to individual-level modelling techniques. Note the software used.
Decision trees are useful for models with short time horizons. General spreadsheet software (eg Excel) or specialist software (eg TreeAge) can be used. Follow good-practice guidelines for using decision trees.33
Cohort-based state transition (or Markov) models
Use cohort-based state transition models to represent longer time horizons for models that can be represented using a manageable number of health states under the constraints of the Markovian (memoryless) assumption. General spreadsheet software (eg Excel) or specialist software (eg TreeAge) can be used.
Follow good-practice guidelines for using state transition models.34 In particular, consider the following questions when implementing a cohort-based state transition model:
- Is it reasonable to assume that transition probabilities from each defined health state are independent of states that may have been experienced before entering each state? Health states that describe pathways through the model can be used to represent the effects of previous events on subsequent transition probabilities.
- Do transition probabilities vary according to how long individuals have remained in each health state? Tunnel states are required to represent time-varying transition probabilities.
- Is the eligible population homogeneous, or is variation in patient variability normally distributed? This issue commonly refers to the age of the eligible population, but may include other factors. If relevant factors are not normally distributed, run separate analyses of the model and aggregate the outputs.
- What is the likely impact of alternative cycle lengths on the model outputs? Describe the factors determining the selected cycle length.
A half-cycle correction is the default approach to representing the time of transition between states, although an alternative correction factor may be proposed with justification.
Individual-level (or microsimulation) models
Use individual-level modelling approaches only when a defined model structure cannot be feasibly implemented as a cohort-based model. Describe the characteristics of the model structure that prevent using a cohort-based model. Potential factors include baseline heterogeneity, continuous disease or condition markers, time-varying event rates and the influence of previous events on subsequent event rates.35 Also describe how incorporation of these features in an individual-level model are expected to produce a more accurate representation of the disease or condition pathways, costs and patient outcomes.
The most common individual-level approaches include individual-based state transition and discrete event simulation models. Follow published guidelines on good research practices for applying these models.34,36 Discuss any requirements for specialist software with the Pharmaceutical Evaluation Branch (PEB) in advance.
Other modelling techniques
If the results from simpler models are robust enough to produce plausible sensitivity and scenario analyses, it is not necessary to use more complex modelling techniques.37 If an alternative modelling technique is used, describe and justify how the approach leads to more accurate and valid results. For example, in the clinical area of infectious diseases, the use of dynamic transition models or agent-based models to represent herd immunity may be justified if a simple nondynamic model will not demonstrate cost-effectiveness accurately enough.
Note that more complex modelling techniques may be less transparent, and the model assumptions less certain. This might result in the PBAC having less confidence in the cost-effectiveness claim. Discuss the use of complex modelling techniques (including any specialist software) with the PEB in advance.
3A.2.4 Fully editable electronic copy of the economic evaluation
Provide access to the electronic copy of the economic evaluation. Ensure that all variables can be changed independently, including allowing the base case of the economic evaluation to be respecified and a new set of sensitivity analyses to be conducted with each respecified base case. Ensure that the economic evaluation can produce results following respecification of variables within reasonable running times.
The following software packages do not need prearrangement with the PEB:
- TreeAge Pro
- Excel 2010, including @RISK®, but not necessarily including all advanced features and plug-ins (eg Crystal Ball).