(a) Structural features | Question to differentiate between attributes: | Typical classification | Definition |
---|---|---|---|

Population resolution | What level is the model arising? | Aggregate (may also be referred to as cohort) | The model is at a macro-level with a population aggregated and run through the model together. Variables represent population averages [8]. Relies on a homogeneity assumption that individuals within a particular health state are homogeneous [10, 12]. To incorporate individual factors or memories into the model, separate health states are required [10, 12]. Interactions are also modelled at an aggregate level |

Individual |
The model is at a micro-level with individuals going through the model separately [8, 10, 13]. This easily incorporates individual factors and memory. Patient characteristics may be retained as continuous variables [4] Permits exploration of first-order uncertainty | ||

First order uncertainty [14] | To what extent is the model capable of incorporating and analysing patient-level variability within its structure? | Deterministic | No variability in the outcomes between identical patients. Within a given sample of patients, individuals facing the same probabilities and outcomes will experience the effects of a disease or intervention identically |

Stochastic | Permits random variability in outcomes between identical patients as there exists uncertainty in patient-level outcomes that is entirely due to chance. Within a given sample of patients, individuals facing the same probabilities and outcomes will experience the effects of a disease or intervention differently. This can be perceived as a form of random error and, with increased sample size, the extent of this uncertainty can be reduced | ||

Interactivity | Are actors in a model or the overall system independent? | Static/independent | No interaction present between or within actors as each actor is independent and no interactions at the system level [9] |

Dynamic/dependent | Interaction exists between or within actors or at the level of the system. Feedback and interdependencies may exist within the modelled system [9] | ||

Resource constraint | Are constrained resources or queuing important to the decision problem? | Unlimited | There exist no constraints in the system |

Constrained | Resource constraints has impacts on features within the model [13] | ||

Dimension of time | How is time handled by the model? | Untimed | Time is not explicitly modelled. Another term used to describe this concept of time is “aggregate” as changes in time are not considered important to the model [13] |

Discrete | Time separated into discrete units with an event occurring during one of the discrete time steps [8, 13]. To handle simultaneous events, requires smaller fixed time intervals [10] | ||

Continuous | Time is continuous with an event occurring at any point in the continuum of time; thereby, permits modelling of multiple simultaneous events [8] |

(b) Practical consideration | Definition |
---|---|

Data availability | The availability of the necessary data to populate the economic model [7] |

End-user Requirement | This considers whether the model meets the need of its end-users and decision-makers. It is dependent on how well the model structure reflects and is able to capture all relevant aspects of the underlying reality and the corresponding uncertainties that exist [6, 10]. End user requirement may capture whether the modelling approach is considered acceptable and whether funding is present to support a particular project |

Experience | The extent to which the modeller has accumulated knowledge and implementation skills to construct the model [7] |

Model error | The degree of imprecision in the model that is deemed acceptable by either the modeller and/or its end-users [12]. Model error can either be systematic or unsystematic. Unsystematic error, synonymous to uncertainty, can be explored through the application of sensitivity analysis. The feasibility of conducting sensitivity analysis is dependent on the model structure and its underlying parameters |

Modelling software availability | The accessibility of the necessary software(s) to construct and evaluate the model. Different software may support different modelling approaches and are associated with licensing fees. Softwares for health economic modelling include Microsoft Excel (for decision trees and Markov cohort models); Treeage (for decision tree, Markov cohort model and Markov microsimulation); Arena (for discrete-event simulation); Any Logic (for discrete-event simulation, agent-based model, system-dynamics and compartmental models); and Berkeley Madonna (for system-dynamics and compartmental models) |

Simplicity | The degree of complexity in a model. This is essentially dependent on the size of the model (e.g. the number of states/transitions in state-transition models) and the number of parameters present [9]. Simpler models are more likely to be understood and accepted by stakeholders [12] |

Time | This considers the speed of model development and captures several aspects including the time required to programme the model (building time), the time required to collect the necessary data to fill the model (data collection) and the time required to generate simulation results (simulation time) [7] |

Transparency | The degree to which the end-user of the model can review the model structure, equations, parameter values and the underlying assumptions. This is considered important by modellers for two reasons: (i) to provide non-quantitative description of the model to those interested in understanding how a model works; and (ii) to provide technical information to those interested in evaluating a model at the higher level mathematical and programming detail, possibly with the interest to replicate the results. Transparency promotes an understanding on the model’s accuracy, limitation and potential application. This is deemed important to build trust and confidence in a model to the appropriate decision-makers [15] |

Validity | The clinical representativeness of a model to the actual decision problem [7, 12]. This addresses how adequately a chosen modelling approach reflects and captures all relevant aspects of the underlying reality and the corresponding uncertainties that exist |